The Notions of Literal and Non-literal Meaning in Semantics and Pragmatics

The Notions of Literal and Non-literalMeaning in Semantics and Pragmatics

Dissertation

zur Erlangung des akademischen Gradesdoctor philosophiae

(Dr. phil)

Eingereicht an derPhilologischen Fakultatder Universitat Leipzig

Kristin Borjesson

Verleihungsbeschluss 10. Oktober 2011

Gutachter:PD Dr. Johannes DollingInstitut fur Linguistik, Universitat Leipzig

Prof. em. Dr. Anita SteubeInstitut fur Linguistik, Universitat Leipzig

Prof. Robyn Carston, PhDResearch Department of Linguistics, University College London

Contents

1 Introduction 11.1 Standard Notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Are the Standard Notions Adequate? . . . . . . . . . . . . . . . . . . 31.3 Interpretation and Levels of Meaning . . . . . . . . . . . . . . . . . . 51.4 Aims of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61.5 Thesis Plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Against the Standard Notions of Literal Meaning and Non-literalMeaning 92.1 Literal Meaning and Context-Independence . . . . . . . . . . . . . . . 9

2.1.1 Literal Meaning as Compositional Meaning? . . . . . . . . . . 102.1.2 Literal Meaning as Context-Independent? . . . . . . . . . . . . 142.1.3 Literal Meaning as Primary to Non-literal Meaning? . . . . . . 19

2.2 Non-literal Meaning and Conventionality . . . . . . . . . . . . . . . . 232.2.1 Empirical Evidence . . . . . . . . . . . . . . . . . . . . . . . . 232.2.2 Theoretical Considerations . . . . . . . . . . . . . . . . . . . . 29

2.3 Consequences for Lexical Meaning . . . . . . . . . . . . . . . . . . . . 352.3.1 Problematic Data . . . . . . . . . . . . . . . . . . . . . . . . . 362.3.2 Approaches to Meaning in the Lexicon . . . . . . . . . . . . . 412.3.3 Semantic Underspecification in the Lexicon . . . . . . . . . . . 51

2.4 Empirical Investigations of Aspects of Semantics . . . . . . . . . . . . 672.4.1 Polysemy vs. Underspecification in the Lexicon . . . . . . . . 672.4.2 Empirical Evidence for Semantic vs. Pragmatic Processing . . 71

2.5 Why the Standard Notions? . . . . . . . . . . . . . . . . . . . . . . . 752.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

3 Utterance Meaning and the Literal/Non-literal Distinction 833.1 Levels of Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

3.1.1 Grice’s Four Types of Meaning . . . . . . . . . . . . . . . . . 843.1.2 Bierwisch’s Three Levels of Meaning . . . . . . . . . . . . . . 883.1.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

3.2 The Problem of Characterising the Level of Utterance Meaning . . . 943.2.1 Explicit/Implicit Meaning . . . . . . . . . . . . . . . . . . . . 953.2.2 Unarticulated Constituents vs. Hidden Indexicals . . . . . . . 1063.2.3 Minimal Semantic Content and Full Propositionality . . . . . 1153.2.4 Minimal Proposition vs. Proposition Expressed . . . . . . . . 123

iv Contents

3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

4 Utterance Meaning and Communicative Sense – Two Levels orOne? 1334.1 Problematic Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . 134

4.1.1 Metaphor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1344.1.2 Irony . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1514.1.3 Conversational Implicatures . . . . . . . . . . . . . . . . . . . 1624.1.4 Speech Acts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

4.2 Differentiating What is Said from What is Meant . . . . . . . . . . . 1844.2.1 What is Said/What is Meant and Indirect Speech Reports . . 1864.2.2 Primary vs. Secondary Pragmatic Processes . . . . . . . . . . 1894.2.3 What is Said/What is Meant and Distinct Knowledge Systems 194

4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

5 Varieties of Meaning, Context and the Semantics/Pragmatics Dis-tinction 2035.1 Towards an Alternative Characterisation of (Non-)Literal Meaning . . 203

5.1.1 Literal Meaning and Types of Non-literal Meaning . . . . . . . 2055.1.2 Literal Meaning as ‘Minimal Meaning’ . . . . . . . . . . . . . 2115.1.3 Nature of the Processes Determining (Non)-Literal Meaning . 2155.1.4 (Non-)Literal Meaning as (Non-)Basic Meaning . . . . . . . . 221

5.2 The Nature of Context in Utterance Interpretation . . . . . . . . . . 2275.2.1 Context and the Interpretation of Implicit Meaning Aspects . 2285.2.2 Context, Semantic Interpretation and the Semantics/ Prag-

matics Distinction . . . . . . . . . . . . . . . . . . . . . . . . 2425.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

6 Summary 255

Bibliography 259

Chapter 1

Introduction

1.1 Standard Notions

One of the major issues in investigating the relation of language and meaning is thequestion of how to characterise and draw the line between what traditionally arecalled semantics and pragmatics. In describing what they take to be the characteris-tics of one or the other system, linguists often make use of the terms literal meaningand non-literal meaning. For example, Lyons (1987) lists a number of propositionsused in the differentiation of semantics from pragmatics, amongst which is the follow-ing: ‘. . . that semantics deals with literal, and pragmatics with non-literal, meaning. . .’ (ibid., p. 157). Similarly, Cole (1981, p. xi) states that semantics ‘. . . is involvedin the determination of conventional (or literal) meaning . . . ’, whereas pragmaticsis concerned with ‘. . . the determination of nonconventional (or nonliteral) meaning. . . ’ and Kadmon (2001, p. 3) writes ‘. . . I think that roughly, semantics only covers“literal meaning.” Pragmatics has to do with language use, and with “going beyondthe literal meaning.”’. More recently, Recanati (2004) summarised (and criticised)the standard view on the division of labour between semantics and pragmatics,starting with the following sentence.

‘Semantics deals with the literal meaning of words and sentences as de-termined by the rules of the language, while pragmatics deals with whatthe users of the language mean by their utterances of words and sen-tences.’ (Recanati 2004, p. 3).

Actually, the pair of terms literal meaning/non-literal meaning is only one ofquite a number of dichotomies used in the characterisation of semantics and prag-matics. Thus, the two systems are often characterised in terms of the differentiationbetween conventional vs. non-conventional meaning, as, e.g. in the quote from Cole(1981) above. Similarly, in Lyons (1987) one finds the proposition ‘. . . that se-mantics has to do with conventional, and pragmatics with the non-conventional,aspects of meaning . . .’ (ibid., p. 157). Another important pair of terms tradition-ally used is context-independent vs. context-dependent meaning. Thus, Lyons (1987,p. 157) states ‘. . . that semantics deals with context-independent, and pragmaticswith context-dependent, meaning’. More specifically, Katz (1977) introduces the no-tion of the ‘anonymous letter situation’ to characterise the kind of meaning captured

2 Introduction

by semantics.

‘[I] draw the theoretical line between semantic interpretation and prag-matic interpretation by taking the semantic component to properly rep-resent only those aspects of the meaning of the sentence that an idealspeaker-hearer of the language would know in an anonymous letter sit-uation, . . . [where there is] no clue whatever about the motive, circum-stances of transmission, or any other factor relevant to understandingthe sentence on the basis of its context of utterance.’ (Ibid., p. 14)

To summarise, whereas semantics is characterised as dealing with literal, conven-tional and context-independent meaning, pragmatics deals with non-literal, non-conventional and context-dependent meaning. More generally, the standard notionsof semantics and pragmatics may be described as follows. Semantics deals withthose aspects of meaning that both simple and complex expressions have, indepen-dent of their use. In contrast, pragmatics deals with those aspects of meaning thatare determined by the actual use of language.1 That is, semantics is concerned withmeaning that is independent of any specific context, whereas pragmatics specificallydraws on contextual information for the interpretation of some expression.

Assuming that the borderline between semantics and pragmatics is fixed and sta-ble, using the dichotomies mentioned above in the characterisation of the respectivesystems suggests that there is a correspondence between literal, conventional andcontext-independent meaning, on the one hand, and non-literal, non-conventionaland context-dependent meaning on the other.2 However, one usually does not findactual characterisations of the kind of meaning picked out by the terms literal mean-ing and non-literal meaning other than such that again (more or less explicitly) relatethe two notions back to the semantics/pragmatics distinction. For instance, Bach(2001a) writes

‘Words do not have nonliteral meanings [. . . ], but they can be used innonliteral ways. [. . . ] In familiar cases, such as metaphor and metonymy,particular expressions are used nonliterally. [. . . ] But there is a differentphenomenon which I call “sentence nonliterality,” [. . . ] Here a wholesentence is used nonliterally, without any of its constituent expressionsbeing so used.’(Ibid., p. 249, my emphasis).

Thus, whereas literal meaning is a feature that expressions are said to have, the non-literal meaning of an expression results from the particular use of that expression.Moreover, not only simple but also complex expressions can be used non-literally. Inboth cases, the non-literal meaning results from the actual use of a certain expressionin a specific context. In addition to the circumscription ‘non-literal use of language’,one finds descriptions such as ‘non-literal utterances’ (cf. Carston 2005) or ‘non-literal interpretation’ (cf. Papafragou 1996, Ariel 2002, Carston 2005).

1Cp., again, Lyons (1987) who mentions the idea that ‘. . . semantics has to do with meaning,and pragmatics with use . . .’ (Ibid., p. 157).

2From the quotes given above, this is especially apparent in Cole’s, who uses the terms literaland non-literal as synonymous to conventional and non-conventional, respectively.

Are the Standard Notions Adequate? 3

Generally, it seems that the two terms literal meaning and non-literal meaningare treated as denoting basic kinds of meaning that are intuitively clear and as suchneed no further description. However, although there are no clear definitions of theterms literal meaning and non-literal meaning given in the literature concerned, thefact that the two terms are used amongst others in a dichotomous characterisationof semantics and pragmatics suggests that these terms may also be used in charac-terising literal meaning and non-literal meaning as such (or the other way aroundas done by Cole). This latter characterisation has led to what might be called thestandard notions of literal meaning and non-literal meaning which are summarisedin what follows.

Literal meaning, on the one hand, is assumed to be conventionalised, that is, itdoes not take any special interpretation effort to arrive at it. The literal meaningof simple expressions is listed in their lexical entries; the literal meaning of complexexpressions is the result of a principled combination of the literal meanings of theirparts. Thus, both the literal meaning of simple as well as complex expressions ischaracterised by the fact that it is context-independent. Non-literal meaning, onthe other hand, is assumed to be non-conventionalised, thus, it does take a specialinterpretation effort to arrive at it. Intuitively, it is considered as deviating fromsome more basic (literal) meaning in a fairly special way. Overall, the term non-literal meaning is used to differentiate from literal meaning a kind of meaning thatis derived from the latter and, in a sense, has a secondary status. Therefore, it istraditionally assumed that in terms of the enfolding of the interpretation process,the literal meaning of an expression is processed first, whereas any potential non-literal meanings are processed afterwards and only if the literal interpretation doesnot fit the given context.

Finally, in accordance with the standard characterisations of semantics and prag-matics given above, literal meaning is traditionally assumed to belong to the domainof semantics (as it can apparently be determined without reference to any particularcontext of utterance), whereas non-literal meaning is assumed to be properly inves-tigated by pragmatics (as it seems it can only be determined by taking into accountthe context of an utterance). Such a classification presumes that the definitions ofthe terms literal meaning and non-literal meaning, as well as the characterisationof what semantics and pragmatics are concerned with are indisputably clear. How-ever, the ongoing debate about the proper demarcation of semantics from pragmaticsshow that this is not at all the case (Carston 1999, Levinson 2000, Bezuidenhout andCutting 2002, Bianchi 2004, Carston 2004a, Recanati 2004, Cappelen and Lepore2005, Dolling 2005, Carston 2007, etc.).

1.2 Are the Standard Notions Adequate?

Note that although there is something like a standard characterisation for the termsliteral meaning and non-literal meaning, they are not consistently used in accordwith these characterisations. For instance, the term literal meaning is not only usedin connection with such phenomena as are considered totally context-independent.Thus, Carston (2007) refers to the ‘. . . literal meaning of [a speaker’s] utterance’(ibid., p. 21). Similarly, Recanati (1995) speaks of ‘. . . the literal interpretation of

4 Introduction

an utterance (the proposition literally expressed by that utterance). . . ’ (ibid, p.2) and Sag (1981) speaks of the ‘. . . propositional content of an utterance (i.e., itsliteral meaning) . . . ’ (ibid, p. 274-5). What makes such descriptions interestingin the present context is that what the authors refer to by the description ‘literalmeaning of an utterance’ does not necessarily correspond to the context-independentmeaning of the expression used to make the particular utterance, since, as we willsee, this need not actually be a full proposition. Rather, they use the term literalhere to differentiate the meaning of some expression in a particular context fromthe meaning that the speaker actually intended to express (or at least is taken bythe hearer to have expressed). Bezuidenhout and Cutting (2002, p. 435) make asimilar observation when they say that ‘[t]he phrases “literal meaning” or “literalinterpretation” have been used to cover both the literal meaning of a sentence andwhat is said by the utterance of a sentence in a context.’ This is also noted by Kortaand Perry (2008), who write that ‘[w]hat is said has been widely identified with theliteral content of the utterance . . .’. Generally, the term literal meaning is used withrespect to a level of meaning which can no longer be said to be context-independent.

It should be noted that it is authors who reject the standard characterisation ofsemantics and pragmatics that use the terms literal meaning and non-literal meaningin a non-standard understanding. The problem is that they do so implicitly. Thatis, these authors do not explicitly say anything new concerning the understandingof the terms literal meaning and non-literal meaning. As Wilson and Sperber (2000,p. 250) put it, ‘[t]he notion of literal meaning, which plays such a central role inmost theories of language use, is unclear in many respects’. Note also that it isnot only the pair of terms literal (meaning)/non-literal (meaning) that is employeddifferently in the various approaches. Such notions as (non)-conventionality andcontext-(in)dependence are problematic as well. Thus, the use of the pair of termsconventional vs. non-conventional as exemplified above suggests that conventional-ity is an all-or-nothing property. However, as is suggested by the results of variousexperiments investigating the nature of the interpretation process on the one hand(cf. Giora 1997, 1999, Gibbs 2002), as well as theoretical considerations within thefield of historical semantics on the other (cf. Busse 1991), this view is an oversimplifi-cation of the facts. Similarly, not all approaches that are characterised as essentiallysemantic by their proponents necessarily share the view that what semantics dealswith is context-independent meaning only (cf. Sag 1981, Borg 2004b, Cappelen andLepore 2005).

The problem that arises is that if the terms in which the notions literal meaningand non-literal meaning are defined are put under closer scrutiny, it might turn outthat they do not capture the aspects that we intuitively and pre-theoretically asso-ciate with the notions of literal meaning and non-literal meaning. Another problemis that with only the standard notions of literal meaning and non-literal meaning torely on, it is no trivial question to ask how these two meaning aspects are related toother kinds of meaning aspects identified in the individual approaches, such as e.g.,explicit/implicit meaning aspects of an utterance, so-called ad-hoc concepts or con-versational implicatures. For instance, metaphor – one particular type of non-literalmeaning – has actually been treated in terms of conversational implicature and sohas irony. In fact, conversational implicatures in general have been characterised as

Interpretation and Levels of Meaning 5

non-literal (cf. Recanati 2004), although, intuitively at least – and especially if oneactually does not count metaphor and irony amongst them – this is a questionableclaim.

1.3 Interpretation and Levels of Meaning

A further point to note is that the assumption that two systems, namely semanticsand pragmatics, are involved in the interpretation of natural language utterancessuggests that there is also only a differentiation of two levels of meaning necessary.Thus, the semantic component takes the lexical meanings of the individual expres-sions used in an utterance and combines them in a principled manner, resultingin the truth-conditional or semantic meaning of the complex expression used in anutterance, traditionally taken to be a proposition. This semantic meaning, then,is the input to the pragmatic component which determines what the speaker whouttered the expression in question actually meant in a particular situation. This isthe view usually taken to underlie Grice’s differentiation of the two levels of mean-ing what is said and what is meant.3 In Grice’s terminology then, literal meaningis part of the level of what is said, whereas non-literal meaning is to be found onlyat the level of what is meant. However, it has been pointed out both in Neo- aswell as Post-Gricean approaches that Grice’s notion of what is said is problematic.More specifically, it has been argued that what is said cannot both be the semanticmeaning of an utterance as well as the basis for drawing conversational implicatures(cf. Sperber and Wilson 1995, Levinson 2000, Carston 2002c, Recanati 2004).

As a consequence of such problems of Grice’s approach, more than two levelsof meaning have been assumed: a level that captures the semantic meaning of anexpression, a context-dependent level that can function as the basis for furtherpragmatic inferences to be drawn (call this what is said for now) and the level ofwhat is meant. What is interesting to note with respect to the differentiation of threelevels of meaning is the various possibilities this opens up for the classification ofliteral meaning and non-literal meaning. Thus, with the differentiation of semanticmeaning from what is said, the question arises where ‘to put’ literal meaning: isit part of the level of semantic meaning or part of the level of what is said? Thequotations above suggest that different linguists hold different views concerning thisquestion. Note also that a similar question can be asked concerning non-literalmeaning. That is, with the differentiation of the two distinct levels of context-sensitive meaning what is said and what is meant, the question arises whether non-literal meaning aspects exclusively belong to the latter level or whether they mightnot already arise at the former. Note that if both literal meaning and non-literalmeaning are assumed to arise at the level of what is said already, then this wouldsuggest that the two types of meaning are not as different from one another astraditionally assumed.

However, the various approaches to the interpretation of natural language differin the characterisations they assume for the individual levels of meaning identified.

3It should be noted that in his efforts to explicate the notions of saying vs. meaning, Griceactually contemplated four different types of meaning. For more on this issue see chapter 3.

6 Introduction

Again, the different characterisations are dependent on the particular views heldregarding the nature of the semantic and pragmatic components, the roles they playin the overall interpretation process as well as the type of information they areassumed to have access to.

1.4 Aims of the Thesis

I want to show that the standard notions of literal meaning and non-literal meaningare not consistent and that, as a result, the two terms cannot be used to differentiatebetween the types of meaning aspects semantics and pragmatics are usually assumedto deal with. In contrast, I want to develop a characterisation of literal meaning andnon-literal meaning that is consistent and captures the various uses the two termsare put to. In addition, I want to defend the view that differentiating between twocontext-dependent levels of meaning what is said/utterance meaning and what ismeant/communicative sense is possible and necessary. In particular, I will proposea modelling of the process that provides implicit meaning aspects to the level ofwhat is said/utterance meaning that allows viewing this process – as the processescontributing to this level of meaning in general – as operating independently ofconsiderations of speaker intentions. Finally, I want to show that a differentiationof semantics from pragmatics that implicitly relies on a logical or temporal orderingof the two systems is inappropriate. Rather, the processes constituting these twosystems should be seen as operating in tandem.

1.5 Thesis Plan

The thesis is structured as follows. In chapter 2, I will argue against the stan-dard notions of literal meaning and non-literal meaning. In particular, I will argueagainst the traditional characterisation of literal meaning and non-literal meaning,according to which the former is taken to be context-independent and the latternon-conventional. Having established that literal meaning does not necessarily haveto be taken to be context-independent and as such semantic in nature, I will discussthe consequences this view has for the nature of lexical meaning. After reviewing anumber of different types of approaches to lexical meaning, I will argue for a viewthat assumes a high degree of underspecification of lexical meaning. Generally, inthe discussions in chapter 1, I will consider both theoretical viewpoints as well asempirical data. In particular, one section is dedicated to empirical studies on as-pects of the semantics component, namely that lexical meaning is characterised byunderspecification and that, generally, semantic processes of meaning constructionshould be differentiated from pragmatically based plausibility checks. In the lastpart of chapter 1, I will try to answer the question of why the standard notions ofliteral meaning and non-literal meaning came to be assumed in the first place. Here,the idea of stereotypical interpretations of linguistic expressions presented ‘out ofcontext’ will be considered.

Having argued against the standard notions in chapter 2, and more specifically,having argued for viewing literal meaning, similarly to non-literal meaning as essen-

Thesis Plan 7

tially context-dependent as well, chapter 3 is dedicated to looking in detail at the firstcontext-dependent level of meaning called what is said by Grice, to see how this hasbeen characterised subsequently and to identify the processes potentially involvedin determining literal meaning at this level of meaning. I will start with Grice’sdifferentiation of four different types of meaning and relate them to the two levelsof meaning Grice introduced: what is said and what is meant. Following that, I willpresent Bierwisch’s threefold differentiation of levels of meaning, based on the dif-ferent knowledge systems made use of in their determination. In the second part ofchapter 3, I will discuss a range of approaches that give alternative characterisationsfor Grice’s level of what is said. The overall aim is to identify the different processesat work in determining what is said, how these processes are characterised and whichtypes of meaning aspects can be found at this level of meaning (appart from po-tentially literal or non-literal meaning). At the same time, the various approachesdiscussed also all offer slightly different views on the nature of the semantics andpragmatics components and how they interact in the process of utterance interpre-tation. While the greater part of chapter 3 is taken up by theoretical considerations,towards the end of that chapter a few empirical results will also be discussed.

Chapter 4, then, is concerned, on the one hand, with phenomena traditionallyassumed to arise at Grice’s level of meaning what is meant, and, on the other hand,with the more basic question of whether a differentiation of two context-dependentlevels of meaning what is said and what is meant actually is necessary/possible.Thus, in the first part of chapter 4, alternative approaches to the phenomena ofmetaphor, irony, (primarily generalised) conversational implicature and (primarilyindirect) speech acts will be reviewed as well as empirical results considered thattest the predictions following from the individual approaches. Here, the aim is toestablish, on the one hand, how these different meaning aspects are determined and,on the other hand, which of the phenomena actually can be usefully considered asnon-literal. More generally, the question is addressed at which level of meaning (i.e.what is said or what is meant) the individual phenomena should be taken to arise.In the second part of chapter 4, various arguments will be presented for and againstdifferentiating the two levels what is said and what is meant from one another. Ihope to make clear that such a differentiation is useful and necessary, although itmight be difficult to decide on the criteria to be used in this differentiation.

Chapter 5, finally, turns back to the basic question that chapter 2 ends with,namely how literal meaning and non-literal meaning actually should be characterisedif one wants to capture the various uses the two terms are put to. I will start out withtwo alternative characterisations of what literal meaning and non-literal meaningshould be taken to be, before presenting my own characterisation, based on thediscussion in the preceding chapters. As a preliminary for my characterisation, Iwill review the various processes identified in the preceding chapters as involvedin the overall interpretation of utterances. The main consequence drawn from mycharacterisation of literal meaning and non-literal meaning will be that these twonotions actually cannot be used in the characterisation of the distinction betweensemantics and pragmatics, if the former, in contrast to the latter, is essentiallytaken to be context-independent. The last part of chapter 5 will take up exactly thispoint, namely the nature of contextual information in utterance interpretation and

8 Introduction

whether the notion of context-(in)dependence actually is useful in differentiatingbetween semantics and pragmatics. Thus, I will first offer a proposal concerningthe nature of the contextual information the process of free enrichment makes useof. Free enrichment is one of the processes assumed to contribute to the level ofutterance meaning and crucially is taken to depend on a consideration of potentialspeaker intentions for its operation. I will show that this assumption is not necessary.In the final section of chapter 5, I will turn back to the characterisation of thesemantics/pragmatics distinction and after discussing a number of views on thatcharacterisation present my own.

Chapter 2

Against the Standard Notions ofLiteral Meaning and Non-literalMeaning

In this chapter, I will argue against the standard notions of literal meaning andnon-literal meaning described in chapter 1 (sections 2.1 and 2.2, respectively). Themain aim of this chapter is to show that the dichotomies traditionally used to differ-entiate literal meaning from non-literal meaning either cannot in fact differentiatethe two meanings (as is the case with the feature of context-(in)dependence) or arenot such ‘all-or-nothing’ concepts as traditionally implied (as is the case with theproperty of conventionality). Generally, the arguments presented point to the crucialconclusion that literal meaning and non-literal meaning are in fact not so differentfrom one another as traditionally assumed. Having argued against viewing literalmeaning as essentially context-independent and non-literal meaning as essentiallynon-conventional, I will consider the consequences this has for the nature of lexicalmeaning (section 2.3). Moreover, I will consider empirical evidence supporting theassumption of underspecification of lexical meaning and, more generally, a distinc-tion between distinctly semantic and pragmatic processes in interpretation (2.4). Inaddition, I will address the question of why the standard assumptions came intoexistence in the first place (section 2.5).

2.1 Literal Meaning and Context-Independence

Traditionally, complex expressions are assumed to have literal meaning in the formof what formal semantics1 calls sentence meaning, which results from the process ofsemantic composition which combines the literal meanings of the simple expressionsthat together constitute the complex expression and which captures the propositionexpressed by that sentence.2 Moreover, during interpretation, the literal meaning of

1Note that in what follows, on the semantics side, I am primarily interested in assumptionsmade in the programme of formal semantics.

2In this section, I will primarely be concerned with two of the three properties literal meaningis standardly claimed to exhibit, namely that it is context-independent and primary to non-literalmeaning. Thus, I will concentrate on complex expressions, leaving the discussion of the nature of

10 Against the Standard Notions of Literal Meaning and Non-literal Meaning

a complex expression is computed first, whereas its potential non-literal meaning iscomputed afterwards and only if the literal meaning does not fit the given context(cf. Grice 1975; Searle 1979).

Intuitively, these characterisations seem to be sound. They give a fairly generaldescription of what we take to be literal meaning with respect to complex expres-sions. However, looking at each of the characteristics in more detail reveals thatthey are not unproblematic. Thus, it is questionable whether what we usually taketo be a complex expression’s literal meaning does in fact correspond to its context-independent, compositional meaning. Put differently, the question is whether theformal semantic notion of sentence meaning can be assumed to both be the sumof the lexical meanings of the simple expressions involved as well as having a fullypropositional form. Furthermore, in computing the ‘speaker-intended’ non-literalmeaning of an expression, it may not actually be necessary to first compute theliteral meaning of the expression the speaker used as an intermediate step.

2.1.1 Literal Meaning as Compositional Meaning?

Concentrating on the traditional characterisation of the programme of formal se-mantics and the role of literal meaning therein reveals that, in a sense, the char-acterisation of literal meaning and non-literal meaning is interdependent on thecharacterisations of semantics and pragmatics. Thus, basically, formal semanticscan be characterised as dealing with the context-independent meaning of simpleand complex expressions.3 More specifically, it aims at formulating truth conditionsfor sentences. That is, it takes as a starting point for analysis the level of sentencemeaning, mainly for two reasons. First, it seems that sentences express proposi-tions, that is, complete thoughts, something of which it makes sense to ask whetherit is true or not. Second, intuitively at least, the meaning of a sentence can begrasped without any reference to an actual utterance of that sentence and is thuscontext-independent. It contrasts with interpretations of a sentence that can onlybe derived by considering the actual context in which that sentence is uttered (e.g.,cases of irony or particularised conversational implicature). Thus, sentence meaningis considered literal in the sense that its derivation is independent of contextual in-formation. Moreover, sentence meaning also is the level from which the meaningsof the individual expressions involved are derived, following the principle of com-positionality. And since sentence meaning is context-independent, the meanings ofthe simple expressions derived from it are context-independent too. They are thelexical meanings of the expressions concerned. Thus, primarily, what the term literalmeaning refers to is a certain type of meaning that, intuitively, seems to differ fromother types of meaning mainly by virtue of the fact that it is context-independentand fully propositional (sentence meaning). Derivatively, the term also refers totypes of meaning which are not propositional, but crucially are context-independentand are derived from a full proposition via the principle of compositionality (lexicalmeaning).

lexical meaning to section 2.3.3See below, however, for formal semantic approaches that also take into account contextual

information.

Literal Meaning and Context-Independence 11

meaning

context-independent context-dependent

lexical m. sentence m. . . . . . . . . .

non-literalliteral

Figure 2.1: Traditional differentiation of types of meaning

So far so good. However, the characterisation of formal semantics as stated abovehas proven to be problematic. And, as we will see, these problems also extend to thecharacterisation given to the notion of literal meaning. Thus, to summarise: in itstraditional form, three of formal semantics’ main assumptions are the following: a)semantics is concerned with the context-independent meaning of natural languageexpressions, b) for sentences, what is determined by the semantic component of anatural language grammar is the proposition expressed by that sentence and c) forsimple expressions their semantics (or lexical meaning) is whatever aspects of theirmeaning remain constant across different uses of that expression.

However, as Sag (1981) points out: a formal semantic theory which does not allowfor any contextual information to be made use of in determining the propositionexpressed by a sentence ‘. . . appears to be falsified by the mere existence of sentencescontaining tense morphemes or other indexical expressions.’ (Ibid, p. 274). Thus,consider the sentence in (1).

(1) He went to the bank yesterday.

For the sentence in example (1), it is clearly not the case, that semantics determinesa truth-evaluable proposition, due to the occurrences of the context-dependent ex-pressions he and yesterday as well as the homonymous noun bank. As is the casefor all indexical expressions, the exact reference of he and yesterday differs with thecontexts in which they are uttered. Thus, for such expressions semantics only givesrules for ’where to look’ in the search for potential referents. In the case of the occur-rence in a sentence of homonymous expressions such as bank, the assumption is thatthe process of semantic composition has to build up as many different structuresfor the sentence, as there are ambiguous expressions in it. Thus, for (1) to expressa full proposition, the references of the occurring indexical expressions have to befixed, that is, recourse has to be taken to the context of the utterance. Moreover,the sentence has to be disambiguated, which, again, is only possible with the helpof contextual information. Even then, the sentence does not express a propositionuntil the reference of the NP the bank to some unique location has been fixed. Whatthis shows is that the proposition expressed by some sentence can only actually be

determined once the context in which it is uttered is taken into consideration. Thus,it seems that the semanticist cannot uphold both assumptions a) and b). If he wantsto rescue assumption a), it seems he has to concede that, in fact, the semantic com-ponent does not determine the truth-evaluable proposition expressed by a sentence;if he wants to rescue the assumption in b), he has to allow for context-sensitiveprocesses to take place during the determination of the proposition expressed by asentence.

However, formal semantic approaches exist which attempt to capture the differ-ence between context-sensitive and context-insensitive expressions and at the sametime uphold assumptions a) and b). One such approach is Kaplan (1989b). Thus,Kaplan proposes to differentiate between, in a sense, two meanings of expressions:their character and their content.4 Consider example (2).

(2) a. Mary: I am hungry.

b. John: I am hungry.

On the one hand, the notion of character captures the intuition that Mary and Johnin a way have said the same thing: both used the same sentence. The notion ofcontent, on the other hand, captures the intuition that, at the same time, Mary andJohn have not expressed the same idea. Kaplan’s suggestion is that a sentence’scharacter is a function that takes a context in order to deliver a proposition or thecontent of that sentence in that context. Thus, although Mary and John use the samesentence, they express different propositions: Mary says that she is hungry, whereasJohn says that he is hungry. This difference is due to the character of I, which can beglossed as ‘referring to the speaker or writer’. Applying I’s character to a particularcontext determines the actual speaker in that context, i.e., the content (or intension)of I in that particular context. Having determined the proposition expressed by asentence in a particular context, the proposition can then be evaluated with respectto a circumstance of use or possible world. Thus, the content (or intension) of asentence in a particular context is a function from possible worlds to truth values.

In Kaplan’s approach, then, context-sensitive expressions are such that theircharacter applied to different contexts yields different contents. However, a context-sensitive expression’s content in turn is a constant function from possible worlds toextensions since regardless of the world at which the content of the expression isevaluated, it will always have the same extension. For example, the content of anexpression such as I varies depending on the context in which it is used. However,once the content is determined, it stays the same for all possible worlds. In contrast,the content of hungry does not depend on the context in which it is used. It alwaysis the property ‘being-hungry’. However, the actual extension of this predicatedepends on the possible world that is assumed. That is, the set of individualsto which the predicate applies may differ across different worlds. Thus, context-insensitive expressions have varying extensions, while their characters are such thatregardless of the context the respective character is applied to, the same contentwill be determined. In a way, for context-insensitive expressions their character andcontent fall together.

4Cf. Chierchia and McConnell-Ginet (2000), Braun (2010) for accessible introductions to thesenotions.

Kaplan’s approach, thus, allows a differentiation of three levels of meaning: char-acter, content or intension and extension. For sentences this means one can differen-tiate between the context-independent sentence meaning, the proposition expressedby a sentence in a context and the truth value of a sentence in a context with respectto a possible world. Thus, implementing these ideas in a model-theoretic semanticapparatus leads to the truth of a sentence not only being determined with respectto a world and time, but also a context of utterance (cf. Sag 1981).

In a way, within such an approach, both assumptions a) and b) can be main-tained. That is, what is determined by the semantic component is the context-independent meaning of a sentence and the conditions under which that sentence istrue. Using the indices w, i and c, the instruction of how to determine the propo-sition expressed by a sentence is also given. However, it should be noted that theproposition expressed only actually is determined, once the functions are applied toa particular world, time and context. In other words, although it is possible withinsuch an approach to formulate conditions under which a particular sentence is true,due to the indices used the sentence’s meaning thus given may be compatible withquite a number of different situations. Thus, it cannot be taken to represent theproposition expressed in a particular utterance situation.

Traditionally, formal semanticists have assumed that the semantic component ofthe language faculty determines the meaning both of simple and complex expressionsand then there are a restricted number of processes (namely, resolving of reference,fixing of indexicals and disambiguation) that lead to the proposition expressed bya sentence. However, these processes are not explicitly referred to as being of apragmatic nature. This is quite obvious in the works of Grice, who mentions theprocesses that lead to what he called what is said, but does not seem to considerthem as pragmatic in the same sense as the processes that result in conversationalimplicatures (Grice 1975) (cf. figure 2.2). However, if pragmatic processes are charac-terised by the fact that they take into account contextual information then, surely,the processes of fixing indexicals, resolving references and disambiguation are ofpragmatic nature.

Semantics lexical meaning

sentence meaning

Pragmatics

semantic composition

basis for furtherpragmatic inferences

reference resolutionfixing indexicalsdisambiguation

what is meant

what is said

conversational implicature. . .

Figure 2.2: Grice’s distinction of what is said and what is meant

Thus, as Strawson (1950) noted, it is not sentences which express something ofwhich it makes sense to ask whether it is true or false but rather the utterances ofthose sentences. Thus, one and the same sentence can be used to express somethingtrue at one point and something false at another. That is, regardless of whethersentences include indexical or ambiguous expressions, it is not a general property ofsentences, but rather of utterances that they express propositions.

If it is not sentences per se that express propositions and are truth-evaluable butrather their utterances, what exactly, then, does the concept of sentence meaningcapture? This is an important question considering that formal semantics takessentence meaning as the starting point from which to deduce the meanings of sim-ple expressions, which presupposes that the notion of sentence meaning is clearlydefined. A possible answer is to still regard both the meaning of simple as wellas complex expressions, in particular sentences, as essentially context-independent.That is, as traditionally assumed, semantics deals with the meaning of both simpleand complex expressions, where the meaning of simple expressions forms part oftheir lexical entries and the meaning of complex expressions is a function of themeanings of their parts and their syntactic combination. However, such a view doesnot claim that sentence meaning necessarily is propositional; it simply assumes thatsentence meaning is context-independent.

2.1.2 Literal Meaning as Context-Independent?

But what about the correlation between sentence meaning and literal meaning sug-gested above? There it was stated that, apparently, literal meaning refers to a levelof meaning identified as sentence meaning by traditional formal semantics and char-acterised as being context-independent and fully propositional. The assumption wasthat the notion of literal meaning mainly captures the fact that sentence meaning iscontext-independent, thus, with the revised characterisation of sentence meaning as‘only’ context-independent but not necessarily fully propositional, the term literalmeaning should still be applicable to that level of meaning.

There are a number of considerations that go against this characterisation. Thus,recall the uses of the term literal meaning mentioned in chapter 1, where the term,on the one hand, is used to refer to a kind of context-dependent but at the sametime in some sense ‘basic’ meaning and, on the other hand, is contrasted with a kindof meaning that is not only context-dependent but crucially in some sense ‘derived’or non-basic (cf. Sag 1981, Recanati 1995, Carston 2007). As mentioned before,such a use calls into question the adequacy of characterising literal meaning ascontext-independent meaning. In fact, already in his (1978) paper, Searle criticisedthis characterisation of literal meaning. He argues that there is no such thing as asolely linguistically determined literal meaning of a complex expression. As regardssentence meaning, one cannot speak of the literal meaning of a sentence in thestandard sense. As Recanati (2004) puts it, Searle holds the view of contextualism,according to which ‘. . . there is no level of meaning which is both (i) propositional(truth-evaluable) and (ii) minimalist, that is, unaffected by top-down factors.’ (Ibid.,p. 90). Thus, Searle assumes that the expression of a determinate proposition takesplace against a set of background assumptions. To illustrate his point of view, Searle

uses the sentence in (3), which, taken out of context, seems to have a quite obviousliteral meaning, which Searle (1978) depicts as in 2.3.

(3) The cat is on the mat.

Figure 2.3: The typical cat-on-the-mat configuration

The problem with this ‘literal sentence meaning’ is that although speakers orhearers are not necessarily aware of the fact, a number of preconditions are assumedto hold5. To show this, Searle constructs a context of utterance for the sentencein (3), where it is questionable whether one would want to say that the sentencecorrectly describes the state of affairs at hand.

‘. . . suppose the cat and the mat are in exactly the relations depicted onlythey are floating freely in outer space, perhaps the Milky Way galaxyaltogether. In such a situation the scene would be just as well depicted ifwe turn the paper on edge or upside down since there is no gravitationalfield relative to which one is above the other. Is the cat still on the mat?’(Searle 1978, cited from Searle 1979, p. 122)

Thus, if what the meaning of a sentence does is determine a set of truth-conditions,Searle argues that for most sentences this determination can only take place againstspecific background assumptions. These background assumptions are not part ofthe semantic structure of the sentence, that is, they are unarticulated. Moreover,due to possible variations in the background assumptions, the same sentence mighthave varying truth-conditions. For any sentence, there is no fixed set of backgroundassumptions of which it could be said that it determines that sentence’s literalmeaning. To illustrate this fact, Searle construes a context of utterance for (3), inwhich it could be used to truthfully describe a situation such as depicted in figure2.46.

A further example for the fact that the literal meaning of a sentence depends onbackground assumptions can be found in Searle (1980). Searle gives a number ofsentences containing the verb to cut; here are the first five.

5 Note that Searle is not referring to the fact that the sentence in (3) additionally containsindexical elements. That is another matter.

6This is Searle’s context: ‘The mat is in its stiff angled position, as in [figure 2.4], and it is partof a row of objects similarly sticking up at odd angles - a board, a fence post, an iron rod, etc.These facts are known to both speaker and hearer. The cat jumps from one of these objects toanother. It is pretty obvious what the correct answer to the question “Where is the cat?” shouldbe when the cat is in the attitude depicted in [figure 2.4]: The cat is on the mat.’ (Searle 1978,cited from Searle 1979, p. 125).

Figure 2.4: A rather unusual cat-on-the-mat configuration

(4) a. Bill cut the grass.

b. The barber cut Tom’s hair.

c. Sally cut the cake.

d. I just cut my skin.

e. The tailor cut the cloth.

As Searle notes, in each of the example sentences in (4) cut occurs in its literalmeaning. There is nothing in these sentences as such that would lead one to interpretthem as metaphorical or figurative. However, although cut occurs in its literalmeaning, the situations that it is used to describe differ conceptually. Thus, althoughcut is used in its literal meaning, for the different sentences in (4), it determinesdifferent truth conditions. This can be seen if one considers what it would mean toobey an order of cutting something. Searle puts it as follows.

If someone tells me to cut the grass and I rush out and stab it witha knife, or if I am ordered to cut the cake and I run over it with alawnmower, in each case I will have failed to obey the order. That isnot what the speaker meant by his literal and serious utterance of thesentence. (Searle 1980, p. 223).

Thus, again, in the examples in (4), the literal meaning of the individual sentences(and of the word cut) is determined against a set of background assumptions, namelywhat we know about lawns and cakes and so on and what are usual actions in whichwe involve with regard to those ‘things’.

Furthermore, in his discussion on the cut examples, Searle points out that it isnot sufficient to assume that the different readings of cut – its different literal mean-ings are due to some intrasentential interaction between the verb and its internalargument. That is, he argues against the view according to which cut together withthe respective argument determines that cut in ‘cut the grass’ will receive a differentinterpretation from the one it receives in ‘cut the cake’. His reasoning is that it ispossible to ‘. . . imagine circumstances in which “cut” in “cut the grass” would havethe same interpretation it has in “cut the cake”. . . ’ (Searle 1980, p. 224).

‘Suppose you and I run a sod farm where we sell strips of grass turf topeople who want a lawn in a hurry. [. . . ] Suppose I say to you, “Cut halfan acre of grass for this customer”; I might mean not that you shouldmow it, but that you should slice it into strips as you could cut a cakeor a loaf of bread.’ (Searle 1980, p. 224–5).

Moreover, he points out that there is a difference to be drawn between what hecalls background assumptions and the special context of utterance for a given utter-ance. While background assumptions are involved in determining a sentence’s literalmeaning or truth-conditions, the context in which a sentence is uttered helps thehearer to decide on whether a speaker intended her utterance to be taken literallyor non-literally. However, since Searle does not explicitly define what constitutesbackground assumptions and what is part of the context of an utterance, the ques-tion arises whether this differentiation really is necessary. From the examples Searleuses to defend his view of what constitutes literal meaning, it could be argued thatthe background assumptions necessary for determining the literal meaning of a sen-tence are in fact part of the specific context in which an utterance takes place. WhatSearle obviously means by background assumptions are certain aspects of knowledgethat we have, namely those aspects which are relevant in the particular utterancesituation. Thus, one could also assume that depending on the situation speakersand hearers find themselves in that situation will make certain aspects of knowledgethey have more prominent (or salient). Those aspects, then, constitute what Searlecalls background assumptions in the sense that speakers and hearers are presumablynormally not aware of basing their utterances and interpretations on such assump-tions. As their name implies, background assumptions are in the background; theyform the basis from which speakers formulate their utterances and hearers intepretthem. Thus, background assumptions depend on the particular context of utteranceand therefore can be said to form part of the contextual information used in inter-preting. Thus, in order to disambiguate whether the expression cut is used with themeaning as in ‘cut the grass’ or with the meaning as in ‘cut the cake’, the hearerneeds to take into account contextual information. That is, even if the backgroundof the utterance is such as Searle gives it, the hearer would still have to decide thatthe reading of cut as in ‘cut the cake’ is the one the speaker intended in that situa-tion. From what has been said about salience of meaning above, it is of course verylikely that the particular utterance situation will speed up the hearer’s unconsciousdecision.

Furthermore, Searle’s argument that there are possible circumstances in whichcut in ‘cut the grass’ may be interpreted as in ‘cut the cake’ actually does not consti-tute an argument against the assumption that the interpretation of cut is influencedby the intrasentential context. That is, one could assume that the co-occurrence ofparticular lexical items does help the hearer to narrow down the possible sentencemeaning. However, this influence on the interpretation might have a default charac-ter. Thus, it only applies where the particular contextual conditions do not preventit from applying. In the context Searle supplies, the interpretation of cut as in ‘cutthe grass’ is rendered less likely as being the intended reading, than the reading asin ‘cut the cake’. As Searle argues, this reading of cut in the given context doesnot seem to be a non-literal reading of cut, since, intuitively at least, it does notseem to be derived from some clear basic, underlying meaning. However, assumingthat the reading only comes about, or is interpretable as intended, in a particularcontext of utterance, suggests once again that literal meaning should not be takento be a phenomenon of context-free sentence meaning. If that is the case, then theconcept of literal meaning is not applicable at Searle’s level of sentence meaning, but

rather at some context-dependent level of meaning. Be that as it may, Searle stillassumes that literal meaning is the basis for any non-literal meaning. It is speakerswho may use some expression or other non-literally. Thus, non-literal meaningshave to be intended and should be expected to be consciously recognisable as such.That is, speakers should have no difficulty identifying some reading as being non-literal, as they have to intentionally use some expression ‘deviantly’ in order for thatexpression to get interpreted non-literally.

Although Searle thus argues against the view according to which literal meaningis determined by the linguistic system alone, he does not want to deny that sen-tences in fact do have literal meanings. ‘Literal meaning, though relative, is stillliteral meaning.’ (Searle 1978, cited from Searle 1979, p. 132). However, he appliesthe concept of literal meaning to those cases in which the speaker means what shesays, contrasting them with those cases in which the speaker means more, or some-thing different from what she said (e.g., cases of irony, conversational implicature orindirect speech acts). Thus, although Searle identified literal meaning as belongingto the level of sentence meaning, actually the differentiation between literal and non-literal utterances seems to hold at the level of utterances. However, he also arguesthat literal meaning is a relative notion. That is, it is rather likely that what wetake to be the literal meaning of the utterance of some sentence will depend heavilyon the specific circumstances in which the utterance of that sentence takes place.This is in agreement with uses of the term literal meaning where it refers to anutterance’s meaning, in contradistinction to the meaning intended by the speaker ofthat utterance.

Similarly to Searle, Bierwisch (1979, 1983) assumes that what is called the lit-eral meaning of an utterance of some (simple or complex) expression is not identicalto the linguistically determined meaning of that expression. Thus, Bierwisch alsoplaces literal meaning at a level of meaning that is no longer independent of con-text, namely the level of utterance meaning. Utterance meaning is the meaning anutterance token of an expression has when it is used in a context. The utterancemeaning can be equivalent to the utterance token’s literal meaning, but it does notnecessarily have to be. Therefore, Bierwisch differentiates the literal meaning of anexpression from its utterance meaning. Crucially, an utterance token of an expres-sion can only have literal (or, for that matter, non-literal) meaning in a context.With respect to this assumption, Bierwisch and Searle hold similar views. Thatis, what is called the literal meaning of an expression is not determined languageinternally, rather, it is dependent on certain background assumptions (Searle) or aparticular context of utterance (Bierwisch). Thus, literal meaning is a special caseof utterance meaning. A consequence of such a view is the assumption that thelexical semantic representations of simple expressions do not encode what we taketo be their literal meaning.

What has been said sofar, corroborates a suspicion expressed at the beginningof this section. That is, one has to ask whether the particular standard characteri-sation of literal meaning might not be very much influenced by our characterisationof the field of semantics. Thus, consider again that, traditionally, semantics takesas its starting point sentence meaning of which it assumes that it is both context-independent as well as propositional. Because it is taken to be context-independent,

it seems to be what the sentence literally expresses. However, we saw that, for alarge number of sentences, it cannot be said that they express propositions, unlesscontextual information is first taken into account (e.g., cases of reference resolution,fixing of indexicals and disambiguation). That is, sentence meaning and ‘proposi-tional content of an utterance’ are not equivalent. Furthermore, as we saw fromSearle’s remarks, actually, what we take to be the literal meaning of a complex ex-pression, is dependent on the context in which that expression is uttered. That is,essentially, literal meaning seems to be context-dependent after all.

Thus, it seems reasonable to posit a partial new characterisation of the termssentence meaning and literal meaning (with respect to sentences). Whereas the for-mer is the meaning of a certain type of complex expression and characterised bythe facts that it is compositional, context-independent and (more often than not)sub-propositional, the latter is a certain type of meaning a sentence may have whenused in a context that allows a literal interpretation and in which that sentenceexpresses a full proposition. What this characterisation of literal meaning suggestsis that whatever differentiates between literal meaning and non-literal meaning can-not be the criterion of context-(in)dependence. Moreover, this characterisation ofliteral meaning (with respect to sentences) makes it equivalent to a particular typeof proposition. Thus, a new question arises, namely, how this particular type ofproposition is characterised. That is, which conditions does a proposition have tofulfil for it to be literal in meaning? This is a question which is very close to thecore of the discussion around the semantics/pragmatics distinction and we will comeback to it in the following chapters.

meaning

context-independent context-dependent

lexical m. sentence m. . . . . . . . . .

non-literalunderspecified literal

Figure 2.5: Revised differentiation of types of meaning

2.1.3 Literal Meaning as Primary to Non-literal Meaning?

The traditional view of the interpretation process assumes that an utterance’s lit-eral meaning is always activated first. Potential non-literal meanings only get acti-vated as a result of the literal meaning’s not fitting in the respective context. Thisview, of course, is based on the traditional assumption about literal meaning beingcontext-independent. Since this allows an interpretation of the literal meaning of an

utterance without taking recourse to contextual information, it will be computedautomatically. If afterwards it becomes apparent that the literal meaning does notfit the contextual circumstances, a reinterpretation will take place, resulting in anon-literal interpretation of the utterance (call this the standard pragmatic view).

This is essentially how Grice must have viewed the relation of literal meaningand non-literal meaning, since he described different kinds of non-literal meaning,such as irony and metapor, as being conversational implicatures, that is, inferencesthat require a prior recovery of what is said by an utterance. Of this latter level ofmeaning, Grice said that it is very closely connected to the conventional meaning ofthe words or the sentence uttered by the speaker. Specifying on this characterisation,what is said has been taken to be the fully propositional semantic form of theutterance resulting from the processes of disambiguation, reference resolution andfixing of indexicals, where these processes already are of pragmatic nature since theyinvolve contextual information (see figure 2.6).7

sentence meaning

what is said

Pragmatics

what is meant

conversational implicaturespeech acts. . .

does not involve anynon-literal meaning

may involvenon-literal meaning

Figure 2.6: (Non-)Literal meaning and Grice’s levels of what is said/what is meant

Thus, what is said, being a full proposition, provides the basis for further infer-ences about what the speaker actually meant with his utterance. That is, conversa-tional implicatures are derived from the fact that the speaker said what he said inthe particular way he did and with respect to a number of conversational maxims,which are taken to underlie human communication. Since what is said includes theconventional meanings of expressions, it is traditionally assumed to be the level atwhich literal meaning is expressed. And since Grice, similarly to Searle, viewed non-literal meaning such as irony or metapher, as an aspect of speaker meaning, what ismeant is the level of meaning at which such non-literal meaning aspects come intoplay. Thus, since for the recovery of what the speaker meant we first have to knowwhat the speaker said, interpretation of the literal meaning of an utterance is priorto the interpretation of a potential non-literal meaning. Moreover, since non-literal

7Note that this fact is independent of the assumption that what is said constitutes the level ofmeaning from which conversational implicatures are determined. That is, even if what is said isnot viewed as wholly semantic, the standard pragmatic view could still hold in that what is saidstill forms the basis for drawing further inferences and thus has to be determined first.

meaning aspects such as irony or metaphor rest on the violation of a conversationalmaxim, they will only get derived if the literal interpretation, that is, what is said,cannot possibly be construed as the meaning intended by the speaker.

Interestingly, although arguing against viewing literal meaning as solely linguis-tically determined, Searle shares the view according to which literal meaning isprior to non-literal meaning. Thus, in his (1979) paper, Searle says of metaphoricaland ironical utterances, that their respective interpretations are arrived at by goingthrough the literal meaning of the sentences used to make the utterances. Thus, heseems to assume that although a sentence’s literal meaning can only be determinedagainst particular background assumptions, in terms of the temporal progression ofthe interpretation process, literal meaning is a necessary intermediate step in theinterpretation of non-literal meaning.

Fortunately, with the methods developed in psycholinguistics, the assumption ofthe primacy of literal meaning as an assumption about the operational sequence ofthe interpretation process has become empirically testable. And in fact, results ofexperiments employing different methods in examining the understanding of varioustypes of non-literal meaning in comparison to literal meaning show that the standardpragmatic view makes the wrong predictions. That is, in terms of cognitive effort,the standard pragmatic view predicts that interpreting non-literal meaning shouldbe cognitively more exacting than the interpretation of literal meaning. Given thatreaction or reading times mirror the relative cognitive effort involved in interpret-ing an utterance, results such as the following suggest that the interpretation ofnon-literal meaning does not necessarily differ from that of literal meaning. Thus,Gibbs (1994) mentions an experiment (Ortony et al. 1978), where subjects werepresented with sentences in a context that was either literal or metaphoric. Thehypothesis that was tested in the experiment was that people may not have to anal-yse the literal interpretation of a metaphorical utterance before actually derivingthe intended metaphorical reading. The hypothesis was confirmed. Thus, althoughsubjects took longer to read metaphorical targets than literal ones in short contexts,in long contexts, there was no difference in reading times for the metaphorical andthe literal target sentences. These results suggest that the richness of contextualinformation available during the interpretation of an utterance has an effect on howdifficult it will be to give that utterance a non-literal interpretation. Another ex-periment showed that utterances may be interpreted non-literally although thereare no conditions that trigger the failure of a literal interpretation, suggesting thatpeople automatically apprehend the metaphorical meaning of an utterance (Glucks-berg et al. 1982). The task was to judge sentences such as Some jobs are jails as totheir literal truth. Thus, it was not necessary to seek a non-literal interpretation forthe sentences, subjects only were asked for the literal truth of each sentence. Now,if for a non-literal interpretation of a sentence a pragmatic triggering condition isrequired, sentences such as Some jobs are jails should simply be considered as false.If, however, people automatically interpret the metaphorical meaning of such sen-tences, then the ‘false’ judgement for the literal reading of the sentence should be inconflict with the ‘true’ judgement for the non-literal reading of the sentence. Andin fact, although subjects correctly judged sentences such as Some jobs are jailsas literally false, if a metaphorical interpretation for the sentence in question was

available, subjects took much longer to make that judgement. Thus, apparently, themetaphorical meanings were automatically interpreted, without the need for somepragmatic triggering condition (i.e. maxim violation). This suggests that the inter-pretation of non-literal meaning does not rely on the violation of some conversationalmaxims or principles. Moreover, it shows that, although context may facilitate theinterpretation of an utterance as non-literal, it is not absolutely necessary.

Note, however, that more can be said with respect to such examples as Somejobs are jails. Thus, although it is true that there does not seem to be a prag-matic triggering condition such as maxim-violation for a metaphoric interpretation,nevertheless, it can be argued that in cases such as this, there is some triggeringcondition, either semantically or pragmatically induced. To repeat, the standardpragmatic view assumes the literal meaning of an utterance is interpreted first andonly if this does not fit the contextual circumstances a non-literal interpretation isdetermined. The problem with sentences such as Some jobs are jails or, for thatmatter, The ham sandwich is sitting at table 7 is that it is not clear what their literalmeaning should be (cp. Stern 2006). Thus, the longer reaction times measured inthe experiment mentioned above might be due to the fact that whatever componentis responsible for this stage in the interpretation process is having problems deter-mining a literal interpretation for the sentences in question. This, in turn may besufficient to trigger an alternative, non-literal interpretation (cp. Dascal 1987).

Keysar (1989) takes up this criticism and shows that even in contexts were aparticular sentence is understood as literally true, that sentence’s potential but falsemetaphorical meaning interfers nevertheless. Thus, subjects were asked to judge testsentences as true or strongly implied to be true after having read small texts. Thetexts consisted of two parts, were one part related to the literal interpretation (L) ofthe target sentence, rendering it either true (L+) or false (L-), and one part relatedto the metaphorical interpretation (M), again rendering this either true (M+) orfalse (M-). Thus, texts were (L+M+), (L-M+), (L+M-) or (L-M-). For example,one of the test sentences was Bob Jones is a magician. An example text for whichthat sentence is interpreted as literally true but metaphorically false (L+M-) is givenbelow.

Bob Jones is an expert at such stunts as sawing a woman in half andpulling rabbits out of hats. He earns his living travelling around theworld with an expensive entourage of equipment and assistants. Al-though Bob tries to budget carefully, it seems to him that money justdisappears into thin air. With such huge audiences, why doesn’t he everbreak even? (Keysar 1989, p. 378)

The results show that subjects are quickest in responding after texts that renderedthe target sentence both literally and metaphorically true (L+M+). Generally, theyare quicker in responding in congruent contexts (i.e. L+M+ and L-M-) than inincongruent ones. Thus, as in (Glucksberg et al. 1982)’s experiment, subjects takelonger judging literally false but metaphorically true sentences. Crucially, however,they also take longer judging literally true but metaphorically false sentences, indi-cating that even in such a situation the potential metaphorical interpretation of thesentence is computed. This result shows that the longer reaction time measured by

Non-literal Meaning and Conventionality 23

(Glucksberg et al. 1982) for literally false but metaphorically true sentences is notdue to the difficulty of determining a literal interpretation for those sentences in thefirst place, as similar results are achieved in situations in which the target sentenceactually is literally true but metaphorically false.

Summing this section up, the conclusion one can draw from examining differentempirical studies is that it is not generally necessary to compute the complete lit-eral meaning of an utterance before deriving that utterance’s intended non-literalmeaning (cf. Gibbs 2002). Thus, it seems that the possible but non-intended literalinterpretation of an utterance does not constitute a necessary intermediate step dur-ing the processing of that utterance’s intended non-literal interpretation. Moreover,contextual information seems to be an important factor in the non-literal interpre-tation of utterances. Thus, if the context in which an utterance takes place pointstoward a literal interpretation, the utterance will be interpreted literally. If, how-ever, the context of an utterance points towards a non-literal interpretation, theutterance will be interpreted non-literally. However, it seems that context is not al-ways necessary for the non-literal interpretation of an utterance. Thus, Glucksberget al. (1982) found that sentences such as Some jobs are jails are interpreted non-literally, even if there is no context given with which the literal meaning could besaid to be conflicting. The criticism that in such cases what may underlie the longertime needed for judgements concerning the literal truth/falsity of such sentencesis not a maxim-violation problem, but rather the problem of determining a literalmeaning at all was refuted by Keysar (1989). He found that even in contexts werea particular sentence gets a straightforward literal interpretation, a potential butnon-fitting metaphorical interpretation interfers leading to longer reaction times insubjects. Thus, non-literal interpretations seem to get considered without the needof a triggering condition. With such results as the above, it becomes questionablewhether it is warranted to assume a difference in conventionality for literal meaningand non-literal meaning.

2.2 Non-literal Meaning and Conventionality

Whereas, traditionally, literal meaning was taken to be semantic, that is, context-independent, conventional and primary, non-literal meaning was characterised aspragmatic, that is, context-dependent, non-conventional, a feature of the intendedspeaker meaning arrived at by a reinterpretation process and thus secondary toliteral meaning. Having in the last section mainly argued against the traditionalcharacterisation of literal meaning as context-independent and thus primary, in thissection, I will argue that non-literal meaning should not be viewed as necessarily non-conventional and thus secondary. Generally, I want to argue that conventionalityshould not be viewed as an ‘all-or-nothing’ property, but rather as being gradual.

2.2.1 Empirical Evidence

Since non-literal meaning usually is considered a feature of language use, it is not sur-prising that the term non-conventional often features in its characterisation, whereits opposite conventional is mostly used in the sense of ‘stored in the lexicon’. How-

ever, it is not necessarily so that a particular instance of non-literal meaning maynot be considered as conventional. Rather, given that conventionality is reflected byfamiliarity and relative ease of interpretation, cases of conventionalised non-literalmeanings of expressions are quite numerous.8 As a consequence, context may notactually play such an important role when it comes to the interpretation of an ex-pression as literal or non-literal. Moreover, and as already suggested in the lastsection, the claim that non-literal meaning is always the result of a reinterpretationof some earlier composed literal meaning can no longer be sustained.

Thus, consider the claims Rachel Giora (1997, 1999, 2002) makes in the con-text of her graded salience hypothesis. Although she does not question the generalimportance of contextual information for utterance interpretation, she argues thatregardless of the richness of the context, it can be shown that in particular casesboth the literal as well as the non-literal meaning of an expression may be initiallyactivated, suggesting that in such cases, the two types of meaning have the samestatus. More specifically she argues that it is the notions of salient vs. non-salientmeaning that play a decisive role for the process of interpreting utterances. Thecrucial idea is that salient meanings are always activated, whether they are actuallyintended or not. Thus, in her (1999) paper, Giora mentions a number of experimentswhich, again, show that the non-literal meaning of an expression is not necessarilyactivated after its literal meaning has been. For example, Giora and Fein (1999)showed that reading times for familiar metaphors9 were equal for literally as wellas metaphorically biasing contexts. That is, for familiar metaphors, both their pos-sible literal and non-literal meanings are activated during interpretation, regardlessof whether they are uttered in literally or metaphorically biasing contexts. This isbecause both meanings are familiar (salient) as for instance for to break ones head.

(5) a. In order to solve the math problem, the student broke her head.

b. Because she was so careless when she jumped into the pool, the studentbroke her head.

Thus, the experiments show that in cases of familiar metaphors literal meaningand non-literal meaning of an expression are computed in parallel. In contrast, forless familiar metaphors only their literal meaning is salient, thus, it should be ac-tivated prior to the less familiar metaphoric meaning. This hypothesis was indeedconfirmed by longer reading times found for less familiar metaphors in metaphor-ically biasing contexts as in contrast to literally biasing contexts. Giora (1999)cites the following sample contexts and target sentences for a less or unfamiliarmetaphor.10

8Having said this, one might wonder whether under such circumstances it is actually still possibleand useful to differentiate literal from non-literal meaning. It is if one wants to capture the fact thatnon-literal meaning, even if conventionalised, is perceived as in some sense non-basic or derived.

9It should be noted that this is a somewhat sloppy use of the term metaphor. What is meantis that expressions where investigated that could either be interpreted metaphorically or literallyand for which the metaphorical interpretation is rather common.

10It should be mentioned that the original experiment was carried out in Hebrew. Unfortunately,neither Giora (1999) nor Giora and Fein (1999) give examples for any of the original material. Thus,the critical points I will make below concerning the material and the differentiation of idiom frommetaphor could just be an artefact of the English translation. However, as an Hebrew informant

(6) Mary: My husband is terribly annoyed by his new boss. Every day he comeshome even more depressed than he had been the day before. Somehow, hecannot adjust himself to the new situation.

Jane: Their bone density is not like ours.

(7) Our granny had a fracture just from falling off a chair and was rushed to thehospital. I told my sister that I never had fractions falling off a chair. Sheexplained to me about elders. She said: Their bone density is not like ours.

Note, however, that at least for the example cited, the longer reading time mayalso at least partly result from a further difficulty encountered during the inter-pretation process. Thus, whereas, arguably, in (5a) and (5b) the binding of theanaphoric expression her is equally difficult, the case is different for their in (6) and(7). Whereas in (7) their can be bound to the overt antecedent elders in the pre-ceding sentence, there is no overt antecedent for their in the preceding sentence in(6). Thus, the longer reading time may reflect the greater difficulty in binding theanaphoric expression their to an appropriate, not explicitly expressed antecedent. Asimilar example of metaphoric material taken from Giora and Fein (1999) suggeststhat this is a general problem of the testing material used to investigate differencesin the interpretation of metaphoric vs. literal interpretations.

(8) The Saturday night party went on for hours. Drinks were poured, and wedanced all night. We were probably less than considerate when, the nextevening, we called on our friends who had been partying with us. When theyopened the door, we realised: Only then did they wake up.

(9) A bloody war has been going on in central Europe for a few years. Thousandsof innocent women, men and children got massacred, and no one budged orlifted a finger. At last, a decision was made to intervene in the fights. Onlythen did they wake up

As in the example before, in (8) the anaphoric expression they in the last sentencecan be bound to an appropriate antecedent in the preceding sentence, namely ourfriends. This, however is not the case for they in (9), where there, once again,does not seem to be an appropriate, overt antecedent given. An attendant, generalproblem is the fact that the context preceding the target sentence may also primethe meaning intended in the target sentence. Thus, the fact that in (7) in thesentences preceding the target sentence, such expression as fracture and fraction areused, facilitates a literal interpretation of bone density in the target sentence. Thereis no such priming effect for the metaphoric interpretation of bone density in (6).

The results of another experiment employing a word fragment completion task(Giora and Fein 1999) also suggests that the interpretation process varies with thefamiliarity of the metaphor processed. In this experiment, subjects were instructedto read little texts, the last sentence of which was the target sentence, involving ametaphor. Depending on the nature of the preceding context, subjects were intentedto be biased either towards a literal interpretation or a metaphorical interpretationof the target sentence. Their task then was to complete two fragmented words, one of

tells me, the criticism also holds when considering the Hebrew equivalents.

which was related to the literal (or rather compositional) interpretation of the targetsentence, and one which related to the metaphorical interpretation. The resultsshowed that in the case of familiar metaphors - regardless of the bias induced bythe preceding context - both the compositional as well as the metaphorical meaningwere activated. In contrast, for less familiar metaphors, there was near to noneactivation of the metaphorical meaning in the literally biasing context, and moreactivation of the literal meaning than the metaphoric meaning in the metaphoricallybiasing context.

Giora and Fein (1999) also tested the activation of meaning during the inter-pretation of familiar and less or unfamiliar idioms. The assumption was that forfamiliar idioms, their idiomatic meaning is more salient than the possible literalmeaning. Thus, the idiomatic meaning should get activated both in idiomaticallybiasing as well as literally biasing contexts, whereas the literal meaning should onlyget activated in literally biasing contexts. As before, the task was to completefragmented words with the first word that came to mind after reading an eitherliterally or idiomatically biasing context with the last sentence being the target (seeexample below). This time, however, subjects were asked to complete only one ofthe two given fragmentations. Again, the results of the experiment confirmed thishypothesis.

(10) a. He told me the whole story on one leg.

b. In the zoo, I saw a stork standing on one leg.

Thus, what the results of the different experiments mentioned in this and thepreceding section show is that literal meaning is not necessarily processed first ininterpretation. Moreover, once a non-literal meaning of an expression has grownfamiliar, it can be activated in parallel to the literal meaning of that expression,suggesting that non-literal meaning should not be seen as a feature of the level ofspeaker meaning, exclusively. Thus, the results from psycholinguistic experimentsreinforce the need to rethink the view of the relation between literal meaning andnon-literal meaning. That is, although historically and with respect to new or adhoc instances of figurative meaning, the latter is derived from literal meaning bycertain processes, such as metaphorical use, synchronically speaking this fact mayno longer play a role. More specifically, in terms of the operational sequence of theinterpretation process, the fact that the non-literal meaning of an expression may beactivated in parallel to its literal meaning rather suggests that they are more similarin nature than traditionally assumed.

However, the results of experiments such as the above cannot be used as evidencefor one view or the other of literal meaning and non-literal meaning without caution,as there are a number of issues that need careful consideration. Generally, a lot ofpsycholinguistic studies are interested in phenomena such as metaphor, idioms andirony, none of which are uncontroversial. That is, there is a lot of debate concerningthe appropriate characterisation of these phenomena. Particularly with respect tometaphor and idioms, it does not seem to be trivial differentiating the one from theother, especially since one way for idioms to get the particular meaning they haveis through a metaphorical use of the phrase in question. Thus, looking again at theexamples for a metaphor and an idiom from Giora given above, I would content that

it is not intuitively clear by which criteria one is a metaphor and the other an idiom.

(11) to break ones head

(12) on one leg

Moreover, if metaphor is taken to differ from idiom by the fact that the metaphoricmeaning is related to a single simple expression rather than to a phrase, the ques-tion arises which expression in to break ones head carries this metaphoric meaning.11

However, judging from the examples Giora gives for metaphors, she does not seemto assume that metaphoric meanings are related to a single expression, which makesthe question, what, then in her view differentiates between idiom and metaphor evenmore pressing. Furthermore, Giora’s use of the terms literal meaning and non-literalmeaning with respect to idioms is rather confusing. It is generally assumed that id-ioms are coded in the lexicon as a whole form associated with the respective (moreor less) idiomatic (or in Giora’s sense non-literal) meaning. This coded meaningis usually taken to be the literal meaning of idioms, simply because the idiom isnot taken to be a decomposable phrase, but as a whole. Idioms are usually notconceived of as having literal meaning in Giora’s sense. It is the phrasal, composi-tional counterparts of idioms that can be said to have literal meaning. The problemis, that it is not totally clear how selection or activation in cases where idioms areinvolved works. That is, when a sentence such as He told me the whole story onone leg is read or heard, the question is at which point the idiomatic meaning isactivated. That is, how many segments must the addressee have heard in order toactivate the idiomatic meaning? It seems rather plausible then, that an addresseewill initially assume that he is dealing with a compositional phrase and then revisethis assumption as soon as it becomes clear that the phrase in question is one that iscoded in the lexicon as such. That is, from the phonological surface form only it isnot obvious whether a specific phrase is used with a compositional meaning or withan idiomatic meaning. In this sense idioms and their compositional counterpartsare similar to homonyms. They have the same surface graphical or phonologicalform, but differ in their meaning and, additionally, in their complexity. Given thisassumption, of course, one would expect that both meanings should be equally acti-vated, which, according to Giora and Fein (1999), they are not. The results in thatstudy show that for familiar idioms, in an idiomatically biasing context, the com-positional meaning is only marginally activated. However, since it does not seemclear on which basis Giora differentiates idioms from metaphors, the results cannotreally be compared, since some of the example expressions that were classified asmetaphors may quite as well be classified as idioms. Moreover, one cannot rule outthat the results for (familiar) metaphors and idioms in the two experiments onlydiffer due to the difference in testing procedure. Thus, recall that in the experimenton metaphors, subjects were asked to complete both of the two given fragmentedwords, whereas in the experiment on idioms, subjects only had to complete one ofthe two given words. The fact that for familiar idioms, the number of activated

11One would probably be inclined to say that break carries the metaphoric meaning, however,then the question arises of whether this reading of break can be described without referring to theexpression head.

idiomatic meanings is so much higher than the number of activated literal meaningsprobably is due to the fact that the idiomatic meaning is the most salient, however,for all we know the possible literal meanings might be activated to some degree aswell, but this simply isn’t captured by the design. In other words, had subjects beenasked to complete both given words, the difference between the extent of activationof the idiomatic and literal meanings might have been smaller and thus more similarto the difference recorded for familiar metaphors. In addition, since the experimentinvolved a task that measures the activation of concepts after the actual interpreta-tion process has already taken place, the results do not tell us anything about early,initial processes. In other words, for all we know, an idiom’s literal meaning (i.e.,the compositional meaning of the idiom’s compositional counterpart) may still beactivated first during the interpretation process, but may be suppressed as soon asit is clear that it does not fit the context. The degree to which such suppressiontakes place may have a significant impact on whether the initially activated literalmeaning is still active enough to get accessed by the time the task is being solved.12

Generally, Giora (and Feins)’s results do not tell us what exactly is goingon during the interpretation process.13 14 Thus, Giora and Fein (1999) used aword fragment completion task, which, it can be argued, is only sensitive to thestate of activation at a time when the interpretation process has already finished.Although reading times may be better reflective of difficulties encountered duringthe interpretation process, it still is not clear what, or which steps, exactly arereflected by that longer reading time. Vice versa, similar reading times do not tellus anything either of what exactly is going on during the interpretation processand whether what we assume to be literal meaning and non-literal meaning areprocessed similarly. However, although the results of such experiments do not clearlycontribute to disentangling the notions of literal meaning and non-literal meaning,at least they do not provide evidence that the traditional view of the operationalsequence of the interpretation process is accurate. On a more positive note, ignoringthe problematic aspects of such experiments for the moment, it nevertheless seemssignificant that all these studies point towards the same conclusion, namely thatliteral meaning and non-literal meaning are not so different in kind as traditionallyassumed.

In fact, Mashal et al. (2008)’s study provides evidence for differential processingof salient vs. non-salient meanings, rather than literal vs. non-literal meanings. Al-though this study, using fMRI, cannot make any claims concerning the time-courseof activation of salient vs. non-salient meanings during interpretation, it does showa difference in the brain areas involved in interpretation related to the salience ofthe interpreted meaning. Thus, Mashal et al. (2008) investigated the interpretationof highly familiar idioms as compared to their possible compositional interpreta-tion and the interpretation of matched literal sentences. Recall that according to

12Cp. Cacciari and Tabossi (1988).13Similarly, Dascal (1989) questions the appropriateness of assuming that reaction times reflect

what happens early on during interpretation.14Although Laurent et al. (2006) used ERPs in their testing of the graded salience hypothesis

concerning the activation of meaning for idioms, they only recorded EEGs during presentation ofthe last word of the respective idioms and the target words following that. Thus, their results donot reflect initial processes either.

the graded salience hypothesis the (idiomatic) meaning of highly familiar idioms ismore salient than the (‘literal’) meaning of their compositional counterparts. Theresults show that while for the interpretation of the idiomatic meaning of idiomsand the interpretation of literal sentences – both assumed to be salient – there is noinvolvement of the right hemisphere (RH), interpreting the literal and less-salientcompositional counterparts of highly salient idioms does involve the RH. Thus, inthis study, involvement of the RH does not correlate with literal/non-literal meaningof the sentence interpreted, but rather with the degree of salience of the meaninginterpreted.

2.2.2 Theoretical Considerations

There are also some theoretical considerations supporting the view that literal mean-ing and non-literal meaning should not be seen as concepts definable in terms ofdichotomies. In particular, literal meaning and non-literal meaning should not bedifferentiated on the grounds that the former is conventional meaning as opposedto the latter, which is non-conventional meaning. Generally, one has to ask, whatproperty exactly the concept of conventionality is supposed to pick out. That is, isan expression’s conventional meaning that which is coded in the lexicon, or is con-ventional meaning the reading with which a certain simple or complex expressionis frequently used, regardless of whether that meaning is pragmatically derived orcoded in the lexicon? In this context, the studies mentioned above, which showedthat there are cases in which a non-literal meaning cannot be supressed, are interest-ing. Thus, the fact that for highly familiar metaphors their metaphoric meaning isactivated even in contexts, where a literal interpretation is intended, lends credit tothe assumption that such metaphoric meaning is as conventional as literal meaning.In addition to such empirical evidence, there are tendencies in theoretical linguistics,in particular with respect to meaning change, to view conventionality as a gradualconcept instead of as a binary notion.15 This actually seems to be necessary, if onewants to explain how meaning change can take place at all, since it does not takeplace abruptly, but gradually. Thus, following Morgan (1978), Busse (1991) differ-entiates between conventions of language and conventions of language use, which hedefines from a functional-pragmatic perspective.16 Thus, conventions of languageare manifest in conventional language use, that is, a use of linguistic items that isgenerally and consistently conceived of as an intuitively and thoroughly ‘literal use’by the members of a language community (Busse 1991, p. 55). In contrast, conven-tions of language use are defined as a costumary and established procedure of use ofa particular linguistic expression for a consistent communicative purpose, where theuse of that expression presupposes a still recognisable literal meaning as a necessarymeans of arriving at that purpose (Ibid.). Busse (1991) argues that different degreesof conventionality can be identified on the basis of a threefold distinction of levels

15But see also Nunberg (1978, esp. ch. 4) for a characterisation of conventionality that is gradient.16Note that Busse also argues against an absolute differentiation between literal meaning and

non-literal meaning, which conceals the fact that in most concrete cases, the dividing line cannotbe drawn according to general, objective features. However, he also acknowledges that the differ-entiation is one that is made by normal speakers on an everyday basis. That is, it corresponds tosome intuition about language and thus should not be ignored by theory.

of communicative use of language. At the first level, language is used ‘literally’, inthat the members of the language community cannot detect any sort of ‘underlyingmeaning’ (i.e. a use of the conventions of language). The second level is the so-called‘indirect’ or ‘deviating’ use, where that use might also already be conventionalisedagain in some sense (e.g. idioms or routine formulas), however, where most speakersmay still recognise a ‘literal’ use of the expression in question. Thus, Busse acknowl-edges the fact that a particular ‘indirect’ meaning may also be conventionalised tosome extent by differentiating between conventions of language and conventions oflanguage use. On the third level of use, speakers use items from the first and secondlevel or both, however, this use constitutes a communicative sense that is only com-prehendable within a particular situation or context of utterance. Referring to thelevels identified above, one can differentiate degrees of conventionality, according towhether speakers can still differentiate literal or ‘underlying’ meanings from a givencommunicative sense.

conventions of language literal meaning

conventionsof language use

non-literal meaning

notc o

particularised CIs

idioms

unconventionallanguage use

Figure 2.7: Busse’s view of degrees of conventionality

Again following Morgan (1978), Busse (1991) recognises three elements of theconventions of language use: situation (here: context), purpose and means. Thus,the clearer the connection between communicative purpose and linguistic meansto the communication partners, that is, the more obvious it is to the participantsthat the literal meaning differs from the intended meaning, the further removedfrom a convention of language a particular use is. It then depends on the epistemicstatus of the purpose-means-relation whether what we are dealing with is a context-dependent pragmatic inference or a form of language use that is already consolidatedto a convention (of use). At the point where the relation between purpose and meansbecomes totally obscure, it may be reinterpreted as completely arbitrary, that is,at this point the convention of language use is reinterpreted as a convention oflanguage (Busse 1991, p. 57). Thus, Busse (1991) shows that Grice’s theory ofconversational implicature together with his theory of meaning can be utilised forthe explanation of how meaning change proceeds. A conversational implicature isthe result of a chain of inferences, which goes from the literal meaning, context,

purpose and further assumptions to the actually intended meaning. The shorterthis chain of inferences becomes, that is, the fewer necessary ‘links’ it has, the moreconventionalised the expression in question is. This may happen, e.g., if certaininferences are no longer necessary, because a particular use has already turned intoan established procedure of use (Busse 1991, p. 58). Busse (1991) argues thatmeaning change starts at the point where speakers are no longer able to reconstructall links within a chain of inferences. The becoming increasingly unnecessary of linkswithin an inference chain then corresponds to increasing degrees of conventionalityof the expression under consideration (cf. Konig 1988, Rolf 1995). However, as longas there is some transparency of purpose and linguistic means, we are still in therealm of what Busse calls convention of language use. Only at the point where therelation between purpose and means becomes absolutely ‘dark’, does the conventionunderlying the use of the expression in question become a convention of language.

Thus, on Busse’s view conventionality and non-conventionality still go along withthe differentiation of literal and non-literal meaning, however, he acknowledges thefact that there is a ‘grey area’, where meanings are to some extent conventionalised(as conventions of language use), but are not yet perceived of as literal meanings(thus, are not conventions of language). Thus, whereas Busse would still characteriseliteral meaning as fully conventional (being part of the conventions of language), non-literal meaning can be characterised as either being non-conventional or as beingconventionalised to a lesser or greater extent. These differences in conventionalitycan be captured by assuming that the process of conventionalisation is a gradualone, leading to different grades of conventionality at different stages of that process.

Looking, once again, at idioms shows that this characterisation still is problem-atic. As mentioned above, idioms are usually characterised as being coded in thelexicon as a whole, since their meaning is not (totally – cf. Nunberg et al. (1994))compositionally determined. Thus, idioms may have compositional counterpartswith meanings built up regularly from the meanings of their parts. Crucially, thereare cases of idioms, where (at least synchronically) there is no transparent con-nection between the compositionally determined meaning and the meaning of theidiom, thus, making idioms and their compositional counterparts rather similar tohomonyms (e.g. kick the bucket). As we saw, such idioms may still be characterisedas having non-literal meaning17 (albeit their meaning is considered conventional),although this seems a rather irritating use of the term non-literal meaning. How-ever, the reason for calling the meaning of an idiom non-literal may simply be thatspeakers are, in principle, able to recognise that these expressions have counterpartsthat have compositional and thus ‘regular’ meanings. In that case, the term non-literal is not only used for a meaning of some expression that is somehow ‘derived’or based on some ‘underlying’ meaning, but also for a meaning of what ‘looks like’ acomplex expression but is characterised by the fact that it does not have a (totally)compositional meaning. Thus, it is questionable whether normal speakers of Englishrecognise a connection between the idiomatic and the compositional meaning of kickthe bucket, in the sense that they take the idiomatic meaning to be derived from thecompositional. One could argue, therefore, that idioms like kick the bucket are to befound on Busse’s first level of language use, the level where no ‘underlying’ meaning

17Cf. e.g. Giora (1999, 2002), Giora et al. (1997)

can be identified.

conventions of language literal meaning

non-literal meaning

notc o

particularised CIs

idioms

Figure 2.8: Idioms in Busse’s framework

In this feature, idioms differ from metaphors in that with metaphors, the ‘under-lying’ meaning always has to be transparent or accessible for us to be able to call thephenomenon in question metaphor. Thus, at the point when the underlying meaningof a metaphor is no longer transparent, the metaphor turns into a dead metaphor,effectively seizing to be a metaphor at all. However, that does not mean that certaintypes of metaphor may not be conventional. It simply means that in order to classifysome phenomenon as a metaphor, recognising an ‘underlying’ meaning is necessary,whereas for idioms it is not.18 19 Thus, it seems that whether we categorise a mean-ing of an expression as non-literal or literal, does not only have to do with whetherwe are consciously aware of some connection between the meaning in question andsome underlying, basic meaning (as in metaphor), but also whether we recognisethat a certain expression form can be used with a different, compositional meaning(as with idioms). Thus, idioms are usually taken to be coded in the lexicon, sincetheir meanings have to be learned and since they only superficially look like complexexpressions with complex meanings associated with them. That is, in Busse’s terms,they are considered as being ‘conventions of language’, although according to Bussethey should not be, since their meaning is perceived to be a so-called non-literalmeaning – hence his classification of idioms belonging to the second level, the levelof conventions of language use. However, there does not seem to be any independentreason why at least some idioms should not be treated similarly to expressions thatare perceived as having no ‘underlying’ meaning. On the contrary, as mentionedabove, from a processing perspective, the interpretation of idioms may not be anymore ‘costly’ than the interpretation of ‘literal language’, where it seems that thelevel of conventionality of the idiomatic meaning plays a decisive role here. Thus,Giora recognises a type of idiom that she explicitly calls conventional idiom, where

18Cp. Stern (2006, p. 250), who makes a similar point.19This is, of course, not to say that there are no idioms which are clearly perceived as ‘derived’

in their meaning.

the crucial point is that in terms of activation their meanings are similar, if notprivileged, to the literal (i.e., compositional) meanings of their compositional coun-terparts. That is, the meanings of highly salient idioms will be activated as early oreven earlier than the meanings of their compositional counterparts, the latter beingthe case if the compositional meaning of some complex expression is actually non-conventional (in the sense of ‘not normally or often used’). Once again, if ‘earlinessof activation’ reflects the level of conventionality of the meaning concerned, thenthere is no reason why one should assume that the meaning of such conventionalidioms is less conventional than the meaning of their compositional counterparts.

conventions of language

highly & less-salientnon-literal meaning

non-salientnon-literal meaning

notc o

highly, less & non-salient literal meaning

Figure 2.9: Salient meanings according to Busse’s characterisation

Generally, with respect to different types of what traditionally is called non-literal meanings such as metaphor or idioms, the experimental results differ with thegrade of salience that these meanings have. However, according to Busse’s char-acterisation of levels of conventionality, Giora’s salient and less-salient non-literalmeanings belong to Busse’s conventions of language use, whereas the non-salientnon-literal meanings are the result of what Busse calls an unconventional languageuse. As regards highly, less and non-salient literal meanings, according to Busse’sclassification criteria, they are all part of the level of conventions of language. Incontrast, Giora does not actually make use of the traditional differentiation betweenliteral and non-literal meaning, precisely because the conventional/non-conventional‘divide’ seems to cut across that distinction. Thus, she assumes that regardless ofwhether some meaning would traditionally be classified as literal or non-literal, ifthat meaning is highly salient, it is reasonable to assume that this meaning is codedin the lexicon and thus, is part of the conventions of language. Moreover, Giora as-sumes not only of highly salient meanings, but also of less-salient meaning that theyare part of the lexical semantics of the relevant expressions. That is, an expression’ssalient and less-salient meanings form part of its lexical entry, whereas possible, butnon-salient readings are pragmatically derived.

Thus, Giora’s differentiation between salient, less- and non-salient meanings dif-fers from Busse’s assumptions in an interesting way. Busse assumes that only fully

conventionalised meaning is lexical meaning (or a convention of language), whereasless conventional or non-conventional meanings are not lexical. The former may beconsidered as a convention of language use, but not of language. Giora, on the otherhand, assumes that both highly and less-salient meanings are coded in the mentallexicon, and thus are part of the conventions of language, whereas non-salient mean-ings are not. In terms of the differentation of what is traditionally called literal andnon-literal meaning this means that whereas Busse only allows for genuine literal,basic, underived meanings to be entered in the lexicon, Giora allows both literal aswell as non-literal meanings in the lexicon, provided they are salient, where saliencecorrelates strongly with conventionality.20 Moreover, she does not only assume ofhighly salient meanings, but also of more and less-salient meanings that they areentered in the lexicon. What is problematic about this assumption is that one hasto ask, how the grade or level of conventionality of an expression’s meaning canactually be measured and thus allows one to identify and differentiate non-salient(and thus non-coded) from less or highly salient (and thus coded) meanings.21

conventions of language

notc o

highly andless-salientmeaning

non-salientmeaning

literal and non-literal meaning

Figure 2.10: Giora’s characterisation within Busse’s framework

Returning to the argument that literal meaning and non-literal meaning shouldnot be characterised in terms of dichotomies: the discussion in the section on lit-eral meaning above already suggested that one dichotomy between literal meaningand non-literal meaning cannot be sustained, namely that of context-independentvs. context-dependent meaning. Instead it was suggested that what traditionally iscalled literal meaning actually is not a complex expression’s context-independent,compositionally derived meaning, but rather a level of meaning that, similarly to

20Giora takes conventionality to be only one graded property among others that influence anexpression’s meaning’s salience, such as frequency of use. Nevertheless, conventionality seems toplay an important role. That is, other things being equal, the more conventional an expression’smeaning is, the more salient it will be. Moreover, from the kind of conventionalisation processthat Busse suggests, conventionality seems to be closely connected to the frequency of use of anexpression with a particular reading. Thus, it can be argued that it is increased frequency of usethat leads to a shortening of the inference chain ending in a conversational implicature.

21This point will be taken up in section 2.3.3.

Consequences for Lexical Meaning 35

non-literal meaning is context-dependent. From what has been said above concern-ing conventionality, it should have become clear that this property cannot be usedeither in the differentiation of literal from non-literal meaning, since, on the onehand, conventionality seems to be gradual and not binary, and on the other hand,there are instances of what is traditionally called non-literal meaning, which may becharacterised as conventional to some greater or lesser degree.

Thus, neither the dichotomy of conventional vs. non-conventional, nor that ofcontext-independent vs. context-dependent can be used to differentiate literal mean-ing from non-literal meaning. As concerns the characterisation of non-literal mean-ing as always being the result of a reinterpretation process, similar points can bemade. If reinterpretation processes are characterised as requiring the trigger of‘non-compatibility of literal meaning with context’ to get started, then the processby which non-literal meanings are generated should not be called a reinterpretationprocess. That is, if non-literal meanings only get generated if the literal meaning ofan expression does not fit the context, we would expect that non-literal meanings arenever activated in contexts where the literal meaning of an expression is intended.As we saw above, this is not the case. Whatever triggers the activation of non-literalmeaning, it does not necessarily have to be the non-compatibility of an expression’sliteral meaning with the context.

2.3 Consequences for Lexical Meaning

As mentioned above, formal semantics traditionally assumed that an expression’slexical meaning can be derived from the meaning of a sentence the expression ispart of via the principle of compositionality. Since sentence meaning was assumedto be context-independent, propositional and literal, lexical meaning is also taken tobe context-independent and literal.

However, in section 2.1 arguments were given against characterising sentencemeaning as both context-independent as well as fully propositional. Rather, sentencemeaning was characterised as context-independent, but not necessarily fully proposi-tional. Moreover, arguments were given for not assuming that sentence meaning assuch is literal, but rather that it is at some context-dependent and fully propositionallevel of utterance meaning that we might speak of a literal meaning, where it is notclear as yet how exactly this particular level of meaning is characterised. The crucialpoint here is that if sentence meaning is no longer considered to be literal but onestill wants to assume that lexical meaning can be derived from sentence meaningvia the principle of compositionality, lexical meaning can no longer be characterisedas being literal either. In fact, I will argue below that the lexical meaning(s) of anexpression should not be taken to be full readings at all.

First, however, I will present some of the problematic data the semantic compo-nent of a grammar may or may not be required to be able to handle (2.3.1). What isimportant to note here is that the stance one takes in this question has consequencesfor the amount of lexically given semantic information assumed. I will then go onand discuss two general types of approaches to lexical meaning which differ mainlyin how rich they assume the lexical semantic component to be (2.3.2). As we willsee, approaches that do not assume underspecification of lexical meaning have quite

a number of problems to face concerning the literal/non-literal meaning distinctionthey assume, the idea of grades of conventionality of meanings and the roles of theprinciple of compositionality and context in semantic composition. Finally, I want toshow that an approach that assumes underspecification for lexical meaning forms anappropriate basis for a view according to which both literal and non-literal meaningare context-dependent and more or less conventional.

2.3.1 Problematic Data

Traditionally, it is assumed that the lexical meaning of an expression captures thosereadings that are ‘basic’ in the sense that they do not depend on an actual utterancecontext. A potential problem with such an approach is that it leads to a high degreeof (non-idiosyncratic) polysemy in the lexicon. Consider example (13).

(13) a. The newspaper was on the table.

b. The newspaper was censored, yesterday.

c. The newspaper was founded three years ago.

It seems intuitively obvious that the expression newspaper is understood differentlyin the individual sentences. Thus, in (13a) it refers to a physical object, in (13b) to amental object and in (13c) to an institution. The difference in meaning is not due toa difference in the utterance context, which is completely ignored. Traditionally, onemight therefore assume that the different readings of newspaper are all coded in thatexpression’s lexical entry and that newspaper has to be considered as polysemous.However, this type of variation in meaning is not characteristic of newspaper alone.That is, it is not an idiosyncratic property of that expression. Compare example(14) (originally due to Bierwisch 1983).

(14) a. Die Schule spendete einen großeren Betrag.

‘The school donated a larger sum.’

b. Die Schule hat ein Flachdach.

‘The school has a flat roof.’

c. Die Schule macht ihm Spaß.

‘He enjoys school.’

d. Die Schule ist eine der Grundlagen der Zivilisation.

‘The school is one of the bases of civilisation.’

Thus, both newspaper and school (or Schule) have in common the ‘physical object’and ‘institution’ reading. The latter reading allows for a metonymical interpretationas ‘person associated with institution’ of the NPs in (15a) and (15b), respectively. Inthis respect newspaper differs from book which, although it does have the ‘physicalobject’ and ‘mental object’ readings, does not have an ‘institution’ reading andtherefore does not allow the metonymical interpretation ‘institution-for-person’.22

22This is not to say that there is no metonymical interpretation possible in principle. Thus, ina suitable context The book in (16d) could, e.g., be interpreted as ‘product-for-producer’.

(15) a. The newspaper called.

b. Die Schule hat angerufen.

‘The school called.’

(16) a. The book was on the table.

b. The book was censored, yesterday.

c. ??The book was founded three years ago.

d. ??The book called.

The different readings given for the individual nouns in (13), (14) and (16a-16b)are a good example of what is traditionally called systematic polysemy. The indi-vidual readings of the respective noun are clearly delimitable, yet at the same timerelate to one another. Also, all the readings seem to have the same status: they allintuitively seem to be basic, or literal, in the sense that no underlying meaning fromwhich they are derived is consciously detectable. Moreover, they are not characteris-tic for only one expression, but rather can be found across sets of expressions. Thus,with respect to the lexical entries of the nouns above, it is questionable, whether oneshould assume that all of the relevant readings are listed as possible literal mean-ings of those nouns (cf. Nunberg 1979). Note also that the variation in meaningexemplified above seems to arise from the intersentential context in which the re-spective expressions occur. Thus, the differences between the readings in (13a-13c),(14a-14d) and (16a-16b) are obvious simply from comparing the respective sentenceswith one another. It is arguable that sentential context, although being a specialtype of context in that it does not depend on a particular utterance situation, stillis a type of context and that, thus, the readings of the expressions in question is nottotally context-free, as it depends on the sentential contexts in which they occur. Infact, one actually needs to use fully sentential examples to make obvious that thereare differences in the understanding of the expressions concerned. Moreover, eventhough the relevant examples are presented out of a particular context, as pointedout above, it is reasonable to assume that hearers or readers always interpret linguis-tic expressions against a number of background assumptions, which may be takento have a default character.

In contrast to the related readings in case of systematic polysemy, strictly speak-ing, the metonymical readings for (15a) and (15b) above are not ‘basic’. That is,our interpretation of such examples relies on general knowledge we have regardinginstitutions, namely, that people (e.g., as employees) are associated with them. Onemight wonder, whether this reading is also part of the lexical meaning of the re-spective nouns, as it can be understood in the absence of a particular utterancecontext. However, as suggested before, the metonymical interpretation ‘person as-sociated with institution’ is generally available for all lexical items which have an‘institution’ reading, rather than being an idiosyncratic lexical feature of individuallexical items. Thus, the NPs in example (17) with such nouns as university, hospital,police get a similar metonymical interpretation.

(17) The university/hospital/police called.

In addition, more often than not context does play an important role in determiningthe exact interpretation of a metonymically used expression.23

(18) a. The ham sandwich in the corner wants some more coffee.

b. Plato is on the top shelf next to Russell.

c. John got a dent in his left fender.

Thus, in (18c), the proper noun John apparently has to be understood as referringto John’s car. In this case the interpretation of John as meaning ‘John’s car’ seems tobe constrained by the sentential context, i.e., its co-occurrence with the expressionsdent and fender. (18b) is a typical example for the metonymical interpretationof names of authors, which can then be used to refer to books written by thoseauthors. However, again the actual context in which the sentence gets utteredplays an important role. Thus, in a suitable context the terms could be referringto the busts of the respective persons just as well. Similarly, depending on theactual context of utterance, ham sandwich in (18a) can be understood as, e.g. ‘ham-sandwich-orderer’, or ‘person-who-prepared-the-ham-sandwiches’. More specifically,in examples like (18a), the sentential context already makes it clear that a shiftin the meaning of the noun ham sandwich from ‘object’ to ‘person’ is necessary(especially due to the verb want). However, the exact interpretation or specificationof the meaning (i.e., ‘person who prepared . . .’, ‘person who ordered . . ., ‘personwho delivered . . . , etc.) depends on the actual utterance context (and, arguably,the speaker’s intentions).

Similarly, to understand an utterance of the sentence in (19a), recourse has tobe taken to the actual context of utterance of that sentence. In addition, generalbackground knowledge is necessary to provide potential interpretations for an ut-terance of the sentence. Thus, depending on the actual context of utterance, (19b -19d) are possible readings for the sentence in (19a).

(19) a. Joyce is difficult to understand.

b. Joyce’s pronunciation is difficult to understand.

c. Joyce’s behaviour is difficult to understand.

d. Joyce’s books are difficult to understand.

That is, in order to correctly interpret an utterance of (19a) in a given situation, thehearer needs to have some general knowledge concerning properties characteristicfor the person named Joyce. Given that, it seems even less plausible that all thepossible readings for a proper name such as Joyce are listed in that expressions’lexical entry.

Note that this example (as did Searle’s cut-example) shows that verbs have avariation potential with respect to their possible meanings as well. In this particular

23The same, of course, holds true for particular cases of systematic polysemy. That is, forexamples such as in (i) and (ii), the sentential context alone does not make it clear which particularreading of newspaper and school, respectively, is intended.

(i) He is not interested in the newspaper.(ii) She paid particular attention to the school.

example, the meaning of understand seems to co-vary with the meaning assigned tothe proper name Joyce. That is, understanding someone’s pronunciation, behaviouror books differs conceptually. Another example for such a variability of verb meaningis given in (20).

(20) a. John opened the door.

b. John opened the exhibition.

c. John opened the file.

d. John opened the parcel.

e. John opened the store.

Again, the processes which are named by the verb to open in (20) are intuitivelyconceived of as being conceptually different. In contrast to the readings of theterm newspaper, which are clearly delimitable, this does not seem to be the case forthe readings of to open or to understand. Although there is some sort of similaritybetween the readings for each expression, the relation between them does not seem tobe of the kind that one would call systematic. Thus, Dolling and Zybatow (2007) callthis particular type of meaning variation of verbs non-systematic and characteriseit as, in a sense, idiosyncratic for the respective verb that exhibits it. Nevertheless,similarly to the readings of systematic polysemous nouns, intuitively at least all thereadings of open given above seem to be basic. Again the question arises, whetherone should assume that all these readings are therefore listed in the verb’s lexicalentry making open – and a large amount of verbs generally – polysemous. As wasthe case for systematic polysemous nouns, the fact that the relevant readings of openare all basic would be masked by assuming that it only has one primary reading fromwhich all other readings are derived.

Moreover, there is a further type of meaning variation with verbs that is moresimilar to the systematic polysemy exhibited by nouns. Thus, consider the examplesin (21).

b. The door opened.

In (21a) open has a causative reading, whereas in (21b) it has an inchoative reading.Thus, (21b) describes a change of state, wheras (21a) describes the bringing aboutof this change of state. This particular type of meaning variation is systematic (cf.Dolling and Zybatow 2007) because, on the one hand, the different readings arerelated to one another and, on the other hand, it can be found with a number ofverbs and, thus, is not idiosyncratic (cf. 22).

(22) a. John melted the ice.

b. The ice melted.

Again, the question is how this particular variation should be treated with respectto the lexical entries of the verbs concerned. On the one hand, one wants to capturethe fact that the different readings of the respective form are related to one another,on the other hand, one wants to capture that this particular variation in meaning isnot idiosyncratic for one verb.

Note that the possible non-systematic readings of a verb also differ dependingon the valency with which it occurs (cf. 23).

(23) a. Karl hat das Geschirr/ die Wasche/ den Mund/ . . . gespult.

‘Karl has washed the dishes/ the cloths/ his mouth . . . .’

b. Karl hat seiner Mutter das Geschirr/ die Wasche/ . . . gespult.

‘Karl has washed the dishes/ the cloths/ . . . for his mother.’

c. (Mit Hinweis auf Geschirr:) Karl hat gerade gespult.

‘(With reference to dishes:) Karl has just washed (the dishes).’

Thus, in (23a) the verb spulen takes two NP arguments, whereas in (23b) it takesan additional benefactive NP argument. Although, in (23c) spulen only takes oneovert argument, it is implicitly understood that there has to be something whichKarl washed, that is, the NP complement has to be present covertly in the semanticstructure of the sentence. In any case, the processes that spulen denotes in the exam-ples in (23) are conceived of as being conceptually different depending on whetherand if, what kind of complement NP occurs. Furthermore, the interpretation ofintransitive spulen seems fairly restricted to specific contexts.

The last set of phenomena I want to mention here are cases of so-called comple-ment coercion (e.g., 24b) and aspectual coercion (e.g., 26b).24 Thus, verbs like tobegin and to finish may occur in two different syntactic constructions: either withan event-denoting VP-complement as in (24a) and (25a) or with an entity-denotingobject NP as in (24b) and (25b).

(24) a. Max began reading a novel.

b. Max began a novel.

(25) a. Nigel finished eating a hamburger.

b. Nigel finished a hamburger.

Note that in the latter case the respective sentences are nevertheless understood aspredicating of the respective subjects that they began/finished some event involvingthe entity denoted by the object NP. However, this meaning aspect is not overtlyexpressed by any of the constituents in the respective sentences. Thus, it seemsthat the two verbs’ selectional restrictions actually are such that they require event-denoting complements. Since the sentences in (24b) and (25b) are grammatical, theusual assumption is that during the semantic composition of the meanings of thesesentences a process intervenes that turns the entity-denoting object into an eventdescription. Hence this process is called complement coercion.

While in complement coercion the ontological sort of the complement is shifted(from entity to event), in aspectual coercion what is changed are the aspectual prop-erties associated with a particular verb in order for it to fit with a given modification.Thus, consider the examples in (26).

24Note that while coercion phenomena, on a first blush, may not seem to be relevant to thequestion of how much semantic information is lexically given, considering how the phenomena areexplained in different approaches, it turns out that they are relevant after all.

(26) a. The light flashed.

b. The light flashed until dawn.

Taken in isolation, the verb to flash is understood as denoting a punctual event. Assuch, a sentence including it as well as a durative modifier such as until dawn shouldbe ungrammatical. However, as (26b) shows, such sentences are grammatical, whenunderstood as involving a repetition of the event denoted by the verb.

A similar contrast can be seen in example (27). In (27a), the VP play a sonatadenotes an event that already has an intrinsic boundary. If such a bounded eventis modified by an adjunct that has the function of introducing a boundary for theevent expressed, a conflict should arise. However, example (27b) is grammaticalbut again only under the interpretation that the subject was involved in the eventdenoted by the VP repeatedly.

(27) a. Amelie played a sonata.

b. Amelie played a sonata for three weeks.

Thus, it seems that complement coercion and aspectual coercion are variants ofone and the same general phenomenon. In both cases a process intervenes duringthe semantic composition of the relevant sentences and adds some meaning aspect.Put differently, the expectation is that the semantic composition of such sentencesshould crash as it apparently involves semantic mismatches. However, what seemsto happen is that such impending (semantic) mismatches are circumvented by theapplication of special operators that mediate between the potentially mismatchingelements and at the same time introduce additional meaning aspects.

All in all, although this is only a very small set of data presented here, it sufficesto indicate the kinds of problems one faces when developing a theory of lexicalmeaning and the role it plays in the overall interpretation process. In the nextsection, we will start to look at possible approaches to meaning in the lexicon andsee how the data given above is treated, as well as how these approaches reflect therevised notions of literal meaning and non-literal meaning as proposed so far.

2.3.2 Approaches to Meaning in the Lexicon

Dolling and Zybatow (2007) identify three broad approaches to meaning in thelexicon: maximalist, intermediate and minimalist approaches. Generally, these ap-proaches have in common the assumption that a lexical entry for an expression atleast contain some phonetic (PHON), syntactic (SYN) and semantic (SEM) infor-mation about that item.25 However, the approaches differ in how much ‘semanticmaterial’ they assume to be given in the lexical entry of an expression, how this ma-terial is structured, how it is made use of in the actual interpretation of an utterancecontaining the relevant expression and how it interacts with the syntactic informa-tion given in the entry. Specifically with respect to the lexical entries of verbs,there is the question, whether information concerning argument structure and eventstructure are part of the lexical information given for a linguistic expression.

25Since here we are mainly interested in the interfaces between syntax, semantics and pragmatics,in the representation of lexical information the phonetic information will sometimes be neglected.

In this section, I will be concerned with the maximalist and intermediate ap-proaches, leaving the discussion of minimalist approaches to the next section.

The Maximalist Approach

Maximalist approaches to meaning in the lexicon can be characterised by the fol-lowing features. All possible literal meaning variants of an expression are listed inthe lexicon. In addition to differences in the valency of a verb, the sortal restric-tions for the individual meaning variants are stated as well. Thus, differences invalency and/or sortal restrictions result in separate lexical entries. Furthermore,the semantic relations holding between the individual meaning variants are stated.In the lexical entries of verbs, information about their argument structure is given aswell as the respective verb’s particular event structure. An example of a maximalistapproach is Engelberg (2000). For example, he gives the lexical entry, or ratherentries, for the verb spulen (see figure 2.11). As one can see from the entry, notonly differences in the valency of the verb, but also different sortal restrictions of itsarguments are taken into account. As a result, Engelberg can account not only forthe different readings of spulen in (28a-28c), but also for those in (28d) and (28e).

spulen1 syn: V, /npakk/npnomsem: λyλxλe. spul1(x, y, e)bpspulen1: � ∀xye[spul1(x, y, e) → geschirr(y) ∨ wasche(y) ∨ . . . ]

spulen2 syn: V, /npakk/npdat/npnomsem: λyλzλxλe. spul2(x, y, z, e)bpspulen2: � ∀xyze[spul2(x, y, z, e) → belebt(x) ∨ belebt(z)]bpspulen3: � ∀xyze[spul2(x, y, z, e) → spul1(x, y, e)]

spulen3 syn: V, /npnomsem: λxλe. spul3(x, y, e)bpspulen4: � ∀xye[spul3(x, y, e) → geschirr(y)]bpspulen5: � ∀xye[spul3(x, y, e) → spul1(x, y, e)]

spulen4 syn: V, /npnomsem: λxλe. spul4(x, y, e)bpspulen6: � ∀xye[spul4(x, y, e) → mensch(x)]bpspulen7: � ∀xye[spul4(x, y, e) → mund(y)]bpspulen8: � ∀xye[spul4(x, y, e) → spul1(x, y, e)]

spulen5 syn: V, /npnomsem: λxλe. spul5(x, y, e)bpspulen9: � ∀xye[spul5(x, y, e) → waschmaschine(x)]bpspulen10: � ∀xye[spul5(x, y, e) → wasche(y)]bpspulen11: � ∀xye[spul5(x, y, e) → spul1(x, y, e)]

Figure 2.11: Lexical entry for spulen according to Engelberg (2000)

(28) a. Karl hat das Geschirr/ die Wasche/ den Mund/ . . . gespult.

‘Karl has washed the dishes/ the cloths/ his mouth . . . .’

b. Karl hat seiner Mutter das Geschirr/ die Wasche/ . . . gespult.

‘Karl has washed the dishes/ the cloths . . . for his mother.’

c. (Mit Hinweis auf Geschirr:) Karl hat gerade gespult.

‘(With reference to dishes:) Karl has just cleaned (the dishes).’

d. (Beim Zahnarzt:) Karl hat gerade gespult4.

‘(At the dentists:) Karl has just rinsed (his mouth).’

e. Die Waschmaschine hat gerade gespult5.

‘The washing machine has just rinsed (the cloths).’

Note that for Engelberg, the sentences in (28c - 28d) are ambiguous. That is,only after having considered the context in which an utterance of the sentence Karlhat gerade gespult takes place, can the hearer disambiguate which literal readingof spulen was intended. However, if the context has to be taken into account inorder to be able to decide which reading of spulen was intended, the principle ofcompositionality cannot be maintained. A possible counter-argument is to say thatsemantic composition actually generates all possible semantic forms for the sentenceand only afterwards, during the process of disambiguation, the context is consideredand those semantic forms that do not fit it are discarded. In this case the principleof compositionality would be adhered to. However, considerations of economy of thelexicon militate against a maximalist organisation. This is especially so consideringthat this kind of ambiguity is characteristic for the majority of lexical items ina language (cf. the newspaper, open and cut examples). Furthermore, Engelbergapparently has not reached his aim of listing all possible literal readings for spulen.Thus, Dolling (2005) argues that a reading for spulen as in (29) is yet another onethat would have to be listed in the lexicon. Note also that this reading cannot besubsumed under Engelberg’s spulen1. Probably even more ’literal’ readings could belisted, if one were to do a corpus analysis to find attested examples (cf. Ruhl 1989,on data won through intuition).

(29) (Mit Hinweis auf die Toilette:) Karl hat gerade gespult.

For maximalist approaches, the process of semantic composition is essentially asassumed by traditional formal semantics. That is, the process follows the princi-ple of compositionality and therefore, does not take into account any contextualinformation. Thus, for an utterance such as Karl hat gerade gespult, three semanticforms are generated by the process of semantic composition, as there are three lexi-cal entries for spulen in its intransitive form. Thus, although the lexicon is very richin what is usually considered world knowledge, this never really comes into play,since the disambiguation process that follows semantic composition simply checks,whether the information given in the individual semantic forms fits the informationprovided by the contextual conditions. In a sense, then, the information stated inthe individual lexical entries for intransitive spulen is redundant.

Moreover, if the lexical entry for an expression only lists its literal meaning(s),one would have to assume a Gricean interpretation process, where necessarily a

literal reading for an utterance has to be derived first via semantic composition,which, in case of a non-fitting context, has to be reinterpreted. Assuming that allliteral readings are listed in the lexicon for each lexical item would make the dis-ambiguation process a very important part of the overall interpretation process andarguably, a rather extended one in terms of the time needed for that process tofinish. If literal meanings are context-independent and coded in the lexicon, theprocess of semantic composition will determine as many possible sentence meaningsas there are possible combinations of the literal meanings of the expressions con-cerned. Only in a second step, when contextual information is considered and thesentence meanings are enriched to the point that they are actually propositional,does the disambiguation process filter out all those propositions that do not fit thegiven context. Even assuming that all of this takes place during the short period oftime that addressees usually only require to understand some utterance, one wouldstill assume that of the time span needed for interpretation, a relative large amountis spend disambiguating potential sentence meanings. However, the fact that poten-tial non-literal meanings of some expression may be activated very early on duringinterpretation goes against such a view.

Note also that an interesting question here is, which semantic form the re-interpretation process would be based on if during the first pragmatic interpretation,and due to a non-fitting context, none of the potential semantic forms derived duringsemantic composition could be chosen as fitting. Thus take once more the examples(15a) and (18), repeated below.

(15a) The newspaper called.

Approaches that assume that semantic composition combines the literal mean-ings of simple expressions and thus produces the resulting complex expression’s lit-eral meaning would be hard pressed to give the literal meaning of (18a) from whichthe metonymical reading is taken to be derived. This is especially so for approachesthat assume that the sortal restrictions for lexical items form part of their lexicalentry. In that case, arguably, the verb want would be specified as requiring an ar-gument that has volition, which, literally, is not the case here. A possible way outwould be to assume that ham sandwich is polysemous, one of its possible readingsbeing ‘ham sandwich orderer’, and that this reading is listed in the lexicon. Thequestion is whether one would like to assume such an entry and thus to make theprocess of metonymic shift a lexical one. As noted above, in view of the context-dependency of the results of an application of the process of metonymic shift, this isunlikely. Thus, as mentioned before, Plato in (18b) might be interpreted as ‘the per-son Plato’, ‘the book(s) written by Plato’ or possibly ’the bust of Plato’, dependingon the particular context of an utterance of (18b).

Regardless of how the question of which semantic form the re-interpretation pro-cess would be based on might be answered, the fact is that a maximalist approachthat only allows literal readings of an expression to be listed in that expression’s

lexical entry cannot account for cases of similar speed in, or even faster, interpreta-tion of certain, conventionalised non-literal meanings. Thus, such a characterisationof meaning in the lexicon seems to assume the traditional view of a dichotomy ofconventionalised vs. non-conventionalised meaning, for which, as we saw, there arearguments that it is not necessarily warranted. That is, a maximalist lexical entrydoes not reflect the fact that there are levels of conventionality rather than just adifferentiation between meanings that are conventional and such that are not.

The Intermediate Approach

Intermediate approaches employ one of two general strategies. Either one specificmeaning variant of an expression serves as the basis from which all other readings arededuced, or an abstract semantic representation provides the basis for all primarymeaning variants of the expression. One example for an intermediate approach,using the first strategy is Pustejovsky (1991, 1995). With respect to the lexicalentry for verbs, he states one specific meaning as their basic meaning, from whichall others have to be deduced. Lexical entries are structured and involve a numberof levels of representation: argument structure, event structure, qualia structure andlexical inheritance structure. For example, the lexical entry for the verb to bakelooks as in figure 2.12.

eventstr =

[e1 = e1: process

Head = e1

argstr =

arg1 = 1

[animate ind

formal = physobj

arg2 = 2

formal = physobj

qualia =

[state change lcp

agentive = bake act(e1, 1 , 2 )

Figure 2.12: Lexical entry for to bake according to Pustejovsky (1995)

Thus, not only the syntactic order of lexical items, but also the semantic contentof their entries may be decisive for how lexical items are combined. That is, thelexicon is actively involved in the semantic composition of sentences. For example,with the help of the information given in the qualia structure of bake, togetherwith information from the lexical entry of the respective complement bake takes,problematic cases as in (30) can be explained.

(30) a. Mary baked a cake.

b. Mary baked a potato.

Whereas in (30a) the verb to bake denotes the act of creating something, in (30b)it denotes a change of state. As can be seen from the lexical entry of bake, its basicmeaning is that of denoting a change of state, thus, the reading that needs to beexplained is that in (30a). Information about the internal argument, in this casethe noun cake, is used to shift the verb’s event type. Intuitively, an object such asa cake is brought about by the activity it is in composition with. This is expressedby the information stated in the qualia structure. Thus, the agentive quale of thenoun cake makes reference to the process of baking.

argstr =

[arg1 = x:food ind

d-arg1 = y:mass

qualia =

const = y

formal = x

telic = eat(e2, z, x)

agentive = bake act(e1, w, y)

Figure 2.13: Lexical entry for cake according to Pustejovsky (1995)

A number of operations take place in combining the lexical structures of bake and(a) cake. Amongst them is the process of co-composition, which results in a semanticstructure for bake a cake that incorporates aspects of both the constituents. Thus,in combining the lexical information of bake with that of cake, the resulting VP isinterpreted like a genuine creation verb such as build. The resulting semantic formof the VP bake a cake looks like in figure 2.14.

Thus, systematic (or as Pustejovsky calls it: logical) polysemy is explained onthe level of expression meaning with the help of lexical semantic processes thatcompose larger semantic structures out of the lexical semantic information of theindividual expressions. However, it is not clear what leads Pustejovsky to assumethat the ‘change of state’ reading is the underlying, basic meaning of bake, especiallysince in the examples in (30), neither the one nor the other seem to have a derivedmeaning.

Generally, an advantage of Pustejovsky (1991, 1995)’s approach is the fact that itprovides a framework to model the role of sentential context in interpretation. Thus,the process of co-composition might be used to explain the phenomenon that thereading of a particular expression may be determined before the context of utteranceis actually considered. That is because the lexical entries of linguistic expressions arerelatively rich, consisting of different types of information, where that informationinteracts with the information coded in the lexical entries of expressions which occurin the same sentence or phrase. Note also, that Pustejovsky (1991, 1995)’s approachallows for a relative straightforward modelling of complement coercion in terms ofthe process of co-composition. A sentence like (24b) – repeated below – for instance,is interpreted by using information stored in the qualia structure of the lexical entryfor novel.

bake a cake

evenstr =

e1 = e1:process

e2 = e2:state

Restr = <∝Head = e1

argstr =

arg1 = 1

[animate ind

formal = physobj

arg2 = 2

artifact

const = 3

formal = physobj

d-arg1 = 3

[material

formal = mass

qualia =

create-lcp

formal = exist(e2, 3 )

agentive = bake act(e1, 1 , 3 )

Figure 2.14: Semantic representation for the VP bake a cake according to Puste-jovsky (1995)

(24) a. Max began reading a novel.

b. Max began a novel.

In case of an interpretation of (24b) as equivalent in meaning to (24a), novel’s telicrole – namely that its purpose is to be read – is made use of. This information isintegrated into the composition of the semantic form of the respective sentence bymeans of a special operator (QT ) which intervenes between the verb to begin andthe NP a novel. Pustejovsky (1991, p. 430) gives the following derivation of themeaning of (24b).26

(31) a. Max began a novel.

b. begin′(QT (a novel)

)(Max)

c. begin′(λx, eT [read(a novel)(x)(eT )]

)(Max)

d. Max{λx[begin′(λx, eT [read(a novel)(x)(eT )](x∗)

)(x∗)]}

e. Max{λx[begin′(λeT [read(a novel)(x∗)(eT )]

)(eT )]}

f. begin′(λeT [read(a novel)(Max)(eT )]

)(Max)

Note, however, that several problematic issues can be identified for Pustejovsky(1991, 1995)’s approach. One such aspect is that the content and formality of thequalia structures are only loosely defined. There seem to be no convincing reasonsfor choosing a particular element over another as being part of the qualia. In fact,

26Actually, his example sentence is John began a novel.

Egg (2005) questions whether it is generally appropriate to include such parts ofbackground knowledge in the lexicon as the coming about of some entity a lexicalitem can be used to refer to. Another problematic aspect is that if systematicreinterpretation becomes necessary during semantic composition, as we saw above,Pustejovsky argues for the application of special transfer functions (such as e.g.,QT ), which are based on conceptual knowledge. It is not clear how the necessaryaccess to conceptual knowledge is controlled. Moreover, if such transfer functionsapply during semantic composition, the principle of compositionality is not adheredto (cf. Dolling 2001). Furthermore, transfer functions making use of material inthe qualia structure of lexical items would have to be overruled when it comes toreinterpretation guided by contextual information. That is, the information that canbe retrieved from the qualia structure can only have the status of default knowledge;otherwise the transfer functions would fail to generate the appropriate number ofpossible reinterpretations for an utterance.

For instance, recall that the information retrieved from the lexical entry of noveland integrated into the semantic form of a sentence such as (24b) in Pustejovsky(1991, 1995)’s approach is taken to be semantic in nature. However, depending onthe context in which (24b) is uttered, the process the speaker might implicitly refer tomay not necessarily be a reading or writing process.27 Generally, the fact that a richcontext may facilitate the interpretation of an expression with a less-conventional,but intended reading without any major noticable effects attributable to extensivere-interpretation processes suggests that the type of information used in cases of‘out-of-context’ interpretation should be treated as having default character insteadof being semantic in nature. Thus, this type of knowledge is applied only if there is nomore specific (contextually provided) information that can be applied. Therefore,it should not be assumed to be semantic, if one wants to assume that semanticinformation always enters into the composition of the meaning of an utterance,regardless of context of or intentions behind some utterance. Thus, the type ofinformation proposed by Pustejovsky in the form of qualia structure may actuallyplay an interesting role when it comes to interpreting some linguistic utterance,however, this structured information is not available at the level of lexical entriesbut rather at the level of encyclopaedic knowledge. Moreover, one could argue thatat this latter level, certain types of information are easier to access or privilegedin the sense that they are the concepts certain expressions are taken to express indefault situations. Also, it is at this level of information where it seems to makemore sense to assume that we store facts such as for example how certain artefactscome about.28

Egg (2003, 2005) notes another problem for Pustejovsky (1991)’s particular for-mulation of the coercion operator. As can be seen in the analysis of (24b) in (31),

27Egg (2005) mentions a context in which it is known that Max is a professor of literature, where(24b) might reasonably be understood as referring to Max’s beginning of a lecturing process onsome novel. Crucially, this interpretation intuitively does not seem to require the revision of apreviously computed reading or writing interpretation.

28Incidentally, similar arguments can be made concerning such information as argument andevent structure. That is, the question is whether the knowledge we seem to have about the numberand type of obligatory or optional participants involved in some event needs to be reflected in ourlexical knowledge.

the verb to begin’s semantics ends up outscoping that of the quantifier in the object-NP. That is, since the coercion operator intervenes between the overt functor andargument, it fixes the scope of the quantifiers involved. However, this does not seemto be what is actually happening. Thus, consider (32).

(32) a. Max began reading every novel.

b. Max began every novel.

Both (32a) and (32b) may be understood to describe a situation in which John readthe initial part of each of a contextually specified range of novels. However, (32a)but not (32b) has another possible reading in which John is described as involvedin the initial phase of an eventuality that is a reading of the contextually specifiedrange of novels. This latter reading is compatible with a situation where John infact is engaged in reading the initial part of just one of the contextually specifiedrange of novels. Thus, whereas in (32a) either of to begin and every may take widescope with respect to one another, in (32b) every always has wider scope. However,this state of affairs is exactly opposite to what is predicted by Pustejovsky (1991)’sanalysis.29

Leaving these issues to the side for the moment, I will turn to an example ofan intermediate approach which uses the strategy of providing an abstract seman-tic representation as the basis for all primary meaning variants of an expression:Bierwisch’s Two-Level-Semantics. As already noted, Bierwisch assumes that an ex-pression’s literal or primary reading is only determined at the level of utterancemeaning. In that sense, an expression’s semantic form has to be quite abstractand underspecified. In order to build up the linguistically determined meaning ofa complex expression, the process of semantic composition combines the individualsemantic forms, following the principle of compositionality. The main feature of thisapproach is the fact that the semantic information given in the lexical entry of anitem is underspecified and only gets specified during the actual interpretation pro-cess of an utterance including that expression. Thus, on the one hand, the semanticform of an expression specifies the contribution of that expression to the semanticrepresentation of utterances of which it is part. On the other hand, the semanticform of an utterance constitutes the interface with the conceptual-intentional sys-tem. That is, an interpretation function maps the linguistically determined semanticform of an utterance onto the utterance meaning, depending on the context in whichit is used. Thus, only when an utterance gets actually interpreted, will its particularreading get fixed.

For instance, for the noun Schule, which, similarly to the noun newspaper, canoccur with different readings, Bierwisch (1983) gives the semantic form stated in(33), which is compatible with any of the readings Schule can be given in (14).

(33) SEM: λx [ zweck x w ]

w = lern- und lehrprozesse

29Note that in Egg (2005)’s approach this problem does not arise since the process of semanticcomposition as such is characterised by the fact that it produces a semantic structure that isunderspecified with respect to scope. For details of that approach, see below.

(14) a. Die Schule spendete einen großeren Betrag.

‘The school donated a larger sum.’

b. Die Schule hat ein Flachdach.

‘The school has a flat roof.’

c. Die Schule macht ihm Spaß.

‘He enjoys school.’

d. Die Schule ist eine der Grundlagen der Zivilisation.

‘The school is one of the bases of civilisation.’

The actual ‘institution’ (14a), ‘physical object’ (14b), ‘assembly of processes’ (14c)or ‘institution as principle’ (14d) readings only get determined, or specified, duringthe interpretation of the respective utterances. Generally, no primary reading isgiven in the semantic form of an expression, however, the semantic form does deter-mine as a possible interpretation a family of concepts for the lexical item it is partof.30 This differentiation reflects Bierwisch’s view of the lexicon as the repository oflinguistically relevant information. Because they are not necessary for any linguisticprocess to function properly, an expression’s possible literal meanings need not all belisted in its lexical entry. Thus, for the noun Schule, the different meaning variantsconstitute a family of concepts, or more specifically, a set of conceptual schemata,in terms of which its semantic form can be interpreted, depending on the particularcontext. The interpretation function then chooses a member of that family as the ac-tual interpretation in a given situation, depending on the particular context. Thus,in Bierwisch’s approach systematic polysemy no longer is treated as a phenomenonat the level of expression meaning, rather, it belongs to the context-dependent levelof utterance meaning.

However, although the semantic form of an expression is abstract and under-specified, similarly to Engelberg’s approach, information on the argument structureis part of a verb’s lexical entry. Thus, the lexical entry for a verb such as to enterlooks as in (34) (adapted from Bierwisch and Schreuder 1992, 27).

(34) PHON: /enter/

SYN: [+V, -N]

SEM: λxλy [y DO [MOVE y] : FIN [y LOC IN x]]

As in Engelberg’s approach, argument structure and semantic form of a particularlexical item are closely interrelated pieces of information. However, in contrastto Engelberg, differences in valency do not automatically lead to separate lexicalentries. This can be seen from the underspecified lexical entry for open, which isintended to reflect that lexical item’s variable valency. That is, rather than assumingseveral lexical entries related by meaning postulates, Bierwisch tries to capture therelation between the different readings of open, by integrating the possibility ofdifferent valencies for a lexical item into its syntactic information, its argument

30However, it is not clear what it means for the SEM of some expression to determine a familyof concepts as possible readings for the expression at hand.

structure as well as its semantic form. Thus, he even considers the possibility ofincluding the adjective open in one and the same entry with the intransitive andtransitive verb forms, in order to keep the lexically given information as minimal aspossible (cf. 35) (adapted from Bierwisch 1997, 244).

(35) PHON: /open/

SYN: [+V, -βN]

SEM: λx (β(α λy) λe [ e INST [ (α y DO-CAUSE) [ BECOME) [ OPEN x ]]]]

2.3.3 Semantic Underspecification in the Lexicon

The Minimalist Approach

Although in Bierwisch’s approach already the semantic form of lexical items wasno longer assumed to be maximally specified, an approach is possible which min-imises the lexical information even more. Thus, minimalist approaches to lexicalmeaning are characterised by the following facts. The lexical meaning of an expres-sion is maximally underspecified. Furthermore, the semantic form of an expressionfunctions as a pointer31 to possible meaning variants of that expression. Thus, sim-ilarly to Bierwisch’s approach, the intended literal (or non-literal) reading is derivedonly during the interpretation process. Moreover, information on possible argumentstructures an expression might occur in are not part of its semantic information. Anexample for a minimalist approach is Dolling (1997, 2005), who gives the followinglexical entry for the noun newspaper.

(36) PHON: /newspaper/

SYN: N

SEM: λx. x Kn newspaper

Kn is an SF32-Parameter which is part of the lexical meaning of all common nouns.The set of possible values for this parameter includes the predicates � (‘instance of’)and ≤ (‘sub-kind of’). Thus, the semantic form in (36) can be specified either to apredicate which denotes the set of instances of newspapers or one which denotes theset of sub-kinds of the kind newspaper. Furthermore, the conceptual entity ‘newspa-per’, which forms part of the lexical entry for newspaper, is an SF-Parameter as well.It identifies a certain kind, namely, the kind that is realised by each newspaper-as-mental-object, newspaper-as-material-thing and newspaper-as-institution. The ax-ioms in (37) make it clear that ‘newspaper’ is a specific combination of newspaper1,newspaper2 and newspaper3, which are sub-kinds of the categories of mental objects,material things and institutions, respectively.

(37) a. newspaper = [newspaper1` newspaper2]

` newspaper3

b. newspaper1 ≤ ment.object

c. newspaper2 ≤ mat.thing

31This term was first introduced in Relevance Theory, more on which see below.32SF is the abbreviation Dolling uses, following Bierwisch, for semantic form, here generally

abbreviated to SEM.

d. newspaper3 ≤ institution

e. ∀x [x ≤ ment.object & x ≤ newspaper → x ≤ newspaper1]

f. ∀x [x ≤ mat.thing & x ≤ newspaper → x ≤ newspaper2]

g. ∀x [x ≤ institution & x ≤ newspaper → x ≤ newspaper3]

The determination of a specific reading takes place only during the interpretation ofan utterance that includes the noun newspaper. Thus, as is the case with Bierwisch’sapproach, systematic polysemy is not a phenomenon of the level of expression mean-ing, as the intended reading for the noun is only determined during interpretation.

The method of maximal underspecification is also used for verbs. Thus, Dolling(2005) states the following lexical entry for the German verb offnen (to open).

(38) SYN: V, [α/NPakk] [β/NPdat] /NPnom

SEM: λe.OFFN(e)

PRAG: If α = - and β = -, then <OFFN> = {OFFNk1, OFFNk2, . . . }If α = + and β = -, then <OFFN> = {OFFNl1, OFFNl2, . . . }If α = + and β = +, then <OFFN> = {OFFNm1, OFFNm2, . . . }

In this entry, only the syntactic information SYN expresses the possible variationin valency for the verb offnen. That is, depending on which value replaces α or β,one gets an inchoative or causative reading, or the causative reading with an addi-tional benefactive complement. In the semantic form SEM, however, there is onlyone variable for the event argument. Thus, argument structure and semantic formare separate parts of information. SEM includes the SF-Parameter OFFN, whichfunctions as a pointer to the set of possible primary meanings of the verb offnen.The fact that specific valencies in which a verb can occur determine its interpreta-tion is reflected in the information given in the PRAG component. However, thisinformation is only accessed during the interpretation process and ensures that therelevant reading is picked, depending on the syntactic environment in which theitem occurs, from the appropriate set of possible primary meanings. Thus, a lexi-cal entry (and as a result the lexicon as a whole) actually consists of two parts: agrammatical one and a pragmatic one. The grammatical part is accessed during thesemantic composition of the meaning of a complex expression, which, as in Bier-wisch’s and Engelberg’s approaches, follows the principle of compositionality. Thepragmatic part only comes into play once the result of semantic composition, thesub-propositional semantic form of an utterance is further processed by the prag-matic component which yields the actual propositional meaning of the utterance.Thus, the lexicon is not only a repository of information required by the individuallinguistic subsystems, but it also constitutes an interface to the pragmatic system,providing information relevant for the correct interpretation of an utterance.

Note that in Dolling (1997, 2005)’s approach, during the semantic compositionof an utterance’s SF, operators may apply that are not the meaning of some syn-tactically provided functional head. However, such operators only provide variablesthat might be filled with values by pragmatic processes after the actual semanticcomposition process is finished. For example, using the operator in (39), Dolling(2005) proposes a treatment of coercion, where that operator is introduced during

the process of semantic composition, providing parameters which have to get fixedor specified in a later step after semantic composition has been completed and re-course may be taken on conceptual knowledge. In fact, the same operator applies toall one-place predicates during semantic composition, regardless of whether there isan impending semantic mismatch. In other words, the operator applies by default.33

(39) metn: λPλx. Qny [Sn(y, x) Cn P(y)]34

Using the example in (40), Dolling (2005) shows how different potential readings ofthat sentence – given in (41) – might be derived from one underlying semantic form(43l). The respective lexically given semantic forms for the expressions/functionalheads used in (40) are given in (42).35

(40) Anna hat die Tur funf Minuten lang geoffnet.

(41) a. ∃p[AG(anna, p) & TH(def TUR, p) & ∀e[CONST(e, p) → OFFN1(e)] &τ(p) ≥ 5min]

b. ∃p[AG(anna, p) & TH(def TUR, p) & ∃e[COMPL(e, p) & OFFN1(e)] &τ(p) ≥ 5min]

c. ∃e[AG(anna, e) & TH(def TUR, e) & OFFN1(e) & ∃s [RESULT(s, e) &τ(s) ≥ 5min]]

(42) a. offnen: λx. OFFN(x)

b. funf Minuten lang: λx. τ(x) ≥ 5min

c. MOD: λQλPλx. P(x) & Q(x)36

d. θakk: λPλyλx. θakk(y, x) & P(x)

e. die Tur: def TUR

f. θnom: λPλyλx. θnom(y, x) $ P(x)

g. Anna: anna

(43) a. λPλx. Qny [Sn(y, x) Cn P(y)] (λx. OFFN(x))

b. λx. Q1y [S1(y, x) C1 OFFN(y)]

c. λPλx. Qny [Sn(y, x) Cn P(y)] (λx. τ(x) ≥ 5min)

d. λx. Q2z [S2(z, x) C2 τ(z) ≥ 5min]

e. λQλPλx. P(x) & Q(x) (λx. Q2z [S2(z, x) C2 τ(z) ≥ 5min]) (λx. Q1y [S1(y,x) C1 OFFN(y)])

f. λx. Q1y [S1(y, x) C1 OFFN(y)] & Q2z[S2(z, x) C2τ(z) ≥ 5min]

33Note that Dolling (2000) uses the same operator (in a slightly different form) to composemetonymic or metaphoric readings of nouns (see section 4.1.1).

34Qn and Cn as well as Sn are SF-Parameter which need to get fixed during interpretation, Qn

by either ∀ or ∃, Cn by → or =, respectively and Sn by a predicate of relations between elementsof two sorts of entity.

35Note that due to the fact that in Dolling’s approach generally verbs are highly underspecifiedregarding their semantics, additional operators are necessary (i.e., θnom, θakk) that introduce thearguments with which the respective verb occurs in a particular sentence.

36This is the operator used to integrate the semantic contribution of potential adjuncts into thesemantic form of the complex expression under consideration. Note that this operator is solelysemantic, i.e., it is not the meaning of some overt syntactic element.

g. λPλyλx. θakk(y, x) & P(x) (λx. Q1y [S1(y, x) C1 OFFN(y)] & Q2z[S2(z,x) C2τ(z) ≥ 5min])

h. λyλx. θakk(y, x) & Q1y [S1(y, x) C1 OFFN(y)] & Q2z [S2(z, x) C2τ(z) ≥5min] (def TUR)

i. λx. θakk(def TUR, x) & Q1y [S1(y, x) C1 OFFN(y)] & Q2z [S2(z, x) C2τ(z)≥ 5min]

j. λPλyλx. θnom(y, x) & P(x) (λx. θakk(def TUR, x) & Q1y [S1(y, x) C1

OFFN(y)] & Q2z [S2(z, x) C2τ(z) ≥ 5min])

k. λyλx. θnom(y, x) & θakk(def TUR, x) & Q1y [S1(y, x) C1 OFFN(y) & Q2z[S2(z, x) C2τ(z) ≥ 5min] (anna)

l. λx. θnom(anna, x) & θakk(def TUR, x) & Q1y [S1(y, x) C1 OFFN(y) & Q2z[S2(z, x) C2τ(z) ≥ 5min]

The result of the process of semantic composition for the sentence in (40) is theunderspecified semantic form given in (43l). As we saw above, (40) might get any ofthe readings stated in (41) (and more). The different readings are derived by differentfixations of the underspecified parameters in (43l) in correlation with a redefinitionof the unbound variables x, y and z as sorted variables for eventualities.37 Thus,the different readings are not assumed to arise as a result of some co-compositionprocess. Rather, the semantic contribution to the interpretation of an utterance of(40) in all three cases is the same. It is only once the compositionally determined SFof that sentence is ‘given over’ to the pragmatics component that the underspecifiedparameters and variables get fixed/sorted.

Concentrating on Dolling (1997, 2005)’s assumptions concerning the nature oflexically stored information; a general point to note is that integrating pragmati-cally relevant information into the lexical entries is not unproblematic. Thus, oneargument for underspecification of semantic information in the lexicon was that itreduces its size and makes it more economic. In the approach suggested by Dolling,however, the reduced semantic information is countervailed by the additional PRAGcomponent. More specifically, the question arises what such labels as OFFNk2 orOFFNl2 refer to. From what Dolling says, they are actual conceptual constantswhich may be used to fix the parameter OFFN. If that is the case, then such alexical entry, actually, would not differ much from those found in maximalist ap-proaches, since the possible readings of an expression, although no longer listed inthe semantic component of an item’s lexical entry would then be listed in its prag-matic component. Moreover, characterising the semantic form of an expression as a‘pointer’ to the set of possible primary readings for that expression is thus superflu-ous, since semantic form as well as all possible primary readings are actually partof one and the same lexical entry. Also, with respect to the subsets identified inthe PRAG component, the question arises, how they are determined and by whichcriteria their members are characterised. Dolling states that he takes the readingsgiven in the PRAG component to be conventionalised and primary in that they arenot derived from some underlying full reading. From this perspective, it seems thatinstances of non-literal readings that are highly salient in Giora’s terms are not partof the readings in PRAG. The question is, whether this is a desirable state of affairs.

37For particulars, please refer to Dolling (2005).

Having said all the above, the major advantage of such an approach that assumesmaximal underspecification of lexical meaning is that it allows for strict semanticcomposition without that process having to generate numerous potential semanticforms for a particular utterance which, in a second step, as it were, have to bechecked against the actual context of utterance for the actual semantic form to bedisambiguated.

Ruhl’s monosemic approach

Another approach to lexical meaning that assumes a high degree of underspecifi-cation of semantic information in the lexicon is put forward by Ruhl (1989), whoassumes that an expression’s lexical meaning, or semantic representation is veryabstract and needs to be enriched by pragmatic processes in order to yield anyfull-fledged reading. He argues that the literal meaning we take an utterance toexpress actually results from an unconscious cognitive process, which makes useof language-external information. Thus, generally, Ruhl does not subscribe to thestandard notion according to which the literal meaning(s) of a lexical item is/arewhat is stated in its lexical entry. Rather, he assumes an abstract semantic repre-sentation which is then specialised or generalised through what he calls pragmaticmodulation, where modulation can take the form of pragmatic specialisation on theone hand, but also of pragmatic generalisation on the other38. Thus, much of whathas been assumed to be semantic aspects of a given word are actually pragmaticallydetermined aspects of the meaning of that word.

Moreover, Ruhl (1989) argues that one should not go by intuition concerningthe possible readings an expression may have, since intuitive listing is very selec-tive. Thus, looking at a dictionary entry for some expression, what is described arepossible uses of an expression, however, dictionary entries by no means include allthe possible readings in which an expression can occur, not even all its so-calledliteral meanings. Ruhl takes to bear as an example. Looking at the readings for tobear given in the Oxford English Dictionary, he argues that they do not exhaust tobear’s possible literal readings. Crucially, readings that cannot be described with thehelp of other expressions are ignored. He compares this to the results of a corpusanalysis, where many more different, but still literal meanings can be differentiated.He then argues that the general problem with examples drawn from intuition is thatthey are the readings that come to mind first and they in turn are stereotypical forthe expression under consideration, but they are by no means the only readings aparticular expression may be used in.

In order to avoid rampant lexical polysemy, Ruhl assumes of simple expressionsthat what is coded in their lexical entries is an abstract semantic representationwhich is worked on by pragmatic processes. As mentioned before, Ruhl uses thenotion of stereotype to explain why, for a given example, we arrive at what we thinkof as the literal meaning of that expression. For example, if we were to give adescription of the meaning of the verb hit, presumably the first thing that wouldcome to our minds would be something along the lines of ‘strong physical contact’.

38 Note that these two processes are very similar in kind to those that in relevance-theoreticterms are called narrowing and loosening, respectively.

We take this to be the ’basic’ sense of the expression. However, Ruhl argues thatthis supposed ‘basic’ sense is the expression’s stereotypical meaning. Like the otherreadings hit may have, the sense ‘strong physical contact’ is a pragmatic inference.It only differs from other possible readings in that it is a ‘“favored” pragmaticinference’ (Ruhl 1989, p. 122).

Unfortunately, Ruhl (1989) does not explain what it means for a pragmaticinference to be ‘favoured’, that is, what it is that makes a reading a stereotypicalone. He does state that stereotypes may be triggered by the immediate sententialcontext (as, for example, the interpretation of take as ‘steal’ in The thief took thejewels.). Very often, however, as in the case of hit, stereotypical meanings seem to notdepend on any concrete context. But, as Ruhl argues, this is only an illusion sinceinterpreters always make assumptions about contexts of utterances of expressions,although not consciously so. Thus, words never ‘mean’ in isolation.

A Cognitive Approach

A further view of the relation between lexical items and the meanings they encode– one not primarily aiming at formalisation, but rather cognitive plausibility – isoffered by Relevance Theory.

Originally, Relevance Theory (RT) assumes that (monomorphemic) expressionsencode atomic concepts. On the one hand, these concepts form constituents ofthe logical form of an utterance the respective lexical item is part of. On theother hand, a concept is taken to be an address in memory, making available avariety of information. This information is structured under three distinct ‘entries’:a lexical, a logical and an encyclopaedic entry. For example, for a concept likerobin, the lexical entry consists of specifically linguistic information (phonologicaland syntactic information), the logical entry consists of inference rules which applyto mental representations that include the concept robin. The encyclopaedic entry,then, includes all kinds of information about robins, that is, about the extension ofthe concept, in form of a set of assumptions and assumption schemas.

Sperber and Wilson (1995) describe the relation of a concept as an address inmemory and the information it makes available as follows.

. . . the distinction between address and entry is a distinction betweenform and content, the address being what actually appears in logicalforms, and the various entries spelling out its logical, lexical and ency-clopaedic content. (Sperber and Wilson 1995, p. 92)

However, if this is how concepts are generally characterised, then the question arisesin which sense they are supposed to be atomic.39 Moreover, under such a characteri-sation of ‘concept’, we get the undesirable situation that a lexical item’s phonologicalform and syntactic information would be part of its meaning, as they are part of thelexical entry that forms part of the concept the lexical item in question is taken toencode (cp. Groefsema 2007). Thus, the concepts encoded by linguistic expressions

39Incidentally, consider also the following quote from Sperber and Wilson (2008, p. 12) talkingabout the interpretation of temperature in I have a temperature: ‘This concept has a parameterthat can take a range of values . . .’ (my emphasis).

should not be taken to include lexical entries. Rather, as Groefsema (2007) suggests‘. . . activation of linguistic information of a word (e.g. activation of its phonologicalform) gives access to a conceptual address, and vice versa.’ (Ibid., p. 138-9).

Carston (2002c) actually proposes that what lexical items encode need not befull-fledged concepts but rather may be concept schemas, that is, very abstract‘meanings’ in that they simply constitute ‘pointers’ which indicate the conceptualspace that is the basis for inferring the actual reading that an expression will get in aparticular utterance situation. This assumption goes well together with the generalassumption that the semantic form of a lexical item is not of a different format thanconcepts in general but rather is a particular kind of conceptual structure. Thus,the lexical meaning of an expression only provides the basis from which pragmaticenrichment processes can take off, resulting in the more specific concepts the expres-sion can be used to express.40 The information available in this ‘conceptual space’is organised with respect to accessibility, where accessibility of certain elements ofthe various types of encyclopaedic information is a function of (amongst others)frequency and recency of use of that element in processing. Thus, for an expressionlike open, the concepts that this expression is used to communicate most frequentlymay be considered to be more or less ready-made retrievable.

Generally, the characterisation of lexical meaning as underspecified is in linewith RT’s overall assumption that literal meaning and non-literal meaning are notas different as traditionally assumed. Thus, an expression’s lexical meaning is neitherliteral nor non-literal; it is not even a full-fledged concept. That is, both the literalas well as potential non-literal meanings of an expression result from the applicationof pragmatic processes in interpretation. Thus, with both literal and non-literalmeaning resulting from similar processes, there is no reason to assume that theformer is necessarily prior to the latter in interpretation.

Figure 2.15: Literal and non-literal meaning in Relevance Theory

In this respect, the relevance-theoretic approach is similar to Giora’s approach,where it is not the traditional differentiation of literal and non-literal meaning, butrather the degree of salience that a certain meaning of a particular expression hasthat is responsible for the relative ease or difficulty of its interpretation. As already

40See also Assimakopoulos (2008) for a discussion of RT’s original conception of encoded meaningand its problems.

mentioned in section 2.2.2, Giora assumes that the salient and less-salient meaningsof an expression are coded in that expression’s lexical entry. To repeat, the salienceof a meaning is a function of its conventionality, familiarity, frequency, or givennessin a certain context (Giora 1997). Salient meanings are always processed first,regardless of the context in which the respective expression is used. In contrast, non-salient meanings of an expression are not coded in that expression’s lexical entry,but rather are the result of a pragmatic interpretation process leading to the speakerintended meaning, which either first accesses the salient meaning, discards it as notbeing intended and then goes on to interpret the intended meaning (traditional viewfollowing Grice), or works in parallel to the process that accesses the salient meaningof that expression.

open: PHON: . . .SYN: . . . SEM: {OPEN-DOOR[10], OPEN-WINDOW[9], OPEN-FILE[8], OPEN-EYES[8], . . .}

degree of salience

Figure 2.16: Lexical entry according to Giora

As mentioned above, RT and Giora’s approach have in common that they con-sider literal meaning and non-literal meaning as more similar than traditionallyassumed. However, whereas under the relevance-theoretic approach both literalmeaning and non-literal meaning are phenomena that occur at a pragmatically de-termined level, in Giora’s approach, some meanings traditionally characterised asliteral are of lexical and some of pragmatic nature. The same holds for non-literalmeaning. Furthermore, Giora characterises salient meanings and less-salient mean-ings as being coded in the lexicon. However, there seems to be no important reasonwhy this should be the case. Rather, assuming that salience is a gradual property,one could in fact ask, by which criteria less-salient meanings are differentiated fromnon-salient meanings. To put it differently, what combination of factors is it thatmakes a specific meaning pass ‘the threshold’ from being a pragmatically derivedreading (i.e. a non-salient meaning) of a certain expression to being lexicalised (i.e. aless-salient to highly salient meaning)?

Assuming, in contrast, that what Giora characterises as salient and less-salientmeanings are actually on the same informational level as non-salient meanings avoidssuch problematic questions. Moreover, the feature of salience, on a first blush, seemsto go well together with what RT calls accessibility. That is, different readingsfor an expression are more or less accessible, depending on their frequency andrecency of use. Moreover, as Carston puts it, concepts that an expression is usedto communicate most frequently may be more or less ready-made retrievable andmaybe considered, in that sense, as conventional. Also, considering that the salienceof a meaning may be influenced by such factors as whether it has been recently

Figure 2.17: Salient and non-salient meaning in Giora’s approach

used/activated makes it even more desirable to assume that it is meanings at somenon-lexical level that are characterised as being salient, less-salient or non-salient.Otherwise one would have to assume that certain meanings are sometimes part ofthe lexical entry of a particular expression and sometimes are not, depending ontheir givenness in a context or recency of use.

Integrating these ideas into a view of a lexical item’s entry where this is not partof the concept encoded by that item, one might assume the following constellation.Part of the lexical entry of a particular lexical item is its semantic form, which ishighly abstract and underspecified and which ‘points’ to what Sperber and Wilson(1995) call the encyclopaedic entry of a concept.41 Thus, the semantic form of anexpression makes available, in the sense of activates, a range of information that maypotentially be important in the particular context of utterance. The informationactivated is structured in the sense that due to differences in the accessibility ofthe various pieces of information, they will be more or less strongly activated. Asalready mentioned, accessibility is taken to be a function of frequency and recencyof use/activation of a particular piece of information in processing. Moreover, as theresults of studies show that tested the role of context in interpretation, it is likelythat accessibility is also sensitive to contextual information available at the timeof utterance. Thus, on the one hand, pieces of information of the encyclopaedicentry activated by the utterance of a particular lexical item may be inhibited intheir strength of activation, if they are incongruent with the contextual informationavailable. On the other hand, contextual information congruent with particularpieces of information may strenghten their level of activation. However, since wesaw that what Giora calls salient meanings are activated regardless of whether ornot they fit the context, it seems that incongruent contextual information is never‘strong’ enough to actually prevent salient meanings from being initially activated.Thus, the impact contextual information may have on the degree of accessibility ofthe readings of an expression seems to be constrained.

41Remember that RT takes the content of a concept – ignoring the lexical entry, which hasalready been identified as problematic – to consist of both an encyclopaedic and a logical entry. Inassuming that it actually is the encyclopaedic entry alone that the semantic form of an expression‘points’ to, I follow a suggestion made by Groefsema (2007), who, in my view, gives a range ofconvincing arguments to this effect.

open: PHON: . . .SYN: . . . SEM: <OPEN>

degree of accessibility{OPEN-DOOR[10], OPEN-WINDOW[9], OPEN-FILE[8], OPEN-EYES[8], . . .}

points to

Figure 2.18: A lexical entry adapted from Relevance Theory

Underspecification and Conventionality

One other issue I would like to discuss is the relation of lexical meaning and theproperty of conventionality in approaches that assume a high degree of semanticunderspecification. Thus, according to Saussure (1931), linguistic signs consist oftwo parts, a form and a meaning, where the relation between those two parts restson a convention. That is, there is no natural connection between the meaning andthe form of an expression. But, what do we mean by ‘convention’? (Lewis 1969)offers a formalised definition of convention.

A regularity R in the behaviour of members of a poluation P when theyare agents in a recurrent situation S is a convention iff it is true that, andit is common knowledge in P that, in almost any instance of S amongmembers of P:

1. almost everyone conforms to R

2. almost everyone expects almost everyone else to conform to R

3. alomst everyone has approximately the same preference regardingall possible combinations of actions

4. almost everyone prefers that any one more conforms to R, on con-dition that almost everyone conforms to R

5. almost everyone would prefer that any one more conform to R’, oncondition that almost everyone conform to R’

where R’ is some possible regularity in the behaviour of members of Pin S, such that almost none in almost any instance of S among membersof P could conform both to R’ and to R.’

Thus, conventions seem to have to do with ‘customariness’ within a group. Thatmeans that a speaker associates a certain expression form with a certain mean-ing, because a sufficiently large enough group of speakers associates that expressionform with that meaning. Under such a view of convention, it seems inappropriateto characterise the abstract underspecified semantic form of an expression as be-ing conventional in the intended sense, since SEMs are not actual readings of anexpression.

As mentioned above, there are a number of psycholinguistic studies which showthat particular types of non-literal meaning may be interpreted as fast as literal

meanings. In particular, it has been shown that certain non-literal meanings of ex-pressions get activated even in contexts where it is obvious that a literal interpreta-tion is intended. Moreover, the results differ with respect to whether the non-literalmeanings are salient, less-salient or non-salient. According to Giora’s definition ofsalience, conventionality of a reading plays an important role. Moreover, I havealready argued above, that conventionality should not be viewed as a binary featureand, thus, cannot be used to differentiate between literal and non-literal meaning.Thus, the fact that under the assumption of underspecification of lexical meaning,conventionality in the intended sense cannot be taken to be a feature of lexicalmeaning, does not pose a problem. As the results of the empirical studies men-tioned suggest, conventionality is a graded notion which applies not only to whattraditionally is called literal meaning, but also to non-literal meaning, in some casesleading to a non-literal meaning actually being conventionally associated with somephonological form to a greater degree than a possible literal reading.

Thus, approaches assuming highly underspecified semantic forms come very closeto what Recanati (2004) calls meaning eliminativism (ME) and which he charac-terises as follows.

‘. . . according to ME, we do not need linguistic meanings even to serve asinput to the construction process [. . .] the senses that are the word’s con-tributions to contents [of utterances] are constructed, but the construc-tion can proceed without the help of conventional, context-independentword meanings.’ (Recanati 2004, p. 147, my emphasis)

Although underspecification-approaches do assume some SEM-component in an ex-pression’s lexical entry that SEM only serves as a ‘pointer’ to potential readings ofthe expression.42 It does not constitute anything like the conventional meaning ofan expression, as assumed under the traditional view of lexical semantics. Moreover,Recanati (2004) says of ME that it computes ‘directly the contextual sense whichan expression assumes on a particular occasion of use on the basis of the contextualsense which that expression had on previous occasions of use . . .’ (Recanati 2004,p. 147). If, however, the contextually determined reading of an expression is com-puted on the basis of the readings of earlier uses of the same expression, there hasto be some sort of connection between the expression and those readings: this is theradically underspecified SEM component.

Underspecification and Semantic Relations

Traditionally, one of the goals of semantics was to describe and explain the meaningrelations that were found to hold between individual expressions. However, it hasbeen noted that this task is not at all easy to accomplish. Thus, under the traditionalperspective of linguistic signs as consisting of a form and a meaning, first, individuallexical units (combinations of a distinctive sound shape and a distinctive meaning)

42For an actual ME-approach, one that denies that words actually consist of both sound form andmeaning but rather assumes that phonological form merely represents conceptual content (hencethe name representational hypothesis), cp. Burton-Roberts (1994, 2000, 2007a), Burton-Robertsand Poole (2006), Chng (1999).

which share the same phonological form, have to be classified as belonging to thesame or different lexemes. Second, a particular lexeme which is characterised asbeing polysemous, thus, incorporating more than one lexical unit, may have severaldistinct meaning relations to several, distinct lexemes.

What is generally problematic about the notion of semantic relation is that thereis the danger that specifically linguistic knowledge is mixed up with general concep-tual knowledge (cp. Fodor and Lepore 1998). Thus, the underlying assumption forthe specification of meaning relations between expressions is that their intensionsshare some feature that allows their extensions to be related in a certain way. Thatis, the fact that the expressions human and woman are taken to be in a hyponymyrelation is due to the observation that the denotation of woman forms a subset ofthe denotation of human. In order to explain this fact about the denotations ofthese expressions, it is assumed that their intensions, or meanings, are in a partic-ular relationship as well, namely, the meaning of woman includes the meaning ofhuman. However, the question is whether, when thinking about such relationshipsbetween denotations, it is admissible to link this relation to a relation between theirlinguistic meanings. I think it is not. Rather, the relationship holding betweenthe denotations of human and woman is due to the relation between the conceptshuman and woman that we have and which, crucially, we have independently ofthe fact that we can use the phonological forms /ju:m�n/ and /w=m�n/, respec-tively, to express them. Thus, semantic relations are much more closely related toour general cognitive ability of categorisation than to linguistically relevant aspectsof meaning – whatever those may be. Moreover, a speaker may be able to use aparticular expression correctly without being aware of the relationships that holdbetween the concept that the expression he uses expresses and other concepts. Thatis, I might use the expression whale successfully to refer to an instance of the kind‘whale’ without knowing that the denotation of the concept whale is a subcategoryof the denotation of the concept mammal.

Thus, one might question Borg (2004b)’s claim that a semantic theory shouldmake clear the relations that hold between individual complex expressions, for exam-ple, which inferential relations possibly hold between them. In fact, one consequenceof assuming underspecification of the linguistic meanings of expressions is that se-mantics is no longer able to capture the various meaning relations an expression mayenter into. In that case, how could semantics capture the inferential relations hold-ing between elements of different domains that may be denoted by one and the sameexpression beyond only very general relations. More specifically, meaning relationsmay be taken to hold between full readings of expressions, but not between theirabstract, underspecified semantic forms. However, from the arguments given above,it seems that one should not expect of semantics in general that it has anything tosay concerning the relations holding between the concepts expressions may be usedto express.

More Underspecification in the Lexicon

It should be mentioned that the idea that the lexical information for some linguisticitem does not include information on that item’s argument structure has not onlybeen pursued for the semantic, but also for the syntactic component of such entries.

Thus, Kratzer (1996) argues against the assumption that the argument that willoccupy the subject position in a sentence is projected by the verb, rather, it comesinto the structure by means of a functional head. Thus, for the sentence in (44),(Kratzer 1996, p. 121) gives the partial syntactic structure as in (2.19).

(44) Mittie feeds the dog.

VoiceP

��

Mittie

Voice’

��

DP�� PPP

the dog

Figure 2.19: Partial syntactic structure for Mittie feeds the dog

Thus, Mittie is base-generated in SpecVoiceP, whereas the head of VoiceP intro-duces the external argument. Note that Kratzer uses the ordered argument notationfor argument structure, which does not clearly separate syntactic from semantic in-formation. However, translating the facts into Dolling’s notation, the syntacticcomponent of the lexical entry of feed would thus only specify that it is a verb andthat it takes one NP (internal) argument. As Dolling (2005) notes, Borer (2004)goes even further in claiming that one should also separate the internal argumentfrom its verb. Ignoring issues concerning the adequacy of such an approach for themoment, what this would result in is a syntactic component for feed, which wouldno longer include information on the number and kind of arguments that item takes.Thus, there are approaches in syntax also that reduce the amount of informationrequired to be given in some linguistic expression’s lexical entry.

In fact, there are attempts at radically reducing the role the traditional lexiconplays in the linguistic system. Thus, Marantz (1997) wants to dispense with thelexicon as traditionally conceived altogether. He assumes three distinct lists whichto draw from in producing an utterance. The first of these, which he calls ‘narrowlexicon’, is the list of units that syntax operates with. This list contains a language’satomic roots, as Marantz calls them, as well as the atomic bundles of grammaticalfeatures. What is interesting about the roots is that they are neutral with respectto word category. That is, a root like

√destroy is neither a noun nor a verb;

depending on the kind of syntactic structure which it is part of it may function as averb or as a noun. Thus, merging a root like

√destroy with ‘D’ (i.e., a nominalising

environment) places it in a nominal context (cf. Figure 2.20). In contrast, mergingthe root with a verbalising functional head (v-1 in 2.21) results in destroy functioningas a verb. True nominalisations, then, contain both a verbalising and a nominalisinghead (cf. 2.22).

What is interesting about this approach in light of the present concerns is thefollowing. If lexical items (or rather roots) actually are characterised by the factthat they are category neutral, the syntactic information one would expect to findfor them might be similarly underspecified as is the case for semantic information.

��

D√destroy

��

√destroy the city

Figure 2.20: e.g. the destruction of the city, the city’s destruction

��

v-1√destroy

��

√destroy the city

Figure 2.21: e.g. (John)destroyed the city

Thus, the lexicon may not provide any distinct information on category membershipof roots. However, and Marantz (1997) explicitly notes this, roots would need to in-clude some type of information which ‘tells’ the syntax in which potential structuresthe root may occur. Thus, Marantz (1997) notes that a root like

√grow differs

from a root like√destroy in that it does not imply an agent. Thus,

√grow

belongs to a different class (or possibly has different features) than√destroy.

Whereas√grow semantically denotes a change of state that is internally caused,

√destroy denotes a change of state that is not. More specifically, with

√destroy

an external cause or agent is implied. However, assuming such a characterisationmay not be unproblematic from a semantic point of view. Generally, it is unclearin which form this type of information enters into the lexical entries of the items inquestion. One plausible possibility is to assume that it is part of an expression’s se-mantic form. However, this is precisely what a minimalist approach wants to avoid.Another possibility is to assume some abstract ‘group’ feature for different classes ofroots. Leaving the particulars to the side, it is still interesting to note the parallelsbetween the minimalist approach and the syntactic/morphological approaches men-tioned above, all of which argue for some underspecification of lexically providedinformation.

Underspecification of Semantic Composition

Similarly to Dolling (1997, 2005), Egg (1994, 2003, 2005) also assumes that lexicallygiven semantic information for expressions is highly underspecified. However, inhis approach, the result of semantic composition is not only underspecified dueto the underspecified nature of the elements that it composes, but in addition,the resulting semantic form only consists of fragments and an indication of thedominance relations between them. Crucially, the fragments are not yet totallyfixed with respect to the scope they take over one another. For an exemplificationof Egg (2003, 2005)’s approach, consider again (32), repeated below.

(32) a. Max began reading every novel.

b. Max began every novel.

��

v-1√destroy

��

√destroy the cityFigure 2.22: e.g. (John’s) destroying the city

Recall that whereas (32a) has two readings depending on the scope the universalquantifier takes, (32b) only has one reading, where the universal quantifier outscopesthe verb to begin. The respective semantic forms of (32a) and (32b) in Egg’s ap-proach resulting from semantic composition are given in figures (2.23) and (2.24).

Figure 2.23: The semantic form of Max began to read every novel.

Figure 2.24: The semantic form of Max began every novel.

The squares indicate ‘holes’ in the semantic representation. These holes haveto yet be specified. The dotted lines indicate dominance relations between theindividual fragments. Thus, in (2.23), both the universal quantifier and the verb tobegin dominate the verb to read, whereas the dominance relation between them is notyet specified. However, either specification is compatible with the relations alreadyfixed in the semantic representation of (32a). This is different for (32b), where inits semantic representation (2.24), the dominance relation between the universal

quantifier and the verb to begin is already fixed such that the former dominates andthus has scope over, the latter.

I do not want to go into too much detail of Egg (2003, 2005)’s analysis here, butnote that the difference in the semantic representations for the sentences (32a) and(32b) is due to differences in the syntax-semantics interface rules that determinehow the respective sentence meanings are composed. Moreover, whereas in the caseof the semantic representation of (32a), there is a so-called constructive solution ofthe underspecification in the semantic representation, there is no such solution inthe case of (32b). A semantic representation has a constructive solution if the under-specified elements (‘holes’) may be resolved simply by integrating the material thatis already present in the semantic representation itself. For example, in the semanticrepresentation for (32a), the semantics of the verb to read can be inserted into thehole immediately dominating it. By functional application, the verb’s semantic typeis converted from <e, <e, <s, t>>> to <s, t>, which makes it of the appropriatetype for insertion either into the hole of the semantics of the verb to begin or thatof the quantifier. Depending on into which of these holes the verb’s semantics isinserted, the resulting representation can then be inserted into the remaining hole,thus determining the scopal relations between the quantifier and the verb to begin.43

The problem with the semantic representation of (32b) is that a constructivesolution is not possible, simply because the verb to begin as it occurs in that rep-resentation is not of the appropriate type for insertion into the hole immediatelydominating it (it would have to be of type <e, <e, <s, t>>> but actually is oftype <<e, <s, t>, <e, <s, t>>). Thus, this is the place where an appropriate rein-terpretation operator needs to be inserted in order to derive a well-typed solutionfor the underspecified semantic representation. One possibility is the operator thatprovides the reading-interpretation of (32b) and which would have to look like in(45). The final semantic representation for (32b) given in (46) is obtained by apply-ing the operator in (45) to the semantics of the verb to begin and only integratingthe result of this operation into the hole originally dominating the verb.

(45) λz.2(λyλxλe. read∗(x, y)(e)(z))

(46) λe′′ ∀y.novel′(y) → begin′(max′, λxλe′.read∗(x, y)(e′)

)(e′′)

Thus, Egg’s approach handles what is traditionally called complement coercionby actually augmenting the verb’s semantics instead of that of the verb comple-ment. Generally, his approach can also handle aspectual coercion as well as casesof metonymy. In all these cases, an underspecified semantic representation is buildup, followed by the attempt of a constructive solution of the underspecifications. Ifno such solution is available, operators are made use of that integrate conceptualinformation which allows for a solution of the semantic representation after all. AsEgg (2003, 2005) himself notes, calling the phenomena his approach handles rein-terpretation phenomena is somewhat misleading, since his analysis actually does notinvolve a revision of some previously build up fully specified semantic form.

43Following Egg (2002) and Egg (2003), I ignore the topmost hole and its dominance relationshere.

Empirical Investigations of Aspects of Semantics 67

2.4 Empirical Investigations of Aspects of Seman-

In this section, I want to review the results of empirical studies that investigateddifferent aspects of the semantic component assumed to form part of the languagefaculty. I will concentate on empirical investigations concerning the nature of lex-ically stored semantic information, on the one hand (2.4.1) and the question ofwhether a distinct neurological correlate of the process of semantic compositionmay be identified, on the other (2.4.2).

2.4.1 Polysemy vs. Underspecification in the Lexicon

As stated in section 2.3.1, one of the challenges for a theory of meaning in the lexiconis the treatment of the phenomenon of systematic polysemy. Recall that systematicpolysemy is characterised by the following facts. The different readings of a particularnoun relate to one another but at the same time are clearly delimitable. There is nounderlying meaning consciously detectable from which the different readings couldbe said to be derived, thus, intuitively they all seem basic. Crucially, the differentreadings are not characteristic for only one particular expression, rather, they arefound across sets of expressions, thus, they are not idiosyncratic. As mentionedbefore, the question with respect to the lexical entries of systematic polysemousnouns is, how much semantic information is given.

From the different approaches to meaning in the lexicon described above, theo-retically, there seem to be at least three possibilities. First, as traditionally assumed,all of the relevant readings of a systematic polysemous noun might be listed as pos-sible literal meanings of that noun. This would amount to a sense enumeratinglexicon, where polysemy is not treated different from homonymy. Second, one of thereadings could be assumed to be the ‘core’, or basic, meaning, from which all othermeanings are derived. At least as far as intuitions are concerned, this view does notseem to capture the facts, however, intuitions might be misleading. Third, none ofthe different readings might be lexically coded, rather, the semantic information inthe lexical entry might be maximally underspecified, allowing for different specifi-cations as the interpretation enfolds, taking into account the specific context of anutterance of the respective noun.

The different views make varying predictions concerning processing. Thus, if thesense enumeration approach is correct, then one should not find any differences inthe processing of homonymous vs. polysemous expressions, as they are representedsimilarly. If, however, the core meaning approach is correct, then one would ex-pect differences in processing, depending on whether a particular expression has ahomonym or whether it is polysemous. Making predictions for the underspecifica-tion approach is a bit more difficult, as it does not seem quite clear what to expect.However, as we will we see below, studies that tested whether homonymy and poly-semy are similar in processing actually found that there are specific differences andthese in turn are explainable if one assumes underspecification of lexical semanticinformation.

Klepousniotou (2002) took up the question of whether homonymy and polysemy

are processed similarly, thus, supporting a sense enumeration approach. In thisstudy, four types of ambiguity were tested, namely, homonyms and three differ-ent types of polysemous expressions (i.e., expressions with a metaphoric reading,words with metonymic count/mass readings and words with metonymic producer-for-product reading). The task was a cross-modal sentence-priming lexical decisiontask. Thus, subjects heard sentences that primed a particular meaning of a poly-semous word or the meaning of a homonym. At the offset of the priming sentence,subjects were presented a target word on a computer screen. Reaction times fordeciding whether the target word was a real or non-word were measured.

The results showed that subjects reacted faster in the lexical decision task whenthe target word was polysemous as compared to homonymous. Klepousniotou (2002)takes this as evidence that different processes are involved in recognising words ex-hibiting different types of ambiguity. Thus, the results do not support a senseenumeration approach. Rather, Klepousniotou (2002) interprets the results as sup-porting a core meaning approach. Generally, she argues that the results are bestexplainable by assuming a basic difference in the representation of homonymouswords as compared to a polysemous word. Thus, words that are homonymous reallyare distinct words in that they are distinctly represented. In contrast, in case ofpolysemy, there is only one entry, which, according to Klepousniotou (2002), givesthe core or basic meaning of the polysemous word. As a consequence, there is a dif-ference in processing. Thus, in case of homonymy, the interpretation process selectsone of the given mental representations or entries, which goes along with a compe-tition effect between the possible entries and results in slower overall processing ofhomonyms. In contrast, in the case of polysemy, no selection is necessary, as thereis only one entry to begin with.44

However, as Beretta et al. (2005) note, a task such as lexical decision does notnecessarily reflect early stages of processing. Thus, the effects attained in a lexicaldecision task could also be due to post-lexical processing. The authors thereforedecided on a MEG experiment in order to find out whether the differences in reactiontime noted by Rodd et al. (2002) for the processing of homonymy vs. polysemy wouldbe paralleled by differences in the MEG data recorded. The component they focusedon is the M350, of which they argue that it has been shown to be an index of initiallexical activation, but – and this is crucial – not of later processing stages (see alsoPylkkanen et al. 2006). The hypotheses investigated are the following. If the senseenumeration approach is correct and homonyms and polysemous words are processedsimilarly after all, one would expect the M350 data to be similar for both conditions.In that case the M350 data would be in contrast to the reaction time data, suggestingthat the latter reflects post-lexical processing. If, however an approach is correctthat assumes a single entry for polysemous words (i.e., either the core meaning oran underspecification approach), the M350 data should be differential for the twoconditions and thus, parallel that of the reaction time studies.

Beretta et al. (2005) used the stimuli and design of Rodd et al. (2002)’s ex-periment, carrying out both the behavioural (reaction time) test and a MEG test.

44Note that although this study did not take into account the possibility of an underspecificationapproach, Klepousniotou (2002)’s arguments for a core meaning approach are compatible with anunderspecification approach as well.

Thus, subjects were visually presented with test words of which they had to decidewhether they were words of the language or not. Test words were either homonyms(two meanings) or non-homonyms (one meaning). Within these two categories afurther distinction was made between meanings that consisted of only a few andmeanings that comprised many senses.45 Beretta et al. (2005) replicated the be-havioural results of Rodd et al. (2002), thus, reaction times to homonymous werelonger than to polysemous words. Moreover, the more senses a word had, the fasterit was recognised (cf. figure 2.25). Crucially, the M350 data recorded paralleled thereaction time data. Thus, polysemous words led to an earlier peak of the M350 com-ponent than homonymous words. Again, the more senses a word had, the earlierthe peak of the M350 component (cf. figure 2.26).

Mean response time (ms): factorial design and main effectsHomonymy Polysemy

Few senses Many senses MeanSingle meaning 626 611 619(non-homonyms)More than one meaning 648 622 635(homonyms)Mean 637 617

Figure 2.25: Reaction times from Beretta et al. (2005)

Mean M350 latency (ms): factorial design and main effectsHomonymy Polysemy

Few senses Many senses MeanSingle meaning 345 328 336(non-homonyms)More than one meaning 359 349 354(homonyms)Mean 352 338

Figure 2.26: M350 latencies Beretta et al. (2005)

The results, thus, support the hypothesis that polysemous words are representedin a single entry in the mental lexicon, since both the reaction time as well as theMEG data show that they are processed differently from homonymous words. Sim-ilar results were attained by Pylkkanen et al. (2006), who also used the M350 com-ponent to investigate the processing of homonymous and polysemous words. Theseauthors reach the same conclusion as Beretta et al. (2005), favouring a single-entryapproach to polysemy. Note, however, that neither of the two studies makes anyclaims as to what type of single entry approach (core meaning or underspecification)is better reflected by the data. In contrast, Frisson (2009) does argue explicitly foran underspecification approach.

45For example, there are two homonymous words bank, one with a meaning that might be glossedas ‘river bank’, the other with a meaning glossed as ‘financial institution’. For the latter, differentsenses are possible like ‘building’ or ‘institution’.

Frisson (2009) recapitulates the results of experiments investigating the process-ing of homonymy and polysemy, concentrating on two factors, namely, the influenceof frequency of occurrence of a meaning/sense and context of utterance on the pro-cessing of homonymous vs. polysemous expressions. Whereas for the former thesefactors have an impact on their processing, for the latter, they do not. More specif-ically, taken out of context, encountering a homonymous phonological/graphicalform will result in strong activation of the meaning that is more frequent and lessactivation of less frequent meanings. Thus, Frisson (2009) speaks of homonym pro-cessing as a ranked parallel process, where the ranking of the individual meaningsis provided by their individual frequency. However, this ordering can be influencedby contextual factors. That is, if the context biases the less frequent meaning of ahomonymous phonological/grapical form, this meaning becomes better accessible.However, context cannot suppress the more frequent meaning, that is, the morefrequent, or dominant meaning, is accessed in any case. Moreover, processing a lessfrequent meaning involves an extra effort. As Frisson (2009) puts it:

. . . when the following context indicates that the subordinate meaningwas the intended one, a great deal of extra processing is observed, in-dicating that the wrong meaning had been selected; when the preced-ing context disambiguates towards the subordinate meaning, increasedfixation times are observed on the target word, indicating competitionbetween the two meanings . . .’ (Frisson 2009, p. 114).

What is important for Frisson (2009)’s argument is that these effects show upvery early during processing and on the homonymous form itself. Thus, it seemsthat for homonymous expressions, a specific interpretation is selected immediatelyon its encounter. In contrast, for polysemous expressions, such frequency or con-textual effects were not perceived. Thus, it seems that polysemous expressions arenot subject to a ranked parallel process, where the various senses are activated inparallel and where the strength of activation depends on the individual senses’ fre-quency. Rather, as Frisson and Pickering (1999) argue, the results for the processingof polysemous expressions seem to be best captured by a model that assumes un-derspecification of semantic meaning. Thus, if it is an underspecified semantic formthat is activated during early processing, one would not expect that the frequencyof the individual potential senses plays a role at this stage. Similarly, the fact thatnone of the potential senses could be established as the dominant reading (althoughsenses do differ as to their frequency of use) is explainable by an underspecificationapproach, but not by a core meaning approach.

Moreover, as Frisson (2009) points out and as already argued above, a furthercharacteristic that makes an underspecification approach attractive is that it allowsfor a similar treatment of so-called literal and non-literal senses of polysemous words.It explains the fact that there may be no differences in the time taken to processa literal vs. non-literal sense of a polysemous word. Thus, Frisson and Pickering(2007) tested whether there are differences in the processing of the (literal) personsense and the (non-literal) familiar metonymic producer-for-product sense of namesfor famous persons such as Dickens, using sentences as in (47).46

46See also Frisson and Pickering (1999) where no significant differences were found in processing

(47) a. I heard that she often met Dickens when she had the time.

b. I heard that she often read Dickens when she had the time.

They found no significant early differences in reading time between the two senses,although overall the metonymic senses were much lower in frequency.

To summarise this section, it seems that the results obtained in studies investi-gating the processing of homonymous vs. polysemous expressions are best explainedby assuming an underspecification approach for expressions with at least two re-lated senses, regardless of whether these senses are literal or non-literal, as long asthe latter are familiar.47

2.4.2 Empirical Evidence for Semantic vs. Pragmatic Pro-cessing

Several studies have been conducted to investigate the processing of sentences whosesemantic composition is assumed to involve coercion processes. One of the questionsinvestigated was whether coerced sentences would be harder to process than controlsentences that do not involve any coercion processes, thus reflecting the insertionof additional operators into the semantic form of the coerced sentences. More gen-erally, there is the question of whether coercion is any different from plausabilityviolations, which are pragmatically based. That is, from a theoretical perspectivecoercion clearly is a semantically based operation and as such one might expectcoerced sentences to have different effects on processing than sentences whose inter-pretation involves pragmatically based plausibility violations. Another question ofinterest is whether there are any differences in the processing of complement coer-cion as compared to aspectual coercion. If no such differences are found, this couldbe taken to indicate that the two phenomena really are two different cases of thesame type of underlying process. If, on the other hand, there are differences inhow these phenomena are processed, this would indicate that they really are twodifferent kinds of phenomena. Furthermore, considering that in Egg (2003, 2005)’stheoretical approach, for instance, metonymy and coercion are treated similarly, thequestion arises whether the two types of phenomena show similar processing effectsin empirical studies.

The prediction that coerced sentences are harder to process follows from the as-sumption that coercion processes, in contrast to more basic semantic compositionoperations such as functional application or predicate modification, are not syntac-tically triggered, but are purely semantic in nature. Thus Pylkkanen and McElree(2006, p. 35) claim that ‘[i]f coercion requires an extra-syntactic process that is notpart of the default repertoire of interpretive rules, coerced sentences should be morecostly to process than non-coerced controls’. This hypothesis has been confirmed by

place-for-institution (i) and place-for-event metonymies (ii) as compared to the ‘literal’ senses ofthe respective expressions.

(i) That blasphemous woman had to answer to the convent.(ii) A lot of Americans protested during Vietnam.

47However, it seems that there is not much difficulty involved in extending the number of sensesof an expression, if the extension is systematic (cf. Frisson and Pickering 2007).

longer readings times (cf. McElree et al. 2001) as well as differences in eye tracking(cf. Pickering et al. 2005, Traxler et al. 2002) for complement-coerced compared tonon-coerced control sentences. However, for aspectual coercion the results are not soclear, with some studies reporting effects in secondary lexical decision and self-pacedstop-making sense tasks (Pinango et al. 1999, 2006, Todorova et al. 2000) and somereporting no reading time or eye movement effects (cf. Pickering et al. 2006). Neu-rolinguistic results are more consistent, with both types of coercion modulating thesame brain activity (cf. Brennan and Pylkkanen 2008), where this, in turn, differsfrom that associated with the ERP component N400, traditionally taken to reflectso-called semantic integration difficulties (cf. Pylkkanen, Oliveri and Smart 2009).

Thus, Pylkkanen et al. (2004) used magnetoencephalography to investigate po-tential effects of complement coercion on processing. Most importantly, the idea wasto test the hypothesis that in the case of complement coercion, semantic composi-tion is more difficult and costly and should lead to some measurable effect differentfrom pragmatically based plausibility effects usually engendering an N400. The un-derlying assumption is that the difficulty in so-called semantic integration taken tobe captured by the N400 ERP component is in fact a reflex of the difficulty of in-tegrating the meaning of a sentence with the hearer’s knowledge of the world. Inother words, what the N400 actually captures is not a difficulty in building up thesemantic form of a sentence, but rather the hearer’s difficulty of interpreting it inaccordance with his knowledge of how the world is. Thus, the anomaly of a sen-tence such as (48b) below is not due to some semantic mismatch, but rather to thefact that its meaning does not match the knowledge we have concerning possiblerelations between journalists and articles.48

(48) a. The journalist began the article after his coffee break.

b. The journalist astonished the article after his coffee break.

c. The journalist wrote the article after his coffee break.

The authors found an anterior midline field (AMF) at 350 – 500ms after the onsetof the critical noun (article in 48a) for all sentences, but with the largest amplitudesin coerced sentences (as, e.g., 48a). Importantly, anomalous sentences (e.g., 48b)and control sentences (e.g., 48c) showed no significant difference in AMF amplitude.Moreover, while the left hemisphere M350 field pattern, taken to be the neuralsource for the N400 ERP component and occuring at 300 – 400ms, showed thelargest amplitudes for anomalous sentences, it was not modulated by the coercedstimuli.

Comparable results were obtained by Brennan and Pylkkanen (2008) for casesof aspectual coercion, indicating that the latter phenomenon is indeed similar tocomplement coercion. The authors were concerned that earlier conflicting resultsconcerning the processing of aspectual coercion might be due to the fact that theplausability of the used verbs as expressing a single event or rather a complex eventconsisting of a number of similar single events was not controlled for. For instance,‘most knocking events involve multiple knocks . . . [i]n contrast, a burping event is

48It should be noted that this argument only holds true for semantic approaches that are entirelytype-driven and where restrictions on the type of semantic role required of the arguments do notinfluence the process of semantic composition.

much more likely to involve only a single burp.’ (Brennan and Pylkkanen 2008,p. 133). Thus, for their experiments, Brennan and Pylkkanen (2008) pre-tested thematerial with respect to the likelihood of the verb describing a single or repeatedevents. Verbs that were rated as strongly punctual were then combined with durative(cf. 49a) or punctual (cf. 49b) adverbs to create critical and control stimuli.

(49) a. Throughout the day the student sneezed in the back of the classroom.

b. After twenty minutes the student sneezed in the back of the classroom.

In a first experiment, subjects’ self-paced reading times were measured as well as theaccuracy of sensicality judgements concerning the stimulus sentences. The resultsindicate that punctual verbs are more difficult to process in the environment ofdurative modifiers, as subjects took longer to read them in the critical conditionthan in the control condition. In a second experiment, using MEG, the authorstested whether the AMF effect found for complement coercion would also be foundfor aspectual coercion. If the same effect were obtained, this would indicate that thetwo phenomena are in fact similar and thus involve comparable processes. An AMFeffect was indeed found, confirming the hypothesis that aspectual coercion involvesprocesses of meaning shift similar to those found in complement coercion.

Pylkkanen, Martin, McElree and Smart (2009) showed that the AMF effect ob-tained by Pylkkanen et al. (2004) (and Brennan and Pylkkanen 2008, for that mat-ter) was not due to the experimental task subjects had to carry out, which involveddecision making. Thus, even when subjects were simply reading for comprehensionwithout having to judge the stimuli, AMF effects were obtained for the coercedstimuli. Moreover, Pylkkanen, Oliveri and Smart (2009) showed that the effect isnot restricted to coercion phenomena, that is, resolvable semantic mismatches, butis also obtained in other cases of clearly semantically based but unresolvable mis-matches between elements in a sentence. More specifically, in this study, Pylkkanen,Oliveri and Smart (2009) investigated the effects of violations of the constraints onverbal un-prefixation, which are much more restrictive than those for adjectival un-prefixation. The stimuli used involved semantic violations (as, e.g., 50b), worldknowledge violations (as, e.g., 50c) and well-formed controls (as, e.g., 50a). Tokeep the syntax constant across conditions, all stimuli were presented in the passivewith progressive aspect, to force a verbal reading of the participle (rather than anadjectival reading).

(50) a. The wine was being uncorked.

b. The wine was being unchilled.

c. The thirst was being uncorked.

While an AMF effect was obtained for the critical semantic violation stimuli, nosuch effect was found for the well-formed and pragmatically deviant controls. Thisindicates that the AMF effect does not only reflect successful semantic composition,but more generally attempts at such a composition. Thus, these results make theAMF an even more likely candidate for a neural correlate of the linguistic operationof semantic composition.

Note, however, that caution is called for when interpreting the AMF effect asdirectly reflecting difficulties encountered during the process of semantic composi-tion. Thus, Brennan and Pylkkanen (2008) take the fact that critical stimuli such as(49a) engender an AMF effect as evidence against assuming that (at least stronglypunctual) verbs are semantically underspecified with respect to aspectual properties.In contrast, Pylkkanen and McElree (2006) had still argued that aspectual coercionin fact differs from complement coercion in that the latter, but not the former, in-volves both a semantic type-shift and a shift in ontological sort, whereas aspectualcoercion only involves a sortal shift which, however, is not semantically based. Inother words, aspectual properties were not taken to be semantically coded. In addi-tion, in their argumentation for a basic differentation between complement coercionand aspectual coercion, Pylkkanen and McElree (2006) refer to a parallel differencebetween metonymy and complement coercion. Thus – as was shown by McElree andFrisson (2006) – whereas interpreting a proper name such as Dickens metonymically(as in 51a) is not more taxing than a literal ‘person’ interpretation (as in 51b), acoerced interpretation (as in 51c) is harder than both a literal and a metonymicinterpretation.

(51) a. The gentleman met Dickens.

b. The gentleman read Dickens.

c. The gentleman began Dickens.

On the basis of such results, Pylkkanen and McElree (2006) make the assumptionthat different types of meaning shifts may involve different degrees of difficulty. Inparticular, sortal mismatches are assumed not to block initial semantic compositionof sentence meaning, whereas ontological mismatches do: ‘. . . ontological mismatch,such as [that] found in complement coercion, blocks basic composition and hence im-mediately mandates the on-line deployment of a costly coercion process (Pylkkanenand McElree 2006, p. 55).

However, the results of Brennan and Pylkkanen (2008) militate against the as-sumptions made about the nature of aspectual coercion in Pylkkanen and McElree(2006). That is, either sortal shifts in fact do block initial composition and thus leadto the on-line integration of a coercion operator, or the AMF effect observed bothfor cases of complement coercion and aspectual coercion actually is not a reflex of amore difficult composition of the semantic representation of the respective sentencesas such. If the first alternative is true, one would expect to see AMF effects also formetonymy, which was considered more similar to aspectual coercion by Pylkkanenand McElree (2006) than to complement coercion. Unfortunately, as far as I cantell, to date, potential AMF effects of metonymy have not been investigated.

Concerning the second alternative, recall that both in Dolling’s as well as inEgg’s theoretical approaches to coercion, the result of semantic composition of asentence’s meaning is in fact a highly underspecified meaning representation which,more often than not, is in need of specification by additional material provided bythe conceptual system. I find it conceivable that the AMF effect reflects the relativedifficulty of integrating such material or possibly of overriding the insertion of defaultvalues for particular variables. In Egg’s approach in particular, the AMF might betaken to signal the unavailability of a constructive solution for an underspecified

Why the Standard Notions? 75

semantic form and the search this triggers for material that might be inserted intothat form to get a solution after all.

2.5 Why the Standard Notions?

Before going on to pursue the hypothesis that, actually literal meaning and non-literal meaning are not so different in kind as traditionally assumed, there is anotherquestion connected with the traditional view that I would like to look at. That is, onemight wonder why exactly the traditional view came about in the first place and why,for many people and a long time, it seemed such a reasonable view to hold. Morespecifically, one might wonder why if literal meaning is context-dependent it stillseems possible to interpret sentences presented out of context. With regard to simpleexpressions one might ask why if literal meaning is context-dependent it neverthelessis possible to give the lexical meaning for that expression when presented out ofcontext.

Ruhl (1989) uses the notion of stereotype to explain why, for a given example, wearrive at what we think of as the literal meaning of that sentence. He argues thateven in situations where no explicit context is given, we never interpret sentenceswithout some assumptions about a possible context of utterance (cp. Searle 1978).The problem is that the consideration of these assumptions is not consciously avail-able to us. Our conscious minds lead us to think that we grasp the meaning of agiven sentence without any context whatsoever, but actually, that is not the case.At least certain background assumptions (Nunberg 1978, Searle 1978) or assump-tions about stereotypical uses of the sentence at hand are necessary to be able toassign it an interpretation. Thus, in cases of so-called ‘zero-contexts’, an expressiongets assigned a default interpretation, which is based on the assumptions that wemake in interpreting that expression. Generally, the (literal) meaning that we takean utterance to express is the result of unconscious cognitive processes, which areapplied to the semantic form of that utterance. Such processes make use of differentkinds of information, e.g. stereotypical uses of an expression, frequency of uses of anexpression in a certain reading, background assumptions that are relevant for thecurrent utterance, etc. Since these processes take place unconsciously, we assume ofliteral meaning that it is context-independent.

Thus, the reason why we interpret the sentence ‘The cat is on the mat’, asdescribing a situation as depicted in figure 2.3 is that the indicated state of affairsis a stereotypical one. This state of affairs seems to be a kind of stereotypicalinterpretation for that sentence. However, stereotypical interpretation is not to beequated with literal semantic meaning of a sentence. Apparently, what is actuallyreferred to by the former term, is a pragmatically modulated meaning of a sentence,which has gained a stereotypical status and of which our conscious mind does not‘see’ that it is not ‘isolated meaning’.

Note that in contrast to Searle, Ruhl does allow for intrasentential influenceson the pragmatically modulated word meanings. Thus, in a situation without aconcrete context of utterance, it is very likely that take gets the interpretation of‘steal’ in a sentence like in (52) due to the fact that its external argument is the NPthe thief and its internal argument the NP the jewels.

(52) The thief took the jewels.

Again, this is the interpretation of the sentence in (52) which one stereotypicallyarrives at, under the assumption that there is no given context for it. That is,‘. . . with no contrary evidence, we tend to assume a particular set of circumstances.The reasons for this assuming are largely unconscious.’ (Ruhl 1989, p. 86) However,of course the sentence could be uttered in a situation far from stereotypical. Thusconsider (53).

(53) Peter hired a gunman and a thief to do the job. After they had delivered thedocument, he met them in headquarters to give them their promised payment.The thief took the jewels. But the gunman was not satisfied with his payment.

In the case of (53), take would not receive the interpretation ‘steal’, despite thespecific NPs in its argument positions, because the sentence would not receive astereotypical interpretation. In a sense, the stereotypical interpretation of a sentenceis its default interpretation; it will only be chosen in those cases where there is nocontrary evidence to its applicability.

This view seems to go back to Putnam (1975) who argued that in explaining tosomeone the meaning of a particular expression, we give either an ostensive definitionor a description, where the latter ‘typically consists of one ore more markers togetherwith a stereotype – a standardized description of features of the kind that are typical,or “normal”, or at any rate stereotypical’ (Putnam 1975, p. 147). The features thatare central to the stereotype are such that in normal situations they are sufficient forrecognising whether some thing belongs to a certain category or not. Putnam takesa stereotype of some X to be a conventional idea that we have of X with respect towhat it looks like, or how it acts, etc. The important difference between the twoviews is that Ruhl explicitly takes stereotypes to be of pragmatic nature, whereasthis is not at all clear in the case of Putnam. Thus, he argues that we would notsay of a person that he knows the meaning of the word tiger if he did not know thatthe stereotypical tiger is striped. Interestingly, although Putnam does not give anexplicit definition of what exactly a stereotype is either, he does mention a number ofassumptions regarding stereotypes which actually are not warranted. Thus, sayingof some feature that it is stereotypical for a certain category does not warrant theinference that most members of that category actually have that feature, or that allnormal members of the category exhibit it. For this reason it makes sense to say ofstereotypes that they are conventional ideas.

Ruhl (1989) also relates stereotype to the concept prototype, arguing that thelatter is a specific kind of stereotype. Thus, the prototype is the best examplar ofa given category. The prototype is selected from the set of denotata an expressionpicks out, where the element that is selected is, in some way, seen as being superiorto the others. This is not so for stereotypes which are taken to be typical denotata.So, to make use of Cruse’s (1990) example: arguably, the prototypical mushroom iswell-formed, whereas the stereotypical mushroom is not. In relating prototypes withstereotypes, one should, however, remember that there is an important distinctionto be made between the two concepts in terms of their general nature; whereasstereotypes are generally seen as sociolinguistic concepts (cf. Putnam’s view onthe nature of meanings), prototype is applied in psycholinguistic approaches to the

Why the Standard Notions? 77

mental organisation of meaning. Thus, more needs to be said on how stereotypicalmeaning comes about and exactly how it can be defined.

Putting such questions to the side for the moment, however, from Ruhl’s pointof view, then, prototypes as well as stereotypes are pragmatic, not semantic. Anexpression’s linguistic meaning, in contrast, is very abstract and probably does notinvolve any fullfledged reading at all. This view goes well together with the generalcriticism against prototype theory as being a theory of word meaning. Thus, proto-type theory assumes that concepts have an internal structure which consists of a setof properties that are typical for members of the category picked out by the concept.The objects one encounters in the world are categorised by recognising the similar-ities between them and the concept that underlies a particular category in terms oftypical properties. However, assuming that complex expressions are composed fromsimple expressions, it is questionable whether the linguistic meanings of expressionsare structured in the way prototype theory assumes. Consider, for example, the factthat the meaning of a complex expression is not necessarily the sum of the proto-typical meanings of its constituent parts. Thus, Fodor (1998, p. 102) argues thatthe meaning of pet fish cannot be claimed to be the result of the combination of theprototype of pet and the prototype of fish. However, if word meanings were proto-typically structured, we would expect the meaning of the complex expression to bea combination of the meanings of its parts and thus as having as its prototype acombination of the prototypes of its parts. However, this is not necessarily the case.Thus, prototype theory as a theory of word meaning makes the wrong predictions.Thus, it is neither stereotypes nor prototypes that form the linguistic meaning of ex-pressions. The question then is, what status these types of meaning have: are theypragmatically derived or possibly stored as part of our encyclopaedic knowledge? Iwill take up this point below.

Turning back to the relation of (non-)literal meaning and context-(in-)dependence:there is the question of why it is that we generally assume of literal meaning thatit is context-independent, whereas we usually think of non-literal meaning as be-ing context-dependent. As already mentioned our intuition that literal meaning iscontext-independent is misleading, because even literal meaning depends on somekind of context. However, the type of context minimally required for the inter-pretation of literal meaning is rather unusual in that it does not need to be anyspecific context, but rather something that Ruhl calls stereotypical and which weapply unconsciously in the interpretation of expressions seemingly produced ‘outof context’. In contrast, our ordinary understanding of non-literal meaning is suchthat this is a very special kind of meaning, that deviates from some underlying basicmeaning and that has a secondary character. Best examples of this type of mean-ing are usually such that depend on a very specific context of utterance for theircorrect interpretation (e.g. cases of irony). However, as already indicated and as wewill see in more detail below, just as it is a mistake to think of literal meaning asstrictly context-independent, it is inaccurate not to allow for gradience of context-dependency in the case of non-literal meaning. That is, not all types of non-literalmeaning depend on context for their interpretation, to the same extent. So, thereason for assuming that literal meaning generally is context-independent whereasnon-literal meaning is generally context-dependent is an overgeneralisation, due to

our inability to consciously recognise cases of non-literal meaning if they are not asclearly non-literal as e.g. cases of irony. On the other hand, there are examples ofan expression having several possible readings within different sentences when thosesentences are produced seemingly ‘out of context’.

(54) a. Peter opened his eyes.

b. The girl opened the book.

c. Jane opened the file.

d. The boy opened the window.

(55) a. The shop assistant took the jewels.

b. The thief took the jewels.

c. He took the job.

As already said, it is questionable whether interpretation ‘out of any context’ everactually occurs. Moreover, what is underestimated in such cases is the contributionmade by the immediate sentential context to selecting a particular reading for anexpression.

2.6 Summary

In this chapter, I argued against the standard notions of literal meaning and non-literal meaning. More specifically, I argued against viewing literal meaning as es-sentially context-independent and non-literal meaning as always non-conventional.Subsequently, I discussed the consequences of these assumptions for the nature oflexical meaning, by reviewing different approaches to meaning in the lexicon andthe problems they face. More specifically, I argued for assuming underspecifica-tion of lexical meaning and a general distinction to be made between semantic andpragmatic processes. Empirical evidence supporting such a view was reviewed. Gen-erally, I hope to have shown that the standard notions are not sufficiently preciseenough to allow the terms literal meaning and non-literal meaning to be used asbasic concepts in definitions of semantic vs. pragmatic aspects of meaning. I willsummarise the arguments below.

Thus, arguments were given against the view according to which sentence mean-ing is at the same time fully propositional, context-independent and thus literal.Rather, either one takes the defining characteristic of sentence meaning to be that itis context-independent, then it cannot (always) be fully propositional, or one takesthe defining characteristic to be that of full propositionality, then sentence mean-ing cannot (always) be context-independent. Whichever alternative one chooses,the question then arises, what it is that we can apply the term literal meaningto: a sentence’s context-independent, subpropositional meaning or the propositionexpressed by a sentence when uttered? There are a few arguments in favour ofthe latter characterisation, according to which literal meaning is not equivalent tocontext-independent meaning, but rather associated with utterances. The crucial ar-gument is that what we intuitively take to be the literal meaning of some sentence,is actually determined with respect to the context in which that sentence gets ut-tered. This holds true even for sentences seemingly uttered ‘out-of-context’, where it

Summary 79

is our background knowledge and assumptions about potential situations in whichan utterance of the sentence at hand would be most appropriate that determinewhat we take to be that sentence’s literal meaning. Thus, actually, one should notspeak of ‘a sentence’s literal meaning’ at all, as it is rather utterances that may beused literally or non-literally. Thus, literal meaning, it was argued, in fact is notcontext-independent, contrary to traditional assumptions.

A related issue is the question of what types of information we take the so-calledcontext in which an utterance is made to involve. Thus, in this chapter, severaltypes of context have been vaguely differentiated: the concrete context of utterance,the context build up during forgoing discourse between participants, backgroundassumptions and aspects of world knowledge and finally the immediate sententialcontext of the elements in an utterance. However, sofar it has not been made clearwhat kinds of information or knowledge these types of contexts involve. With respectto the assumption that literal meaning actually is context-dependent, the questionis what kinds of information have to be available for a hearer to be able to interpreta particular utterance as having literal (or, for that matter, non-literal) meaning.49

Having argued for a view according to which literal meaning is actually context-dependent, it does not come as much of a surprise that in empirical studies theassumption that literal meaning always and necessarily has to be processed firstcould not be confirmed. On the contrary, the results of some experiments even showthat a potential non-literal meaning of an expression may be activated as early as,if not earlier than, that expression’s supposed literal meaning. Thus, it seems thatdifferences in the difficulty of interpreting some expression are not correlated towhether that expression was intended in a literal or non-literal meaning. Note thatunder the assumption that literal meaning and non-literal meaning both are context-dependent, one would actually expect results where there is no necessary differencein the interpretation of literal meaning as compared to non-literal meaning. In addi-tion, some of the experimental results suggest that another feature traditionally usedto differentiate literal meaning from non-literal meaning in fact does not differentiatethe two types of meaning: conventionality. Thus, the experimental results generallycall into question the standard notion of non-literal meaning as non-conventional,speaker-intended meaning that results from a reinterpretation process. If one as-sumes that conventionality is reflected in relative ease of interpretation, then one hasto allow for quite a number of conventional non-literal meanings. Also, since thereare cases in which the so-called non-literal meaning of an expression is activated asearly as, if not earlier than, that expression’s literal meaning, the assumption thatnon-literal meaning is always the result of some reinterpretation process cannot beappropriate. Moreover, since there are cases in which non-literal meanings cannotbe suppressed even though it is clear that they are not speaker intended, the as-sumption that non-literal meaning generally is a feature of speaker meaning cannotbe sustained. In addition to empirical evidence, we saw that there are also theoret-ical reasons for assuming that conventionality is a gradual property. Thus, such anassumption makes the description of semantic change much more straightforward,since such changes do not take place abruptly, but rather over a period of time.

The revised characterisations of literal meaning and non-literal meaning have

49This point will be taken up throughout the discussion both in chapter 3 and chapter 4.

consequences for the assumptions made about the nature of lexical meaning. Thus,the meaning(s) stored in the lexical entry of some expression can no longer be viewedas literal, since literal meaning is taken to be context-dependent, whereas lexicalmeaning is not. Several reasons were given that actually make this a desirableoutcome. First, if one wants to assume that all an expression’s literal readings,i.e., those which seem to be basic, are coded in that expression’s lexical entry, onehas to also necessarily assume that the lexicon as such is characterised by a highdegree of polysemy and (if one tends to extremes) as almost indefinite in size. Ifone wants to assume of the lexicon that it is highly economic and minimal, this isan unsatisfactory result. Second, listing all possible literal meanings an expressionmay have in its lexical entry might be problematic, on the one hand due to thesheer number of potential literal readings (cf. readings for the verb to open), andon the other hand, since one misses generalisations (such as the similarities betweenpossible literal readings for expressions like newspaper, university, book etc.) bytreating certain meaning aspects as idionsyncratic when they are not. Third, if onlyliteral meanings are coded in the lexicon and if the disambiguation process first hasto go through all potential literal sentence meanings, rejecting them as not fittingthe context, one would not expect to get evidence of an early activation of non-literalmeaning, as this would be the result of some process that follows the unsuccessfuland protracted disambiguation process.

From the various approaches to meaning in the lexicon reviewed, it seems that therevised characterisations are best reflected by an approach that assumes a high de-gree of underspecification for lexical meaning. Generally, one important aspect thatmakes such an approach superior to the maximalist and intermediate approaches isthe fact that the minimally specified semantic form allows for both literal as wellas non-literal ‘completion’. Thus, such an approach, generally, is compatible withthe empirical results according to which under particular circumstances non-literalmeaning may be interpreted as fast as – if not faster than – the literal meaning ofthe utterance in question. Another important advantage of an underspecificationapproach is that it can explain the differences in processing of homonymous vs. pol-ysemous expressions. More specifically, an underspecification approach can explainthe pecularities found for the processing of polysemous expressions. Moreover, it isalso compatible with the assumption of some type of default readings for an expres-sion, since these, as noted above, seem to be part of a level distinct from the purelysemantic, that is, context-independent level. Thus, literalness neither is a propertyof lexical, nor of sentence meaning. Rather, these latter two types of meaning arecharacterised by being very abstract and heavily underspecified, generally allowingfor a literal or non-literal interpretation/usage.

As a conclusion to this chapter, the following points can be made. Both theconcepts of literal meaning and non-literal meaning in their traditional characteri-sation are not precisely defined. Moreover, the properties captured by the standardnotions of literal meaning and non-literal meaning are problematic at the least.The evidence gathered so far suggests that literal meaning and non-literal meaningshould not be characterised in terms of dichotomies, as they do not seem to be sovery different from one another as traditionally assumed. Still the question remains:what is literal meaning and what is non-literal meaning? Put differently, how are

Summary 81

the processes characterised that lead to what we intuitively call literal meaning ornon-literal meaning? Recall the fact mentioned in chapter 1 that the term literalmeaning traditionally has not only been employed for the level of sentence mean-ing or lexical meaning, but also to describe a particular type of context-dependentmeaning, namely the level of what is said. Thus, in the next chapter, we will takea closer look at the characterisation of this particular level of meaning and whereliteral (and other types of) meaning aspects might be found at it.

Chapter 3

Utterance Meaning and theLiteral/Non-literal Distinction

In chapter 2, I argued for a characterisation of the terms sentence meaning and literalmeaning along the following lines. Whereas sentence meaning is compositional,context-independent and (more often than not) sub-propositional, literal meaningis a particular type of meaning a sentence (or complex expressions generally) mayhave when used in a context. Thus, literal meaning is a term applicable at a level ofmeaning, at least partly resulting from context-sensitive processes. The question iswhich factors have to hold for the thus resulting meaning to be literal. Before thisquestion can be answered, however, the more basic question needs to be addressedof how such a context-sensitive level of meaning generally is characterised, that is,which processes are characteristic for this level of meaning and how it is distinguishedfrom other levels of meaning.

Thus, in this chapter, I first want to review two particular approaches to thedifferentiation of levels of meaning in interpretation, namely those of Grice andBierwisch (section 3.1). Apart from a level of context-independent meaning, bothapproaches actually differentiate two levels of meaning that can be characterised asbeing context-sensitive: what is said (Grice) or utterance meaning (Bierwisch) onthe one hand, and what is meant (Grice) or communicative sense (Bierwisch) onthe other. For the purposes of the present chapter, we will especially concentrateon the characterisations of the level of what is said or utterance meaning in the twoapproaches. However, since assumptions about the relation of this level of meaningto the level of what is meant/communicative sense will be important for the discus-sion in chapter 4, we will look at those as well. The bigger part of this chapter isdedicated to the discussion of various alternative approaches to the characterisationof the level of what is said/ utterance meaning and the particular processes that areassumed to contribute to it. As we will see, the discussion mainly centers aroundthe issue of whether the processes contributing to what is said/utterance meaninghave to be linguistically mandated and whether they should be taken to be inde-pendent of speaker intentions (section 3.2.1 – 3.2.3). In a sense, the issue here is notonly the nature of the processes used in the interpretation of utterances, but alsothe nature of the context against which this interpretation takes place. A relatedissue is the question of whether it is possible and useful to differentiate another

84 Utterance Meaning and the Literal/Non-literal Distinction

context-dependent level of meaning apart from what is said/utterance meaning. Inother words, can ‘saying’ be clearly differentiated from ‘meaning’ and if so, by whichcriteria (section 3.2.4). This point, then, leads over to chapter 4.

3.1 Levels of Meaning

In this section, I want to take a closer look at two approaches to the distinctionof levels of meaning, namely those proposed by Grice and Bierwisch, respectively.Although Grice’s distinction between what is said and what is meant is probablybest known, this is not to say that Grice generally only differentiated between twolevels of meaning. In fact, in trying to specify what exactly his notion of ‘saying’was, he considered a fourfold distinction of levels or types of meaning. Similarly,Bierwisch also differentiates more than just two levels of meaning. Moreover, heand Grice make a number of similar assumptions. However, as we will see, theirapproaches also differ in some crucial aspects.

3.1.1 Grice’s Four Types of Meaning

Grice (1975) distinguishes between two general levels of meaning, namely what issaid and what is meant. What is said is the truth-conditional meaning of complexexpressions as generated by semantic composition plus such information as is theresult of a number of specific processes, namely resolving reference, fixing indexicalexpressions and disambiguation. What is meant is the meaning the speaker intendedto communicate in making a particular utterance. As noted already, Grice does notcall the processes that lead to what is said pragmatic, rather, he reserves this latterterm for such processes as lead to what is meant and are based on the co-operativeprinciple and the maxims of conversation. In his view, pragmatic processes operateon what is said and deliver meaning aspects that are implied rather than ‘said’ andthat together with what is said result in what is meant by a speaker of a particularutterance. Thus, what is said is the proposition explicitly expressed by an utterance.It forms the basis for drawing further inferences (e.g. conversational implicatures),that is, for determining the implicit meaning content of an utterance in a particularcontext. What is meant, on the other hand, comprises both explicit as well asimplicit meaning aspects. Moreover, in contrast to the level of what is said, whichis taken to be independent of speaker intentions, for the determination of what ismeant speaker intentions are essential. Note that although speaker intentions do notplay a role for determining what is said by an utterance, Grice still assumes thatwhat is said by an utterance forms part of what is meant by a speaker.

Thus, on a first blush, it seems as if Grice only identified two levels of meaningas playing an important role in the overall interpretation process: what is said andwhat is meant. However, as Grice himself points out, he uses saying in a ‘favoured’,somewhat ‘artificial’ sense (Grice 1989, p. 118). One of the aims of his investigationsis to make it clear how exactly this sense (and the derived notion of what is said)is to be understood. Thus, Grice (1969, 1989) actually contemplates four types orlevels of meaning: timeless meaning of utterance-types, applied timeless meaning ofutterance-types, occasion meaning of utterance-types and utterer’s occasion meaning.

Levels of Meaning 85

sentence meaning

what is said

Pragmatics

what is meant

sub-propositional

fully propositional

+ narrow context

+ broad contextis part of

Figure 3.1: Grice’s distinction of what is said and what is meant more detailed

To illustrate the differences between these types of meaning, Grice (1969, 1989) usesthe following example sentence.

(56) If I shall then be helping the grass to grow, I shall have no time for reading.

The timeless meaning of utterance-types1 (x [utterance-type] means ‘. . . ’) can bestated both for complete as well as incomplete utterance-types, that is, both forsentences as well as phrases or words. Thus, in the case of (56), there are (at least)two possible specifications of the timeless meaning of a (complete) utterance-type,namely: ‘If I shall then be assisting the kind of thing of which lawns are composedto mature, I shall have no time for reading’ and ‘If I shall then be assisting themarijuana to mature, I shall have no time for reading’. Similarly, for the expressiongrass, the timeless (incomplete) utterance-type meaning can be specified: one of itsmeanings being something like ‘lawn-material’, the other ‘marijuana’.

As the example in (56) shows, a complete utterance-type x may have more thanone timeless meaning. However, one would assume that (at least in most circum-stances) for a particular utterance of x, x only carries one of those possible meanings.Thus, the applied timeless meaning of utterance-types (x [utterance-type] meant here‘. . . ’) specifies what x means on this particular occasion; in the case of (56) thisbeing: ‘If I shall then be assisting the kind of thing of which lawns are composed tomature, I shall have no time for reading’ and in the case of grass ‘lawn-material’.

The occasion meaning of an utterance-type (U meant by x [utterance-type] ‘. . . ’)is supposed to capture what a particular utterer (U) meant by the words of (56); forexample: (i) ‘If I shall then be dead, I shall not know what is going on in the world’and possibly (ii) ‘One advantage of being dead will be that I shall be protected fromthe horrors of the world’2. If a particular U meant by (56) (i), he would certainly

1Note that by utterance-type Grice not only referred to sentences, phrases or words; rather,he intended his explications to apply to non-linguistic utterance-types as well. However, in whatfollows, I will only mention specifically linguistic utterance-types.

2Note, however, that in the original typescript of the William James Lectures, Grice writes of(ii) that one would maybe prefer that the utterer implied it, which means that it actually belongsto the utterer’s occasion meaning.

also be taken to have meant by the words: ‘I shall then be helping the grass togrow’, ‘I shall then be dead’. However, what is important to note here is that since‘helping the grass to grow’ is not a recognised idiom, the specifications given for theoccasion meaning, or of what U meant by (56), are neither admissible specificationsof the timeless meaning, nor the applied timeless meaning of (56). As Grice putsit: ‘The words “I shall be helping the grass to grow” neither mean nor mean here“I shall be dead.” ’ (Grice 1989, p. 90, emphasis in the original). Note that here,although Grice does not use the terms, there is a differentiation between what anutterance-type literally means, contrasting to what a speaker might mean by it.

The utterer’s occasion meaning (U meant by uttering x that . . . ) captures thefact that there is a difference in that U meant by (56) (i), whereas in uttering (56)U meant that if he would then be dead he would not know what was going on in theworld. Additionally, U, in uttering (56) meant (or conversationally implicated) thatone advantage of being dead would be that he would be protected from the horrors ofthe world. As Wharton (2002) notes, the difference between the occasion meaning ofan utterance-type and utterer’s occasion meaning is very subtle and it is not entirelyclear what exactly Grice wanted to capture by this notion. Both types of meaningapparently involve speaker intentions (‘U meant by x’ vs. ‘U in uttering x meantthat’). The only difference between them seems to be that the latter additionallycomprises aspects of meaning that are contributed by conversational implicatures.

Thus, it seems that for Grice the applied timeless meaning of utterance-typescorresponds to what is said.3 It is the level at which disambiguation, the resolvingof indexical expressions and reference assignment has taken place. What is meant,then, corresponds to the utterer’s occasion meaning, as it includes such meaningaspects as conversational implicature. Moreover, Grice also recognises a type ofmeaning that is totally independent of any context of utterance, namely the timelessmeaning of utterance-types. From his characterisation of this latter type or level ofmeaning, it can be assumed that he took this to be the level at which literal meaning– in the traditional sense of conventional meaning – is to be found. In addition, Gricedistinguishes a further type of meaning, namely occasion meaning of utterance-type.However, it is not quite obvious in which respect this is supposed to differ fromutterer’s occasion meaning.

As already hinted at, a problematic issue with respect to the relation of whatis said to what is meant is that Grice actually assumes that the former constitutesa part of the latter. However, there seems to be an important distinction betweensaying something and meaning something in that a speaker in uttering a certainlinguistic string can be saying one thing, but might mean more than or somethingdifferent from what is included in the meanings of the linguistic expressions she usesin her utterance. Thus, there are cases, as in example (56) above, where it seemsthat a speaker means something quite different from what she literally says. Thisis especially obvious in instances of irony such as in (57), where, very crudely, whatcan be assumed of the speaker is that she meant something like ‘the opposite’ ofwhat she said.

(57) (ironically) It’s a great day for a picnic.

3See, however, Wharton (2002), who argues that from the way Grice differentiated the fourtypes of meaning, he might have had another notion of saying (and thus, what is said) in mind.

Semantics (simple) timeless meaning

(complex) timeless meaning

applied timeless meaning

Pragmatics

utterer's occasion meaning

?+ narrow context

+ broad context

Figure 3.2: Grice’s types of meaning and their relation to what is said/what is meant

On Grice’s analysis, this leads to the problem that since the speaker of (57) ap-parently did not mean what her words mean, she has not actually said anything.Similar problems arise in case of metaphor. Thus, the speaker of (58) cannot literallybe taken to mean what his utterance expresses. But neither is it ‘the opposite’ ofwhat his utterance expresses that is what is intended here. In any case, what thespeaker means, again, is not what her words mean, so once again she has not saidanything.

(58) You are the cream in my coffee.

Examples such as in (57) and (58) force Grice to differentiate between saying andmaking as if to say. Thus, the speaker in uttering (57) does not ‘say’ the propositionexpressed by her utterance. (Carston 2002c) points out that the move from ‘saying’to ‘making as if to say’ is highly problematic. To give only one of the problems shementions:

‘The Gricean account of the tropes depends on the flouting of the firstmaxim of quality: do not say what you believe to be false. However, inthese cases nothing has been said so, trivially, the maxim is not violatedand the account on which an implicature is required in order to preservethe supermaxim of quality (try to make your contribution one that istrue) cannot get off the ground . . . ’ (Ibid., p. 115; emphasis in theoriginal).

Similarly, Bach (1994b) wonders why Grice analysed irony and metaphor in terms ofconversational implicature at all, precisely because in their case he speaks of ‘makingas if to say’. Moreover, Bach (1994b, p. 272) argues that ‘[i]ntuitively, one thinksof implicating as stating or meaning one thing . . . and meaning something else aswell, not as meaning something else instead’, as is the case in both metaphor andirony. Note also that Grice’s treatment of metaphor and irony suggests that thesetwo phenomena are similar in terms of the effort involved in their interpretation. Inboth cases, the assumption is that the literal meaning of the sentences used in making

a metaphorical or ironic utterance, is rejected as not fitting in the context and therespective conversational implicatures are drawn instead. However, as suggestedby the results of empirical studies, it seems that the interpretation of metaphor andirony actually differs in terms of cognitive effort (e.g. Happe 1993, Colston and Gibbs2002, Eviatar and Just 2006). Considering the problems arising from an analysisof metaphor and irony in terms of conversational implicature, these two types ofnon-literal meaning should be – and as we will see, actually have been – analyseddifferently (see sections 4.1.1 and 4.1.2).

Another aspect that might be viewed as being problematic has to do with thenature of conversational implicatures. Here, Grice distinguished two basic types.Thus, whereas particularised conversational implicatures (PCI) depend on the spe-cific context in which an utterance is made, generalised conversational implicatures(GCI) seem to be drawn in any kind of context, unless the latter is such that itprevents it.4 Thus, one interesting aspect of GCIs is the fact that they are indepen-dent of the speaker’s intention in making an utterance (cp. Burton-Roberts 2006).In other words, if the context of some utterance is such that it allows a GCI tobe drawn, the speaker of that utterance actually would have to make it explicit ifhe did not want the hearer to draw this inference. This is an interesting fact tonote since Grice identified conversational implicatures in general as belonging to thelevel of utterer’s occasion meaning, which is characterised as the level of meaning atwhich speaker intentions come into play. It might be for this reason that Neo- andPost-Gricean theories of interpretation handled this phenomenon differently, bothfrom Grice as well as from one another (see section 4.1.3).

3.1.2 Bierwisch’s Three Levels of Meaning

Bierwisch (1979, 1983) differentiates between three levels of meaning: expressionmeaning, utterance meaning and communicative sense. The expression meaningSEM (E) is the linguistically determined meaning of an expression E.5 For a complexexpression, its expression meaning is compositionally build up from the meaningsof the simple expressions contained in it and their syntactic combination. That is,meaning on the level of expressions is the context-independent meaning of simpleand complex expressions. Thus, similarly to Grice, Bierwisch assumes a level ofmeaning that is fully linguistically determined and therefore context free.

The utterance meaning M (t) is the meaning an utterance token t of an expressionE has when it is used in a context C. The utterance meaning M (t) can be equivalentto the utterance token’s literal meaning, but it does not necessarily have to be.Thus, contrary to Grice and similarly to Searle, Bierwisch assumes that what iscalled the literal meaning LM (t) of an utterance token t of a simple or complexexpression E is not identical to the linguistically determined meaning SEM (E) of

4Thus, the GCI (ii) of the utterance in (i) is only drawn, if the hearer does not already knowthat (ii) is not the fact.

(i) Some of the students went to the party.(ii) Not all of the students went to the party.

5Note that in different articles, Bierwisch uses different abbreviation terminology. For the sakeof consistency, I shall use one set of terminology throughout.

that expression. Rather, an utterance token of an expression can only have a literal(or, for that matter, non-literal) meaning in a context. Thus, literal meaning LM (t)is a special case of utterance meaning M (t). As already mentioned, a consequenceof this view is the assumption that the lexical semantic representations of simpleexpressions do not encode what we take to be their literal meaning, rather, theirsemantic representations are maximally underspecified. Thus, in contrast to Grice,Bierwisch assumes that the level of expression meaning is characterised by a highdegree of underspecification.

The third level of meaning Bierwisch differentiates is that of the communicativesense CS (t) of an utterance token t of an expression E.6 The communicative sensecaptures what a speaker meant by the utterance of an expression. As such, it includesimplied meaning aspects such as conversational implicature. Thus, this third level ofmeaning can be equated with Grice’s level of utterer’s occasion meaning. Moreover,similarly to Grice, Bierwisch reserves the term pragmatic for processes that leadfrom the level of utterance meaning to that of communicative sense.

Semantics (simple) expression meaning

(complex) expression meaning

utterance meaning

Pragmatics

reference resolutionfixing indexicalsdisambiguation. . .

communicative sense

ConceptualSystem

Figure 3.3: Bierwisch’s three levels of meaning

Bierwisch (1979) illustrates the three levels of meaning by way of the examplesentence in (59).

(59) Das habe ich nur mit der linken Hand gemacht.

‘I made this only using my left hand.’

A speaker uttering a token of this sentence, might mean with it one of the following,depending on the context of utterance:

6With respect to the discussion so far, the term Two-Level-Semantics which came to be appliedto Bierwisch’s particular approach seems rather misleading since he actually differentiates threelevels of meaning. (see Taylor 1994, 1995). However, what that term does reflect is the fact thatBierwisch differentiated two types of cognitive structure: semantic structure (SS) and conceptualstructure (CS), more on which see section 2.3.2.

(60) a. Ich bin stolz, dass ich eine so komplizierte Handarbeit sogar mit der linkenHand anfertigen kann.

‘I am proud that I can produce such a complicated piece of handicraft, even with

my left hand.’

b. Ich entschuldige mich fur die Fehler der Handarbeit, die daher ruhren, dassich sie mit der linken Hand anfertigen musste.

‘I apologise for the defects of the piece of handicraft, which result from the fact

that I was obliged to produce it with my left hand.’

c. Von diesem Artikel halte ich nicht viel, denn ich habe auf seine Anfertigungnicht viel Muhe verwendet.

‘I don’t think much of this piece of handicraft, because I didn’t pay much atten-

tion to its production.’

What is given in (60) are three of the possible communicative senses CS (t) of therespective utterance tokens t of (59). Bierwisch claims that for the production ofthe CS (t) s in (60a) and (60b), (59) is used in its literal meaning, whereas for (60c)it is not. A rough paraphrase of the literal meaning of (59) in the context in whichit is used to communicate (60a) and (60b) is given in (61a); the non-literal meaningwith which (59) is used to communicate the CS (t) (60c) is given in (61b), in bothof which X represents the result of the production.

(61) a. Ich habe X nur mit der linken Hand gemacht.

‘I produced X only with my left hand.’

b. Ich habe X nur nebenbei gemacht.

‘I produced X without paying much attention to it.’

Obviously, the meanings given in (61a) and (61b) are two possible utterance mean-ings M (t) for utterance tokens t of the sentence in (59) in a context C. As alreadymentioned, the utterance meaning of an utterance exemplar t can be equivalent toits literal meaning with respect to C, as in (61a), but does not have to be, as in(61b). Thus, (61a) is the literal meaning LM (t) of an utterance exemplar t of thesentence in (59). t in turn is an utterance exemplar of an expression E, which has alinguistically determined meaning SEM (E).

As mentioned above, Bierwisch assumes of expression meaning that it is charac-terised by underspecification. One advantage of such an approach is that it allowsa non-lexical treatment of systematic polysemy as well as of the non-systematicprimary meaning variation of verbs. Thus, Bierwisch specifically argues for two pro-cesses he calls conceptual shift and conceptual differentiation. Whereas the formerleads to the different readings such nouns as newspaper in (13) can have, the latteraccounts for the different readings of verbs like open in (20).

(13) a. The newspaper was on the table.

b. The newspaper was censored, yesterday.

c. The newspaper was founded three years ago.

b. John opened the exhibition.

c. John opened the file.

d. John opened the parcel.

e. John opened the store.

To repeat, the SEM of a lexical item like newspaper determines a family of concepts aspossible readings for it. As the SEM is underspecified, there is no primary meaningvariant stated in the lexical entry of newspaper. Depending on the context in whichnewspaper is uttered, an interpretation function chooses one member of the family ofconcepts as the reading in that context. In a sense then, the interpretation functionshifts between different concepts, according to the relevant context. Thus, Bierwischcalls this type of interpretation process conceptual shift.

In contrast, conceptual differentiation is a process in which an unspecified SEMvariable is given a specific value. The variable is bound by an existential quantifier,which determines a specific variation possibility by inserting an appropriate entityfrom the interpretation context. Each specification of the variable within the contextresults in a possible differentiation of the meaning of the respective word.

In contrast to expression meaning, the meaning of an utterance can be examinedas to whether it is true or false. However, specific kinds of meaning transfer mightbe necessary to arrive at the utterance meaning.7 Thus, utterances such as in (15a)and (18) can only be assigned a truth value after a metonymic shift of the meaningof the italicised expressions has taken place.

(15a) The newspaper called.

Thus, Bierwisch assumes that an expression gets assigned its literal or, if appli-cable, its non-literal meaning only at the level of utterance meaning. Under thisview, systematic polysemy, the differentiation of meaning variants of verbs as wellas metonymic meaning shift are all phenomena occurring at the level of utterancemeaning. However, the last phenomenon is treated differently from the first two inthat a metonymic meaning shift presupposes an already determined meaning thatcan be shifted.

Note that Bierwisch recognises a further type of process, namely selection, whichoperates in cases where in determining the meaning of an utterance both processesof conceptual shift and conceptual differentiation apply but produce more than onepossible reading for the utterance.

(62) John left the university.

Selection will consider the broader context of utterance in order to disambiguatewhether (62) is interpreted as meaning ‘John moved his position from inside the

7Cp. Nunberg (1995, 2002)

university building to outside of it’ or ‘John severed his affiliation with the university-as-institution’. The nature of this process raises the question of how much and whatkind of contextual information is available to the processes of conceptual shift andconceptual differentiation in the first place. Apparently, the kind of contextualinformation that these processes can make use of is restricted, as otherwise theprocess of selection would be superfluous.

Incidentally, Bierwisch assumes that at the level of utterance meaning, speakerintentions do not yet play a role. They are necessary only when it comes to deter-mining the communicative sense of an utterance. This reflects his very traditionalview of pragmatics, as being concerned only with phenomena such as speech actsand conversational implicature, which he takes to hold at the level of communicativesense. Thus, the processes which lead from the level of expression meaning to thatof utterance meaning Bierwisch calls conceptual, not pragmatic. Note also that al-though Bierwisch allows for such non-literal meaning aspects as metonymy to figureat the level of utterance meaning, it is not quite clear whether he would also assumemetaphors to be already determined at this level of meaning.

In contrast to Grice, Bierwisch does not assume that utterance meaning necessar-ily forms part of the communicative sense. However, although in principle utterancemeaning and the communicative sense of an utterance may be independent of eachother, in the normal course of communication, an utterance’s meaning does serve asthe basis for a recovery of what the speaker meant. Thus, in example (59) above,Bierwisch assumes that the possible communicative senses in (60a) and (60b) arebased on the utterance meaning in (61a), whereas the possible communicative sense(60c) is based on the utterance meaning as given in (61b). The question here, onceagain, is whether deriving the utterance meaning in (61b) includes the derivationof the literal utterance meaning in (61a) as a necessary intermediate step. Fromwhat Bierwisch says about literal and non-literal utterance meanings, this does notseem to have to be the case. Moreover, it should be noted that since there is nodirect relation between the expressions used to convey an intended meaning and theintended meaning as such, it is of course possible that a speaker could have conveyedhis intended meaning by the use of other expressions. That is, it has to be madeclear what it means for the CS of an utterance to be based on that utterance’s M.

3.1.3 Summary

In this section, two different views on the differentiation of levels of meaning wereintroduced. Grice differentiated between four levels of meaning, timeless meaningof utterance-type, applied timeless meaning of utterance type, occasion meaning ofutterance-type and utterer’s occasion meaning.8 On Grice’s differentiation, literalmeaning is determined by the language system, that is, it is already determined atthe level of timeless meaning of utterance-type and therefore totally independent ofany context. At the level of applied timeless meaning of utterance-type, contextualfactors of the utterance of an expression are used, on the one hand, to disambiguate

8However, as mentioned above, what Grice intended the notion of occasion meaning ofutterance-type to capture as different from utterer’s occasion meaning, is far from clear. Therefore,in my summary, I will only refer to the notion of utterer’s occasion meaning.

between different possible literal meanings and on the other hand, to fix indexicalsand assign reference. To get to the level of utterer’s occasion meaning, the speaker’sintentions in making a specific utterance are considered, leading to what he meantin uttering it. Furthermore, based on a number of conversational maxims, at thislevel, implicit aspects of the intended meaning of an utterance are inferred. Thus,the level of applied timeless meaning of utterance-type corresponds to Grice’s whatis said, which he takes to be the basis for drawing further pragmatic inferences,leading to what is meant, which, in turn, can be assumed to correspond to Grice’sutterer’s occasion meaning.

Similarly to Grice, Bierwisch also assumes more than two levels of meaning,namely the levels of expression meaning, utterance meaning and communicativesense. Whereas expression meaning is that level of meaning, which is solely de-termined by the language system, it is not the level at which the literal meaningof an expression is determined. That is, Bierwisch agrees with Grice in assuminga specifically context-free, linguistically determined level of meaning, however, healso agrees with Searle in assuming that the determination of the literal meaning ofan expression is not free of extra-linguistically determined factors. Rather, literalmeaning is determined at the level of utterance meaning, where contextual factorsare considered. As we saw, this view allows a non-lexical treatment of systematicpolysemy and the (non-systematic) meaning variation of verbs, which is not onlypreferable if one wants to assume an economical lexicon, but which also seems tobe more in line with experimental results concerning lexical access of the respectiveexpressions. Furthermore, Bierwisch differs from Grice (and from Searle, for thatmatter) in assuming that just like the literal meaning of an expression, its possiblenon-literal meanings (at least such as idiom and metonymy) can be determined atthe level of utterance meaning as well, due to the fact that these are independentof the speaker’s intention behind the utterance of that expression (cf. the differencebetween the utterance meaning (61b) of (59) and the related communicative sense(60c)). Although more will have to be said concerning this issue, generally speaking,such a viewpoint seems reasonable as it can be used to explain experimental resultswhich show that non-literal meaning is not harder to process than literal meaning.Thus, Bierwisch’s characterisation of the level of utterance meaning and its relationto the other meaning levels identified seems to best capture the data and empiricalresults reviewed so far.

Both Grice and Bierwisch reserve the term pragmatic for the processes thatcontribute further inferences based on what is said/utterance meaning and leadingto what is meant/the communicative sense. In fact, one could argue that undersuch a view pragmatics is not simply concerned with context-dependent information,but rather with such information as is determined based on a consideration of thespeaker’s intention in making a particular utterance. Thus, recall that for bothauthors, speaker intentions only play a role after what is said/the utterance meaninghas already been determined. However, as we will see below, there are argumentsthat such a view of the role of speaker intentions in the overall interpretation processis too restrictive. More specifically, it seems that speaker intentions already have tobe considered in order to resolve reference and fix indexical expressions.9 Thus, I

9Cp. Kaplan (1989a) on the role of directing intentions for fixing the referent of true demon-

will continue to use the term pragmatic to refer to processes and meaning aspectsthat essentially are context-dependent, unless indicated otherwise.

There is a great deal of discussion concerning the level of what is said as Gricecharacterised it, since this level results from an interaction of the semantics com-ponent and pragmatic10 system. Therefore, it is taken to reflect the semantics/pragmatics interface, the nature of which is highly controversial. Thus, in the nextsection, we will be concerned with various problematic aspects that have been raisedwith respect to Grice’s level of what is said, most importantly the claim that thislevel is not rich enough in meaning to form the basis from which further pragmaticinferences are drawn. Note also that although Bierwisch’s characterisations of thelevels of expression meaning and utterance meaning seem to better capture the dataand empirical results reviewed above, generally his level of utterance meaning facesthe same problem as Grice’s, when it comes to forming the basis for pragmatic pro-cesses leading to communicative sense. It is important to consider these problemsas we have to be clear about the character of what is said/utterance meaning beforewe can consider the question of which of the processes involved in determining itactually lead to what we would like to call the literal meaning of an utterance.

3.2 The Problem of Characterising the Level of

Utterance Meaning

Taking Grice’s (characterisation of the) level of what is said as a starting point,there actually are a number of divergent views on exactly how the level of utterancemeaning11 should be characterised and what kinds of processes take place whenmoving from expression meaning to utterance meaning. Thus, as we will see, thereis a lot of discussion about whether only such processes should be assumed thatprovide the values to variables already present in the semantic form of an utterance(i.e. linguistically mandatory processes; roughly Grice’s view), or whether, in orderto account for the facts, processes should be allowed that add pragmatic materialfreely. This issue is closely connected with the view one takes on how much and whattype of contextual information is available when moving from expression meaningto utterance meaning. Both in Grice’s and Bierwisch’s approaches, it can be argued,context is taken to encompass such information as identified by Kaplan (1989b) to

stratives.10in the sense used here11At the risk of seeming inconsistent, in this section, I use the terms expression meaning and

utterance meaning with a less specific characterisation than intended by Bierwisch. Thus, ex-pression meaning is the context-independent meaning of simple and complex expressions, with noassumptions made as to whether this level of meaning is characterised by a high degree of under-specification. Utterance meaning is that level of meaning which forms the basis for drawing furtherpragmatic inferences. As such, I assume that it is propositional, however, there are no particularassumptions made as to the kinds of processes involved in determining this level. My reason fordoing so is that the different approaches I discuss here all assume (slightly) different character-isations for the level of meaning that forms the basis for further pragmatic inferences and theyalso have different views concerning the exact nature of the level of context-independent meaning.Therefore, I want to be able to refer to those two general levels of meaning without yet makingcommitments as to their specific aspects.

The Problem of Characterising the Level of Utterance Meaning 95

be necessary for fixing the reference of indexicals. That is, the speaker/produceras well as the hearer/addressee of the utterance, the time and place at which theutterance takes place and the world of the utterance have to be (mutually) known.This particular set of information is often called narrow context. It differs from broadcontext in that the latter additionally includes information resulting from reasoningabout the speaker’s intention in making the particular utterance he did. Rememberthat Grice as well as Bierwisch assumed speaker intentions to be relevant only atthe level of what is meant/communicative sense. However, a case can be made thatspeaker intentions already have to be taken into account when resolving the referenceof indexical expressions, which would mean that broad context already is at play indetermining utterance meaning.

In section (3.2.1), I will first look at RT’s notion of explicature, which is intendedto replace Grice’s what is said as the level of meaning that forms the basis for furtherpragmatic inferences. As we will see, Bach’s notion of impliciture is very similar toRT’s explicature, however, the two approaches differ in some crucial aspects. Insection (3.2.2), I will discuss Stanley’s characterisation of what is said as resultingexclusively from linguistically mandated processes. Section (3.2.3) is a discussion oftwo notedly semantic approaches – that of Borg and that of Cappelen and Lepore –to the determination of fully propositional sentence meaning as opposed to the twoapproaches discussed in section (3.2.1), which both assume that sentence meaning(more or less often) is sub-propositional. Finally, section (3.2.4) reviews a numberof experimental studies aimed at answering the question whether or not the levelof what is said as characterised by Grice has any significant role to play in theinterpretation process.

3.2.1 Explicit/Implicit Meaning

Explicatures

Recall that although the processes of resolving reference, fixing indexicals and dis-ambiguation clearly require contextual information, Grice apparently did not thinkof them as being pragmatic in nature. However, as has been pointed out especiallyin the relevance-theoretic literature, these processes are similar in nature to the pro-cesses that result in the derivation of such phenomena as conversational implicature(e.g., Sperber and Wilson 1995, Carston 1999, 2002a,c, 2004a). Thus, these twotypes of processes cannot be differentiated on the grounds that one type is prag-matic (i.e., context-dependent) in nature whereas the other is not. Another criterionhas to be found. The criterion used by Relevance Theory is that of explicitness ofmeaning. That is, pragmatic processes are differentiated as to whether they resultin explicitly or in implicitly expressed meaning. Thus, on the one hand, there arepragmatic processes that apply to the expression meaning of an utterance and re-sult in a fully propositional form. The propositional form that is derived by theseprocesses is the meaning which is explicitly expressed by an utterance (or rather bythe speaker of that utterance, as we will see). On the other hand, there are prag-matic processes that apply to the fully propositional form of the utterance of someexpression and derive further pragmatic inferences, which are taken to be implicitlyconveyed by that particular utterance of the expression. That is, what is explicitly

expressed by an utterance is the basis for inferring what is implicitly conveyed. Be-fore going into more detail concerning the difference between explicit and implicitmeaning in Relevance Theory, let us first look at some basic assumptions made inthat approach.

Relevance Theory (RT) basically differentiates two kinds of processes: the decod-ing process of the language system and the pragmatic processes of inference. Thisdifferentiation reflects the understanding of the semantics/pragmatics distinction inRT. Semantics deals with linguistically encoded meaning, totally free from contextand context invariant. Pragmatics, essentially, is the recovery of speaker meaningand is entirely context-dependent. Thus, semantics is nothing more than the sourceof evidence required by the pragmatic system in order to achieve an interpretationof an utterance. The distinction is not one of representational format, that is, thereare no pragmatic representations as such, only pragmatic processes for the deriva-tion of meaning representations. The output of representations from the languageprocessor, the semantic representations make no claims about language externalreality. They simply are of the appropriate format that allows ‘. . . for integrationwith representations from other information sources’ (Carston 2004a, p. 820). Asa result of the integration processes, fully propositional representations are derived.These full propositions, then, may represent possible circumstances and can thus bejudged against an external reality concerning their truth or falsity. From a relevance-theoretic point of view, then, the only semantic part of an expression is its semanticform, which is the output of the context-independent linguistic encoding, but noth-ing of which could be said that it was the intended speaker meaning. In deliveringthe schematic, sub-propositional, decoded SF of an utterance, the semantics com-ponent is exhausted. This SF is then supplemented with contextual informationsupplied by the pragmatic system (cf. Carston 2004a).

What is interesting here is that RT assumes that more processes have to takeplace when moving from SF to an utterance’s meaning than those traditionallyassumed by Grice. Thus, there are cases in which a saturation process has to supplymissing material before the utterance of some sentence can be fully propositional.Thus, saturation ensures that utterances of sentences as in (63) are understood bya hearer as expressing propositions.

(63) a. Paracetamol is better. [than what?]

b. It’s the same. [as what?]

c. He’s too young. [for what?]

d. She’s leaving. [from where?]

Similarly to the process of fixing indexicals, saturation makes use of contextualinformation. For the examples in (63) this means that the missing material askedfor in the brackets, is supplied. Moreover, the assumption is that, like the former,the latter process is linguistically mandated, namely by some variable present in thesemantic form of the sentence, which is assigned a value during interpretation.

Assuming that the most important function of the level of meaning Grice iden-tified as what is said is to form the basis for further pragmatic inferences, RT notesthat what is usually taken to be that level, in many cases, cannot be the basis fromwhich further inferences are drawn (Sperber and Wilson 1995, Carston 1997, 1999,

2002c). For example, Carston (2002c) shows that Grice’s level of what is said is notrich enough in meaning to actually form the basis for such further pragmatic infer-ences as conversational implicatures. Thus, consider the examples in (64) (chosenfrom Carston 2002c, 2004b).

(64) a. It’ll take time for this wound to heal.

b. She has a brain.

c. I haven’t eaten.

d. There’s nothing on telly tonight.

Utterances of the sentences in (64) all have full propositions at the level of what issaid, that is, after reference resolution, disambiguation and possibly other saturationprocesses have taken place. However, these propositions are banal tautologies (64a- 64b), or banal falsehoods (64c - 64d). That is, they are uninformative in that theydo not add to or revise the hearer’s existing assumptions concerning the world. Assuch, they will not trigger any further pragmatic inferences and thus an utterance ofsentences such as in (64) would have no cognitive effects at all. However, in realitythis clearly is not the case. Rather, what the speaker uttering sentences such asin (64) would intuitively be taken to have expressed, depending on the particularcontext, might be the following.

(65) a. It’ll take a considerable amount of time for this wound to heal.

b. She has a high-functioning brain.

c. I haven’t eaten yet/today.

d. There’s nothing worth watching on telly tonight.

As a consequence, RT assumes that there are processes involved in determining whatwas said by an utterance which crucially are not linguistically mandated. One ofthese is the process of free enrichment, which provides constituents that are notpresent in the semantic form of the uttered sentence. Thus, a proposition that isexplicitly communicated by an utterance may contain what might be called unar-ticulated constituents.12 The proposition determined by the various linguisticallymandated saturation processes plus the process of free enrichment, then, actually isrich enough to form the basis for further inferences. Thus, in a particular context,from (65a), the addressee might infer the implicature that he will be restricted inhis usual mobility. From (65b), in an appropriate context, the addressee might inferthat the person referred to is a good candidate for a particular position. Similarly,from (65c), the addressee might infer that the speaker is likely to want somethingto eat and from (65d) the addressee might infer that the speaker does not want tospend the evening watching TV.

Carston (1999) describes another process that might take place when movingfrom an utterance’s SF to that utterance’s meaning. This process, however, does notinsert additional material or provides the value for some variable, rather, it adjuststhe concept picked out by some particular lexical item. Thus, Carston discusses theexample in (66).

12Cf. Perry (1998) for a first introduction to the phenomenon of unarticulated constituent.

(66) Kato (of O. J. Simpson, at his trial):He was upset, but he wasn’t upset.[He was upset’ but he wasn’t upset”]

Assuming that the expression upset is used to express one and the same concept,Kato’s utterance would have to be taken as blatantly contradictory. However, asCarston notes, it was not intended, nor was it understood as such. Rather, whatKato is taken to have intended is to express two different concepts by his use of theform upset, one of which he states that it did apply to Simpson at a particular time,and one of which he expresses that it did not hold of Simpson at that same time.13

Relevance Theory calls this process ad-hoc concept formation. Thus, there are casesin which some lexical item is not used to express the concept that it encodes, butsome pragmatically adjusted concept, which may be narrower or looser than theoriginal concept and, thus, merely overlap with the original concept’s denotation.Concepts that result from the process of ad-hoc concept formation are not capturedby lexical approaches, since their defining property is that they are the result ofan online, ad-hoc interpretation process. The idea is that such concepts are notpermanently stored. That is, they are usually highly specific, picking out situationsthat are very roughly characterised as being non-normal. Thus, in example (66), theidea is that this ad-hoc concept is such that there is no economic or other reason forstoring it in long term memory because it is rather unlikely that it will be neededagain in that form. Having said that, the idea of ad-hoc concepts is also interestingfrom a diachronic point of view, as some cases of nonce-formation arguably are ad-hoc concepts, which may or may not become institutionalised. Be that as it may,ad-hoc concept formation is one of the processes that may apply when determiningan utterance’s meaning and it is a process that results in what can be called anothertype of non-literal meaning, in the sense that the process results in a concept thatdeviates from and replaces the concept the expression is normally taken to encode.

Having said this, it should be noted that the characterisation of ad-hoc conceptformation is not unproblematic when considering RT’s general view of the nature oflexically encoded meaning. Remember that Carston (2002c) proposes that what isencoded by lexical items are not full-fledged concepts, but rather abstract meaningsin that they merely are pointers to concepts. From this viewpoint, a lexical item isnever used to express what it encodes and it does not make sense to speak of adjust-ing what is lexically encoded, as this is not a full concept.14 If what an expressionencodes is only a concept schema, then this schema needs to be filled in before it canbe adjusted. In that case ad-hoc concept formation would be similar (although moregeneral and less systematic in its output) to the process of metonymic shift men-tioned above. Note further that it is not entirely clear how ad-hoc concept formationactually works, since, generally in RT, it is assumed that concepts are atomic. If

13Thus, a rough paraphrase of what Kato might have intended to convey is: ‘O.J. Simpson wassad and angry to a certain degree, but he was not sad and angry to a degree that led him to killinghis wife.’

14Cp. Burton-Roberts (2005, p. 404): ‘. . . if “word meaning” is not a concept but merely apointer to concepts, do we need – in fact, can we have – any notion of either “narrowing” or“loosening”? Surely, those notions only make sense if “word meaning”is taken to consist in “aconcept” . . .’.

that is so, then the question arises how, e.g., upset* can be an adjusted upsetconcept. That is, if all concepts are atomic, it is not possible to relate an adjustedconcept to the original concept by claiming that they share a certain amount offeatures. Having said that, recall that RT actually associates three distinct entrieswith a concept and thus, allows for correspondence/deviation of features in thoseentries. However, as pointed out already, with the assumption of such entries, it isnot clear in which sense concepts are still atomic.

Leaving these problematic issues to the side for the moment, let us turn back tothe propositional form resulting from the pragmatic saturation, enrichment and ad-hoc concept formation processes, which RT calls explicature in order to differentiateit from the traditional Gricean what is said. Explicatures differ from Grice’s what issaid in a number of respects. Thus, explicatures are taken to be the proposition ex-pressed by the speaker, rather than merely a full, but in a sense minimal proposition(cp. figure 3.4)15. In other words, explicatures are determined not just on the basisof processes that make use of speaker-independent, ‘narrow’ contextual information,but are the result of a number of pragmatic processes, some of which do take intoaccount assumptions about what the speaker wanted to express by making a certainutterance. Furthermore, some of these processes insert material into the SF of anutterance that is not linguistically mandated, whereas in the determination of tra-ditional what is said, only processes are involved that provide values for variablesalready present in the respective SF (cf. Bierwisch’s approach above).

sentence meaning

what is said/minimal proposition

Pragmatics

basis for further pragmatic inferences

explicature/proposition expressed

free enrichment. . .

Figure 3.4: Minimal proposition vs. proposition expressed

Thus, in RT, it is explicatures which are the basis for drawing further pragmaticinferences; the basis for implicit meaning content. Thus, in example (67), the con-versational implicature is not drawn on the basis of Gricean what is said by B, butrather on the basis of the explicature of B’s utterance.

(67) A: Would you like to join us for lunch?

B: I had a very large breakfast.

15The term minimal proposition, more on which below, is taken from Recanati (2004).

EX: I had a very large breakfast today.

CI: B doesn’t want lunch.

In fact, in RT’s cognitively oriented approach to interpretation, the traditional levelof what is said has no psychologically significant role to play. In other words, thereseems to be no reason why the level of meaning that results from the applicationof the processes of disambiguation, reference assignment and fixing of indexicalsshould be recognised as cognitively significant, since this level has been shown notto be rich enough to form the basis for further inferences. This is why RT actuallydoes not recognise what is said as an individual level in addition to, and distinctfrom explicature. Thus, it is explicatures which are taken to be explicitly expressedby the speaker of some utterance, whereas implicatures are implicitly communicated(cp. figure 3.5).16

sentence meaning

explicature

Pragmatics

saturation processesfree enrichmentad-hoc concept form.. . .

what is meant

sub-propositional

fully propositional

+ broad context

actual truth-conditions

underspecified

Figure 3.5: The Relevance-theoretic View

However, since pragmatic processes are involved both in the determination ofexplicatures as well as implicatures, the question is, how these two types of meaningare actually differentiated. Carston (2002c, p. 124) offers the following definition forexplicature.

An assumption (proposition) communicated by an utterance is an ‘ex-plicature’ of the utterance if and only if it is a development of (a) alinguistically coded logical form of the utterance, or of (b) a sententialsub-part of a logical form.

Thus, in order to differentiate explicatures from implicatures, one has to checkwhether the communicated proposition under consideration is a development ofthe SF of the utterance in question. As Burton-Roberts (2005, 2006) notes, theproblem here is the term ‘development’, which itself does not get a clear definition.He suggests the following definition: ‘. . . a communicated proposition P is a “devel-opment” of the encoded logical form L of the sentence uttered – and thus explicated

16For a critique of RT’s use of the terms implicit and explicit see below.

– if and only if P (asymmetrically) entails L’ (Burton-Roberts 2005, p. 397). Thus,in the examples below, the explicatures in (b) entail the logical form of the sentenceuttered in (a).

(68) a. He shrugged and left.

b. He shrugged and then left.

(69) a. He has three kids.

b. He has exactly three kids.

(70) a. I’ve had breakfast.

b. I’ve had breakfast today.

Intuitively, this criterion seems to work, however, Burton-Roberts (2005) points outthat it actually is problematic. Thus, for an explicature of an uttered sentence toentail the logical form of that sentence, the latter would have to be propositional,which RT claims it is not. Another problem is that the entailment characterisationof explicature cannot be applied in the case of negated sentences.

(71) a. I’ve not had breakfast.

b. I’ve not had breakfast today.

(72) a. You won’t die.

b. You won’t die from that cut.

In these examples the explicature in (b) does not entail the logical form of thesentence used to make the respective utterance, rather, the entailment relation isreversed (the logical form entails the putative explicature).

There is yet a further aspect about explicatures that is controversial. As notedabove, the level of explicature is taken to be that level at which an utterance becomestruth-evaluable. In this respect, it is similar to traditional what is said, which alsowas taken to be the level of meaning at which an utterance is truth-evaluable. Morespecifically, what is said was that level of meaning to the ‘expressing’ of whichthe speaker of an utterance was taken to be committed. However, explicature isthe result of pragmatic processes which are similar in kind to those processes thatderive conversational implicatures, where one of their characteristics is that they arecancellable. In consequence, Carston assumes that explicatures should be cancellableas well. However, as Burton-Roberts (2005, 2006) points out, an explicature of someutterance cannot be cancelled without contradiction.17 He suggests that explicaturesmay be clarified, but they may not be cancelled. That is because explicatures aretaken to be the meaning the speaker intended to express by making the utteranceshe did. Burton-Roberts (2006) argues that such implemented intentions cannot becancellable. For an illustration, take the following example from Carston (2002c,p. 138), which Burton-Roberts (2005, 2006) elaborates on.

(73) a. She’s ready.

b. Karen is ready to leave for the airport.

17For another defence of the idea that explicatures are not cancellable, see Capone (2006, 2009)

c. She’s ready but Karen isn’t ready to leave for the airport.

Thus, the sentence in (73a) may be used by a speaker, in an appropriate context,to express what is given in (73b). Now, Carston (2002c) goes on to claim that thisexplicature is pragmatically cancellable without contradiction by an utterance ofthe speaker of (73c). As Burton-Roberts (2006) convincingly argues, this cannot bethe case. To make sure that it is clear which possible explicature for (73a) we areinterested in here, Burton-Roberts (2006) states it in a slightly more formal way asin (74).

(74) Karen[K] is ready at time[U ] to leave for the airport[A]

The explicature of interest here for the second conjunct in (73c), Burton-Robertsstates as in (75).

(75) Karen[K] is not ready at time[U ] to leave for the airport[A]

But then, clearly, the second conjunct in (73c) cannot be used to cancel the expli-cature of the first conjunct without contradiction, if that is taken to be the speakerintended (74). The reason for this is very simple: what the speaker is assumedto have intended to express, he cannot take back without contradiction. The onlything a speaker can do is make it clear to the hearer what she actually intended toexpress, in case the hearer got her wrong. Thus, another example Burton-Robertslooks at is the following.

(76) Ann: That fellow’s playing is lamentable.

Bill: Too right. Cruelty to cellos, I call it.

Ann: Not the cellist - the trombonist!

Thus, Bill, with his reply to Ann’s first utterance, makes it clear that he assumedAnn was referring to the cellist when she said that fellow. Moreover, he indicatesthat he agrees with, what he thinks, she expressed about that person’s ability ofplaying his instrument. However, as Ann’s second utterance makes clear, Bill, un-fortunately, simply ‘got her meaning wrong’. That is, the reference of that fellowwhich Bill’s pragmatic interpretation process supplied, simply was not the referenceAnn had intended. Thus, Ann, with her second utterance clarifies what she actuallyintended to express by her first utterance. More specifically, she makes it clear toBill that by her use of that fellow she did not intend to refer to the cellist, butrather to the trombonist. However, Ann does not ‘cancel’ the explicature that Billthinks she expressed, because, in fact, she never did express that explicature. Thus,although explicatures are pragmatically derived, in contrast to implicatures, they arenot cancellable. That is because a speaker cannot take back, without contradiction,what she did in fact intend to express.

Implicitures

Another approach that tries to bridge the gap between the semantic form of anutterance and what was meant by the speaker uttering the expression in question,is that advocated by Kent Bach (e.g. Bach 1994a, 1997, 2004, 2005, 2006a). In

contrast to RT, Bach does recognise what is said as a distinct level of meaning,although he agrees with RT in that Grice’s notion of what is said does not playthe role in utterance interpretation traditionally assumed. That is, Grice’s notionof what is said (and for that matter Bach’s as we will see) does not provide arich enough basis from which to draw such pragmatic inferences as conversationalimplicatures. Nevertheless, Bach (1994a) recognises the level of what is said as thelevel of meaning explicitly expressed by a speaker, where this contrasts, on the onehand, with Grice’s level of implicated meaning what is meant and on the other handwith a level of meaning Bach (1994a) characterises as implicit and accordingly callsimpliciture. As regards Grice’s level what is said, Bach (2001b) assumes that it issubject to what he calls the Syntactic Correlation constraint. This means that ‘. . .if any element of the content of an utterance, i.e., of what the speaker intends toconvey, does not correspond to any element of the sentence being uttered, it is notpart of what is said.’ (Bach 2001b, p. 15). Thus, what is said is the result of thesemantic interpretation of the sentence uttered, that is, the meaning of the utteredsentence relative to the context of utterance, where the latter includes only a limitedamount of information (comparable to the information allowed for by Kaplan) anddoes not include speaker intentions. As a consequence, what is said may not befully propositional and it may not be the meaning the speaker is taken to haveexpressed. However, what is said forms the basis on which full propositions and,more specifically, the proposition intended by the speaker may be determined. Bach(1994a) assumes two processes which accomplish this task, namely completion onthe one hand, and expansion on the other. Thus, completion comes into play if whatis said is only of sub-propositional form, or what Bach calls a propositional radical.This process adds to what is said by an utterance content which is only implicitlycommunicated. Similarly, in cases where what is said is fully propositional or wherecompletion already has successfully applied, the resulting proposition may still notbe the one the speaker is taken to have intended to communicate. In such cases, theprocess of expansion will provide conceptual material which, again, was intended tobe communicated, but which is only implicit. Since the material that is added towhat is said by completion and expansion is not explicitly expressed, Bach calls theresulting propositions implicitures.

As Bach (2006a) notes, RT’s characterisation of explicature as being what is ex-plicitly expressed by an utterance is somewhat irritating. Thus, what differentiatesan explicature of some utterance from that utterance’s semantic form is that theformer may include material that is in fact not explicitly expressed at all by theSF of the utterance. In stating an explicature of some utterance, material that hasbeen implicitly conveyed is being made explicit. In contrast, Bach uses the termimpliciture to differentiate a level of meaning that is distinct from what is said, butalso different from the level of meaning Grice calls what is meant. However, as hepoints out in Bach (2006a), his notion of impliciture and RT’s notion of explicatureare rather similar in that an impliciture may result from an expansion of an utter-ance’s SF, that is, some kind of free enrichment process. Moreover, what makesexplicatures and implicitures similar, and differentiates both from implicatures, isthat they are conveyed directly. Thus, with implicatures, a speaker means one thingby saying what he says and conveys something in addition (and thus indirectly),

sentence meaning

impliciture

Pragmatics

saturation processes

what is meant

sub-propositional

fully propositional

+ narrow context

+ broad context

what is saidcompletionexpansion + broad context

Figure 3.6: Bach’s view

whereas implicitures include what has been said, but also include material thatremains implicit.

Note that Bach (2006a) claims that there are more differences in the two frame-works that the notions of impliciture and explicature are part of than between thenotions themselves. From this viewpoint, one major difference between the two no-tions is that RT’s explicature is taken to be the property of an utterance, whereasBach takes implicitures to be more akin to implicatures ‘. . . in that both are thingsthat speakers mean in saying what they say’ (Bach 2006a, p. 6). That is, in contrastto RT, Bach assumes that a differentiation between two levels of meaning (sentencemeaning18 and speaker meaning) is sufficient and that there is no need for a thirdintermediate level (utterance meaning). Thus, ‘[u]tterances themselves . . . have nocontent apart from the semantic content of the uttered sentence and the content ofthe speaker’s intention in uttering it’ (Bach 2001b, p. 31).

Interestingly, Bach (2000), like Carston with explicatures, also assumes that whatis implicit in an utterance can be taken back, that is, implicitures are cancellable.Thus, consider the example below (from Bach 2000).

(77) a. Not every bottle is empty.

b. Not every bottle I bought today is empty.

c. Not every bottle is empty – but every bottle I bought today is empty.

Depending on whether one believes in quantifier domain restriction (cp. Bach 2000,Stanley 2000, Stanley and Szabo 2000, Recanati 2002, Carston 2004a, Pagin 2005),the material in italics in (77b) is, in Bach’s terms either the result of a completion oran expansion of what is said by the sentence (77a) as it is used in a particular context.Since it is possible for the speaker of (77a) to continue as indicated in (77c) withoutcontradicting herself, what is left implicit in her utterance of (77a) is not part of whatis said. However, the same criticism of this view as was given by Burton-Roberts(2006) for explicatures also holds here. That is, if the speaker of (77a) intends her

18that is, the meaning of a sentence relative to a context of utterance as Bach conceives of it

utterance to be taken as expressing (77b), she would be contradicting herself bysaying something like (77c). This is because what is intended cannot be cancelled.

Bach notes that implicitures differ from such phenomena as metaphor in thatthere is no use of particular words in a figurative way. Rather, implicitures capturethe fact that one may communicate with an utterance more than what is explicitlyexpressed, namely information that is implicit in what one actually says. One par-ticular type of such implicit meaning, namely the one resulting from the process ofexpansion, leads to what Bach calls sentence nonliterality. Thus, consider example(78), in the context of a mother talking to her son, who has just cut himself and iscrying loudly. (78a) is what the mother says to him, (78b) is what, in this particularsituation, she may be taken to have intended to express.

(78) a. You’re not going to die.

b. You’re not going to die from this cut.

Bach says of this example:

‘. . . the mother is using each of her words literally but is omitting anadditional phrase that could have made what she meant fully explicit.If her son had replied, “You mean I’m going to live forever, Mom?”, itwould not be because she was being obscure but because he was beingobtuse–he would be taking her utterance strictly and literally, not as shemeant it. But if she were using each of her words literally, how couldshe not be speaking literally? This example illustrates a common butnot widely recognized kind of nonliterality, whereby a sentence is usednonliterally without any of its constituents being so used.’(Bach 1994a,p. 135)

Thus, Bach recognises a further type of non-literal meaning, namely one that dif-fers from metaphor or metonymy in not being associated with particular constituentsof an utterance, but rather with the utterance as such. Interestingly, concerning theson’s possible reply, Bach says that the son, in this case, simply was not able tofigure out what his mother meant. However, in a slightly different context, assum-ing, for example, that the son is not all that young, he might actually reply withYou mean I’m going to live forever, Mom?, while being perfectly aware of what hismother intended. One reason for doing so might be to achieve some comic effect.What is interesting here is that this presupposes that the son both understandswhat his mother intended to express by saying (78a) as well as, possibly, a moreminimal proposition, which, in fact, she cannot plausibly be taken as intending toexpress. This is an interesting aspect to keep in mind, and we will come back to itbelow, when it comes to assumptions about whether minimal propositions actuallyplay a role during the interpretation process.

As mentioned above, Bach retains Grice’s distinction between what is said andwhat is implicated in that he assumes that there is a gap between the explicitlyexpressed (what is said) and the implicitly and additionally conveyed. However,if an impliciture is, in a sense, an expanded what is said, including implicit butnonetheless communicated material, the question arises, what differentiates between

implicitures and implicatures. Similarly to explicatures implicitures are also theresult of pragmatic processes, which are assumed to be similar to the processesleading to implicatures. Bach argues that implicitures in contrast to implicaturesare built up from what is said, that is, the explicit meaning content of an utterance.An impliciture of some utterance includes material that might have been expressedexplicitly ‘if the appropriate lexical material had been included in the utterance’(Bach 1994a, p. 140). In contrast, implicatures arise from the saying of what issaid. Thus, implicatures are conceptually independent of both what is said and theimpliciture of an utterance.

Note that Bach’s characterisation of implicitures as being ‘built up from whatis said’ is as imprecise as RT’s notion of explicature being a ‘development’ of alogical form. Incidentally, neither RT nor Bach can characterise the level of implic-itly expressed meaning (explicature/impliciture) as resulting from the ‘filling in’ oflinguistically mandated ‘slots’ in the semantic form of some expression, since bothallow for the insertion of information into the semantic form of an utterance with-out there being a linguistically predetermined ‘slot’, e.g., a variable for it. In thisrespect, both RT and Bach’s approach differ crucially from Stanley’s approach (e.g.Stanley 2000), who assumes that all the information that is part of what is saidby an utterance needs to be mandated by something in the semantic form of thatutterance, as produced by the process of semantic composition.

3.2.2 Unarticulated Constituents vs. Hidden Indexicals

According to Stanley, in arriving at what is said by some utterance, extra-linguisticcontext only plays a role insofar as it provides the values for variables already presentin the semantic form of that utterance, or in supplying the content of indexicalexpressions and resolving ambiguities. That is, only saturation processes take placein the determination of what is said. In other words, he assumes that all truth-conditional effects that the extra-linguistic context in which an utterance is madehas on what is said by that utterance, are linguistically mandated.

sentence meaning rel. context/what is said

Pragmatics

semantic compositionsaturation processes

what is meant

fully-propositional

+ broad context

Figure 3.7: Stanley’s view

With respect to the role of variables in the SF of utterances, Stanley’s approachis similar to Bierwisch’s view of the level of utterance meaning and the nature of

the processes that contribute to it. That is, processes like conceptual shift andconceptual differentiation also rely on the presence of variables in the semantic formof the expressions concerned. Thus, these processes are linguistically mandated.Both in Stanley’s and in Bierwisch’s approaches, the variables are assumed to bepart of the semantic form of the respective utterances, that is, they are contributedby one or another of the individual expressions used to make the utterance.

In his (2000) article, Stanley argues against the existence of non-sentential as-sertions and generally against unarticulated constituents in the semantic form of anutterance, both of which phenomena have been used to argue for the application ofpragmatic processes that are not linguistically mandated as, e.g. free enrichment.Thus, he notes that for examples such as (79) below, B’s utterance of Bill does notconstitute a non-sentential assertion but rather is a case of syntactic ellipsis. Byuttering ‘Bill’, B will be taken to have expressed the proposition Bill bought thebottle. Thus, the logical form of B’s utterance actually includes information whichsimply is not phonologically expressed.

(79) A: Who bought the bottle?

B: Bill.

B can use his utterance of ‘Bill’ to express the proposition that Bill bought the bottlebecause A’s question, in a sense, provides the frame within which B’s utterance isinterpreted. Thus, cases of syntactic ellipsis do not occur discourse-initially, becausethe correct interpretation of elliptical sentences depends on the context in which theyare uttered. Looking once again at the examples given above, where RT assumes thatthe sentences do not express full propositions without some process of saturation,one could argue that these are cases of syntactic ellipsis, especially since an utteranceof any of them would seem odd discourse-initially.

(63) a. Paracetamol is better. [than what?]

b. It’s the same. [as what?]

c. He’s too young. [for what?]

d. She’s leaving. [from where?]

Generally, Stanley assumes that syntactically elliptical sentences need a linguisticantecendent. However, he argues that explicitly mentioning that antecedent is onlyone way of providing that antecendent. It should be noted, that Stanley has a verybroad conception of what counts as a discourse and as such provides the necessarycontextual information for the correct interpretation of a syntactically elliptic sen-tence. Thus, consider one of his examples, given in (80), for which Stanley providesthe following context.

‘Suppose Bill walks into a room in which a woman in the corner is at-tracting an undue amount of attention. Turning quizzically to John,he arches his eyebrow and gestures towards the woman. John replies:’(Stanley 2000, p. 404)

(80) A world famous topologist.

Stanley assumes that the context preceding John’s utterance, namely Bill’s quizzi-cally turning to John, his arching of eyebrows and gesturing towards a specificwoman, make salient a question that Bill might have uttered but did not, namelyWho is that? Because the context makes this question salient, John can utter (80),which Bill will be able to recognise as the answer, as it where, to his implicitlyasked question. Thus, in the example above, the necessary linguistic antecendet(Who is that?) is implicitly provided by the situation in which John’s utteranceof (80) takes place. That is, it is plausible to assume that in this situation, Billtries to make explicit his intention by using non-verbal clues such as gestures andfacial expressions which are ostensively directed towards John. However, I thinkit questionable whether these non-verbal clues actually are of such a definite anddeterminate nature as to make salient a specific linguistic complex expression suchas Who is that?

Thus, consider another example, taken from Carston (2002c), where the contextis the following. B is coming into the kitchen and sees A, apparently searching forsomething, as he is looking around the lower shelfs of a cupboard in the kitchen. Itis breakfast time and B knows that A likes to have marmalade for breakfast, andhe assumes that that is what A is looking for. Thus, B says the following.

(81) On the top shelf

Now, Carston (2002c) assumes that in order for A to interpret correctly that whatB intended to express was in fact a full proposition (namely the proposition thatthe marmalade is on the top shelf of the cupboard), he has to pragmatically enrichB’s utterance to arrive at that proposition. According to Stanley, however, thereare no such things as non-sentential assertions which are not syntactically ellipticalor ‘shorthands’19. However, (81) clearly qualifies as an assertion. Thus, Stanleywould have to explain this example along the lines of the example given above.That is, the context of B’s utterance of (81) makes salient the question Where isthe marmalade, a question A could have uttered but did not explicitly so. However,the context of B’s uttering (81) differs in some important aspects from the contextof (80). Thus, there are no facial expressions or gestures on the part of A whichare intentionally directed towards B and which could function as clues as to thedeterminate linguistic expression A could have but did not utter. In fact, in sucha situation it is likely that A does not entertain any possible linguistic complexexpressions applicable to the situation at hand at all. Thus, if cases of syntacticellipsis need a determinate linguistic antecedent for the recovery of the missingmaterial, it is not clear to me, how in examples as that given above, this implicitbut determinate linguistic antecedent can be made salient by the context. Thisquestion is even more pressing in cases where the interlocutor who is attributed asentertaining this implicit question, by his behaviour, cannot even be taken to haveintended to convey any question whatsoever.

Stainton (2005) states a connected problem for Stanley’s assumption that thecontext in which a syntactically elliptic utterance is made may be taken to provide

19For his notion of ‘shorthand’, see Stanley (2000). An example is an utterance of ‘nice dress’ bysome passer-by on the street to some woman, where this can plausibly be taken to be a ‘shorthand’of the statement ‘This is a nice dress.’. However, as noted by Stainton (2005), this is a very vaguenotion and according to Stainton (2005, p. 404), Stanley has given it up.

the necessary linguistic antecedent. He assumes that true syntactic ellipsis is re-solved wholly linguistically (a plausible assumption considering that a determinatelinguistic antecedent is necessary). That is, cases of syntactic ellipsis are resolvedwithin the language module. Apparently, recovering the missing material is a purelylinguistic process, since it relies on the availability of a determinate linguistic an-tecedent. However, if that is the case, then pragmatic processes should not play arole here. Providing a linguistic antecedent by way of making it salient in a con-text, would require pragmatic processes to take place. Thus, resolving syntacticellipsis could no longer be treated as a solely linguistic phenomenon. Moreover, asmentioned above, it is not even clear whether, and even less how, a determinatelinguistic antecedent can be made salient.

As a last point concerning Stanley’s hypothesis of the non-existence of genuinenon-sentential expressions: Pagin (2005) states that he finds it implausible to assumethat whenever a speaker manages to communicate a thought using some linguisticexpression, this implies that he used a fully articulated sentence, if only as far as theexpression’s logical form is concerned. Rather, he thinks that ‘. . . it is not too rarethat a speaker manages to get across a thought with poor linguistic means, fallingshort of what would be needed to get the thought across to an audience less in tunewith the speaker’ (Pagin 2005, p. 330).

Let us now turn to Stanley (2000)’s more general argument against the presenceof pragmatically provided unarticulated constituents in the proposition expressed byan utterance. Stanley (2000, p. 410) gives the following definition for unarticulatedconstituents.

‘x is an unarticulated constituent of an utterance u iff (1) x is an elementsupplied by context to the truth-conditions of u, and (2) x is not thesemantic value of any constituent of the logical form of the sentenceuttered.’ (emphasis as in the original)

From what has been said above about non-sentential assertions, it should be clearthat if one assumes that there are genuine non-sentential assertions which are notcases of syntactic ellipsis and which can be used to express a full proposition,then one has to also assume the presence of pragmatically provided unarticulatedconstituents in the semantic form of such assertions. That is, unarticulated con-stituents constitute the material that needs to be integrated with the semantic formof non-sentential assertions in order for that semantic form to become fully propo-sitional. However, even if one accepts Stanley’s argument against the existence ofnon-sentential assertions that are not cases of syntactic ellipsis, unarticulated con-stituents may still be at play where the semantic form of some utterance already is afull proposition, but not the proposition the speaker can be plausibly taken to haveintended to express, as in cases of free enrichment. However, Stanley (2000) andStanley and Szabo (2000) argue that what seems to be an unarticulated constituentpragmatically provided and somehow inserted into the semantic form of an utter-ance is in fact simply the pragmatically provided value of some variable that waspresent in the semantic form all along. The argument they pursue is that if someconstituent is wholly pragmatically provided, it should not be able to be boundby some explicit quantifier occurring in the utterance in question. This is because

the operators within a sentence can only interact with variables that are also partof the same sentence and which lie within the operators’ scope. The idea is thatpragmatically provided constituents cannot lie within the scope of such operators.Let us look at some of the relevant examples.

(82) It’s raining.

Provided the speaker uttering (82) does so in a neutral context, she would normallybe understood as expressing the proposition that it is raining at the time of utteranceat the place where she is making that utterance. Thus, assuming that the semanticform of the utterance does include a temporal, but not locational variable, the truthconditions for an utterance of (82), according to RT, look as follows, where thelocation of the raining is pragmatically determined.

(83) An utterance of ‘it is raining (t)’ is true in a context c iff it is raining at t andat l, where l is the contextually salient location in c.

Now consider the possible interpretations of an embedded version of the simplesentence.

(84) Every time John lights a cigarette, it rains.

(85) a. For every time t at which John lights a cigarette, it rains at t at the locationl in which John lights a cigarette at t.

b. For every time t at which John lights a cigarette, it rains at t at somelocation which is salient in the context of utterance.

Whereas RT can account for the reading in (85b), it cannot account for the readingin (85a), which is the preferred reading in this case. This is because in the preferredreading the location variable is within the scope of the universal quantifier. Accord-ing to Stanley, however, a variable that is introduced by some pragmatic processcannot come within the scope of an explicit quantifier. Thus, the argument is thatsince in (84) the location variable is within the scope of the quantifier and thusvaries with varying values for the temporal variable, the location variable is alreadypart of the compositionally built up semantic form of the utterance at hand. Thus,there is no unarticulated constituent present here.

As Carston (2002c) states, Stanley’s argument rests on the assumption thatthe pragmatic mechanisms cannot provide a reading that corresponds to a boundvariable reading. Stanley’s arguments against such a possibility are based on whathe assumes to be the nature of denotations and his view of what constitutes context.

‘. . . denotations of bound variables are odd, theoretically complex en-tities. It is difficult, if not impossible, to see how, on any account ofsalience, such an entity could be salient in a context. Certainly, nei-ther it, nor instances of it, could be perceptually present in the context.’(Stanley 2000, p. 414)

Thus, it seems that for Stanley, only perceptible objects and properties constitutethe context of an utterance. In contrast, RT assumes that, essentially, context

consists of a set of mentally represented assumptions, where only some of those arerepresentations of perceptible environmental features. Thus, for the case at hand,Carston (2002c, p. 200) suggests that the addressee might access a ‘. . . generalknowledge assumption about the way in which times and places pair up when acertain type of event (such as “raining”)’ occurs. This will allow the preferred‘bound-variable’ interpretation, unless the context is such that it prevents it.

To be sure, the hidden indexicals analysis explains the preferred reading of (84)more elegantly, in assuming that there actually is a variable present in the semanticform of the utterance, which, in case of the preferred reading, gets bound by the uni-versal quantifier. However, an argument can be made against the semantic necessityof such a location variable. For instance, Recanati (2004) argues that there is nosuch necessity for assuming a semantically provided location variable in the case ofrain. Thus, he claims that one can imagine situations in which no particular place isrequired for a successful interpretation of an utterance such as (82).20 Since this isthe case, there is no reason to assume that the semantics has to provide a locationvariable, requiring a value in every context. However, if there is no location variablein the semantic form of (82), how is it that we get a bound reading in (84)? Asmentioned above, Stanley claims that such a reading is only possible, if one assumesthat there actually is some variable in SF which gets bound by the quantifier. Morespecifically, he assumes that this variable is provided by the semantics of the verb torain. Recanati (2004) offers an alternative analysis, where the variable is providedby the quantificational phrase. He claims that such quantificational phrases maybe treated in parallel to singular phrases functioning as modifiers, where the lat-ter introduce what are called variadic functions. ‘A variadic function is a functionfrom relations to relations, where the output relation differs from the input relationonly by its decreased or increased adicity’ (cf. Recanati 2002, p. 319). Thus, for asentence such as (86), ‘. . . the phrase “everywhere I go” contributes both the adicity-increasing variadic function and the operator which binds the extra argument-role.’(cf. Recanati 2002, p. 330).

(86) Everywhere I go, it rains.

Thus, (86) is properly represented by (87), where the quantifier binds the free vari-able in the subformula ‘in l (it rains)’, which it has supplied to it in the first place.

(87) [For every place l such that I go to l ](in l (it rains))

The advantage of this approach is that a variable only has to be assumed in caseswhere the simple sentence It’s raining is embedded, whereas the simple sentence assuch does not have such a variable.21

20This is Recanati’s example context: ‘. . . imagine a situation in which rain has become extremelyrare and important, and rain detectors have been diposed all over the territory (whatever theterritory – possibly the whole Earth). In the imagined scenario, each detector triggers an alarmbell in the Monitoring Room when it detects rain. There is a single bell; the location of thetriggering detector is indicated by a light on a board in the Monitoring room. After weeks of totaldraught, the bell eventually rings in the Monitoring Room. Hearing it, the weatherman on dutyin the adjacent room shouts: “It’s raining!!”’

21However, see Pagin (2005) for a critique to the effect that even in the case of an occurrence ofthe simple sentence it is raining the prefixed form ‘in l (it rains)’ has to be assumed as its logicalform, in which case the additional variable would no longer be optional.

Note, however, that if what the verb to rain is taken to express is a particulartype of event, conceptually speaking there is no question as to the status of the lo-cation variable. Thus, events in general are conceptually conceived of as necessarilyinvolving a time (span) and a place at which they take place, although neither timenor place necessarily needs to be made explicit or be specified. In other words, araining event of necessity involves a location at which the event takes place. Now,it seems that when expressing the taking place of a raining event as in (82), whatthe speaker standardly is taken to also express is that the location of the rainingevent is in close approximation to the speaker. That is, the location variable thatevents have in general, in this case is specified to something like in close proxim-ity to speaker. However, the specification of the location variable as in closeproximity to speaker is not part of the semantics of to rain but rather is a stan-dardised, though defeasible meaning aspect. In fact, it reminds one of the notionof deictic centre that plays a role in the interpretation of deictic expressions. Thus,turning back to the example in (84) and its preferred interpretation as ‘wheneverJohn lights a cigarette, it rains where John is when he lights a cigarette’: here thelocation variable associated with the raining event is not bound to where the speakeris, but rather to where the subject of the embedding sentence (John) is whenever heengages in a cigarette-lightning event. In other words, the location variable of theraining event co-varies with the location variable of the cigarette-lightning event.And since John is said to be involved in the cigarette-lightning event, the locationof the raining-event seems to co-vary with where John is. However, again this seemsto be a preferred interpretation, because if more thematic roles are introduced, theinterpretation of the location of the raining event shifts.

(88) a. Every time Susann calls Mary, it rains.

b. Susann: Every time I call Mary, it rains.

c. Mary: Every time Susann calls me, it rains.

Thus, in an utterance of (88a), I would content, the location of the raining eventcannot easily be interpreted as the location at which either Susann or Mary or evenboth are, when the calling-event takes place. However, the location variable of theraining event is still understood as varying with the location of the calling-event.The problem here is that it is much more difficult to indentify the location of thatcalling-event, as it involves two participants and the two participants may be in verydifferent locations. Interestingly, as regards utterances of (88b) and (88c) by Susannand Mary, respectively, it seems that the location variable of the raining event isbound to where the respective speaker is when a calling event by Susann of Marytakes place.

What these examples show is that even if one assumes that a location variable issemantically supplied due to the fact that the verb to rain is understood as expressinga particular type of event, supplying a value to that variable is not a trivial matterand depends very much on the overall structure of the utterance the verb is part of aswell as on the specific circumstances holding in the particular utterance situation.22

22See, however, section 5.2.1 for a view where the location variable is not taken to be part ofthe semantics of the verb but may still enter into the interpretation of an utterance involving thatverb, albeit not ’freely’.

Moreover, the location at which a particular event is assumed to take place doesnot necessarily have to be specified. Thus, note that in (84) no specific locationis inferred for the event expressed in the embedding sentence and still the locationvariable of the event expressed in the embedded sentence is assumed to co-vary withthat unspecified location variable of the embedding sentence.23

Turning back to Stanley’s hidden indexicals approach: a general criticism is thefact that it is not clear how many such variables should be assumed to be presentin the semantic form of a particular utterance. Thus, Wilson and Sperber (2000,pp. 238) discuss the example below, where the context is that Lisa looks in at herneighbours, who are about to start supper.

(89) Alan: Do you want to join us for dinner.

Lisa: No, thanks. I’ve eaten.

According to a standard semantics analysis, Lisa’s utterance of I’ve eaten has thetruth-conditions stated below.

(90) At some point in a time span whose endpoint is the time of utterance, Lisa haseaten something.

However, in the context at hand, this clearly is not what Lisa will be taken to haveintended to express. Rather, what she probably intends to express in this situationand what Alan will understand her to have expressed is something like the following.

(91) Lisa has eaten supper this evening.

That is, Lisa both communicates the object of the eating-process as well as a tempo-ral specification of when the eating-process took place. Wilson and Sperber (2000),and following them Carston (2002c), assume that under the hidden indexical ap-proach the semantic form of Lisa’s utterance includes two variables – one for theobject eaten and one for a specification of the time span.

(92) I have eaten x at t.

Wilson and Sperber (2000) go on to show that in other circumstances, the propo-sition expressed by a speaker uttering I’ve eaten, might also give a specification ofthe place or manner of eating and possibly others.

(93) I’ve often been to their parties, but I’ve never eaten anything [there].

(94) I must wash my hands. I’ve eaten [using my hands].

To capture the propositions expressed by these utterances, even more hidden in-dexicals would have to be postulated, yielding a semantic form for I’ve eaten as in(95), although, as Carston (2002c, p. 204) notes, in many cases the variables presentwould not get a definite specification at all (as in the example of Lisa’s utteranceabove)(cf. 96).

(95) I’ve eaten [object x] [in manner y] [at location l] [within time span t].

23The discussion of these examples will be taken up once more in section 5.2.1.

(96) I’ve eaten supper in some manner at some location this evening.

However, in the context given above, (96) is not what we take Lisa to have expressed.That is, if she had chosen to make explicit the allegedly implicit material, theproposition expressed would have differed considerably from the one expressed byher simply uttering I’ve eaten.

There are several points that have to be made concerning this criticism and thespecific examples used to make it. First of all, conceptually, the act of eating clearlydoes involve something that is being eaten. Thus, ‘eating’ is a relation betweensomeone who eats and something that gets eaten. Therefore, it is not unreasonableto assume that there is a slot for the eaten object in the semantic form of the verb,even though that slot does not need to get specified and it does not need to beovertly expressed. When Lisa says ‘I’ve eaten’, she is understood to have expressedthat she has eaten something or other before the time of utterance. Moreover, inthis particular situation, Lisa can be taken to use ‘I’ve eaten’ in order to give anexplanation for her declining the invitation. It is questionable whether Alan inhis interpretation of Lisa’s utterance has to go as far as to interpret her as havingspecifically expressed (91).

While it is true that Lisa will be taken to have expressed that she has eatenrecently, there is no reason to assume that she intended to express the specificobject of the eating process she engaged in. Having said this, even if Alan doesinterpret her as expressing that she has eaten dinner, all that is necessary to arriveat that interpretation is to supply a value to the semantically supplied eating-objectvariable. Similarly, all that is needed for an interpretation of Lisa’s utterance asexpressing that she has eaten recently (from which Alan can infer that it mightmake it undesirable for her to join Alan for dinner), is a pragmatically determinedrestriction of the time span semantically given by her use of the past tense.24

Similarly, the criticism that the semantic form of I’ve eaten, based on examplessuch as (93) and (94) would have to be assumed to include as many variables asgiven in (95), actually is not warranted. Thus, note that both examples (93) and (94)involve not just one utterance, but rather small texts. Thus, the [there] used in (93)to indicate a particular location of the eating event expressed by the second sentenceactually is a shorthand for ‘at the contextually relevant party’ in this particularsetting. The idea is that in order to interpret the second utterance as involvinga particular location, no semantic location value needs to be posited, rather, theparticular interpretation arises due to the particular discourse context in whichthe utterance is made. Part of that context is the proposition added to it by theutterance of the first sentence in (93). Similar points can be made regarding (94).

Thus, I would assume that the examples stated above do not force Stanley toposit semantic variables for manner or location of eating.25 Rather what they showis that the discourse context in which an utterance takes place influences how that

24This is similar to the today interpretation of ‘I’ve had breakfast’, where when this standardinterpretation is cancelled by saying ‘I’ve had breakfast, but not today’, the speaker will still beunderstood to have expressed that he has had breakfast in the past, but will be understood asnot including the day of utterance in that time span. Thus, in this example, the time span ispragmatically adjusted as well, but it was originally introduced by the past tense and not totallypragmatically supplied.

25Although, as noted for to rain, one might actually argue that the location variable does form

utterance is understood. Moreover, the fact that an utterance by Lisa of (96) wouldhave been interpreted differently from the utterance she actually made, should notbe seen as a problem of the hidden indexicals approach, but rather as being groundedin pragmatic considerations. That is, overtly expressing the proposition that seemsto be covertly expressed by uttering ‘I’ve eaten’ raises questions concerning Lisa’sreasons for doing so.26

Having said this, although there are cases of variables in the SF of an utterancewhich are ‘only’ provided a value by pragmatic saturation processes, it seems thatone has to allow for pragmatic processes that provide constituents which are notarticulated by the utterance’s SF in any way. However, these constituents do notarise out of the blue, rather, one has to turn to the context in which the utterancethat seems to include them is made. This holds also for non-sentential utterances,which can be understood to express propositions because they are interpreted withinthe context in which they are made, even though that context may not necessarilyinvolve previous verbal interaction.

3.2.3 Minimal Semantic Content and Full Propositionality

In chapter 2, I argued that the traditional view of semantics as both being con-cerned with context-independent meaning as well as delivering full propositions forsentences cannot be upheld. I argued for a view of semantics as essentially dealingwith context-independent meaning, but not necessarily determining fully proposi-tional sentence meanings. However, an alternative view is possible, in which thedetermination of fully propositional sentence meanings is taken to be an essentialcharacteristic of semantics. Under such a view, semantics cannot be assumed to beonly concerned with context-independent meaning and, accordingly, the differentia-tion between semantics and pragmatics has to be based on some other criterion. Infact, there are approaches which assume that what the semantic component returnsas output is not some sub-propositional semantic form, but rather, as tradition-ally assumed, a full proposition, that is, something which is truth-evaluable. Twosuch approaches are those suggested by Borg (2004b) and by Cappelen and Lepore(2005), respectively.27 28

Thus, Emma Borg (Borg 2004b,a, 2007) argues for what she calls a minimalsemantic component which, however, retains one of the most characteristic prop-erties of the traditional approach: Borg claims that what the semantic componentreturns as output actually is fully propositional/truth-evaluable, although it maynot be identical to the proposition expressed by a particular utterance of the sen-

part of the verb’s semantic form – and Stanley probably would do so in any case – if it is understoodas expressing a particular type of event.

26Thus, compare this to Levinson’s M-heuristic (more on which in section 4.1.3): what is saidin an abnormal way, is not normal (marked messages indicate marked situations).

27Strictly speaking, Stanley’s approach also falls into the category of approaches discussed inthis section. Thus, the characterisations of semantics and pragmatics given here also hold for hisapproach. However, there are some crucial aspects in which his general approach differs from Borg’sand Cappelen and Lepore’s, which, I think, justify treating the respective approaches separately.

28Bach’s approach does not fit in this section, because, although he also assumes that semanticsis not concerned with context-independent meaning exclusively, he does not require of sentencemeaning even relative to a context of utterance that it be fully propositional.

tence/linguistic string in question. Also, like Stanley, she assumes that only syntacti-cally mandated processes take place when determining the truth-conditional content,or sentence meaning, of an utterance. Thus, there are no unarticulated constituentsand there is no such process as free enrichment at this stage in the interpretationprocess. In fact, she denies that speaker intentions play any role whatsoever atthis stage. Cappelen and Lepore (2005)’s approach differs from Borg’s, and fromStanley’s, in that they deny the context-sensitivity of any other than the expres-sions in, as they call it, the ‘basic set’, that is, such expressions as traditionallywere considered context-sensitive (cf. Kaplan 1989b). Thus, while they also onlyallow linguistically mandated processes to take place when building up the semanticcontent of a sentence, they argue that there are considerably fewer such processesthan assumed by dual pragmatists29 or contextualists, as they call them.

Generally, Borg argues that there are (only) two tasks that an appropriate formalsemantic theory has to fulfill. On the one hand, it has to capture how the meaningof a complex expression is determined on the basis of the meanings of its componentexpressions and the way they are combined, on the other hand it should make clearthe relations that hold between individual complex expressions, for example, whichinferential relations possibly hold between them. Borg claims that other possibletasks one might expect a semantic theory to fulfil actually lie beyond the reach of anappropriate formal semantic theory, such as explaining our communicative abilities,or being ‘. . . sensitive to the kinds of epistemic relations we bear to objects . . .’ (Borg2004b, p. 2). From what Cappelen and Lepore say about the role of semantics, itcan be assumed that they generally share this view.

Interestingly, both approaches agree with dual pragmatists that in order to deliverthe fully propositional semantic content of a sentence/utterance, contextual infor-mation has to be considered. Thus, neither Borg nor Cappelen and Lepore draw thedividing line between the semantics component and pragmatics along the traditionaldichotomy ‘context-independent vs. context-dependent’, rather, what differentiatesbetween semantics and pragmatics is the nature of the contextual information inte-grated in the semantic content of a sentence (especially for Borg – as for Stanley)and the conditions under which such integration takes place (for both Borg and Cap-pelen and Lepore). In other words, what the semantic component deals with aresentence-types relativised to a context of utterance. However, as mentioned already,for Borg, the contextual information used in determining the semantic content ofsuch a sentence-type is restricted to purely objectively available data, such as infor-mation concerning the speaker and addressee of the uttered sentence as well as thetime and place of utterance. Information that requires theory of mind reasoning,such as for example the possible intention behind the speaker’s uttering the partic-ular sentence he did, are not part of this kind of contextual information, as this isbased on subjective data, resulting from abductive mind-reading processes.

This is another aspect in which Borg’s approach differs from Cappelen and Lep-ore’s. While Borg rules out any involvement of such non-demonstrative information,Cappelen and Lepore seem not to mind that such information is necessary when,e.g. determining the actual referent of some indexical expression. However, what is

29Borg’s term for the proponents of approaches such as put forward by Sperber and Wilson,Carston or Recanati.

problematic about this latter viewpoint is that it allows non-objective informationto figure in determining the semantics of an expression, when the assumption is thatsemantics is part of our general language system and as such only involves processesthat are formal in nature, syntax-driven and computational. To repeat, the natureof the semantic component is usually taken to be such that it reflects (a particularpart of) our knowledge of language. Reasoning about plausible intentions behinda speaker’s use of a particular expression does not form part of this kind of knowl-edge. For Borg in particular, there is another reason for not assuming that speakerintentions play a role in the determination of semantic meaning, since she assumesof semantics that it exhibits the characteristics of a Fodorian mental module (cf.Fodor 1983). One such characteristic is informational encapsulation. That is, thesemantic component only has access to a restricted amount and type of information.Thus, in computing the meaning of a sentence, the interpreter’s assumptions aboutwhy the speaker made a particular utterance, are not taken into consideration, asthey constitute aspects of knowledge which is not specifically linguistic.

As already mentioned, Borg claims that what the semantic component deliversare truth conditions for sentences (relativised to a context of utterance). However,the truth conditions figuring in the semantic component are what she calls liberaltruth conditions. In this respect her approach differs from Stanley’s, who seemsto assume that semantics delivers the actual truth conditions of a sentence relativeto a context. Liberal truth conditions capture the conditions that have to hold forthe sentence under consideration to be true. Thus, they differ from actual truthconditions (or the proposition expressed by a speaker’s utterance of a particularsentence), where the latter capture what a speaker intended his utterance to meanand are therefore a solely pragmatic matter. Borg (2004b) says of liberal truthconditions that they may be satisfied in various possible, more specific contexts.Moreover, ‘[a] liberal truth-condition posits ‘extra’ syntactic material . . . only whenit is intuitively compelling to do so, or when there is good empirical evidence tosupport the move.’ (Borg 2004b, p. 230). She gives the following examples of liberaltruth conditions.

(97) a. If u is an utterance of ‘Jane can’t continue’ in a context c then u is true iffJane can’t continue something in c.

b. If u is an utterance of ‘Steel isn’t strong enough’ in a context c then u istrue iff steel isn’t strong enough for something in c.

c. If u is an utterance of ‘The apple is red’ in a context c then u is true iff theapple is red in c.

Looking in more detail at the example sentence Jill can’t continue: in a contextwhere it is clear that Jill did not pass enough university course units to qualify foradmission to second year study, it is plausible that what the speaker intended herutterance to mean is something along the lines of (98c). However, the semanticcontent of the uttered sentence is simply (98b).

(98) a. Jill can’t continue.

b. Jill can’t continue something.

c. Jill can’t continue university education.

Thus, actual truth conditions capture the conditions under which the utterancemeaning30 under consideration would be true, where utterance meaning is not iden-tical to sentence meaning relativised to a context of utterance, as the determinationof the former may rely on non-objective contextual information, more specifically, aconsideration of possible speaker intentions. Thus, while speaker intentions for Borgdo not figure in the semantic meaning of an utterance, they do already play a rolewhen moving to its utterance meaning. In this sense, everything that goes beyondthe semantic content of a sentence may be termed non-literal meaning, whereas thesemantic component delivers a sentence’s literal meaning (its liberal truth conditions)– this is how Borg uses the terminology. Whereas utterance meaning may includeunarticulated constituents, resulting from processes of free enrichment, Borg arguesthat sentence meaning does not allow for unarticulated constituents as the semanticcomponent only deals with syntactical objects. Thus, variables that are given in thesyntactic structure of a sentence do figure in determining the semantic content of asentence, but purely pragmatically supplied constituents, naturally, do not.

Note that whereas Borg allows for covert syntactic constituents in the case ofcontinue and enough, Cappelen and Lepore argue that these are not context-sensitiveexpressions in the sense of those in the ‘basic set’. That is, Cappelen and Lepore areeven more strict about which expressions may be taken to count as being context-sensitive as Borg is, who cautions that covert syntactic constituents should only beassumed if there is empirical evidence of a very strong intuition that it is appropriateto do so. What this means for Cappelen and Lepore’s approach is that, in a sense,the semantic content of a sentence is taken to be even more minimal, at least inone sense, than assumed by Borg. Thus, Cappelen and Lepore claim that the truthconditions for a sentence such as Jill can’t continue or Steel isn’t strong enoughsimply are as follows31.

(99) a. If u is an utterance of ‘Jane can’t continue’ in a context c then u is true iffJane can’t continue in c.

b. If u is an utterance of ‘Steel isn’t strong enough’ in a context c then u istrue iff steel isn’t strong enough in c.

However, as Recanati (n.d.) points out, sentences such as the above invite the ques-tion ‘for what?’ which only gets answered once one takes the context of utteranceinto account. Nevertheless, Cappelen and Lepore insist that the truth conditionsfor such sentences are as given above and that they are propositional. However, asBach (2006b) comments ‘Cappelen and Lepore evidently assume that producing aT-sentence automatically provides a truth condition. [. . .] After all, if a sentenceis semantically incomplete, the corresponding T-sentence will be incomplete too.’(Bach 2006b, p. 4). Furthermore, both Recanati as well as Bach criticise that Cap-pelen and Lepore – other than providing T-sentences such as above – do not explainwhat the propositions such sentences as Jane can’t continue or Steel isn’t strongenough express, are. Cappelen and Lepore (2005) in turn argue that that’s beyond

30Borg (2004b) does not use the term utterance meaning, but rather sticks to the traditionalGricean term what is said.

31Note that throughout their book Cappelen and Lepore (2005) ignore tense, thus, their versionof the truth-conditions for these sentences actually do not refer to a context of utterance.

the semanticist to tell and something that has to be clarified by metaphysics. How-ever, if that is so then one simply has to take the authors’ word for it that the giventruth conditions are actually such that they allow one to evaluate the truth of therespective sentence (for further discussion, see Bach n.d.b,n, Cappelen and Leporen.d.b,n).

In addition, Borg (2007) suggests that for Cappelen and Lepore’s account, pre-serving the law of non-contradiction is problematic.32

For instance, consider the sentence ‘Jill is ready’: according to [Cappe-len and Lepore (2005)] this expresses the minimal proposition that Jillis ready and presumably the sentence expresses something true in anysituation in which Jill is ready. On the other hand, the sentence ‘Jill isnot ready’ expresses the minimal proposition that Jill is not ready andthis presumably expresses something true in any situation in which Jillis not ready. Obviously however there will be innumerable situations inwhich both sentences are true together (for instance, where Jill is readyto go to the party but not ready to take the exam). Yet, since, for [Cap-pelen and Lepore (2005)], the second sentence is just the negation of thefirst, this looks problematic.

Another point Borg (2007) raises is that Cappelen and Lepore (2005) seem to beexclusively interested in ‘bottom-up’ context-sensitivity, that is, the type of context-sensitivity exhibited by the members of the ‘basic set’. As she argues, from theircharacterisation of radical vs. moderate contextualists, Cappelen and Lepore do notseem to recognise a difference between syntactically triggered context-sensitivity andwhat might be called ‘top-down’ context-sensitivity. The latter is manifested in theclaim by, e.g., RT that even with sentences which do express full propositions (ofthe type You are not going to die) some kind of pragmatic enrichment is necessaryin order to determine the proposition that was intended by the speaker. However,Cappelen and Lepore (2005) classify RT as a form of radical contextualism, whichthey characterise as follows:

Radical contextualists: claim that every expression of a natural languageis context sensitive (Cappelen and Lepore 2005, 5-6)

Borg (2007) claims that this characterisation does not apply to RT or other ap-proaches categorised as radical contextualist, at least not if context-sensitivity isunderstood in the ‘bottom-up’ sense. As Borg (2007) argues, this can be seen bythe fact that RT allows for processes such as free enrichment, which are specificallycharacterised as introducing material into the semantic form of an utterance whichis not syntactically triggered. In other words, the material free enrichment addsdoes not constitute the value of some context-dependent variable that forms partof some linguistic expression’s semantic form. However, considering RT’s view oflexical meaning as simply constituting a pointer to the meaning an expression willbe assigned on a particular occasion of use, Cappelen and Lepore (2005)’s charac-terisation of RT as radical contextualist does not seem totally inappropriate.

32For a similar concern cp. Pagin and Pelletier (2007).

Turning to the notion of what is said, Borg’s view of semantics as deliveringsentence meaning relativised to a context of utterance actually corresponds to thetraditional Gricean characterisation of the level of what is said. However, this is nothow Borg views this latter notion, nor do Cappelen and Lepore. As should havebecome obvious, both Borg as well as Cappelen and Lepore clearly take what is saidto be pragmatic33 although their characterisations differ in detail. Borg takes whatis said to be pragmatic principally because it involves such non-objectively obtainedcomponent parts as the intended referent of indexical expressions like that, whereasthis particular type of information is allowed to figure on the level of sentence mean-ing relativised to a context of utterance by Cappelen and Lepore. However, bothapproaches have in common that they assume that what is said, in contrast to sen-tence meaning may include unarticulated constituents. Thus, what is said is partof speaker meaning, where the latter is independent of sentence meaning. That is,what a speaker means by making a particular utterance may be totally independentof what the sentence he utters means.

In contrast to Borg, Cappelen and Lepore (2005) claim that the semantic contentof a sentence, actually does form a proper subpart of the content of the speech actcarried out by the speaker of that sentence. However, as Borg (2007) points out,this assumption is problematic.

. . . we should note that speakers could normally only count as assertingminimal propositions given a quite specialised notion of assertion, onewhich does not rely on speakers being consciously aware of what theyare asserting. For it is clear that the speaker who looks in the fridge andsays ‘There is nothing to eat’ would be surprised to learn that they hadasserted the quite general proposition that there is nothing to eat (in theuniverse). (Borg 2007, p. 15)

And, as Borg (2007) remarks further, there are of course cases in which the speakerclearly does not want to be taken to assert the minimal semantic content of thesentence she utters, e.g., cases of irony. It should be noted, however, that Cappelenand Lepore (2005) do not claim that speakers in carrying out a particular speechact actually assert the minimal semantic content.

It is important for us (a) not to identify the proposition semanticallyexpressed with the propositions asserted (or said), and (b) for that reasonnot to reserve the label ‘what is said’ for the proposition semanticallyexpressed. (Cappelen and Lepore 2005, p. 150)

This would mean, however, that speech act content does not only capture whatthe speaker intentionally asserted and intentionally conveyed, but may also includemeaning aspects that he distinctly does not want to assert. It is questionable whether

33Remember that for Borg as well as Cappelen and Lepore pragmatic does not simply meancontext-dependent, as they assume of semantics that it deals with context-dependent meaning as-pects at least up to a certain degree. Moreover, and as mentioned above, they use the term mainlyto pick out Grice’s functional characterisation of what is said, namely that it is the level of meaningthat is the basis for further pragmatic inferences (e.g. my utterance meaning).

this is desirable. On the one hand, in cases as given above by Borg, this wouldmean that the speech act content may be contradictory. On the other hand, in thosecircumstances in which the addressee has no particular difficulty in understandingwhat the speaker said and meant, there is no reason to assume that he, nevertheless,takes the speech act content to consist of contradictory information.

As Borg states, Cappelen and Lepore are led to this particular assumption be-cause they claim that, contrary to assumptions made by contextualists, the minimalsemantic content or minimal proposition of a sentence does have a unique role toplay in the comprehension process. It can serve as a ‘fall back’ in those cases wherefor some reason it is not clear what the speaker intended (cf. Ariel’s (2002) char-acterisation of linguistic meaning, discussed in section 5.1.2). The authors arguethat assuming that minimal semantic content is part of any possible speech act con-tent the sentence in question might be used to express has the advantage that incases where only little is known about the situation in which an utterance is made orabout the speaker of that utterance, a hearer may still be able to at least capture theminimal semantic content. Borg, however, argues against this view. She mentions anumber of experiments which show that it is likely that the minimal proposition ofa sentence may not even get processed during comprehension, namely in such cases,where pragmatics operates locally. It should be noted that Borg does not want todeny that semantic content may play the role of a ‘fall back’ in particular cases.What she wants to deny is that it is therefore necessary to assume that semanticcontent need be part of speech act content.

sentence meaning rel. context

what is said

Pragmatics

what is meant

fully-propositional

+ broad context

+ narrow context

liberal truth-conditions

Figure 3.8: Borg’s view

Thus, although Borg’s and Cappelen and Lepore’s approaches differ in a num-ber of details, there still are some basic assumptions they have in common. Bothapproaches assume that the semantic component delivers the truth-conditions ofthe sentence used in an utterance. That is, the results of semantic composition arenot sub-propositional forms, but full propositions which are truth-evaluable. How-ever, these truth conditions are in some sense liberal in that they do not necessarilyreflect what the speaker of an utterance intended to express by using a particularsentence. Thus, the truth-conditions determined for sentences relative to contexts

sentence meaning rel. context

what is said

Pragmatics

what is meant

fully-propositional

+ broad context

liberal truth-conditions

Figure 3.9: Cappelen and Lepore’s view

are not identical to the truth-conditions of the utterance meaning of that sentence.The advantage of such an approach to the nature of the semantic component isthat it can be assumed to be rather restricted in its operation, while still fulfillingthe role it traditionally was supposed to have, namely to account for the logicalrelations holding between particular (simple as well as complex) expressions. How-ever, a disadvantage is that it is unclear, which role such minimal propositions playin the overall interpretation process. As mentioned above, it seems that they donot necessarily have to be computed during the interpretation of a particular ut-terance. Furthermore, they are actually irrelevant (unless they correspond to theactual truth-conditions), in a sense, because what the hearer is normally interestedin is the actual not the liberal truth-conditions of some utterance. Moreover, it wasassumed that they may serve as a kind of fall back in the case of misunderstanding orinsufficient information concerning the context of an utterance, etc. However, thereseems to be no reason why a hearer trying to reconstruct what the speaker mighthave intended to express should fall back to this particular level of meaning. Afterall, in a situation in which the hearer is not sure about the contextual conditionsunder which an utterance including indexicals was uttered, the semantic form thathe falls back on is going to be sub-propositional in form. Note also that this is notjust so in cases of misunderstanding but also holds in all such situations in which thespeaker utters a sub-sentential syntactic form. As mentioned already, in the case ofsub-sentential utterances, the semantic component simply cannot provide a truth-evaluable semantic form just on the basis of the expressions used in the utteranceand their possibly attending context-sensitivity. Thus, liberal truth conditions maynot actually have any role to play in the interpretation process after all. Borg wouldconcur in this conclusion. However, she argues that although liberal truth conditionsmay not have a role to play in the actual interpretation process, they are still animportant and necessary level of meaning, as they establish the inferential relationsholding between expressions in a language. As mentioned already in section 2.3.3,the question is whether this type of inference is actually useful at this level. On theone hand, hearers draw inferences that matter to them in communication based on

the actual truth conditions of an utterance. On the other hand, capturing the logicalrelations between particular readings of particular expressions within the semanticcomponent seems only possible if the lexical meanings of expressions are taken to befull readings. However, we saw that there are arguments against generally assumingfull-fledged readings as part of the lexical entries of linguistic expressions. Thus, ifsemantic meaning is taken to be characterised by a high degree of underspecification,then capturing the logical relations between particular full readings should not beviewed as a task of linguistic semantics anyway, but possibly rather a property ofthe language of thought (cp. Burton-Roberts 2005).

3.2.4 Minimal Proposition vs. Proposition Expressed

Looking again at the difference between minimal proposition (corresponding toGrice’s what is said and Borg’s liberal truth conditions) and proposition expressed(corresponding to RT’s explicature or Bach’s impliciture and Borg’s actual truthconditions): although an utterance of the sentence in (64a) necessarily expresses aproposition, it is questionable whether it is the one the speaker of (64a) intended.Imagine a situation in which a mother says to her small son, who has just had acrash with his bike:

(64a) It’ll take time for this wound to heal.

It is more likely that what the speaker of (64a) wanted to express by her utteranceis the proposition that ‘it’ll take a considerable amount of time for the contextu-ally relevant wound to heal’. This is especially important considering that what issaid is supposed to be the basis for further inferences that lead to what the speakeractually meant by her utterance, that is, the basis for conversational implicatures.However, from the apparent tautology expressed by an utterance of (64a), no fur-ther implicatures seem plausible. In contrast, depending on the particular contextualcircumstances, from the speaker intended proposition expressed ‘it’ll take a consid-erable amount of time for the wound to heal’ further inferences may be drawn. Forexample, the child might draw the inference that he will not be able to ride his biketogether with his friends the next day.

In consequence, it has been claimed (specifically by RT proponents) that theminimal proposition or Gricean what is said by an utterance, actually does not haveany role to play in the interpretation process. In fact, this level of meaning is claimedto not be psychologically real. Thus, if what is said has no psychological reality, theassumption is that it will not be consciously available to interpreters. There area number of experiments that have tested that claim (e.g. Gibbs and Moise 1997,Nicolle and Clark 1999, Bezuidenhout and Cutting 2002).

Thus, Gibbs and Moise (1997) report a number of experiments they carried outin order to verify whether what is taken to be an utterance’s minimal proposition(or what is said), is what naive speakers actually do take to be what the speaker ofthat utterance has said. Thus, they assume Recanati’s availability principle34 and

34‘In deciding whether a pragmatically determined aspect of utterance meaning is part of whatis said, we should always try to preserve our pre-theoretic intuitions on the matter’ (Recanati 1989,p. 310).

accordingly were interested in testing naive speakers’ intuitions on the matter of whatis said and whether they are able to consciously recognise a level of meaning belowthe level of communicative sense or what is meant.35 Thus, their first experimentasked subjects to choose one of two given paraphrases of a particular utterance theywere presented. One such utterance is that given in (100a), where subjects wereasked to choose either (100b) (the minimal proposition) or (100c) (the propositionexpressed) as what they thought best captured what the speaker said.

(100) a. Jane has three children.

b. Jane has at least three children but may have more than three.

c. Jane has exactly three children.

The hypothesis was that if interpreters use pragmatic information in order to de-termine what the speaker said with his utterance, they should choose the enrichedparaphrase. This hypothesis was verified by the results of that experiment, as sub-jects chose the enriched paraphrase significantly more often as what they thoughtthe speaker had said with his utterance.

In a second experiment, Gibbs and Moise (1997) checked whether subjects pos-sibly do access the minimal proposition during their interpretation, but simply arenot consciously aware of that fact. Thus, before presenting subjects with the ac-tual data, they were informed about the difference that can be made between whata speaker says and what he implicates by making an utterance. Although in thisexperiment, subjects chose the minimal proposition as what they thought best re-flected what the speaker said slightly more often, overall they still preferred theenriched proposition over the minimal proposition.

A third experiment tested whether naive speakers are actually able to differen-tiate between what a speaker says and what he implicates by his utterance. Thus,in this experiment, the critical sentences were preceded by short contexts whichsuggested certain implicatures as arising from the speaker’s utterance of the criticalsentence.

(101) a. Bill wanted to date his co-worker Jane. But Bill really didn’t know muchabout her. Being a bit shy, he first talked to another person, Fred. Fredknew Jane; fairly well. Bill wondered if Jane was single. Fred replied, Janehas three children.

b. Jane has exactly three children, but no more than three.

c. Jane is married.

Once again, participants were asked to chose the paraphrase that they thoughtcaptured what the speaker had said. This time, one of the paraphrases includedthe suggested implicature, the other the enriched form of what is said. As before,subjects overwhelmingly chose the enriched paraphrase over the paraphrase thatincluded the implicature, which shows that they are able to differentiate between alevel of meaning that does not include implicatures and one that does. Furthermore,

35Note that Gibbs and Moise (1997), following RT, assume that GCIs are part of what is saidrather than ‘proper’ implicatures which are only part of what is meant by an utterance.

it shows that subjects in the first two experiments did not simply choose the mostenriched paraphrase available.

The general conclusion that can be drawn from the results of this experimentis that, apparently, the level of meaning representend by the minimal propositionexpressed by an utterance is one which speakers do not consciously make use ofwhen they interpret an utterance. Although the experiments do not prove thatspeakers do not access such a level of meaning at all, from a psychological point ofview, the level of minimal proposition seems to have no role to play in the overallinterpretation process. It is not consciously accessed by speakers, nor is it a helpfullevel of meaning when it comes to hearer’s inferring implicated aspects of meaning.Rather, what hearers consciously take a speaker to have said is a more enrichedproposition, the proposition expressed. Moreover, subjects choosing as what is saidby an utterance the enriched proposition that included what traditionally are calledGCIs shows that these pragmatically determined aspects of meaning are part of thelevel of what is said rather than what is meant. This is especially apparent whenlooking at Gibbs and Moise (1997)’s third experiment, which tested subjects’ abilityto differentiate between those two levels of meaning.

However, Nicolle and Clark (1999) point out quite a number of critical aspectsconcerning Gibbs and Moise (1997)’s experiments. For example, they are concernedthat Gibbs and Moise (1997)’s data includes quite a number of utterance types whichmay not actually be amenable to a minimal proposition analysis. Moreover, theycriticised Gibbs and Moise (1997)’s method of instructing subjects to choose whatthey thought best captured what a speaker said over the Gricean characterisationof what is said they were informed of in the second experiment. Most importantly,Nicolle and Clark (1999) claim that it is not clear whether subjects, especially inthe third experiment, actually did distinguish between the two levels of propositionexpressed and what is meant or whether their choices may have been influenced bysome other criterion. Thus, they put forward the hypothesis that speakers choosethat paraphrase which is most similar with respect to the way in which the originalutterance achieved relevance. Using an original example from Gibbs and Moise(1997), given in (101) above, they explain what this underlying criterion meansfor subjects’ selection of the best paraphrase of what the speaker said (in a non-technical sense). Thus, they point out that on the basis of the explicature of thecritical utterance together with contextual information given in the preceding text,the addressee computes the intended implicatures of that utterance, which mayinclude any of the following.

‘Jane is married; Jane has a partner (but is not necessarily married);Jane is experienced with men (whereas, Bill is too shy even to approachher, hence they may not be compatible); Bill should only date Jane if helikes children; Jane has family responsibilities and so may not want toenter a relationship’ (Nicolle and Clark 1999, p. 350).

Since (101c) only expresses one of the possible communicated implicatures, it doesnot qualify as an ‘. . . optimally relevant paraphrase’ of the original utterance (Nicolleand Clark 1999, p. 350). Thus, speakers choose the paraphrase which most likely al-lows the same amount of cognitive effects as the original utterance. If, in a particular

instance, this paraphrase happens to be an implicature rather than an explicatureof the utterance, subjects will choose the implicature paraphrase. Thus, considerexample (102).

(102) a. An informal five-a-side football match had been arranged for Saturdaymorning at 11:30h. That morning was warm and sunny, and over twentypeople showed up. ‘Can we play two five-a-side matches at the same time?’asked Steve. John answered, ‘Billy’s got two footballs’.

b. Billy has got exactly two footballs and no more than two.

c. There are enough footballs to play two matches.

The relevant cognitive effect of John’s answer in this particular situation is, of course,that there are enough footballs to play two matches. Although there may be otherinferences that may be derived from John’s utterance (e.g. Billy is an organisedperson, Billy likes football), in the given context it seems unlikely that they wouldbe relevant. Thus, the second paraphrase makes the relevant cognitive effect im-mediately accessible. Therefore, the assumption is that subjects presented with anexample such as (102) will choose the second paraphrase as best reflecting what thespeaker had said.

This hypothesis was tested in a first experiment, which had three conditions. Inthe first condition, subjects were instructed to choose the paraphrase they thoughtbest reflected what the utterance said. In condition two, subjects were explicitlyasked to choose the paraphrase that best captured what the words the speaker usedmean, taking into account that naive speakers use the phrase ‘what is said’ differentlyfrom its theoretical use. In the last condition, subjects were asked to choose theparaphrase that best reflected what the speaker wanted to communicate. The resultsof all three conditions show that speakers mostly choose relevant implicatures overenriched explicatures, even when they are instructed to determine what the speakerhad said. Moreover, the results were similar across the three different conditions,suggesting that subjects did not differentiate between what is said by an utterance,what is meant by the speaker of an utterance or what the words used in an utterancemean.

Thus, Nicolle and Clark (1999)’s results refute the assumption that naive speak-ers intuitively differentiate between the two theoretical levels of meaning what is said(or, more accurately here, explicature) and what is meant.36 Rather, the criterion bywhich they determine the best paraphrase of a speaker’s utterance is the criterion ofrelevance. That is, they choose the paraphrase that is most similar to the originalutterance in the amount of cognitive effects that it allows the hearer to gain from it.Having said this, Nicolle and Clark (1999)’s and Gibbs and Moise (1997)’s resultsboth show that speakers do not usually opt for the original Gricean level of whatis said, that is, they do not decide on the minimal proposition as what is said bysome utterance. However, as pointed out by Bezuidenhout and Cutting (2002), thismay simply be due to the fact that they concentrate on what is most relevant in acontext and often that is not what, strictly speaking, has been said. Thus, althoughin the experiments ‘. . . the minimal paraphrase was rarely chosen, this result by

36Cp. Bezuidenhout and Cutting (2002),who replicated Nicolle and Clark (1999)’s results.

itself does not show that hearers do not recover the minimal proposition at somelevel of processing’ (Bezuidenhout and Cutting 2002, p. 443).

In a last experiment, Nicolle and Clark (1999) tested whether naive speakerscan be made aware of the difference between theoretical what is said and what isimplicated, although their choice of paraphrases in the experiments before seems notto have been based on an intuitive differentiation of these levels. Thus, as in Gibbsand Moise (1997)’s experiment, subjects were given a tutorial on the respectivedifferentiation. They were then presented the material from the first experiment.However, in contrast to Gibbs and Moise (1997)’s experiment, subjects here wereexplicitly asked to make their choices based on the technical definitions of what issaid and what is implicated. The results show that speakers can be made aware ofthe technical difference, although normally, they do not understand what is said inthe technical sense.

Thus, as shown already by Gibbs and Moise (1997)’s experiment, naive speakersseem to have quite a different understanding of the expression ‘what the speakersaid’ from what is captured by the technical definition. What is more, none ofthe experiments in Nicolle and Clark (1999)’s study verify Recanati’s availabilityprinciple, that is, speakers do not seem to have reliable intuitions concerning thetechnical differentiation between what is said by an utterance in contrast to what ismeant. Therefore, it does not seem very helpful to rely on intuitions in characterisingthe levels of what is said and what is meant.37 Thus, the fact that subjects in Gibbsand Moise (1997)’s experiment chose the enriched proposition including so-calledGCIs as what is said by the speaker, may be due to the particular design of theexperimental material used. In other words, the results of that experiment may infact tell us nothing about whether GCIs belong to the level of what is said/explicatureor whether they are genuine implicatures and as such belong to the level of whatis meant. Be that as it may, Nicolle and Clark (1999)’s results do corroborateRT’s assumption that the processes that derive explicatures are not so very differentfrom those deriving implicatures. If this assumption is correct, one would actuallypredict that it should be difficult to find a particular level of meaning which includesexplicatures but does not include implicatures.

However, Bezuidenhout and Cutting (2002) claim that neither the experimentsreported in Gibbs and Moise (1997) nor those in Nicolle and Clark (1999) actuallytell us what exactly happens during the interpretation process, since both are onlyconcerned with the products of that process. More specifically, they do not allowus to draw conclusions about potential stages in the overall interpretation process,and therefore, do not provide an answer to the question of whether or not a minimallevel of meaning such as Grice’s notion of what is said plays a role during interpreta-tion. In order to shed some light on what exactly is going on during interpretation,Bezuidenhout and Cutting (2002) designed the following experiment. They con-structed little stories in which the last sentence was the target sentence. The storieswere constructed so as to either bias a minimal or an enriched interpretation of thetarget sentence. Time taken to read the target sentence was measured, as well as the

37Note that this also makes the appropriateness of Recanati’s distinction between different typesof non-literal meaning doubtful (more on which below), since the distinction again is based on theavailability principle. For a general critique of assuming such a principle, see e.g., Bach (2001b).

time taken to make a match/mismatch decision on a sentence that was shown afterthe target sentence. This sentence never matched the target sentence, but rathercorresponded to the Gricean, and thus minimal, notion of what is said by the targetsentence. The stories were presented line by line on a computer screen and subjectshad to press a button for the next line to appear. After the target sentence, they sawa sentence of which they had to decide whether it exactly matched the respectivetarget sentence they had seen just before.

Bezuidenhout and Cutting (2002) tested the predictions made by three differentmodels of processing for target sentence reading times in the two contexts. Accord-ing to the Literal-First Serial (LFS) model, target sentences in minimal contextsshould be read faster than in enriched contexts, since their minimal proposition iscomputed in both contexts and in the minimal context it is the only propositioncomputed. This is in contrast to the Local Pragmatic Processing (LPP) model,which assumes that computation of minimal propositions is not necessary, but onlytakes place if warranted by the context. Thus, reading times for target sentencesshould be equivalent for the two contexts. The third processing model tested is theRanked Parallel (RP) model, which assumes that all interpretations are determinedin parallel, but that the interpretations are ranked in terms of their accessibility.This model assumes that enriched interpretations generally are higher ranked andthus more accessible than minimal interpretations.

The results show that subjects took longer reading the target sentences in thosecontexts which biased a minimal interpretation, than in the contexts which biasedan enriched interpretation. As regards the time taken for the match/mismatch de-cision, whether or not the preceding context was biasing a minimal or enrichedinterpretation of the target sentence did not make any significant difference. Gen-erally, Bezuidenhout and Cutting (2002) interpret their results as supporting theRP model of the interpretation process, in which both the minimal as well as theenriched interpretation are determined in parallel, but where the enriched interpre-tation is more accessible than the minimal one. Since the enriched interpretationis more accessible, it interfers with the minimal one in the contexts that are biasedtowards the latter. However, as the authors themselves point out, there are someproblems with such an approach. Thus, that the enriched interpretation generallyis more accessible than the minimal one is a questionable assumption. Moreover, itis not clear whether there is only one single highly accessible enriched interpretationin a particular situation, which seems to be implied by such an approach.

Finally, the authors also are concerned that their match/mismatch task actuallyis inconclusive concerning the question of whether during interpretation a minimalproposition is accessed (cp. Garrett and Harnish 2008, p. 71). As already mentioned,the time taken to make a decision did not significantly differ for the two types ofcontexts. As Bezuidenhout and Cutting (2002, p. 454) state, ‘[i]t is necessary to havesome baseline time for this sort of match/mismatch judgment task, to show thatthere was in fact the slowing in judgment times that would be (indirect) evidenceof interference from the minimal interpretation’. Thus, although the authors chosean experimental design of which they argue that it should allow conclusions to bedrawn concerning the question of whether a minimal proposition is accessed duringthe interpretation of some utterance, the result was, at least partly, a null effect.

Summary 129

3.3 Summary

For a summary of the approaches to the characterisation of utterance meaning con-sidered in this chapter, let us start by recapitulating the relevant aspects of Grice’slevel of what is said. Grice took what is said to be a full, and therefore truth-evaluable proposition, determined on the basis of narrow contextual information.It is the proposition the speaker asserted and thus can be taken to be committedto. Moreover, it forms the basis for further pragmatic inferences. One problematicaspect of Grice’s characterisation of the processes that lead to what is said is thatthey may only make use of narrow context. Thus, it has been argued that thismight result in what is said actually not being fully propositional. Moreover, even ifwhat these processes determine is a full proposition, it has been claimed that Grice’swhat is said actually is not informationally rich enough to form the basis for furtherpragmatic inferences. That is, if what is said is a proposition at all, very often itis only a minimal one and not what the speaker will be taken to have intended toexpress. However, it seems that further relevant pragmatic inferences can only bedrawn on the basis of the proposition actually expressed by the speaker of a par-ticular utterance. Concerning the determination of the proposition expressed, thequestion arises whether only linguistically mandated processes should be assumed tooperate, or whether one should allow for unarticulated constituents to be integratedby pragmatic processes.

RT’s as well as Bach’s solution to the problems Grice’s level of what is said facesis to assume that more processes than traditionally recognised are involved in thedetermination of the level of utterance meaning, which level RT calls explicature andBach impliciture. Thus, in order for explicatures/implicitures to get determined,processes are assumed that take into account the broad context of utterance. Theexplicature/impliciture of an utterance, then, is the proposition the speaker intendedto express by his utterance. Its determination involves not only linguistically man-dated processes, but also unarticulated constituents. Moreover, it is taken to be theproposition the speaker is committed to by his utterance. In contrast to Bach, RTdoes not recognise Grice’s what is said as a psychologically real level of meaning.Rather, pragmatic processes apply directly to an utterance’s sub-propositional log-ical form and determine the proposition expressed by the speaker. Although Bachdoes recognise Grice’s what is said in that it forms the basis for implicitures, heagrees that the former may be of sub-propositional form, since for its determinationonly narrow context is considered. What is problematic about the respective char-acterisations of explicature and impliciture is that they are taken to be cancellable.If explicature/impliciture is the proposition the speaker is taken to have intended toexpress, the question is how this can be cancelled without contradiction?

In contrast to both RT and Bach, Stanley claims that only linguistically man-dated processes are at work in the determination of utterance meaning. This meansthat there are no such things as unarticulated constituents at this level of meaning,only hidden indexicals. Generally, Stanley’s approach can be taken as an attemptto rescue the traditional Gricean notion of what is said, in arguing that particularmeaning aspects that can be found at the level of utterance meaning, but are notpart of the semantic form of an utterance, are actually the result of the process that

resolves indexicals. The major critique levelled at the hidden indexical approachis that such hidden indexicals would have to be taken to pervade semantic forms,inspite of the fact that in many cases they will not even be provided a value bythe relevant pragmatic process. One advantage of this approach, however, is that itallows a clear definition of the level of utterance meaning or what is said as the levelof meaning resulting from the application of saturation processes to the semanticform of an utterance. In contrast, both RT and Bach’s approach suffer from animprecise definition of explicature and impliciture, respectively, as the ‘development’of or being based on a semantic form.

Both Borg’s and Cappelen and Lepore’s approaches are similar to Stanley’s inthat they only assume linguistically mandated processes to be involved in the de-termination of the truth-conditions of a sentence relative to a context of utterance.However, one major point they differ on is that the former take what is said tobe a level of meaning distinct from that of sentence meaning relative to a contextof utterance. More specifically, both these approaches assume that the semanticscomponent determines the proposition expressed by a particular sentence in a par-ticular context, where this proposition does not necessarily have to correspond tothe proposition intended to be expressed by the speaker of that sentence. Thus,whereas Borg’s approach is similar to Stanley’s in that she assumes that there aremore context-sensitive expressions than those in the ‘basic set’, the former differsfrom the latter approach in that Borg does not allow for information based onreasoning about the speaker’s intentions in making a particular utterance to enterinto the determination of the proposition expressed by that sentence. In contrast,Cappelen and Lepore do not rule out that speaker intentions might play a role in de-termining the proposition expressed by a sentence, however, they argue that the setof context-sensitive expressions as such is restricted and consists only of those in the‘basic set’. In both approaches, what the semantic component determines are liberalas opposed to actual truth conditions for utterances. The question here is what rolethese liberal truth conditions play in the mental life of speaker-hearers and, thus,why they should be determined in the first place. That is, it seems rather plausiblethat it is the actual truth conditions that a speaker-hearer would (unconsciously)base any further inferences on actually relevant for him.

One of the questions concerning Grice’s level of what is said that came up inseveral of the approaches discussed is whether it is psychologically real, that is,whether hearers actually compute this level of meaning when engaging in utteranceinterpretation. Recanati’s assumption based on his availability principle is that ifhearers access this level of meaning, they should be consciously aware of it andthey should be able to reliably differentiate this level of meaning from the level ofwhat is meant, since the latter is assumed to be based on what is said. However, asthe results of experiments show, the levels of meaning computed during utteranceinterpretation cannot be verified by relying on hearers’ intuitions. One of the reasonsis that naive speaker-hearers do not understand the term ‘saying’ as envisaged bytheorists. Another reason is that the fact that hearers may not consciously be awareof a particular level of meaning as such is not evidence for that level not beingaccessed at all.

Having looked at different approaches to the nature of utterance meaning – under-

Summary 131

stood as that level of meaning which forms the basis for further pragmatic inferences– the question of how this level of meaning relates to the revised characterisationsof literal and non-literal meaning can be approached. To repeat, from what hasbeen said in chapter 2 concerning the time-course of the interpretation of literal andnon-literal meaning, it should be clear that we want literal meaning and non-literalmeaning to be present at the same level(s) of meaning. Thus, it is not possibleto identify what is said as an utterance’s literal meaning on which any non-literalmeaning is based. In fact, the discussion in the present chapter concerning the ex-act nature of the level of meaning what is said and the kinds of meaning aspectsidentified as belonging to it make such a characterisation seem even more inappropri-ate. For instance, intuitively at least, one would probably not want to characterisemeaning aspects due to ad-hoc concept formation or free enrichment as literal.

One possibility of capturing similarities and differences between literal meaningand non-literal meaning is to assume that of the processes that are involved indetermining what is said, a subset can be identified which results in the literalmeaning of a particular utterance. In this case, one has to make sure that theprocesses involved need not be assumed to take place in a particular order, since inthat case one might again end up predicting that a particular type of meaning istemporally prior to another. Another possibility is to assume that literal meaningand non-literal meaning can not only be found at the same level of meaning, butthat they actually result from the same set of processes. The differences betweenthe two types of meaning would then have to be explained in terms of differences inthe information used by those processes.38

Before we can turn to an investigation of these ideas in more detail, however,there are a few more issues that I would like to look at first. One point concernsthe treatment of metaphor as a phenomenon arising at the level of what is meantand thus being treated as rather different from such phenomena as metonymy oridioms and more similar to irony and indirect speech acts. Another point concernsthe possible treatment of generalised conversational implicatures as already arisingat the level of utterance meaning due to the fact that they differ in some crucial as-pects from particularised conversational implicatures. Recall that traditionally, onecriterion that seems to differentiate phenomena such as metaphor, irony, CI, speechacts from those found at the level of utterance meaning – or rather Grice’s what issaid – was that a consideration of speaker intentions is necessary to determine theirinterpretation. However, as we saw in the present chapter, there are theories that as-sume processes to contribute to the level of utterance meaning, where these processesalready take into account the potential speaker intentions in making the respectiveutterance. Thus, the question arises whether the theoretical differentiation betweenthe two levels of what is said/utterance meaning and what is meant/communicativesense actually is warranted. These issues are the topic of the next chapter.

38This idea will be taken up in chapter 5.

Chapter 4

Utterance Meaning andCommunicative Sense – TwoLevels or One?

The aim of chapter 3 was to characterise the level of what is said/utterance meaningas a preliminary to specifying the conditions that have to hold when some utteranceis said to have literal meaning. The question of how what is said differs from what ismeant was touched but not elaborated on. However, this is an important question toinvestigate, since, as with the level of what is said, Grice’s assumptions concerningthe nature of and phenomena found at the level of what is meant have not goneundisputed. Thus, recall for instance that at least for metaphor – traditionallyassumed to belong to the level of what is meant – we already saw that apparentlya prior determination of a literal meaning on which the metaphoric meaning can besaid to be based is not always necessary.

According to Grice, to get from the level of what is said to the level of what ismeant of an utterance, further, implied aspects of meaning get inferred on the basisof the former level. That is, what is meant combines both directly expressed andindirectly conveyed aspects of meaning content (cp. Bach 2006a). More specifically,what the level of what is meant is supposed to capture is the whole of the meaningthe speaker of some utterance wanted to convey. In other words, it is speakers whomean something by their utterances.1 Since speaker intentions play an importantrole in getting to this level, the meaning at the level of what is meant is also oftencalled speaker’s meaning or speaker’s intended meaning.

The phenomena traditionally assumed to belong to the level of what is meant im-portantly include conversational implicatures and speech acts.2 Recall further thatGrice originally took metaphor and irony to be instances of conversational impli-cature and thus to belong to the level of what is meant. However, as mentionedalready, there are arguments both against a treatment of metaphor and irony interms of conversational implicature as well as against generally treating metaphor

1Remember Grice’s differentiation of the four levels of meaning and the formulation ‘U meantby uttering x that . . . ’.

2Thus, although, e.g., conventional implicatures also play a role at this level, the discussion willcenter on the notion of conversational implicature and the phenomena originally subsumed underit.

134 Utterance Meaning and Communicative Sense – Two Levels or One?

and irony as involving similar comprehension processes. In fact, there are reasonsto assume that metaphor, in contrast to irony, actually belongs to the level of (ex-plicated) utterance meaning. Similarly, of the two general types of conversationalimplicature identified by Grice – particularised conversational implicature (PCI) andgeneralised conversational implicature (GCI) – the latter has received a considerableamount of attention leading to the question of whether it actually should be assumedas contributing to the level of what is said rather than to the level of what is meant.As regards PCIs, it has been suggested by RT that their inference is not necessarilybased on a fully propositional utterance meaning/explicature, which suggestion callsinto question the usefulness/possibility/necessity of differentiating between the twopragmatically determined levels of meaning what is said and what is meant at all. Infact, it has been claimed that there actually is no principled way of differentiatingwhat is said from what is meant (Cappelen and Lepore 2005).

Thus, in this chapter, I will review the phenomena originally assumed to belongto the level of what is meant, concentrating on metaphor (section 4.1.1), irony (sec-tion 4.1.2) and (mainly generalised) conversational implicature (section 4.1.3). I willonly mention a few issues concerning speech acts and here specifically indirect speechacts, since in their treatment the notions of literal and non-literal meaning have beenmade use of and since they have also been treated as a type of conversational impli-cature (section 4.1.4). Finally, I will turn to the discussion of the question whetherit is actually useful/possible/necessary to differentiate between two pragmaticallydetermined levels of meaning (section 4.2).

The overall aim of this chapter, then, is to find out which of the phenomenatraditionally identified as belonging to the level of what is meant actually do needa fully propositional utterance meaning as a basis and also how the former levelof meaning is characterised. Thus, the discussion here will also, again, contributeto the understanding of the level of utterance meaning, by possibly adding somephenomena to it that are not traditionally assumed to arise at it. Moreover, thequestion will be pursued what kind of context the individual phenomena discussedrequire for their interpretation, that is, what types of information are necessary forthe interpretation process to determine the various phenomena presented. This is animportant question for the discussion in section 4.2 concerning the differentiation oftwo levels of context-dependent meaning, on the one hand, as well as for determiningthe relations between the individual phenomena traditionally identified at the twolevels, on the other.

4.1 Problematic Phenomena

4.1.1 Metaphor

Traditional Characterisation and its Problems

As already mentioned in section 3.1.1, Grice analysed metaphor and irony similarlyin that he assumed that both are cases of a violation of the first maxim of quality:‘Do not say what you believe to be false’. Thus, recall the treatment of example(58) repeated below.

Problematic Phenomena 135

(58) You are the cream in my coffee.

The idea is that in case of an utterance of (58) by a speaker, it is clear that thespeaker cannot be trying to mislead the hearer, due to the blatant falsity of theproposition expressed by the utterance the speaker uses. Thus, the literal propositioncannot be what the speaker intended to express. Assuming that the speaker isadhering to the co-operative principle, the hearer will try to make sense of hisutterance, by drawing an appropriate implicature. For instance, depending on theparticular context of utterance of (58), the hearer might infer that the speaker meantto convey that the addressee is very important to him.

There are several problems with Grice’s approach to metaphor understanding.First, as already mentioned, his view that what is said is entailed by what is meantforces him to argue that in case of metaphor, the speaker is not actually sayinganything, as the proposition expressed by the sentence used is not assumed to bepart of what the speaker meant in making his utterance. If that is the case, then itactually seems that metaphor cannot be treated in terms of conversational maximviolation (since the maxims refer to what is said). Second, in Grice’s framework, thehearer is traditionally taken as rejecting the literal meaning of the expressions usedin a metaphoric utterance. However, it is arguable that the literal meaning playsan important role in determining the exact non-literal or metaphoric meaning thespeaker did in fact intend to convey. Third, and as mentioned several times before,Grice’s treatment of such non-literal meaning aspects as metaphor suggests that interms of the temporal enfolding of the interpretation process, non-literal meaningaspects will always be interpreted after the potential literal meaning of some complexexpression has been processed and rejected as not fitting the particular context ofthe utterance. We already saw that there is empirical evidence against such aview. Fourth, there are both theoretical considerations as well as empirical evidencesuggesting that understanding of metaphor involves processes different from thoseat play in understanding irony.3

Thus, in this section, I will discuss several alternative approaches to the under-standing of metaphor as well as the results of empirical studies aimed at investigatingthe interpretation of metaphor. Whereas Grice treated metaphor as a phenomenonarising at the level of what is meant, RT takes metaphorical meaning to belong tothe explicit meaning expressed by an utterance (Carston 2002b, Sperber and Wilson2008) and thus, as part of utterance meaning. In fact, RT claims that the interpreta-tion of metaphor does not essentially differ from that of (other types of not strictly)literal language. The underlying process involved in metaphor interpretation is ad-hoc concept formation, which we have already come across.

In a way similar to RT, Glucksberg and McGlone (1999) assume that in (nominal)metaphor interpretation, a ‘new’ category is created by the interpreter on the basisof particular properties of the metaphor vehicle that are taken to be relevant forthe particular metaphor topic. This new category, then, is what is attributed of themetaphor topic.

Thus, both in RT as well as in Glucksberg and McGlone (1999)’s approach, theliteral meaning of the metaphor vehicle plays a crucial role, as it forms the basis

3This point will be taken up in section (4.1.2)

from which the metaphoric meaning of the metaphor vehicle is inferred. Thus, incontrast to the traditional view of metaphor, the literal meaning of the metaphor isnot simply rejected or discarded. This is verified by the results of empirical studiesinvestigating the interpretation of metaphors. Thus, Rubio Fernandez (2007) showedthat irrelevant literal meaning aspects of a metaphor vehicle are activated initiallyand remain active for some time before being suppressed.

There are also formal approaches to the interpretation of metaphor which aimat offering an explanation of how a metaphorical meaning is determined withoutcompromising the principle of compositionality and generally, the autonomy of thesemantic component. Stern (2006, 2009), for instance, treats metaphor interpreta-tion as paralleling that of indexicals and demonstratives. Thus, he assumes a specialoperator in the semantics of natural languages, which indicates that its argument isto be understood metaphorically. Dolling (2001) proposes an operator that obliga-torily applies to all first-order predicates and provides them with structure that canbe, but does not have to be exploited in the further interpretation process.

Metaphor and The Similarity of Various Types of Meaning

Recall that one of the processes assumed by RT to play a role when moving fromexpression meaning to utterance meaning is ad-hoc concept formation. Thus, thisprocess adjusts the concept lexically encoded by a particular expression, so thatthe expression is understood in that particular situation as expressing the adjustedconcept.4 Sperber and Wilson (2008) argue that it is this general process thatunderlies the interpretation of various types of not strictly literal language use,metaphor amongst them.

Generally, in RT literal meaning as well as different types of non-literal meaningare assumed to be similar in that they result from the same overall interpretationprocess, which is geared towards achieving maximal relevance. In addition, thereare types of meaning which are not easily characterised as belonging to one or theother of the traditional categories, so called loose uses as in (103).

(103) a. Holland is flat.

b. France is hexagonal.

c. I must run to the bank before it closes.

Taking utterances of the sentences in (103) as constituting statements, it is not clearwhether they should be judged true or false. That is, strictly speaking, Holland, ofcourse, is not flat. Nor is France, strictly speaking, hexagonal. And the speaker of(103c) might utter it in a context where he actually starts walking rather fast indirection of the bank in question, but does not, strictly speaking, run. However, theaddressee of such an utterance would not intuitively and consciously assume thatthe speaker had said something false. Rather, such uses of the terms flat, hexagonaland run are ‘loose uses’ in that the idea or concept they are used to convey doesnot fully correspond to the strict concept they are taken to encode. That is, the

4See, however, Stern (2006), who argues against RT’s treatment of metaphor in terms of adhoc concept-formation and generally the view according to which metaphor is only one point in acontinuum of strictly non-literal meanings.

concepts they are understood as expressing are the result of the process of ad-hocconcept formation.

The existence of such a phenomenon as loose use together with the problematicnature of Grice’s analysis of such non-literal meaning as metaphor and irony as beingcases in which the speaker actually only ‘makes as if to say’ something and, thus,in effect would have to be taken to not having said anything at all, Wilson andSperber (2000) use to argue against the suitability of a maxim of truthfulness. Thatis, since there apparently are many cases in which speakers make utterances which,strictly speaking, are not literally true and where this fact does not seem to irritatethe addressee much, they conclude that interpretation of a speaker’s utterance isnot guided by some maxim of truthfulness, but rather by considerations of whichinterpretation would make the speaker’s contribution most relevant to the addresseeat the time and in the situation in which it occurs. Moreover, since, accordingto RT, what is encoded by the speaker’s utterance only forms the basis for theactual interpretation of what the speaker tried to convey, addressees do not haveany particular expectations that the encoded material expressed by the speaker betrue.

Thus, since under RT’s view, the only principle guiding an addressee in inter-preting an utterance is the principle of relevance, one would actually expect differentkinds of meaning (i.e. literal, non-literal, ‘loose’, etc.) not to differ from one anotherin terms of the processes involved in deriving them. ‘The very same process of in-terpretation gives rise to literal, loose, hyperbolical or metaphorical interpretations,and there is a continuum of cases which cross-cut these categories’ (Wilson andSperber 2000, p. 245).

Consider a situation in which Peter and Mary are planning their next holiday,which is supposed to be a cycling trip. Peter has just warned Mary that he doesnot feel in very good shape, whereupon Mary replies:

(104) We could go to Holland. Holland is flat.

This, of course, would be an example of ‘loose use’ of the expression flat. WhatWilson and Sperber (2000) claim is that the interpretation process Peter engages into derive Mary’s meaning is not much different from the process that would havebeen triggered by a slightly different utterance on Mary’s part, namely one involvingmetaphor.

(105) We could go to Holland. Holland is a picnic.

What makes the two interpretation processes similar is that they start from the sameassumption, namely that Peter assumes Mary’s utterance to be relevant. Moreover,in both cases, what seems to be involved in interpreting the respective utterances isan adjustment of the concepts expressed by flat on the one hand and picnic on theother. Generally, the resulting ad-hoc concepts might be narrower or broader (or acombination of both) than the encoded concept the process started out from.5

In Sperber and Wilson (2002) the authors compare a metaphorical to a literalinterpretation of the sentence in (106), where the important point, once again, is

5To be precise, actually what is narrower or broader due to ad-hoc concept formation is thedenotation of the resulting adjusted concept.

that the interpretation procedure is similar for the two interpretations and where,crucially, in case of the metaphorical interpretation of (106), the literal interpretationof the sentence used does not figure as an intermediate, much less, necessary step.

(106) John is a soldier!

Thus, imagine the following exchange between Peter and Mary.

(107) Peter: Can we trust John to do as we tell him and defend the interests of theLinguistics Department in the University Council?

Mary: John is a soldier!

Due to the particular context in which Mary makes her utterance, particular aspectsof Peter’s mental concept soldier that is activated by Mary’s utterance will be moreaccessible than others. This difference in accessibility of particular aspects of thesoldier concept correspond to a difference in the accessibility of possible implica-tions the interpretation process will deliver based on Mary’s utterance, cf. (108).

(108) a. John is devoted to his duty.

b. John willingly follows orders.

c. John does not question authority.

d. John identifies with the goals of his team.

e. John is a patriot.

f. John earns a soldier’s pay.

g. John is a member of the military.

According to the relevance-theoretic comprehension procedure, Peter, unconsciously,considers these implications in the order of their accessibility in this particular con-text. Since the idea is to stop the interpretation process when an interpretation isreached that satisfies the interpreter’s expectations of relevance, Peter will considerthe implications down to (108d). Note that Peter, in this particular situation willnot consider the implication in (108g), that is, what may be considered the literalinterpretation of Mary’s utterance. However, in a different situation he may do.

(109) Peter: What does John do for a living?

Mary: John is a soldier!

Now, Mary’s utterance in this particular context, may lead to a different degree ofaccessibility for the various aspects that are part of Peter’s mental concept of soldier.Most importantly, it may lead to a difference in accessibility of the implications Petermay draw on the basis of Mary’s utterance, such that in this particular context, theimplication (108g) might be the most and (108d) the least accessible implication.Thus, in this particular situation, Peter might stop his interpretation of Mary’sutterance after having considered implications (108g) and (108f). As Sperber andWilson (2002) note:

. . . by applying exactly the same comprehension procedure (i.e. followinga path of least effort and stopping when his expectations of relevance aresatisfied), Peter arrives in the one case at a metaphorical interpretation,and in the other at a literal one. (Sperber and Wilson 2002, p. 20)

Thus, Relevance Theory rejects the traditional view according to which literalmeaning and non-literal meaning are the results of different interpretation processes.More specifically, and similarly to Giora (1997, 1999, 2002) and Gibbs (1994), RTclaims that the interpretation of the literal meaning of an utterance need not form anintermediate step in the determination of its intended non-literal meaning. However,assuming that both literal meaning as well as non-literal meaning essentially arecomputed by the same process does not explain why in clear cases of metaphor ormetonymy, these types of meaning are perceived as deviating from some underlyingliteral meaning.6 Moreover, the same problems concerning the nature of encodedconcepts and the process of ad-hoc concept formation mentioned already also applyhere. Thus, if one assumes that what is encoded by an expression is only a pointerto potential concepts, then the process of ad-hoc concept formation cannot be takento operate on encoded meaning. Furthermore, if one assumes that concepts areatomic, then it is not clear how the process of ad-hoc concept formation derives acontextually fitting concept from the concept it applies to. Similarly, if concepts areatomic, it is not clear in which sense aspects of it can differ in accessibility.

Metaphor and Attributive Categories

Glucksberg and McGlone (1999) investigate the question of how people understandsuch metaphors as in (110).

(110) a. My job is a jail.

b. Sam is a pig.

c. Our love has become a filing cabinet.

Consider, for example, (110a). Here, the metaphor topic is my job, whereas themetaphor vehicle is jail. What seems to be asserted by an utterance of the sentencein (110a) is the idea that what is denoted by my job is part of what is denoted byjail, that is, jail is taken to be the superordinate category to which my job belongs.Literally speaking, we know that this is not the case. Nevertheless, Glucksberg andMcGlone (1999) claim, what superficially looks like a category-inclusion statement(i.e. 110a) is in fact a category-inclusion statement, only that the properties of thecategory denoted by the metaphor vehicle are changed such that the metaphor topicmay be included in the category denoted by the vehicle. ‘When someone says that“my job is a jail”, job and jail are cast into a common category, viz., situations thatare confining, difficult to get out of, unpleasant, etc.’ (Glucksberg and McGlone1999, p. 1542). In fact, what is ‘normally’ denoted by the expression jail, actuallycan be seen as the stereotypical subset of the newly created category. Thus, in(110a) jail gives the name to the superordinate category to which both the literaljail as well as the metaphor topic, my job, belong. Glucksberg and McGlone (1999,p. 1543) call such a category that ‘. . . is used to characterize a metaphor topic . . .’an attributive category.

As regards the question of how such an attributive category is created, Glucksbergand McGlone (1999) argue that the properties of metaphor topic and vehicle that

6Cp. Stern (2006, p. 248, footnote 3), who argues that RT ‘plays down the intuitive figurativityof metaphor by placing it on a continuum with loose use’.

may be relevant in this ‘creation’ have to be inferred by the interpreter. Thus, incontrast to Lakoff (1987), the authors do not assume that the creation of attributivecategories or the interpretation of metaphors, generally, relies on the representationof conceptual mappings between a source and a target domain. Rather, the hearerof such sentences as in (110) determines an interpretation of the respective sentence,based on the properties of the metaphor vehicle taken to be relevant for the appro-priate characterisation of the particular metaphor topic. Thus, even in the case of analready conventionalised metaphoric vehicle, the properties that will be importantin a particular instance of use will depend on the metaphor topic and contextualspecifics.

The authors’ view of metaphor interpretation is supported by the results of anexperiment they carried out, which was intended to investigate whether for their in-terpretation of cases of metaphor interpreters rely on conceptual mappings assumedto be already established. Thus, subjects were presented three love-metaphors,which they were asked to paraphrase. According to Lakoff’s view, the abstract con-cept love (the target domain) is understood in terms of more concrete concepts,for instance, as a journey or container (source domains). Specific properties ofthe source domain are mapped onto specific properties of the target domain. Thesemappings, it is argued, enable people to interpret such metaphors as in (111).

(111) a. Our love is a bumpy rollercoaster ride.

b. Our love is a voyage to the bottom of the sea.

c. Our love is a filing cabinet.

However, the paraphrases subjects offered for the metaphors in (111) do not supportsuch a view. Thus, in the paraphrases for the metaphor in (111a), no journey-related references were made. Glucksberg and McGlone (1999) consider the possibil-ity that the meaning of the metaphor vehicle bumpy rollercoaster ride might actuallynot entail any of the necessary journey-love mappings. This, however, should notbe the case for (111b), where explicit reference is made to journey by the expres-sion voyage. Thus, in this case, if Lakoff’s view is correct and subjects interpretthe metaphor in terms of the conceptual mappings between the concepts love andjourney, one can expect their paraphrases to be consistent. In fact, although someof the paraphrases made reference to journey-related aspects, not all of them didand not in the same way. Similar results were obtained for (111c), which was de-signed as a non-conventional instance of the conceptual metaphor relationshipsare containers.

Thus, overall the results indicate that interpreters do not automatically andinvariably use hypothesised knowledge about domain-to-domain mappings in inter-pretation. Rather, what seems to guide their interpretation of novel metaphors isthe question which properties of the metaphor vehicle can plausibly be attributedto the particular metaphor topic.7

7For similar results, supporting the attributive category approach to metaphor interpretation, seeMcGlone (1996). For an attributive category-supporting study investigating the neural correlatesof metaphor interpretation, see Stringaris et al. (2007).

Empirical Results Concerning Metaphor Interpretation

We already looked at a number of empirical studies which suggest that the tra-ditional view of metaphor as being secondary to literal meaning and necessarilynon-conventional is mistaken. Thus, recall the results attained by Glucksberget al. (1982) and Keysar (1989), which indicated that if some sentence can havea metaphorical interpretation, it will be processed, regardless of whether the sen-tence’s potential literal meaning is false or not and even in cases where subjects werespecifically asked to simply judge sentences as to their truth or falsity and, thus, ametaphorical interpretation was assumed not to be necessary. Thus, it seems thatthe processing of metaphor is not optional as suggested by the traditional approachto the understanding of non-literal meaning, but rather obligatory.

This is also assumed by Giora’s graded salience hypothesis. More specifically,recall that she assumes differences in processing depending on the familiarity statusof the meaning in question. Thus, for familiar metaphors she assumes that boththe expression’s literal as well as its metaphorical meaning are activated initially,regardless of the nature of the preceding context (i.e., whether it is literally ormetaphorically biasing) (Giora and Fein 1999). However, as also mentioned already,her results have to be treated with caution, since it is not clear on which basis shedifferentiated between metaphors on the one hand, and idioms on the other.8

Having said that, Blasko and Connine (1993) carried out a range of experimentsin which they tested the influence both of familiarity as well as aptness on metaphorinterpretation. The method they used was a cross-modal lexical priming task. Thus,subjects were auditorily presented utterances involving nominal metaphors (e.g., Xis Y) and had to carry out a lexical decision task on words presented visually atdifferent time points during the auditory presentation, namely at the offset of themetaphor vehicle (Experiment 1) or 300ms afterwards (Experiment 2). Blasko andConnine (1993) found that for highly familiar metaphors, their figurative interpre-tation was available immediately at offset of the metaphor vehicle as evidenced bysimilar reaction times to target words either related to a literal or metaphorical inter-pretation of the metaphor vehicle, both of which were reacted to more quickly thanto unrelated control words. However, for low-familiar metaphors, the metaphoricallyrelated target word was more slowely responded to than both the literally relatedand the unrelated control word. Similar results were obtained with the 300ms inter-stimulus intervall. In a third experiment, the authors tested whether aptness playeda role in the interpretation of low-familiar metaphors.9 The results suggest thatit does, as the response times to the metaphorically related targets in the case ofhighly-apt metaphors were significantly faster than for moderate-apt metaphors.

A further point Blasko and Connine (1993)’s results suggest is that the literalmeaning of a metaphorically interpreted expression is not simply and not immedi-

8See however Schmidt and Seger (2009, p. 385) who investigated the effect of figurativeness,familiarity and difficulty on metaphor processing, using fMRI. They too suggest that ‘. . . [t]hequality of being figurative may not be the crucial factor in determining the neural activationduring metaphor processing.’

9A post hoc analysis of the second experiment had suggested that aptness might play a role inthe case of low-familiar metaphors, but apparently does not have any effect on the processing ofhighly familiar metaphors.

ately suppressed during interpretation. Generally, the results of empirical studiesinvestigating the role of the literal meaning of a metaphorically used expression givesupport to the idea that during interpretation, aspects of the literal meaning of themetaphorically used expressions are enhanced, whereas inappropriate aspects aresuppressed. Thus, Rubio Fernandez (2007) makes explicit reference to both theattributive category and RT’s approach to metaphor. Although the two approachesdiffer in the specific theoretical assumptions they make concerning what is going onin metaphor interpretation, they actually make similar predictions concerning theactivation levels of relevant vs. irrelevant meaning aspects of the metaphor vehicleused in actual metaphoric utterances.

Thus, according to the attributive category approach, the interpreter has to con-struct an attributive category, which ideally the metaphor vehicle stereotypicallyexemplifies. In order to engage in this creation, the interpreter has to infer the rele-vant properties of the metaphor vehicle, with respect to the metaphor topic and thespecific context. In processing terms, this means that properties of the metaphorvehicle that are relevant are enhanced, whereas properties that are inappropriate orirrelevant are suppressed.10

Similarly, in ad-hoc concept formation, the encoded concept is adjusted by beingnarrowed and/or broadened. Recalling RT’s characterisation of concepts as consist-ing of a logical, lexical and encyclopaedic entry, ad-hoc concept formation can beunderstood to involve the following changes to the ‘organisation’ of elements in therespective entries of a concept. In cases of narrowing, a meaning aspect in the en-cyclopaedic entry of the concept is ‘promoted’ to the status of content-constitutive,i.e., it is transferred to the logical entry. In contrast, in cases of broadening, a logicalaspect is discarded. Rubio Fernandez (2007) suggests that in cases of narrowing, themeaning aspect that is promoted becomes and stays activated during interpretation,whereas in broadening, the meaning aspect that is eventually discarded is initiallyactivated but then suppressed.

Using the cross-modal lexical priming paradigm, Rubio Fernandez (2007) testedthis hypothesis. Subjects heard little texts, which always ended in a nominalmetaphor of the form X is a Y, Y being the metaphor vehicle and, thus, herethe prime. A target word was visually presented at 0, 400 or 1000ms after theword-recognition point of the respective prime and had to be judged as to whetherit is a word of the language. Target words either refered to the superordinate cate-gory or to distinctive properties of the metaphor-vehicle. In the control condition,the targets where unrelated to the primes. The context in which the metaphorwas presented was designed so as to render the superordinate category metaphor-irrelevant and the distinctive properties metaphor-relevant. Thus, if the hypothesisthat metaphor-irrelevant properties of the metaphor-vehicle are suppressed, whilemetaphor-relevant features are enhanced is correct, one would expect differences inthe reaction times of subjects when presented with the superordinate category targetas compared to the distinctive property target during the interpretation process.

The results support the tested hypothesis. That is, at 0 and 400ms after word-recognition point of the prime, both the superordinate category as well as the dis-

10Two further experimental studies in which this hypothesis was tested and confirmed are Gerns-bacher and Robertson (1999), Gernsbacher et al. (2001).

tinctive properties of the metaphor vehicle were equally active. However, at 1000msinter-stimulus intervall, subjects took significantly longer to judge superordinatecategory targets than to judge distinctive properties targets, suggesting that themetaphor-irrelevant superordinate category was suppressed during interpretation.This result is corroborated by results on the activation of superordinates in literalneutral contexts, where they are still active at 1000ms ISI (cf. Rubio Fernandezet al. 2003).

Note that the same paradigm has been used to investigate the activation ofthe meanings of homonyms, where, depending on the context given one meaningwould be intended and the other irrelevant. The results there parallel the re-sults for metaphor interpretation with one important exception. Thus, similarlyto metaphor interpretation, in homonymy interpretation both potential meaningsof the homonyms are activated initially, but the irrelevant one is suppressed lateron. However, this suppression is already manifest at a 200ms ISI. In other words,although both in homonymy and metaphor interpretation suppression plays a role,it does so at different time points. More specifically, Rubio Fernandez (2007) argues,whereas in homonymy interpretation, suppression occurs early and thus, in a sense,takes place automatically, in metaphor interpretation, suppression takes place laterand seems rather due to attentional processes. She explains the difference as follows.In homonymy interpretation, the potential readings are immediately available to theinterpretation process, which simply has to select one as the appropriate reading.In novel metaphor interpretation, the potential readings of the metaphor vehicle arenot immediately available, since they are actually only created during interpreta-tion. Thus, irrelevant aspects of the literal meaning of the metaphor vehicle stayactivated longer, because the actual metaphorical meaning has to be created beforeit can enter into the overall utterance meaning. Rubio Fernandez (2007, p. 364)suggests that it is ‘possible that metaphor awareness is related to the fact thatmore attentional resources may be involved in metaphor interpretation as opposedto disambiguation . . .’.

It would be interesting to see how the results of a similar study with, in Giora’sterms, familiar or highly salient metaphors as material would fall out. That is,are familiar metaphors similar to homonyms in that both their literal as well astheir non-literal meanings are stored and immediately available to the interpretationprocess, or are they more like novel metaphors in that both their literal and non-literal meanings are initially activated and retained beyond 400ms ISI, or are theysomewhere in between?

Formal approaches to metaphor interpretation

Stern (2000, 2006, 2009) proposes a treatment of metaphor along the lines of in-dexicals and demonstratives. Thus, he assumes an operator, ‘Mthat’, in analogyto Kaplan (1989b)’s ‘Dthat-operator’, which lexicalises the potential metaphoricalinterpretations for a particular expression type. The operator applies at the levelof logical form and constrains the interpretation of its argument. Thus, similarlyto RT and Glucksberg and McGlone (1999), Stern (2000, 2006, 2009) does not as-sume that a metaphorical interpretation of some utterance depends on the priordetermination of that utterance’s literal proposition, rather, what is metaphorically

interpreted is the metaphor vehicle (which may be simple or complex) and it is theliteral meaning of the metaphor vehicle, its character, that plays an important rolein the metaphoric interpretation.

Stern (2006, p. 262) characterises the ‘Mthat’ operator as follows.

. . . when prefixed to a (literal) expression φ, [the operator ‘Mthat’] yieldsa context-sensitive expression ‘Mthat[φ]’ whose tokens in each contextc express a set of properties presupposed in c to be m(etaphorically)-associated with the expression φ, such that the proposition < . . . {Mthat[φ]}(c) . . . > is either true or false at a circumstance.

Thus, the character of ‘Mthat[φ]’ is similar to the character of indexicals in thatit yields different contents when applied to different contexts of utterance. Notethat, in order for the context to function as an argument to the character of themetaphorical expression ‘Mthat[. . .]’, it has to be richer than the Kaplanian con-text for demonstratives and indexicals in that it additionally has to include sets ofpresuppositions or presupposed properties. In other words, the properties or propo-sitions that are taken to be metaphorically associated with a particular expression,have to be determined before the content of that expression in a particular contextcan be established. Unfortunately, Stern does not address the question of how thisis achieved.11 Obviously, determining the set of m-associated propositions for a par-ticular expression is not a task that can draw on objectively accessible information,as is the case in determining who are the speaker and addressee of an utterance andwhere and when that utterance takes place.

Stern (2009) draws support for his view from the fact that metaphor can be shownto pass the tests on context-sensitivity employed by Cappelen and Lepore (2005) toidentify truly context-sensitive expressions.12 For example, one test is the Inter-Contextual Disquotational Indirect Report Test (ICDIR). Here two contexts C andC* are assumed, where the latter differs from the former in some relevant aspect(s).A sentence S is uttered in context C and this utterance of S is disquotationallyindirectly reported in context C*. If any such report is true, then that is taken to beevidence for the context-insensitivity of the sentence S. For an illustration, considerthe following example, where the assumption is that (112a) and (112b) are utteredin contexts that differ at least in the speaker of the respective contexts.

(112) a. Mary: I am too tired to go to the cinema.

b. Peter: Mary said I am too tired to go to the cinema.

The idea is that Peter by his utterance of (112b) cannot report what Mary hassaid by using the same sentence Mary did, because in his context I will pick out anindividual different from the individual picked out in Mary’s context of utterance.Now consider example (113), which involves a metaphorical interpretation of thepredicate is a fish.

11And he does not seem particularly interested in this question either: ‘From the perspectiveof the semantics, all that is significant is that they are presupposed, abstracting away from theirpsychological sources’ (Stern 2009, p. 17).

12It should be noted, however, that the tests put forward by Cappelen and Lepore (2005) havebeen claimed to actually not differentiate between the expressions the authors claim are context-sensitive and those of which they want to show that they are not. Cf., e.g., Bezuidenhout (2006).

(113) a. Tip: Nixon is a fish.

b. John: Tip said Nixon is a fish.

The context of Tip’s utterance is the time of the hearings after Watergate inCongress. A journalist asks how the hearings are going and Tip responds with(113a) ‘. . . in a context C in which it is commonly presupposed that fish are slip-pery, slimy, and hard to catch’ (Stern 2009, p. 6). However, if Tip’s utterance isreported in a relevantly different context, namely, in a context in which differentpresuppositions hold concerning the metaphorical interpretation of is a fish, thenthe metaphorical interpretation will in fact be different from that intended by Tip.Thus, if John, who utters (113b), is an experienced fisherman and associates withthe metaphorical is a fish ‘. . . the salient presupposition that fish are easy to catchwith a little patience and bait’, then the truth-conditions of the embedded sentence(114a) will differ from those of Tip’s original utterance (114b).

(114) a. < Nixon, being easy to catch with a little patience and bait >

b. < Nixon, being slippery, slimy, and hard to catch >

Thus, it seems that metaphorically interpreted expressions behave similarly to suchindexical expressions as I in that they do not allow an ICDIR. In both cases, thereason is that the semantic content in the two relevantly different contexts is notidentical.13 Stern can explain this behaviour since he assumes that the ‘Mthat’operator turns the expression that it applies to into a context-sensitive expression,whose content varies with the contexts in which it is used. Thus, both Tip and Johnuse the same expression type, namely that given in (115).

(115) Nixon Mthat[is a fish].

However, they make their utterances in different contexts and, thus, the contextualinformation relevant for the interpretation of the metaphorically used expressiondiffers and leads to the different propositions given above being expressed.

To repeat, Stern (2009) assumes that there is a set of properties presupposed inthe particular context of utterance to be metaphorically-associated with the argu-ment of ‘Mthat’. Note though that, at least for the example discussed above, theproperties that figure in the propositions expressed by the respective utterances ofTip and John seem to me to be properties that are literally associated with the con-cept fish. This is especially obvious in the proposition taken to be expressed by Tip,which seems to me to be just a metaphorical paraphrase of what is expressed by hisutterance, but does not capture the actual proposition expressed. That is, (114b) isno more true in the particular context of utterance than the literal interpretation ofTip’s utterance is. Generally, I think that it is a simplification of what is actuallyrequired in metaphor interpretation to assume that the metaphorically-associatedproperties of a metaphor vehicle are given and part of the context before the in-terpretation of the metaphorically intended expression even begins. Thus, recallGlucksberg and McGlone (1999)’s assumption that the interpreter’s identification

13Of course, this argument only works if one accepts that the propositions given in (114a) and(114b) actually are the semantic contents of the respective utterances. See Stern (2009, pp. 9) fora defence of this view.

of the properties of the metaphor vehicle that can plausibly be attributed to themetaphor topic actually is part of the interpretation of a metaphor and cannot betaken to be presupposed.

Another critical point concerns the predictions Stern’s approach seems to be mak-ing conerning the relative difficulty of metaphor interpretation. That is, since Stern(2000, 2006, 2009) takes metaphors to work similarly to indexicals and demonstra-tives, one might expect that they are similarly difficult to interpret. However, resultsfrom neuropathological investigations go against that expectation. Thus, de Villierset al. (2007, 2010), for instance, show that autistic people have no great difficultyinterpreting indexical expressions, but it is well known that they do have problemsinterpreting cases of metaphor (e.g. Happe 1993, Gold et al. 2010, Rundblad andAnnaz 2010, Wearing 2010). Thus, treating metaphor as similar to indexicals, wheretheir content is, in a sense retrieved from the context of utterance, cannot capturewhat it is that leads to differences in the difficulty of interpreting metaphor in con-trast to indexicals in autistic people, unless one wants to argue that although thegeneral mechanism is the same, what is difficult for autistic people is to provide thepresuppositions necessary for successful interpretation.14 However, as I have alreadysaid, I think the idea of potential metaphoric content being provided by the contextis a problematic assumption itself.

Another approach that grounds metaphoric interpretation of utterances in theirsemantics is that proposed by Dolling (2000). Thus, Dolling (2000) assumes an op-erator, which, generally, allows both for metonymic and metaphorical interpretationsof the argument it applies to. The operator applies as part of the semantic com-position of an utterance’s logical form and enhances the general variation potentialof the meaning of the expression it applies to by introducing a range of parameterswhich have to get fixed during interpretation.

(116) met: λP λx.S3n(x) & ∃y [R3

n(y)(x) & S2n(y) & ∃z [R2

n(z)(y) & S1n(z) &

Qnv [R1n(v)(x)][P(v)]]]15 16

Note that the application of the met-operator is obligatory, that is, it applies to thesemantic form of all predicates that are of the appropriate type (namely <e,t>), re-gardless of whether that predicate will or will not be used with a shifted meaning ina particular utterance. In fact, Dolling (2000)’s approach ensures semantic compo-sitionality in two ways. On the one hand, the met-operator is not inserted into thesemantic form only once a conflict arises, but rather by default. On the other hand,the actual content of the met-operator (and, for that matter, of all SF-parameters,generally) is only supplied after semantic composition is finished. In case the en-hanced variation potential is not realised in a particular utterance interpretation,

14This line of argument is taken by Wearing (2010) in her defence of RT’s approach to metaphorinterpretation. Thus, she writes ‘. . . the difficulties with metaphor exhibited by some speakerswith autism can be explained as the result of having to interpret while lacking crucial information,rather than as a defect in the processing abilities required to understand a metaphor correctly’(Wearing 2010, p. 207).

15Sn and Rn as well as Qn are SF-Parameter which need to get fixed during interpretation, Sn

by a predicate of sorts of entities, Rn by a predicate of relations between elements of two sorts ofentity and Qn by some quantifier.

16The complexity of the operator is due to the fact that there can be chains of metonymy whichneed to be captured.

default values will be supplied to fix the SF-Parameters in the structure provided bythe met-operator and, as a result, the contribution of the operator will be reducedto zero.17 Thus, in contrast to Stern (2006, 2009)’s approach, the interpreter doesnot have to disambiguate, as it were, whether the speaker used a non-metaphoric ora metaphoric expression type.18 Moreover, Dolling (2000)’s met-operator is clearlyrestricted in its application to one-place predicates, where this does not seem to bethe case for Stern’s ‘Mthat’-operator.

Let us look at some examples involving the lexical item Lamm (lamb). Dolling(2000) gives the following semantic form (SEM) for Lamm.

(117) λx. x Kn lamm

Recall that Kn is an SF-Parameter which may be fixed by � (‘instance of’), ≤ (‘sub-kind of’) or = (‘is a’) and thus abstracts from the different readings of a systematicpolysemous expression. Thus, for the occurrences of Lamm in (118a – 118c), theparameter will be fixed as in (119a – 119c), respectively.

(118) a. Das Lamm ist ein Tier.

‘The lamb is an animal.’

b. Das Lamm springt uber den Zaun.

‘The lamb jumps over the fence.’

c. Das Jungtier eines Hausschafes ist ein Lamm.

‘The offspring of the domestic sheep is a lamb.’

(119) a. λx. x = lamm

b. λx. x � lamm

c. λx. x ≤ lamm

That is, in (118a), Lamm is understood as denoting the kind, whereas in (118b) itdenotes an instance of that kind and in (118c), a subkind of the kind.

Lamm, of course, can also be used metonymically as, for example, in (120).

(120) a. Hans mochte heute Lamm essen.

‘Hans wants to eat lamb today.’

b. Maria tragt seit gestern Lamm.

‘Since yesterday, Maria wears lamb.’

c. Anna weigert sich, Lamm zu essen oder zu tragen.

‘Anna refuses to eat or wear lamb.’

17Thus, the default value for Qn is ∃, that for Sn is ENT (for ‘entity’) and that for Rn is =.18That is, in Stern’s approach the utterance of (i) might be a token either of the expression type

in (ii) or the metaphoric expression type in (iii).

(i) Juliet is the sun.(ii) Juliet is the sun.(iii) Juliet Mthat[is the sun].

It is in such cases that the material contributed by the met-operator comes to bear.Thus, applied to the semantic form of Lamm, the parameters available will be fixedfor the respective occurrences in (120) as indicated in (121).19

(121) a. λx. FLEISCH(x) & ∃y [MAT(y)(x) & (y) � (lamm)]

b. λx. FELL(x) & ∃y [MAT(y)(x) & (y) � (lamm)]

c. λx. FELL∨FLEISCH(x) & ∃y [MAT(y)(x) & (y) � (lamm)]

That is, in (120a), Lamm is understood as denoting a particular type of meat, in(120b), a particular type of fur and in (120c) something that is either or both aparticular type of meat/fur.

Note again that although in Dolling (2000)’s approach metaphorical as well asmetonymic interpretations are grounded in semantics, the actual content is suppliedonly after semantic composition is finished. That is, the parameters provided bythe met-operator need to be fixed just as the parameters that may be involved inthe SEM of the individual lexical items involved in an utterance (i.e., Kn). Theprocess of fixing parameters, then, takes into account contextual and backgroundconceptual information and results in a truth-evaluable full proposition. Thus, thepredicates that are inserted as values for the parameters are based on a range ofconceptual axioms that capture the relations holding between them. For instance,for the predicates involved in (121), at least the following axioms have to be assumed.

(122) a. ∀x [DING(x) → ∃y [STOFF(y) & MAT(x)(y)]]

b. ∀x [LAMM(x) → DING(x)]

c. ∀x [TIER*(x) → ∃y [FLEISCH(y) & MAT(x)(y) & TYP ESSBAR(x)]]

d. ∀x [TIER**(x) → ∃y [FELL(y) & MAT(x)(y) & TYP PELZBAR]]

e. ∀x [LAMM(x) → TIER*(x) & TIER**(x)]20

Finally, the same met-operator is also involved in possible metaphorical uses ofLamm, such as in (123), where Lamm seems to be used with the meaning indicatedin (124).

(123) Annas Mann ist ein Lamm.

‘Anna’s husband is a lamb.’

(124) λx. PERSON(x) & ∃y [INST(y)(x) & ART(y) & ∃z [AHNLM(z)(y) & (z) =(lamm)]]21

Thus, in (123) Lamm denotes a person that instantiates a particular kind, wherethat kind is similar to the kind of lambs with respect to the set of features M.

Note that although there is only one operator involved in metonymic and meta-phorical interpretations of utterances, the difference between the two is still capturedby the type and nature of the information that is involved in fixing the parameters

19The material fixed by default values and thus, in a sense, cancelled out, is not stated.20TIER* and TIER** denote particularly restricted sets of animals, namely such that their flesh

is typically edible (TIER*) and such that their fur can typically be used for coats (TIER**).21ART is a predicate of the domain of kinds and AHNL is a predicate for the relation of similarity

relative to the set of features M.

provided by the operator in the two interpretation situations. Thus, in metonymythe ontological relations between the predicates that serve as values are capturedby axioms that can be taken to be part of the background knowledge. That is, therelevant conceptual relations for a metonymic interpretation are, in a sense, given.In contrast, part of metaphor interpretation seems to be the establishment of newconceptual entities (ART) and relations (AHNLM), usually based on an assumptionof similarity of features between the kind denoted by the metaphor vehicle and themetaphor target.

One question that arises at this point is how the members of the set M are de-termined. Dolling (2000) acknowledges the importance of this question, but doesnot answer it. ‘Die kognitiven bzw. pragmatischen Bedingungen, die zum einen dieFestlegung der jeweiligen Merkmale und zum anderen die Auswahl des metaphorischverwendeten Ausdrucks bestimmen, gilt es zu untersuchen’ (Dolling 2000, p. 50).22

Another question is how, generally, the conceptual values are determined that areinserted for the respective SF-Parameters. Dolling (2000) mentions abductive inter-pretation as an appropriate mechanism (cp. Hobbs et al. 1993).

Generally, assuming one basic operator that ensures unproblematic integrationof conceptual material during utterance interpretation appropriately captures theamount and variety of meaning variation exhibited by many lexical items. In fact,the same operator may also be involved in cases of what RT calls ‘loose use’ of anexpression. The differences in perceived non-literality could be correlated to thedifferences in the type of information that is inserted as the value of the parametersprovided by the operator or in the effort of providing/creating that information inthe first place.

Thus, as in Stern (2000)’s approach, in Dolling (2000)’s approach, the difficul-ties autistic people exhibit in understanding of metaphor (or metonymy for thatmatter, cp. Rundblad and Annaz 2010) cannot be captured by a general differencein the semantic operations involved in the two types of figurative language use asboth are based on one and the same operator. Rather, it has to be assumed thatautistic people are not able to provide (or in the case of metaphor to create) thenecessary conceptual information. A similar point has to be made concerning typ-ically developing children, where there also seems to be a difference in metonymyas compared to metaphor understanding, in that the former is generally easier thanthe latter and competency for both types of figurative language use increases withchronological age (cf. Rundblad and Annaz in press). Assuming one operator in thesemantics that is involved both in cases of metonymy and metaphor, the differencesin understanding these two types of figurative language use have to be explainedby differences in providing/creating the conceptual information that is integratedby that operator. As we saw above, generally, Dolling (2000)’s approach allows forsuch a difference to be made.

22‘The cognitive or pragmatic conditions, which determine the selection of the respective features,on the one hand, and the choice of the metaphorically used expression, on the other, have to beinvestigated.’

Summary

None of the alternative approaches discussed here assume that metaphor interpre-tation is based on the prior recovery of utterance meaning, rather, it is simple (RT,Glucksberg and McGlone 1999, Dolling 2000) or more complex expressions (Stern2000, 2006, 2009) within some utterance that are interpreted metaphorically.23 Thatis, metaphorical interpretation seems to be based on the lexical meaning of themetaphor vehicle. In fact, if one assumes – and I argued above that one should –that lexical meaning is characterised by a high degree of underspecification, thenmetaphor is not actually based on lexical meaning, since this is not even a full con-cept, rather it is based on, in Dolling (2001)’s terms, one of the set of possible valuesfor the SF-parameter in the SEM of the expression under consideration.

The important point to note is that in all the approaches discussed here, metaphorinterpretation is taken to be part of the processes that determine utterance meaningrather than being based on the latter. In other words, contrary to Grice’s assump-tions, metaphor is not taken to be a phenomenon of the level of what is meant. Infact, treating metaphor as arising at the level of utterance meaning or what is saidactually is compatible with Grice’s assumption that what is said is entailed by whatis meant, since the metaphoric meaning will be treated as forming part of what issaid by the respective utterance. Thus, there is no need to assume that in caseof metaphor, the speaker only ‘makes as if to say’ some literal proposition which,in ‘non-metaphorical circumstances’, actually would be taken to constitute what issaid. Rather, in the case of metaphor, there is no ‘literal proposition expressed’ tobe determined in the first place.

Having said this, we saw that the meaning the metaphoric interpretation isbased on does play a role in the actual interpretation, at least in the case of novelmetaphors. Thus, it seems that metaphor interpretation involves the creation of anew concept on the basis of relevant features of the ‘literal’ concept the interpreta-tion starts out from. In case of the interpretation of novel metaphors, therefore, onewould expect that people first access the metaphor vehicle’s literal meaning and thatthe metaphorical reading becomes available afterwards. This is the pattern observedin the experiments cited in this section.

Generally, it seems that metaphor interpretation proceeds independently of theintentions the speaker had in making the utterance that includes the metaphor.24

This assumption is implicit in Glucksberg and McGlone (1999)’s as well as Stern’sand Dolling’s approaches to metaphor interpretation. Thus, in Glucksberg and Mc-Glone (1999)’s approach, an attributive category is created on the basis of relevantfeatures of the metaphor vehicle, where the relevant features are determined bycomparing the concepts associated with both the metaphor topic and the metaphorvehicle. This process seems to be free of considerations of speaker intentions. Sim-ilarly, in Stern’s approach, it is simply assumed that the potential metaphorically-associated meaning aspects are presupposed by the context, thus, it seems they arenot determined based on a consideration of what the speaker intended with his ut-

23Compare this to Borg (2001), who seems to assume that metaphor is based on the propositionliterally expressed by an utterance – i.e., its minimal semantic content – and involves a set ofassociated propositions.

24Cf. Borg (2001)

terance. In Dolling (2001)’s approach, the metaphoric interpretation of a particularexpression is also constituted by the creation of a new concept (ART) and a relation(AHNLM) holding between that new concept and the concept associated with themetaphor vehicle. Again, the features with respect to which the two concepts aretaken to be similar seem to be determined independently of the speaker’s intentions.In contrast, in RT, metaphor interpretation cannot strictly be characterised as be-ing independent of speaker intentions, since it results from an overall interpretationprocess which crucially is assumed to make reference to speaker intentions.

Recall that in chapter 2, I cited experiments investigating the interpretation offamiliar metaphors, which indicated that their figurative meaning is available asearly as their literal meaning in the interpretation process. Thus, in such cases itseems that during interpretation the metaphoric meaning no longer has to be createdbut is already stored as a potential reading for the expression in question. In otherwords, in case of familiar metaphors, their metaphoric reading might have becomea member of the set of values of the SF-parameters in the SEM of the metaphorvehicle.

4.1.2 Irony

Traditional Characterisation and its Problems

Traditionally, verbal irony is characterised as a figure of speech which involves thesubstitution of a literal by a figurative meaning. More specifically, in irony the figu-rative meaning is characterised as being the opposite of the original literal meaning.Thus, consider the example in (125).

(125) Mary: Paul really is a fine friend.[uttered after Peter has told Mary that Paul has repeatedly run down Marybehind her back.]

In the context in which this utterance is made, it is clear that its utterance meaning– determined on the basis of narrow context – would be judged false. However, oncethe speaker’s intentions are taken into account and the hearer has realised that thespeaker intended her utterance to be taken ironic, he will consider what was meantor the communicative sense as being true.

In his treatment of irony, Grice essentially adopts the classical view. His inno-vation is to consider irony along with other types of non-literal meaning such asmetaphor as implicatures. Thus, his analysis of example (125) involves the floutingof the maxim of quality and a calculation of the irony-implicature by the heareralong the following lines.

(126) a. The speaker has expressed the proposition that Paul really is a fine friend.

b. However, the context in which she expressed that proposition makes itobvious that this proposition is blatantly false.

c. Thus, it is clear that the speaker cannot have intended to mislead the hearer.

d. If the speaker is to be taken to adhere to the co-operative principle, insaying that Paul really is a fine friend, she must have meant the oppositeof what she said.

As has been pointed out by different authors (e.g., Wilson and Sperber 1992,Happe 1993, Dews and Winner 1999, Colston 2000, Schwoebel et al. 2000, Colston2002, Colston and Gibbs 2002, Giora et al. 2007), there are several problems withthe traditional view of irony, based on classical rhetoric and Grice’s reanalysis ofit in terms of conversational implicature. For instance, in irony, speakers do notalways and simply mean the opposite of what they say. Moreover, there are boththeoretical considerations as well as empirical evidence that shows that the process-ing of metaphor and irony differs. Also, there is empirical evidence that shows thatpeople do not simply replace and thus disregard the inappropriate literal meaningof an ironic utterance.

Irony as echoic interpretive use

Wilson and Sperber (1992) give a range of examples which show that irony cannot becharacterised simply as involving a meaning that is the opposite of what has actuallybeen said. The examples include ironical understatements (127a) and non-ironicalfalse-hoods (128a).

(127) a. You can tell he’s upset.[uttered by A to B in a situation in which a customer C complains in ashop, blind with rage and making a public exhibition of himself]

b. You can’t tell he’s upset.

c. You can tell he’s not upset.

Wilson and Sperber (1992) claim that contrary to what is suggested by the tradi-tional view of irony, (127a) neither communicates (127b) nor (127c). Rather, whatis communicated by (127a) – as is typical of understatements in general – is some-how more than what is actually said. Thus, this is an example where the traditionalcharacterisation of irony is not applicable but which, nevertheless, intuitively seemsto be ironic.

Let us now turn to an example, which conforms to the traditional definition ofirony, but fails to be ironic. A and B are out for a walk. They pass a car with abroken window and A says (128a) to B.

(128) a. Look, that car has all its windows intact.

b. That car has one of its windows broken.

Assuming that B, in saying (128a) actually intended to communicate the opposite(128b), B’s utterance meets the criterion of being ironic, but intuitively it fails tobe so.

Thus, the traditional characterisation of irony cannot account for all cases ofirony and it subsumes cases which actually are not ironic. Moreover, the traditionalview cannot explain why ironic utterances should be used at all. That is, if in ironythe speaker simply says the opposite of what he means, the question arises whatrational explanation there is for the fact that the speaker did not simply say whathe meant in the first place.

Wilson and Sperber (1992) offer an alternative characterisation of irony in termsof RT based on the idea that in irony, a speaker makes an utterance which echoes

a thought she implicitly attributes to someone else and from which she wants todistance herself. Generally, an utterance may be used descriptively or interpretively.In the first case, the proposition expressed by the utterance is a true description of anactual state of affairs. In the second case, the proposition expressed by the utteranceresembles another proposition that is the content of some other representation, forinstance, an attributed thought. In cases of irony, Wilson and Sperber (1992) claim,utterances are used interpretively. To illustrate the difference, let us turn back tothe sentence in example (125).

(125) Paul really is a fine friend.

A speaker using this sentence descriptively in an utterance is expressing the propo-sition that Paul really is a fine friend, which, in principle, can be judged againstexternal reality as to whether it is true or not. Suppose, now that John claimedthat Paul really is a fine friend and Mary, thus, entertains the thought verbalised in(129).

(129) John believes that Paul really is a fine friend.

Mary can use the sentence in (125) interpretively and thereby not express the propo-sition that Paul really is a fine friend, but rather express the proposition verbalisedin (129) and at the same time dissociate herself from the proposition that Paul reallyis a fine friend. In other words, Mary can both believe that (129) and at the sametime believe that Paul really isn’t a fine friend. Thus, according to RT, interpretingirony involves second-order metarepresentation of thoughts attributed to others. Inthe example above this means that for Peter to understand that Mary’s utterance isironic, he has to be able to determine the proposition that Mary believes that Johnbelieves that Paul really is a fine friend and that Mary, at the same time, dissociatesherself from the proposition that Paul really is a fine friend.

Characterising irony as a form of echoic interpretive use of an utterance allows foran explanation of why ironical understatements are understood as ironic althoughin these cases the speaker does not say the opposite of what he meant and why non-ironical false-hoods are not taken to be ironic although here the speaker putativelysaid the opposite of what he meant. Moreover, with such a characterisation, thereis no assumption that there is any simple other way in which the speaker couldhave expressed what he intended to. That is because the ironical speaker does notsimply mean the opposite of what he says. In other words, it is a simplification ofwhat is going on in irony to merely assume that it involves a figurative, or non-literal meaning. What it actually involves is the echoing by the speaker of a thoughtshe implicitly attributes to someone else and from which she, at the same time,dissociates herself.

Irony as a Form of Indirect Negation

In contrast to Wilson and Sperber (1992), Giora (1995) defends a view that isbased on the traditional view of irony as involving negation. More specifically, shecharacterises irony as a form of indirect negation, which differs from direct negationin a range of features. For instance, indirect negation does not lead to the type

of scalar interpretation found for direct negation. Thus, the meaning expressedby a directly negated sentence such as in (130a) can be paraphrased as in (130b),which, in principle should allow for a more specific interpretation such as in (130c).However, that is usually not what is communicated by an utterance of a sentencesuch as (130a). Rather, it is a member of the scale closer to the member which hasbeen negated that is communicated to hold (130d).

(130) a. The water isn’t hot.

b. The water is less than hot.

c. The water is cold.

d. The water is warm.

Giora (1995) argues that such a scalar interpretation is not found with indirectnegation. Another aspect of direct negation that indirect negation does not showis its linguistic and psychological markedness. Obviously, indirect negation is notmorphologically marked as such. In addition, it seems that indirect negation, andthus irony, does not have direct negation’s ‘unpleasant value’. That is, forms ofindirect negation are perceived as less aggressive than direct negation.25

Note further that in Giora’s approach irony is not a case of flouting some maximof quality. However, it does violate what Giora calls the graded informativenesscondition26 since the speaker by his utterance in the context at hand introduces aproposition which is highly improbable (and as such perceived as ‘marked’) giventhat context. This violation leads the hearer to ‘. . . evoke an unmarked interpre-tation (i.e., the implicature) comparable with the marked message whereby thedifference between them becomes perceivable’ (Giora 1995, p. 245). To illustrateGiora (1995)’s approach, let us look once more at example (125).

(125) Mary: Paul really is a fine friend.[uttered after Peter has told Mary that Paul has repeatedly run down Marybehind her back.]

With respect to the context in which Mary makes her utterance, the proposition herutterance expresses is highly improbable, however, she actually uses it to evoke a lessimprobable interpretation. Moreover, the proposition expressed by her utterance isnot simply disregarded as not fitting the context, rather ‘. . . it is used to implicatethat a specific state of affairs is far from or very much unlike the expected/desiredstate of affairs indicated by the comparison’ (Giora 1995, p. 245). Thus, in Giora(1995)’s approach, the ‘literal’ utterance meaning is not simply disregarded butrather is necessary in order to compute the intended irony. In terms of processingeffort, this suggests that irony should be more difficult to interpret than ‘literal’language, since it involves the determination of both a literal as well as a non-literalmeaning.

In contrast to the more traditional views of irony involving the notion of negation,Giora (1995) also offers an explanation as to why a speaker should choose a form

25Cf. example (131) below.26According to this condition on well-formed discourses, each proposition that is expressed in a

discourse should be either as or more informative than the proposition preceding it.

of indirect negation over overt negation. Not surprisingly, the reasons given referback to the differences stated for the two types of negation. Thus, in using indirectnegation, the effect of scalar interpretation as well as the psychological markednessof direct negation can be avoided. In fact, Giora (1995) considers irony ‘. . . “apoliteness strategy” which enables the speaker to negate . . . by using an affirmativeexpression’ (Giora 1995, p. 260).27 Thus, consider the example in (131), where theironical speaker in using the affirmative (131a) conveys the rather negative (131b),which will be understood as criticism but is not as agressive as if the speaker hadsaid (131c).

(131) a. (Ironically:) What a lovely party!

b. The party is far from being lovely.

c. The party isn’t lovely at all!

Note also that in Giora’s approach no reference is made to higher-order metarep-resentation being necessarily involved in irony interpretation. In fact, Giora (1995)criticises Wilson and Sperber (1992)’s view of irony as echoic interpretive use of anutterance on several accounts. She argues that, on the one hand, the approach doesnot capture all cases of irony and, on the other hand, it predicts irony to be involvedin cases where it actually is not present. More specifically, Giora (1995) claims thatthere are both instances of intuitive irony which do not involve an echoic interpretiveuse of an utterance and instances of echoic interpretive use of an utterance, whichfail to be ironic in the context in which they occur. Thus, consider Giora’s exampleconversation in (132), assumed to take place between two friends who actually knowthe person talked about quite well.

(132) A: Do you know any G. M.?

B: Rings a bell.

Giora (1995) argues that in this case of irony, one need not necessarily assume thatthe utterances made need to be attributed to another speaker. They thus fail tobe echoic. However, as Curco (2000) points out, assuming that irony involves thespeaker’s echoing of a thought implicitly attributed to some other speaker actuallyis to narrow a characterisation of echoic use and she argues that it is not howWilson and Sperber (1992) view the notion. Rather, ‘[e]choic utterances . . . includethose cases where the speaker attributes a thought or an utterance to herself at atime different from the time of utterance . . .’ (Curco 2000, p. 262. emphasis in theoriginal). Thus, in the example above, the thought uttered by B is attributable toB in a context where it is not mutually known to the conversation partners involvedthat B knows G. M. very well.

For an example involving an echoic interpretive use of an utterance which failsto be ironic, Giora (1995) gives the conversation in (133).

(133) Dina: I missed the last news broadcast. What did the Prime Minister sayabout the Palestinians?

27Compare this to Dews and Winner (1999)’s Tinge Hypothesis, according to which ‘. . . ironymutes the evaluative meaning conveyed in comparison to literal language. Thus, ironic criticism isless critical than is literal criticism, and ironic praise is less praising than is literal praise’ (Dewsand Winner 1999, p. 1580).

Mira: (with ridiculing aversion) That we should deport them.

The idea is that with her utterance, Mira both reports on what the Prime Ministersaid and at the same time expresses her negative attitude to what she reports.Nonetheless, her utterance does not come across as ironic. In fact, this is preciselyas it should be, Curco (2000) argues, because actually, Mira’s utterance is ellipticaland the proposition her utterance really expresses is

(134) The Prime Minister of Israel said that we should deport the Palestinians.

Thus, Mira actually uses her utterance descriptively in that she states a fact. Thatis, she says something true about a state of affairs in the real world, namely, that thePrime Minister of Israel said that the Palestinians should be deported. Since Miraused her utterance descriptively, according to Wilson and Sperber (1992), it cannotbe understood as being ironic. Moreover, what the speaker is taken to dissociateherself from is only a subpart of the complex proposition expressed by her utterance,namely ‘that we should deport the Palestinians’. However, as Curco (2000) pointsout, the fact that the speaker implicitly distances herself from the speech she reportsis not sufficient to make her utterance ironic.

A further claim Giora (1995) makes against Wilson and Sperber (1992)’s viewconcerns the processing of irony. She argues that under the ‘echoic interpretive use’view, irony should not be more difficult to interpret than ‘literal’ language, sinceno norm or maxim is violated and no non-literal meaning is derived in consequence.In contrast, treating irony as a form of indirect negation, where a communicativenorm is violated, leading the hearer to compute a non-literal meaning in the form ofan implicature, suggests that the interpretation of ironic utterances should be moredifficult than the interpretation of ‘literal’ language. Moreover, since the ‘ironyas indirect negation view’ assumes that an ironic utterance’s ‘literal’ meaning isnecessary for the hearer to compute the difference holding between it and the derivedimplicature, it suggests that in processing an ironic utterance, its ‘literal meaning’should remain activated during the interpretation process.

However, there is no reason why difficulties in processing should only arise be-cause some communicative norm has been violated and, as a consequence, somefurther meaning has to be computed. In fact, and as we will see below, Wilsonand Sperber (1992)’s claim that irony interpretation involves second-order metarep-resentation of an attributed thought on the part of the hearer, leads to the samehypothesis as posited by Giora (1995), namely that processing of irony should bemore difficult than that of ‘literal’ language. Moreover, since in Wilson and Sperber(1992)’s account there is no non-literal meaning replacing the literal utterance mean-ing, their approach also predicts that an ironic utterance’s literal meaning shouldbe and remain activated during interpretation.

Empirical Results Concerning Irony Interpretation

As we already saw above, concerning the relative difficulty of processing irony incontrast to ‘literal’ language, Wilson and Sperber (1992) and Giora (1995) makepredictions similar to that suggested by Grice’s framework. However, both Wilsonand Sperber (1992)’s as well as Giora (1995)’s view are in contrast to Grice’s view

of irony and metaphor, where the two phenomena are treated as types of non-literalmeaning whose interpretation essentially involves similar processes. According toGrice, in both cases the literal meaning of the respective utterances is rejected asnot fitting the context and the non-literal meanings are calculated as implicatures.However, there are both theoretical considerations as well as empirical evidence thatirony and metaphor differ in some essential properties. On the one hand, Wilson andSperber (1992) argue that the interpretation of irony seems to require the abilityof second-order metarepresentation, while in case of the interpretation of metaphoran ability of first-order metarepresentation seems to be sufficient. Moreover, theseauthors assume that in contrast to irony, in cases of metaphor, utterances are useddescriptively. On the other hand, Giora (1995) suggests that comprehending ironyis more exacting than understanding metaphor, since the former is a type of indirectnegation and negation, generally, has been shown to take more time in processingthan positive utterances do (cf. Clark 1973). Thus, as concerns the processing ofirony and other types of meaning, the two approaches make similar assumptions,although for different underlying reasons. Similarly, in contrast to Grice, both theseapproaches assume that in interpreting an ironic utterance, its literal meaning isnot simply discarded. However, whereas Giora (1995) holds this assumption dueto the fact that she assumes the literal utterance meaning is necessary for a hearerto compute the irony as such, Wilson and Sperber (1992) hold it because in theirapproach to irony comprehension there is no derivation of a non-literal meaning inthe first place.

Empirical investigations of these hypotheses are overall favourable to Wilsonand Sperber (1992)’s and Giora (1995)’s view concerning irony interpretation ingeneral and differences to metaphor interpretation in particular. Happe (1993),for instance, provides data that corroborates Wilson and Sperber (1992)’s viewthat irony interpretation involves second-order metarepresentation. Colston andGibbs (2002) is an investigation of the interpretation of metaphor vs. irony, again,showing that there are differences between them. Finally, Giora et al. (2007) show(as do Dews and Winner 1999, Schwoebel et al. 2000) that during the interpretationof ironic utterances their so-called literal meaning is not simply disregarded andsubstituted by a non-literal meaning.

Thus, Happe (1993) investigated whether the interpretation of irony does involvesecond-order metarepresentation. In order to do so, Happe (1993) chose as subjectgroup autistic children, since they are independently assumed to have problemswith higher-order metarepresentation (cf. Baron-Cohen et al. 1985). Happe (1993,p. 103) assumed that ‘. . . an inability to represent thoughts about thoughts wouldpreclude . . . utterances which . . . are interpretations of the speaker’s thought aboutan attributed or desirable thought’. In other words, if autistic children cannot formsecond-order metarepresentations, then they should have problems identifying casesof irony if Wilson and Sperber (1992)’s assumption is correct that irony interpreta-tion involves such second-order metarepresentation.

Happe (1993) divided the autistic subjects into three groups, according to theresults of tests designed to assess their theory of mind abilities. Thus, one group ofsubjects consisted of children who had passed all first-order metarepresentation tasksgiven them but none of the higher-order tasks (i.e. first-order ToM group). Another

group consisted of children who also had passed the first-order tasks and in additionhad scored high (6 out of 8 correct) on the second-order task (i.e. second-order ToMgroup). Finally, one group of subjects consisted of children who had failed the first-order false belief tasks (i.e. no-ToM group). A last group of subjects, functioning asa control to the autistic subject groups, consisted of children with moderate learningdifficulties, who had passed both first- and second-order metarepresentation tasks,but whose verbal IQ was considerably lower than that of the first- and second-orderToM groups while comparable to that of the no-ToM group. This group was includedto exclude the possibility that the autistic childrens’ performance was simply due toa general intellectual impairment.

The results of Happe (1993)’s experiments show that autistic children who passthe second-order metarepresentation tasks also are able to understand ironic utter-ances. The first-order ToM subject group – who failed at attributing false beliefsto others – do considerably worse in understanding ironically intended utterances.However, these same subjects (as well as the subjects of the second-order ToMgroup) seem to have no problem understanding metaphoric utterances. This resultcan be taken as evidence that metaphor interpretation does not involve as complexcognitive tasks as does the interpretation of irony. In contrast to both the first-and second-order ToM groups, the No-ToM group has severe problems understand-ing both metaphor and irony. Note that the results for the non-autistic controlgroup showed that they did better than both the no-ToM group in understandingmetaphor as well as the first-order ToM group in understanding irony, although thecontrol group’s verbal ability was similar to the no-ToM group’s and lower than thefirst-order ToM group’s. This suggests that the difficulty for the no-ToM group ofunderstanding metaphor and for the first-order ToM group of understanding ironyis not due to a generally low verbal ability of these subjects.

Thus, Happe (1993)’s results support Wilson and Sperber (1992)’s assumptionthat the interpretation of irony involves second-order metarepresentation and thatthe interpretation of irony differs from that of metaphor, which only involves first-order metarepresentation.28 Similar results were obtained by Colston and Gibbs(2002), investigating the interpretation of both metaphor and irony in ‘normal’ sub-jects. In their first experiment, Colston and Gibbs (2002) tested identical statementswhich, depending on the context in which they were placed, either had a metaphoricor an ironic meaning. Subjects read at their own pace through small stories pre-sented on a computer screen, the last sentence of which was the target sentence andwas either intended as metaphoric or ironic. Following the target sentence, subjectssaw a paraphrase sentence of the target, intended to capture the speaker’s meaning.Their task was to decide whether the paraphrase was good or bad and their reactiontimes were measured. The results show that subjects took longer to comprehendironic than metaphoric utterances. Assuming that longer reaction times reflect alarger amount of inferencing, these results again suggest that irony is harder tointerpret than metaphor.

28However, this is not to say that the ability of second-order metarepresentation is the onlyrelevant factor in irony understanding. Thus, Pexman and Glenwright (2007) point out that suchfactors as neural maturation and social learning also play a role in children’s development of ironyunderstanding.

Although the results of their first experiment suggest that processing irony in-volves more complex inferencing than required for interpreting cases of metaphor,reaction times as such do not tell us exactly what kinds of inferences subjects madewhen they interpreted metaphoric as compared to ironic utterances. Thus, in asecond experiment, Colston and Gibbs (2002) investigated whether ‘. . . people ac-tually engage in more complex metarepresentational reasoning when reading ironicremarks than they did when reading metaphoric statements’ (Colston and Gibbs2002, p. 64). To this end, the authors had subjects read the same stories withtheir respective final ironic or metaphoric utterances as in the first experiment. Af-ter reading the stories, subjects were asked to rate their agreement with a numberof inference statements. These statements were designed to probe for the variousmetarepresentational inferences people are assumed to draw when interpreting ironicor metaphoric utterances. Thus, one statement probed subjects’ assessment of thebeliefs of the speaker when interpreting ironic or metaphoric utterances. Anotherstatement was intended to show whether the notion of pretense plays a role in under-standing irony or metaphor.29 The third statement was designed to reveal whethersubjects recognised some echoing by the speaker of a thought attributed to someoneelse. Colston and Gibbs (2002) hypothesised that if irony interpretation in con-trast to metaphor interpretation involved higher-order metarepresentation, subjectsshould give higher ratings to the inference statements referring to such higher-orderreasoning after having read stories ending in ironic utterances than after storiesending in metaphoric utterances. This hypothesis was confirmed.

In addition, Colston and Gibbs (2002) also investigated whether there are dif-ferences between understanding utterances that involve irony and such that involveboth metaphor and irony. The authors predicted that metaphoric irony should bemore difficult to understand than ‘simple’ irony. The idea is that there is a ‘. . .mesh between these two types of figurative meaning . . .’ that leads to longer pro-cessing times in cases where both metaphor and irony are involved. They also notethat this view is compatible with the assumption that different processing modesare required in metaphor vs. irony interpretation, as suggested, e.g., by Wilson andSperber (1992), Sperber and Wilson (1995) (i.e., descriptive vs. interpetive use ofutterances). The results of that experiment confirmed their prediction. That is,interpreting cases of metaphoric irony took subjects longer than interpreting casesof simple irony, suggesting that the former is a cognitively more exacting process.

There are several studies which investigated whether or not in irony compre-hension, the non-intended literal meaning of the ironically intended utterance isprocessed and if it is to what extend. Dews and Winner (1999), for instance, inves-tigated the processing of ironic criticism (cf. 135a) and ironic praise (cf. 135b) com-pared to ‘literal’ uses of the same expressions (cf. 136a constituting literal praise and136b constituting literal criticism) (examples taken from Dews and Winner 1999).

(135) a. Good move![to a person who has just done something foolish]

b. We never do anything fun.

29It has been suggested, e.g. by Leslie (1987), that engaging in pretense requires the capacityfor second-order metarepresentation.

[to a companion on a dream vacation]

(136) a. Good move![to a person who has just done something wise]

b. We never do anything fun.[to a roommate who has been studying hard for finals]

The hypothesis tested was that subjects should take longer judging the evaluativetone of the ironically intended utterances, if their unintended literal meaning, whichis opposite in evaluative tone, is computed and thus interferes. Thus, subjects werepresented stories on a computer which ended in a character making either a literalor ironic comment. They then had to judge the evaluative tone of the commentby pressing a designated key. Reaction times were measured. The results showthat subjects both made more errors in judging the evaluative tone of ironicallyintended utterances and they took longer to make those judgements than they didto judge literally intended utterances. Thus, they are interpretable as supporting thehypothesis that in irony interpretation at least some aspects of the literal utterancemeaning are processed.

Similarly, Schwoebel et al. (2000) also investigated the interpretation of ironiccriticism and ironic praise and their literal counterparts. However, since one of theiraims was to pin-point the exact time during interpretation at which the ironicallyintended utterances are recognised as such, they measured subjects’ reading timesof target utterances. In a pretest, they had subjects indicate at which expression(s)in the target utterance, the utterance was recognised as ironic. The informationgleaned from this test was used to divide the target sentences into three phrases forpresentation to subjects. The first phrase gave no indication of intended irony. Thesecond phrase ended in the expression(s) that had been identified as indicating theintended irony and the last phrase, again gave no indication of irony. An examplestory with a target sentence and an indication of its division is given below.

(137) A new professor was hired to teach philosophy. The professor was supposedto be really sharp. When Allen asked several questions, the professor offerednaive and ignorant answers. Allen said: That guy/is brilliant/at answeringquestions.

Each story was followed by a comprehension true/false question which subjectshad to answer by pressing one of two designated keys. Accuracy of answers tothese questions as well as the reading times for each of the three phrases of targetsentences was measured. The results show that subjects took longer to read thecritical (second) phrase when it followed an ironically biasing story, than when itfollowed a literally biasing story. Schwoebel et al. (2000) interpret these results asshowing, once again, that the ‘literal’ meaning of an ironically used utterance isprocessed, even when it is not intended. Moreover, the results indicate that theironic interpretation of an utterance is accomplished ‘online’, that is, while thecritical utterance is being processed.

More recently, Giora et al. (2007) tested in how far contextual information avail-able when processing an ironically intended utterance influences the interpretationprocess. More specifically, it was investigated what effect an expectation of irony

on the part of the subjects had for their interpretation of critical utterances. Inaddition, the authors also were interested in the question of when the ironic inter-pretation of an ironically intended utterance becomes available to subjects. Moregenerally, the experiments reported are aimed at supporting Giora’s graded saliencehypothesis and the predictions it makes concerning the interpretation of irony.

Thus, according to the graded salience hypothesis, apart from conventional irony,ironic interpretations of utterances are not salient. As we saw earlier, Giora assumesof salient meanings that they are always activated, even in contexts in which theyare not intended. For cases of intended irony, this means that the salient mean-ings of the expressions used in the ironically intended utterance should be activatedalthough they are inappropriate in the given context. In experiments 3 and 4 es-pecially, it was tested whether given a strongly supportive context, an intendedironic interpretation would be determined without activation of the non-intendedbut salient utterance meaning, built up on the basis of the salient meanings of itsparts. Materials consisted of small stories ending in the target sentence, which, de-pending on the preceding context, was either literally or ironically interpreted. Toraise the expectation for irony, one set of materials actually only included storiesending in ironically intended utterances. An example context together with theironically intended target utterance is given below (taken from Giora et al. 2007).

(138) Sarit worked as a waitress in a small restaurant in central Naharia. Theevening was slow, and even the few customers she did wait on left negligibletips. She didn’t think that the elderly man who walked in alone and orderedjust a couple of small sandwiches would be any different. Indeed, after makingher run back and forth throughout the meal, he left, and she collected his payfor the meal from off his table and found 2.5 NIS tip! When she showed herfriends how much she got, Orna commented: “That was real noble of him!”

The availability of the non-intended but salient utterance meaning and the intendedbut non-salient ironic meaning was elicited by presenting subjects with probe wordsrelated to one or the other of the respective utterance-level meanings at differentinter-stimulus intervals (ISI).30 The results show that even in contexts where sub-jects can reasonably be taken to expect ironic comments from the speaker, it seemsthe target ironic interpretation takes longer to determine than the unintended, con-textually incompatible but salient utterance meaning. In fact, even after an ISI of1000 msec, probe words relating to the ironic interpretation of an utterance weremore slowly responded to than probe words related to the salient utterance meaning.

Overall, the results support the graded salience hypothesis and more generallythose accounts, which assume that the literal (or rather salient) meaning of an iron-ically intended utterance is not simply discarded, rejected or suppressed. Moreover,they show that determining the ironic interpretation of an utterance is a time-consuming process, which is not shortened even in highly supportive contexts.31

30In the case of the example above, probe words were generous (literally related), stingy (ironi-cally related) and sleepy (unrelated control).

31This is also shown by the results of Giora et al. (2009)’s study, which investigated the influenceof providing an ‘ironic situation’ on the interpretation of the target utterance as ironical.

Summary

Although the two post-Gricean approaches to irony discussed above differ in vari-ous details from one another, both assume that in order to correctly interpret anironically intended utterance, that utterance’s ‘basic’ meaning – where this mightin fact already involve figurative meaning aspects such as metonymy or metaphor –has to be processed. In RT this is so because the ‘would-be expressed’ propositionplays an integral part in irony, as it is interpretively used by the speaker to conveyhis dissociative attitude. In Giora’s approach, the utterance’s ‘basic’ meaning needsto be determined first as it is necessary to compute the intended irony.

Contrary to Grice, who treated metaphor and irony as on a par, both post-Gricean approaches to irony discussed above assume that the understanding ofmetaphor and irony differs. Thus, whereas in the case of metaphor, it is the ‘basic’meaning of the metaphor vehicle the metaphoric interpretation is based on, in caseof irony it is the ‘basic’ utterance meaning that forms the basis for an ironic inter-pretation. In fact, as we saw in section 4.1.1, metaphor is actually taken to be aphenomenon of the level of utterance meaning, whereas irony clearly belongs to thelevel of speaker meaning.

Generally, the two post-Gricean approaches to irony we looked at both assumethat irony processing is cognitively more exacting than metaphor understanding.This assumption is supported by the results attained in experiments investigatingthe processing of metaphor and irony both in healthy subjects (e.g Colston and Gibbs2002, Giora et al. 2007) as well as in autistic children (e.g. Happe 1993, MacKayand Shaw 2004).

4.1.3 Conversational Implicatures

Generalised vs. Particularised Conversational Implicature – TheoreticalApproaches

As already mentioned, Grice characterised conversational implicatures as being in-ferences that are drawn on the basis of what is said, but which are not entailed bywhat is said. Thus, conversational implicatures are communicated in addition towhat is explicitly said. In this sense, they constitute implied meaning aspects ofwhat is meant by the speaker. On the Gricean view, then, it makes sense to claimthat conversational implicature is a phenomenon that belongs to a level of meaningdistinct from the level of what is said. On the one hand, what is said actually formsthe basis on which CIs are drawn, on the other hand, CIs differ from the kindsof implicit (or in RTs terminology explicit) meaning aspects that may be argued toform part of utterance meaning, as the former arise due to the existence of particularconversational maxims and the general cooperative principle.

As we already saw, Grice differentiated two types of conversational implicature,namely particularised vs. generalised conversational implicature. Recall that thedifference between these two types of CI is that the former depends on the contextin a way the latter does not. Thus, PCIs are computed only if the context is suchthat they are necessary to make sense of why the speaker said what he did. GCIs,on the other hand, occur unless the context is such that it prevents them from

occurring. That is, while PCIs clearly are recognised as belonging to the level ofwhat is meant by a speaker, this seems not so clear for GCIs. Thus, with respectto speaker intentions, PCIs have to be intended to arise, whereas GCIs seem to beindependent of speaker intentions. In fact, a speaker might actually have to makesure ostensively that a hearer does not compute a particular GCI, if he does notintend the hearer to do so. Due to these characteristics of GCIs, they have receiveda treatment different from that of PCIs. Thus, Levinson (2000) suggests that GCIsactually do not arise at the level of speaker meaning or communicative sense, butrather that they are already involved in the determination of the truth-conditionalcontent of the speaker’s utterance, that is, in determining what is said or utterancemeaning. Chierchia (2004) takes this idea further in that he actually claims GCIs aregrammatically induced meaning aspects (see also Chierchia et al. 2010). RT treatssome types of what Levinson calls GCIs as part of the explicature of an utteranceand some as genuine implicatures. In fact, RT argues against viewing the formermeaning aspects as implicatures at all. This, in turn, is in contrast to Borg (2009)as well as Burton-Roberts (2006), who both take GCIs to be implicatures and assuch not contributing to an utterance’s truth-conditional content, albeit, as we willsee, for different reasons.

Levinson (2000) argues that GCIs belong to a level of meaning distinct bothfrom the level of utterance meaning as well as the level of communicative sense asdiscussed above.32 Whereas he takes the latter to be the level at which an utter-ance’s utterance-token-meaning is determined, he argues that there are phenomena,like GCIs, which should not be treated as once-off inferences, like PCIs are. This isbecause PCIs and GCIs are so different in nature, the former only being triggered inparticular contexts, the latter always triggered unless the particularities of the con-text prevent them from being inferred. This property makes GCIs a kind of defaultinference, or, as Levinson puts it, a preferred interpretation. Thus, Levinson arguesfor a level of meaning he calls utterance-type-meaning33, that is, a level of meaningthat depends on the structure of language, but not on any particular context. Forexample, he takes GCIs, being one particular type of utterance-type meaning, to bethe result of a number of heuristics which do not take into account such matters asspeaker’s intentions or encyclopaedic knowledge of the domain talked about. Thus,utterance-type-meaning is also distinct from the level of utterance meaning, sincecontextual information does not have a role to play in the generation of such defaultmeaning aspects as GCIs. Generally, utterance-type meaning can be taken to bea preferred interpretation for an utterance, which may be overridden. This levelof meaning is based ‘. . . on general expectations about how language is normallyused.’ (Levinson 2000, p. 22). To point out the significance of this level of meaning,Levinson argues that it is the basis for a distinction between convention of languageand convention of language use.

‘Why is it that I can introduce myself with My name is Steve, but notI was given the name Steve, that I can express sympathy with you with

32In fact, Levinson refers to the traditional two-fold distinction between sentence meaning andspeaker meaning, of which he argues that it cannot properly accommodate the notion of GCIs andother related phenomena.

33Note that he uses the term not in the sense Grice did.

I am sorry but not conventionally with That saddens me; that I expressoutrage with Really! but not with In truth!; that I can say I am delightedto meet you but not idiomatically I am gratified to meet you; that I canchoose a pastry by saying I’d like that one but not I’d admire that one,and so on.’ (Levinson 2000, p. 23)

Thus, Levinson claims that a theory that does not assume a level of utterance-type-meaning cannot account for the regularity, systematicity and recurrence of particulartypes of pragmatic inference. Although particularly interested in GCIs, Levinsonmentions other types of phenomena that may be considered as belonging to the levelof utterance-type-meaning, for example, standardised inferences and short-circuitedimplicatures. These two phenomena differ from GCIs in that their determination isnot based on heuristics and, moreover, in that the coming about of these particulartypes of utterance-type-meaning depends on routinisation, which notion, once again,can be taken to reflect (amongst other measures) frequency of use of a particularexpression in a particular (or rather what gradually comes to be considered a normal)situation with a particular reading. Moreover, he takes the level of utterance-type-meaning to capture idiomatic uses of expressions as well as partially conventionalisedreadings. Thus, this may be the level at which collocations of particular expressionsare associated with the concepts they are normally used to refer to (e.g., white wine).Levinson conceives of the level of utterance-type-meaning as a level at which ‘. . .sentences are systematically paired with preferred interpretations’ (Levinson 2000,p. 27). What is retrieved at the level of utterance-type-meaning are, e.g., schemasof particular sentence constructions as well as the preferred or default readingsassociated with particular, more or less complex expressions.

Note that Levinson’s level of utterance-type-meaning plays a role in the deter-mination of the utterance meaning of an utterance. That is, in the absence of aparticular context, or in the absence of particularities in a given context, the inter-pretation process will determine on a reading for a particular utterance that can beconsidered normal or a default interpretation and in that sense, a preferred inter-pretation. Only if the context of utterance is such that it is not compatible witha default interpretation will the interpretation process determine a reading that is‘special’ in the sense that it does not constitute a preferred or default interpreta-tion. This suggests, of course, that in cases where there is no given context at all,a default interpretation will be determined, either on the basis of former uses ofthe expressions at hand or, in the case of GCIs, on the basis of some underlyingheuristics.

More specifically, Levinson assumes that GCIs result from the interaction ofthree general heuristics that underlie the interpretation of utterances. Thus, the Q-heuristic (‘What is not said, is not the case.’) derives from Grice’s first submaximof Quantity: ‘Make your contribution as informative as is required’. This heuristicgives rise to such GCIs as scalar and clausal implicatures. This can be shown usingthe familiar example below, where the GCI derived from the uttering of the sentencein (139a) is the proposition spelled out in (139b).

(139) a. Some of the students went to the party.

b. Not all the students went to the party.

Thus, since all is the stronger lexical item on the scale that some belongs to, andsince the speaker did not use that stronger element, the hearer is licensed to inferfrom that non-use that the speaker actually was not in a position to use it. Thus,although, semantically, the meaning of some is taken to be compatible with themeaning of all, the pragmatic inference, due to the Q-heuristic, is such that aninterpretation of some as meaning ‘all’ actually is not warranted.

The I-heuristic (‘What is expressed simply is stereotyptically exemplified.’) canbe related to Grice’s second submaxim of Quantity: ‘Do not make your contributionmore informative than is required’. This heuristic makes use of the fact that partic-ular linguistic expressions are stereotypically used in particular situations, where theexpression at hand does not fully encode what characterises that situation. How-ever, we are able to use a minimal amount of expressions to describe a particularsituation, because our interlocutor will, due to this second heuristic, determine ona stereotypical and much richer interpretation. Thus, in the example below, an ut-terance of (140a) normally is understood as expressing the proposition spelled outin (140b).

(140) a. Peter broke a finger, yesterday.

b. Peter broke one of his fingers, yesterday.

The third heuristic Levinson identifies is the M-heuristic (‘What is said in anabnormal way is not normal.’), which goes back to Grice’s maxim of Manner. Thislast heuristic is, in a sense, in opposition to the I-heuristic, in that it ensures a non-stereotypical interpretation for a non-simple description of a particular situation.Whereas an utterance of the sentence in (141a) is normally understood as describ-ing a situation in which John directly caused Peter’s death, this is not the case foran utterance of (141b). The idea is that the speaker employed a marked (complex)expression (cause to die) instead of a possible unmarked expression (to kill), indi-cating, due to the M-heuristic that the unmarked expression was not warranted, asit would have led the intepreter, due to the I-heuristic, to interpret the utteranceas describing a ‘normal’, that is direct killing-situation. Thus, an utterance of thesentence in (141c) usually is understood as describing a situation in which John didnot directly cause the death of Peter.

(141) a. John killed Peter.

b. John caused Peter to die.

c. John caused Peter’s death indirectly.

Generally, Levinson’s utterance-type-meaning seems similar to what Ruhl (1989)called default interpretation and tried to explain using the notion of stereotype. How-ever, Ruhl did not assume that there are any default readings that arise only on thebasis of particular conversational maxims or heuristics. Rather, from what he saysabout default interpretation, it exclusively relies on stereotypical uses of a particular(simple or complex) expression, where stereotypical may be taken to capture signifi-cantly high frequency uses. In fact, one might ask whether it is actually possible todifferentiate between types of utterance-type-meaning being determined on the basisof heuristics and such as are determined on the basis of former uses of the expression

in question. In other words, the question is whether the results of the interpretationprocess in case of particular utterances giving rise to GCIs can only be accountedfor by applying to some underlying general heuristics or whether these are preferredinterpretations that have emerged over repeated uses of the expressions in question.This latter view seems to be endorsed by Borg (2009, p. 79), who writes that ‘[a]GCI can be recovered just by knowing that a speaker who says “some” often conveyssome but not all . . . this information could be gathered as the result of exposure topast conversational exchanges . . .’.

So far, Levinson’s approach to GCIs can be taken to be Neo-Gricean in thathe, similarly to Grice, assumes a number of underlying guidelines that lead to thederivation of various types of (generalised) conversational implicature. However,Levinson’s theory of GCIs goes further than Grice’s account of conversational impli-catures in general, in that the former argues that GCIs are involved in the determi-nation of what was said or the utterance meaning of an utterance. More specifically,he claims that the same processes that derive GCIs are responsible for the resolv-ing of references, fixing of indexicals and disambiguation, which are necessary toarrive at the proposition expressed by an utterance. In other words, rather thanbeing derived on the basis of what is said and thus playing no truth-conditional role,Levinson assumes that GCIs are necessary to arrive at what is said and that theydo play a truth-conditional role. This observation leads Levinson to posit what hecalls Grice’s circle and, eventually, a particular view on the interaction of semanticand pragmatic processes which is characterised by what he calls pragmatic intrusioninto semantics.

Thus, he differentiates two semantics components, as it were, namely Semantics 1and Semantics 2. Essentially, this view follows the traditional definition of semantics,which is supposed to state the truth conditions for sentences and whose outputtherefore should be truth-evaluable propositions. Semantics 1, then, comprises theprocesses that produce the semantic form of a sentence from its surface structure.However, as we saw before and as Levinson agrees, the semantic representationsresulting from semantic composition are not fully propositional, that is, they arenot truth-evaluable. Thus, before semantic interpretation (e.g., Semantics 2) ofthe sentence in question can begin, the sub-propositional form has to be enriched.That is, particular pragmatic processes that take into account the context in which aparticular sentence is uttered, have to provide contextually derived material to enrichthe semantic form of the sentence to the proposition expressed in that particularcontext by that sentence. More specifically, Levinson argues that in determining theproposition expressed, pragmatic processes are inserted which are involved in thederivation of generalised conversational implicatures. These processes are supposedto guarantee the resolving of reference, fixing of indexicals and disambiguation andthus result in a truth-evaluable proposition, the proposition expressed. Once theproposition expressed is determined, semantic interpretation of that proposition cantake place. Thus, what we have between Semantics 1 and Semantics 2, is an intrusionof a number of restricted pragmatic processes.

What is problematic about this assumption is that, contrary to Grice’s origi-nal definition of conversational implicature, GCIs are taken to contribute to thetruth-conditional content of an utterance (since they contribute to the proposition

SEM 1: sentence meaning

Pragmatics

saturation processesGCIs

what is meant

PCIspeech acts. . .

sub-propositional

fully propositional

+ narrow context

+ broad context

SEM 2: what is said

pragmaticintrusion

Figure 4.1: Levinson’s view

expressed by an utterance, or what is said, which then is interpreted by Semantics2). Furthermore, since GCIs contribute to what is said by an utterance, we runinto trouble with Grice’s original characterisation of CIs being inferred on the basisof what is said. This latter problem is what Levinson calls Grice’s circle: in orderto infer CIs from a particular utterance, we need as a basis what is said by thatparticular utterance, which in turn is, at least partly, determined by processes thatare employed to derive GCIs. As a solution to Grice’s circle, Levinson proposes thatthe processes which derive GCIs actually are local processes. That is, GCIs are notderived only once the interpretation process has yielded a full proposition (which,in fact, would not be possible if they are taken to be essential contributions to thatproposition), rather, the processes that derive GCIs apply locally. In other words,the process that derives GCIs is triggered by particular expressions or constructionsand it does not depend on a full proposition to apply.

Note that Levinson’s assumptions concerning the processes or stages involvedin the interpretation of utterances are actually quite similar to what we have seenalready. Thus, the interpretation of some utterance starts with the semantic com-position of the semantic form of the complex expression uttered. This semanticform is then enriched by pragmatically determined aspects of meaning to yield afull proposition which can then be checked as to its truth or falsity. Whether ornot one calls the component responsible for semantic interpretation ‘Semantics 2’seems to simply have to do with whether one is interested in somehow upholding thetraditional view of semantics being concerned with full propositions. In a sense, RTalso makes reference to this traditional view of semantics by differentiating betweenlinguistic semantics and real semantics. While the former refers to the semanticcomponent in the language faculty, the latter refers to the ‘language of thought’.Thus, the role of linguistic semantics is to provide, on the one hand, the meaningsencoded by linguistic expressions and on the other hand, the rules of compositionfor a combination of those meanings. In contrast, it is in real semantics that thelinguistic-semantically composed and pragmatically enriched semantic forms of ut-

terances are interpreted.34

Whereas Levinson assumes that GCIs are based on pragmatic heuristics but‘intrude’ into the semantic form of the utterance giving rise to them prior to itssemantic interpretation, Chierchia (2004) assumes that scalar implicatures (SI) atleast actually are computed by the grammar.35 However, similarly to Levinson,Chierchia (2004) also views SIs as default interpretations and assumes that theyare locally drawn. More specifically, Chierchia (2004) argues that in computingthe meaning of an utterance, both its ‘plain’ (without potential SIs) as well asits ‘strengthened’ meaning (with potential SIs) are computed. In particular, indetermining the strengthened meaning of a given utterance, the set of alternativepropositions the speaker might have expressed but didn’t plays a crucial role: ‘. . .the relevant set of alternatives is constituted by propositions built up by using theother items on the scale.’ (Chierchia 2004, p. 42). Thus, for a sentence such asgiven in (142a), the corresponding logical form is as in (142b) and the set of thatproposition’s alternatives is (partially) given in (142c), where the arrow indicatesthe direction in which the entailment relation holds.

(142) a. John earns $200 an hour.

b. earn(j, $200)

c. { . . . earn(j, $100), , earn(j, $300), earn(j, $400) . . . }entailment: ←

Now, if someone utters (142a) in a ‘neutral’ context, he is normally understood asnot ‘plainly’ expressing the proposition given in (142b), but rather the ‘stronger’proposition in (143), which involves a SI.

(143) earn(j, $200) ∧ ¬ earn(j, $n), where $200 < n

Chierchia (2004) develops a definition for the operation of functional application(FA) that allows for an incorporation of SIs into an utterance’s semantic form whilethis is being built up (cf. 144).36

(144) Suppose α = [ β γ ], where β is of type <a,b> and γ is of type a. Then

‖ [ β γ ]‖S =

‖β ‖S

(‖γ ‖S

), if ‖β ‖ is not DE

‖β ‖S(‖γ ‖

), otherwise

In order to account for the fact that in certain contexts (so-called downward-entailingones) SIs do not arise/survive, Chierchia posits a constraint (the Strength Condition)that ensures that the plain meaning of a given utterance will be preferred, as insuch cases it will be logically stronger than the ‘strengthened’ meaning.37 One

34Thus, this second semantics component, then, does capture the semantic relations holdingbetween the concepts a speaker expresses in a particular utterance situation.

35Thus, Chierchia’s proposal does not apply to GCIs in general, but only to one subtype, namelyscalar implicatures.

36The superscript S stands for ‘strong’; DE abbreviates downward-entailing.37Strength Condition: ‘. . . the strong value cannot become weaker than the plain value.’ (Chier-

chia 2004, p. 62).

advantage of Chierchia’s proposal over Levinson’s is that it can account for the factthat SIs might play a truth-functional role when the sentence including a SI triggeris embedded into a larger structure without having to assume pragmatic intrusioninto semantics. That is, the process that computes SIs actually is taken to be partof the grammar.

Turning back to Levinson’s characterisation of GCIs: this has been criticisedby proponents of RT on several grounds. First, RT actually does not differentiatebetween different types of implicature, that is, there is no differentiation betweenGCIs and PCIs. Second, proponents of RT claim that the GCIs Levinson assumescontribute to what is said or the utterance meaning of an utterance in fact are notimplicatures, but pragmatic enrichments of semantic form. Third, contrary to bothLevinson and Chierchia, RT denies that GCIs have a default status and, thus areautomatically and locally triggered.

Proponents of RT argue against Levinson’s assumption that GCIs belong to aspecial level of utterance-type meaning. In RT, there is in fact no differentiationmade between GCIs and PCIs. This is because all implicatures are taken to besubject to the same principle – the principle of relevance – and thus to result fromsimilar types of processes. GCIs therefore cannot be taken to have a default statusand thus to belong to some level of meaning distinct from the level of meaning thatPCIs belong to. Thus, Carston claims that ‘No implicatures are a matter of defaultinference; rather, all must be warranted by contextual relevance’ (Carston 2004c,p. 8). This is not to say that implicatures do not vary in their generality, but themore general implicatures are not accorded the status of default interpretations.

In addition, Carston (2004c) criticises Levinson’s view of GCIs as contributingto what is said. She claims that there is no circularity in Grice’s definition of CI.Rather, what Levinson claims to be GCIs are in fact no implicatures at all butrather pragmatically contributed aspects of utterance meaning. The crucial pointhere is that these pragmatically contributed aspects are not implicatures, where therelevant characteristic of implicatures is that they are inferred on the basis of com-plete propositions. Thus, the pragmatic processes that Levinson takes to contributeto what is said are in fact not such that they derive implicatures, as they do notneed full propositions as their basis, but rather may operate on sub-propositionalforms. Thus, it is not surprising that the phenomenon Levinson calls GCI veryoften contributes to the truth-conditional content of some utterance, because, actu-ally, those aspects of meaning are part of what RT takes to be explicitly expressedby an utterance.

Note, however, that RT claims the determination of explicature as well as im-plicature – both being pragmatically informed aspects of meaning – is characterisedby mutual parallel adjustment. This means that it is not necessarily the case thatthe explicature of an utterance is determined prior to the determination of an im-plicature and implicatures, therefore, may be derived before a full proposition isavailable. More specifically, a particular explicature may be taken to be expressedby a particular utterance due to the influence of the hearer’s expectation of a specificimplicature. On the one hand, this seems to be a reasonable assumption. Consider-ing that communication proceeds so fast and that hearers usually are rather quickin interpreting what speakers meant, it seems unlikely that a hearer ‘waits’ until the

end of an utterance, before he starts inferring what the speaker may have meant.On the other hand, assuming mutual parallel adjustment makes the claims for dif-ferentiating explicature as a separate level of meaning that is different from Grice’swhat is said and necessary for the derivation of implicatures appear less convincing.That is, if the temporally preceding determination of a level of meaning that can becharacterised as the proposition expressed by the speaker in fact is not a necessarybasis for the drawing of further inferences, then why make the point that Grice’slevel of what is said is not rich enough in meaning to form that basis? If, contraryto Grice’s assumptions, the basis for CIs does not even have to be a full proposition,why does RT argue for explicatures in the first place? Moreoever, it is not clearhow individual steps of the determination of explicatures and implicatures in termsof mutual adjustment are to be characterised. Unfortunately, Carston’s remarks onthe point are not very clarifying either:

‘The mechanism that mediates the inferences from logical form to com-municated propositions is one of “mutual parallel adjustment” of ex-plicatures and implicatures, constrained by the comprehension strategy.The result should consist of (sets of) premises and conclusions makingup valid arguments, but the important point is that the process need notprogress strictly logically from the accessing of premises to the drawing ofconclusions. For instance, a particular conclusion, or type of conlusion,might be expected on the basis of considerations of relevance and, viaa backwards inference process, premises constructed (explicatures andimplicatures) which will make for a sound inference to the conclusion.’(Carston 2002a, p. 139)

Note that it is not obvious how the comprehension strategy constrains the mechanismof mutual parallel adjustment. Moreover, and as already noted above, the very notionof explicature and the criteria by which it may be differentiated from implicatures, arenot clear. Thus, RT replaces the condition for something to count as an implicature,namely that it be determined on the basis of a fully-propositional what is said, bythe condition that it be not a development of the semantic form of the utterance.38

One way of testing that this is not the case is by application of the scope criterion39:implicatures are assumed not to fall within the scope of logical operators. If somepragmatically determined aspect of meaning does fall within the scope of somelogical operator, it is taken not to be an implicature, but rather part of what issaid or the explicature of an utterance. Thus, consider, for instance, the sentencein (145a) (taken from Borg 2009), which is typically understood to mean (145b),unless the context prevents this.

(145) a. Jill blew the whistle on poor practices at work and was sacked.

b. Jill blew the whistle on poor practices at work and was sacked because ofthat

38We already saw that ‘development’ cannot be understood in terms of entailment (cf. Burton-Roberts 2005).

39Recanati (1989, p. 325) calls this scope principle.

Thus, the causal relation holding between what is expressed by the co-ordinatedclauses is pragmatically contributed and constitutes, in Grice’s terms, a truth-conditionally irrelevant GCI. However, if the sentence in (145a) is embedded in alarger structure involving a logical operator such as a conditional, the GCI appearsto be truth-conditional after all.

(146) a. If Jill blew the whistle on poor practices at work and was sacked, then sheis entitled to compensation.

b. If Jill was sacked and blew the whistle on poor practices at work, then sheis entitled to compensation.

Thus, according to RT and based on the scope principle, such GCIs actually are notimplicatures but rather part of what is explicitly expressed by an utterance. How-ever, and as already mentioned, it is an assumption only that implicatures cannotfall within the scope of logical operators. Having said this, note that according toChierchia (2004) SIs – when they are present – do fall within the scope of logicaloperators. However, under his approach SIs as such actually are no longer a prag-matic phenomenon, but rather are grammatically determined. Thus, the optionalityof SIs is built into the grammar by the modification of the definition for functionalapplication.40 In other words, if SIs are present, they are already part of the SFdelivered by the semantics component. As such, SIs in Chierchia’s theory are notimplicatures in the usual sense.

Generally, Levinson’s account and RT are similar in that both assume the phe-nomenon Levinson calls GCIs to contribute to the truth-conditional content of anutterance. This is also true of Chierchia with respect to SIs. The major differencebetween the two former approaches with respect to GCIs is that while Levinsonassumes of GCIs that they are implicatures, RT takes these meaning aspects to be-long to the explicature of an utterance. Thus, RT argues that although Levinson’sGCIs are pragmatically derived meaning aspects, they are not implicatures (sincethey are subject to the Scope Criterion). In contrast, for Chierchia, SIs at least,already form part of the semantic form of an utterance as determined by the se-mantics component. Moreover, RT both rejects Levinson’s assumptions that GCIsbelong to a separate level of meaning distinct from both utterance meaning andcommunicative sense as well as Levinson’s and Chierchia’s claim that they are, asit were, automatically and locally triggered.

(Mostly) Empirical Evidence Concerning GCIs

The claims made about the nature of the interpretation process by the two types oftheory (i.e. default and local vs. contextually driven and global generation of GCIs)have actually been empirically tested and it seems that the assumption about GCIsas being default interpretations that are locally triggered cannot be verified (e.g.Noveck and Posada 2003, Bezuidenhout and Morris 2004, Reboul 2004, Brehenyet al. 2006). Thus, Noveck and Posada (2003) carried out an experiment thatmeasured reaction times as well as ERPs for subjects’ truth/falsity judgements ofsentences such as the following.

40Recall the definition of FA in (144) which is an ‘if x then y else z’ instruction.

(147) a. Some houses have bricks.

b. Some elephants have trunks.

c. Some crows have radios.

Of particular interest are sentences such as in (147b), as these are the sentenceswhich are assumed to give rise to the scalar implicature ‘not all elephants’. Thus,they are underdeterminate in that they can either be understood logically as ‘At leastsome elephants have trunks.’, or as pragmatically implicating that ‘Not all elephantshave trunks.’ The results show that subjects overwhelmingly judged patently truesentences as in (147a) as true and patently false sentences as in (147c) as false.However, with sentences of the type in (147b), judgements were equivocal, withsome subjects judging these sentences with respect to their assumed semantic orlogical meaning as true, and others judging them based on the implicature carriedby the sentence as false. However, there was clear within-subject consistency, witha subject either judging underdeterminate sentences as false or as true.

Moreover, what was interesting with respect to this last group of sentences wasthe fact that, regardless of the judgement given by the subjects, there was no differ-ence in the ERP measured. The ERP of interest here was the N400, which is takento indicate semantic integration.41 Whether or not a subject judged an underde-terminate sentence as true or false, the resulting N400 was flat in both cases. Incontrast, both the N400s for patently true and patently false sentences were steeper,which indicates that these items take more semantic integration than the underde-terminate items. Even when concentrating exclusively on the performance of thosesubjects who judged the underdeterminate sentences as false – thus, presumablyderiving the implicature ‘not all’ – the results remain the same. That is, with thatsubject group the N400 for the patently true and patently false sentences was sig-nificantly larger than the N400 for the underdeterminate sentences, showing that noparticularly strenuous semantic integration of a GCI was taking place. In contrast,there was considerable difference in the time subjects took to make a true vs. afalse judgement with respect to underdeterminate sentences, with false judgementstaking almost twice as much time as true judgements. This result can be taken toreflect the fact that, contrary to what Levinson assumes, deriving a GCI is in facteffortful and that it does not take place as early during the interpretation processas to allow it to be reflected by an N400, signalling semantic integration.42

While Noveck and Posada (2003) use a single sentence verification task, Brehenyet al. (2006) take into account the role of context for the derivation of GCIs. Inparticular, an experiment was carried out that tested whether or not in a neutralcontext – thus a context that neither facilitates nor precludes a particular scalarimplicature – scalar implicatures are computed. According to RT they should notbe, as their derivation depends on the context just as much as for PCIs. In contrast,a default approach such as Chierchia’s or Levinson’s would predict that the respec-tive GCI is triggered, namely by default. With respect to the notion of ‘neutral

41The EEG in the experiment was recorded for 1100ms, starting 100ms before presentation ofthe last word in each sentence.

42Note, however, that this generalisation may not hold, since what was tested in the experimentwere scalar implicatures only. Thus, it might be that other types of GCI behave differently andmight lead to different results (cp. Garrett and Harnish 2008).

context’, Breheny et al. (2006) point out that a neutral test concerning the ‘some X’construction using isolated sentences (as was done by Noveck and Posada 2003) ac-tually is not possible, due to potential topic effects. Thus, in many languages, thereis a tendency to give old information sentence initially. Therefore, presenting anisolated sentence such as (147b) to subjects might trigger contrastive assumptionsand thus lead to the inference of the implicature.

(147b) Some elephants have trunks.

This tendency, however, is rather mild for English, since here word order is fixed.Therefore, the experiments were carried out in Greek, which has flexible word or-der, thus giving rise to ‘. . . a far stronger tendency . . . for grammatical subjectsin sentence-initial position to be construed as old information and objects in anon-initial, post-verbal position to be construed as unrelated to a contextual issue’(Breheny et al. 2006, p. 446). Thus, consider the English translations of examplestimuli below, where the ‘some x’ phrase occurred either sentence initially or sen-tence finally; the former triggering the topic effect for the critical phrase, whereasthe latter did not. Moreover, to generally control for potential order effects, Brehenyet al. (2006) designed stimuli that included the explicit trigger only for a some butnot all reading.

(148) a. Some of the consultants had a meeting with the director. The rest did notmanage to attend.

b. Only some of the consultants had a meeting with the director. The rest didnot manage to attend.

c. The director had a meeting with some of the consultants. The rest did notmanage to attend.

d. The director had a meeting with only some of the consultants. The rest didnot manage to attend.

The phrase the rest at the beginning of the sentence following the sentence with the‘some x’ construction was used to force the some but not all reading. The assumptionwas that if GCIs are context-driven rather than interpreted in default, then for asentence including a ‘some x’ construction and presented in a truly neutral context,the GCI should only arise once the forcing phrase is encountered, leading to longerreading times for that phrase. Breheny et al. (2006)’s results favour RT’s approachto GCIs in that they are interpretable as showing that GCIs are not triggered inneutral contexts.43 Thus, subjects took longer reading the rest in the non-topicimplicit condition as compared to all other conditions.

Thus, contrary to Chierchia’s and Levinsons’ assumptions, the experiments men-tioned above could neither verify the assumption that the processes that create GCIsapply locally nor that they apply by default, that is, automatically. Rather, the re-sults seem to support a context-driven appoach, such as RT. However, Borg (2009)argues that it is not all that clear whether that really is the case. On the one hand,the time delay found by Noveck and Posada (2003) for the processing of scalar

43As before, this particular experiment tested only scalar implicatures.

implicatures may pose a problem for a general explanation in terms of a relevance-theoretic comprehension procedure. On the other hand, it seems that RT cannotexplain the differences found concerning adults’ vs. children’s recovery of scalar im-plicatures (Noveck 2001). Recall that RT actually treats scalar implicatures not asimplicatures but rather as part of what is explicated by an utterance of a particularsentence involving scalar expressions. Borg argues that

. . . if the recovery of enriched scalar terms is held to be effortful and time-consuming, so too should the recovery of other kinds of pragmaticallyenriched explicatures. Yet it seems unlikely that RT theorists wouldwant explicature recovery in general to be marked by time-delay (Borg2009, p. 76).

Another problem for treating scalar implicatures as part of the explicature of anutterance is the fact that children give utterances involving scalar expressions muchless often a pragmatic interpretation than adults (e.g., Noveck 2001). Given that‘. . . both adults and children are in possession of the right reasoning processes togenerate both explicatures and implicatures (i.e. both possess the same relevance-directed mechanism)’ (Borg 2009, p. 76), it is not clear why these differences shouldarise. This question becomes even more pressing, considering that children performsimilarly to adults when it comes to employing pragmatic enrichment processes inovercoming semantic underdeterminacy, which also results in enriched explicaturesfor utterances.

Borg (2009) offers an alternative explanation of GCI data, and scalar implicaturesin particular, which is based on her minimal conception of semantics (see section3.2.3). Due to Borg’s minimal characterisation of semantics, it is clear that GCIscannot be semantic. However, she also recognises the fact that there is some intuitivedifference between GCIs on the one hand and PCIs on the other. Thus, GCIs shouldnot be treated as on a par with once-off inferences, ‘. . . since they do not rely onaccess to the current speaker’s state of mind’ (Borg 2009, p. 79). The differencebetween PCIs and GCIs, then, is captured by the fact that they result from distinctkinds of cognitive processing. More specifically, Borg (2009) suggests that during theinterpretation of an utterance, the output of the minimal semantic component servesas input to two further systems: the general pragmatic system that derives PCIs anda more limited system that recovers GCIs. Of this latter system she assumes thatit makes use of statistical facts concerning what speakers intented when they usedparticular expressions on former occasions. Borg (2009, p. 79, footnote 20) suggeststhat ‘. . . GCIs are a kind of abstraction from . . . PCIs: an agent starts by graspinga pragmatically enhanced reading of a scalar term as a PCI but later they come tograsp such readings simply through knowing how words are commonly used’. Thus,similarly to Levinson, Borg assumes that GCIs belong to a level of meaning thatis distinct both from the level of utterance meaning as well as the level of what ismeant. However, in contrast to Levinson, Borg neither assumes that GCIs arise onthe basis of particular heuristics nor that they arise locally.

Moreover, Borg (2009) has to concede that with such an approach, the truth-conditional effects of GCIs on utterance meaning have to be treated as rather ap-parent than real. However, the approach does offer an explanation as to why we

intuitively take GCIs to have a truth-conditional effect on utterance meaning. Thatis because these meaning aspects are, in a sense, automatically derived, both inde-pendently of the actual context as well as the actual speaker intentions in makinga particular utterance. Furthermore, Borg (2009) argues that her account of GCIsalso fits the experimental results mentioned before. Thus, the time delay noted forthe processing of scalar implicatures is explained by the assumption that they are in-ferences that are drawn after the truth-conditional content of an utterance has beencomputed, so, contrary to Levinson’s assumption, they are not local.44 Moreover,the fact that the younger a child is, the less likely it is to interpret scalar implicaturespragmatically is explained by the assumption that the younger a child is, the lessexposure it will have had to ‘usual uses’ of scalar terms, thus lacking the necessarypremise for an automatic recovery of GCIs that is free of speaker intentions.

The independence of speaker intentions of GCIs is also central to Burton-Roberts(2006)’s discussion of the notion of cancellation. Thus, recall that traditionally, oneof the properties of CIs is that they are assumed to be cancellable. As Burton-Roberts (2006) convincingly argues, this property actually only holds of GCIs. Thatis, a speaker cannot explicitly cancel a PCI, the reason being that PCIs are alwaysintended. They only ever arise in particular contexts. Thus, consider the followingsmall conversation.

(149) Max: Do you ever speak to Charles?

Ann: I never speak to plagiarists.

In that particular context, Max will take Ann to have conveyed both that Charlesis a plagiarist and that she never speaks to him. The question is, whether she couldactually explicitly cancel these PCIs without seeming contradictory. Burton-Roberts(2006) argues that she could not add ‘I’m not suggesting that Charles is a plagiarist,mind you – in fact, I love talking to him’.

Similar points must be made with respect to RT’s category of explicature, ofwhich we already saw that they may be clarified but not cancelled.45 Incidentally,Burton-Roberts (2006) suggests that RT’s assumption that (at least some) GCIsactually are part of an utterance’s explicature is the reason why RT has to assumethat the latter are cancellable in the first place – because they may involve suchpragmatically determined meaning aspects as Grice called GCI and of which weindependently know that they have the property of being cancellable.46 However, wealready saw that there is a good reason against assuming that explicatures generallyare cancellable and that is that they are taken to be the proposition the speakerintended to express. Burton-Roberts (2006)’s consequence is to assume that GCIsare a ‘real’ phenomenon and that they are implicated rather than explicated by anutterance.

Thus, Burton-Roberts (2006) assumes that GCIs are different from both expli-catures as well as PCIs in that the communication of the latter two by a speaker

44For experimental results that support such a view, see the discussion of Huang and Snedeker(2009, in press) below.

45For a similar suggestion with respect to the level of what is said and using the term ‘revisability’,see Korta (1997).

46Note, however, that originally the idea was that it is a characteristic of all pragmaticallyinformed inferences that they are cancellable.

is always taken to have been intended. In contrast, GCIs arise independently fromand regardless of the speaker’s intentions. In characterising GCIs, Burton-Roberts(2006) refers to Gazdar (1979)’s notion of im-plicatures, that is, potential implica-tures that only become actual implicatures of an actual utterance, if they are notinconsistent with contextual information. Thus, potential GCIs are assigned auto-matically to the relevant linguistic expressions. However, they only become actualGCIs when these expressions are uttered and if there is no evidence that the speakerdid not want to communicate them. Such evidence may be provided either by thespeaker explicitly cancelling the potential GCI or by contextual information that isinconsistent with the potential GCI. In those cases where the speaker does want toconvey a particular GCI, she does not have to do anything special, as the hearer willassume that the GCI is part of what the speaker intended to communicate since thespeaker did nothing to prevent him from making that assumption.

Similarly to Borg (2009), Burton-Roberts also assumes that GCIs are implica-tures, that is, they are drawn on the basis of the utterance meaning of a speaker’sutterance. They are cancellable in principle, however, if they are not cancelled –either explicitly or by the context – they will become actual implicatures and as suchform part of what the speaker is taken to have intended to convey. Arguably, thatis why they seem to have a truth-conditional effect in complex sentences. However,as Borg (2009) points out, they do not in fact contribute to the truth-conditions ofthe uttered sentence, but rather to speaker meaning. Thus, there is no circularityin Grice’s definition of CIs. Nevertheless, GCIs are special in that they are, in somesense, drawn automatically, although this does not happen locally.

Note that the results of experiments reported by Huang and Snedeker (2009, inpress) can be taken to support such a view, at least as far as scalar implicaturesare concerned. In Huang and Snedeker (2009), the authors report experimentsmaking use of the visual world paradigm. Their subjects heard stories, which whereaccompanied by a visual display. The stories where about four characters (twogirls and two boys) and two types of objects, which were distributed amongst thecharacters. For instance, one girl might end up with a proper subset of one type ofobject (e.g., socks), whereas the other girl was given all of the instances of the othertype of object (e.g., soccer balls). After hearing the stories, subjects were giveninstructions such as the following.

(150) Point to the girl with two/three/some/all of the socks.

Subjects’ gaze direction and time were measured while they were given instructionsand carried them out. Note that when the subjects hear the expression girl, the po-tential referent is restricted to two possible entities. Once subjects heard a numeral,the intended referent was immediately disambiguated, as the girls in the storiesnever had the same amount of objects. The same happened with the quantifier all,since there was always only one girl with all the instances of one type of object.The critical condition, however, was that with the quantifier some. The hypothesesHuang and Snedeker (2009) wanted to test are the following. If some is interpreted‘semantically’ (meaning ‘some and possibly all’), the intended referent would onlybe disambiguated at the final noun of the sentence. If, however, some is locally givenits ‘pragmatic’ interpretation (‘some but not all’), the intended referent should also

be immediately disambiguated (it cannot be the girl with all of the instances of oneobject type). Thus, either subjects show no difference in the time they take to fixon the intended referent in the different conditions, thus suggesting that they locallyretrieve the interpretation of some as ‘some but not all, or they take longer in thesome condition, showing that they initially interpret some semantically as ‘some andpossibly all’ and thus have to ‘wait’ in order to disambiguate the intended referent.

The results show that the potential pragmatic interpretation of some is notcomputed locally. Thus, while subjects take approximately 400ms from the onsetof the quantifier to lock their gaze on the intended referent in the two, three andall conditions, they take about 800ms in the some condition. However, the timeat which they identify the intended referent is not clearly the time at which theyinterpret the final noun in the sentence. That is, it is possible that subjects fixed ona referent before the disambiguating information from the final noun in the sentencebecomes available. The experiment reported in Huang and Snedeker (in press) wasdesigned to further investigate this possibility. Thus, the time between onset of thequantifier and disambiguating expression was increased by using compound nounpairs.

(151) Point to the girl with two/three/some/all of the ice cream sandwiches/cones.

The results of this experiment clearly showed that although scalar implicatures arenot computed as part of initial sentence processing, they are drawn online, before thedisambiguating information becomes available.47 While these results, once again,disconfirm Chierchia’s and Levinson’s assumption that GCIs are locally retrieveddefault interpretations, Huang and Snedeker (2009, in press) note that they arenot totally unproblematic for RT either. Recall that in RT drawing pragmaticinferences are assumed to involve as little cognitive effort as possible while at thesame time resulting in as many cognitive effects as possible. However, as Huangand Snedeker (2009) point out, in their experimental task, it seems that little isgained by drawing the critical inference, since ‘. . . the referent of the quantifiedphrase was always lexically disambiguated, making the implicature unnecessary forfull comprehension’ (Huang and Snedeker 2009, p. 409). Nevertheless, subjectsseem to have drawn the inference, suggesting that it involves low cognitive costs.Thus, Huang and Snedeker (2009, p. 409) caution that ‘. . . these data suggest thatRelevance theory may need to acknowledge the possibility that scalar implicatureshave a preferred status relative to other pragmatic inferences’.

Note that the time-course of scalar implicature interpretation observed by Huangand Snedeker (2009) may allow for an alternative interpretation of the results at-tained by Breheny et al. (2006). Thus, recall that in the critical ‘neutral context’condition, the scalar expression occurs sentence-finally in the first sentence, followedby the ‘forcing phrase’ sentence-initially in the second sentence (cf. 148c), repeatedbelow.

(148c) The director had a meeting with some of the consultants. The rest did notmanage to attend.

47Cp. Storto and Tanenhaus (2005) for similar results concerning the inclusive/exclusive inter-pretation of or.

Breheny et al. (2006) predicted that if GCIs are interpreted locally by default,the reading time for the forcing phrase should not be any different from that inthe other conditions, as the GCI would be drawn immediately on the encounterof some. In fact, the reading time for the forcing phrase in the ‘neutral context’condition turns out to be longer than in the other conditions. Breheny et al. (2006)interpret this result as showing that the GCI was in fact only drawn at the timethe forcing phrase was encountered. More importantly, they assume that it is theforcing phrase that triggers the drawing of the scalar implicature. However, althoughHuang and Snedeker (2009)’s results confirm that the drawing of scalar implicaturesis not a local process, nevertheless, it does seem to take place automatically althoughdelayed with respect to the onset of the scalar expression. Thus, it is possible thatthe longer reading time for the forcing phrase – which, in terms of presentationimmediately followed the phrase involving the scalar expression – simply reflects thedelay in drawing the scalar implicature, rather than indicating that the implicaturewas actually triggered by the forcing phrase.48

While most of the studies discussed so far concentrated on scalar implicature,Garrett and Harnish (2007) investigated the interpretation of expressions argued byLevinson to give rise to GCIs based on the I-heuristic. More specifically, the authorsconcentrated on three types of implicit meaning aspects – which, following Bach(1994a) they call implicitures – namely ‘location’ (cp. 152a), ‘possession’ (cp. 152b)and ‘time’ (152c) implicitures.

(152) a. It’s raining.

b. John cut a finger.

c. I’ve had breakfast.

In a first experiment, Garrett and Harnish (2007) investigated the strength of thethree types of impliciture associated with particular complex expressions out ofcontext. Subjects were presented stimulus sentences auditively. After each stimu-lus sentence, a single question word was visually presented, with two probe wordsprinted below. The question word was appropriate to the respective implicituretype. Similarly, the probe words were such that one was the word generally used tomake explicit the presumed impliciture and the other a word that is not part of theimpliciture but syntactically and semantically comparable to the impliciture probeword. Thus, subjects heard ‘Somebody said: It’s raining.’, followed by the questionword WHERE? on a screen with two possible answers from which to choose, in thiscase HERE and THERE. Both the choice subjects made as well as the time theytook to make it were noted. Moreover, in addition to the impliciture sentences,other similarly structured sentences which, however, did not give rise to implicitureswere tested as a control. Results showed that for impliciture sentences subjectsoverwhelmingly chose the impliciture probe word as the appropriate answer to the

48Compare Garrett and Harnish (2009, p. 100) who make a similar general point concerningthe interpretation Breheny et al. (2006) give their results: ‘Is the [scalar implicature] processingfollowing the explicit trigger [i.e. only K. B.] faster than it is following the implicit trigger? If thelatter is delayed (but does occur), the interpretation of the increased readings times at the forcingregion [i.e., the rest K. B.] would be affected.’

question posed and they did so quickly. Thus, ‘. . . [t]he results confirmed the specialstatus of the implicitures’ (Garrett and Harnish 2008, p. 81).

In a second experiment, the authors used the materials from the first, however,the critical sentences were now placed at the end of small stories, intended to eitherenable or cancel the respective impliciture that had been elicited out of contextbefore. In this experiment, the stimulus texts were presented visually and subjectsread through them at their own pace. Reading times for stimulus texts were mea-sured as was response time for question answering as in the first experiment. Therewere no statistically significant differences in the reading times for the two contextsor the target sentences in the respective contexts. Response times, however, differedin that subjects responded more quickly to the probe questions in enabling contextsthan they did in the cancelling contexts. Overall, accuracy of responses was veryhigh for both contexts. These results suggest that the impliciture was computed inboth contexts, but had to be cancelled in the cancelling context. More generally,the results of this experiment once again suggest that ‘. . . contextual penetration ofsentence interpretation should be distinguished from the interpretation that relieson sentence form . . .’ (Garrett and Harnish 2008, p. 84).

Summary

In summary of the section on conversational implicatures, the following points canbe made. It is useful to differentiate GCIs from PCIs, both for theoretical as wellas empirical reasons. A theoretical differentiation is based on two intuitive factsabout GCIs, which make them different from PCIs, namely, that they seem to ariseindependently of a particular context as well as regardless of speaker intentions.However, as we saw, both these intuitive facts have to be treated with caution.Thus, on the one hand, there are arguments that scalar implicatures at least doneed a ‘special context’ to be triggered, if information-structural aspects such asthe topic effect are to be counted among contextual properties. Similarly, a pointcan be made that GCIs have to be consistent with speaker intentions in order to‘survive’ or become ‘actual’ implicatures of an utterance, in which case they will betaken to have been intended by the speaker after all (cp. Burton-Roberts 2006). Asalready mentioned, assuming such a behaviour of GCIs might explain the fact thatthey seem to have a truth-conditional effect on utterances that give rise to them,without necessarily having to assume that they arise locally and are already part ofthe utterance meaning of an utterance.

Empirical investigations of GCIs also suggest that they are ‘special’. Acrossa range of experiments using different testing methodologies, it has been shownthat, on the one hand, the influence of GCIs on the interpretation process is notimmediate but rather delayed, but on the other hand, they ‘arise’ even in cancellingcontexts. While the former result suggests that GCIs are not locally drawn defaultinterpretations, the latter result suggests that although their inference is effortfuland time-consuming, GCIs still differ from PCIs in that they seem to influence theinterpretation process even in those cases where they are not intended by a speaker.

As concerns the coming about of GCIs, we saw that Levinson assumes that theyare based on a number of heuristics, whereas Borg and Bach assume they are, in asense, abstracted or standardised from PCIs as a result of their frequent occurrence

in a particular reading. As was argued above, this latter view is straightforwardlycompatible with the different results found for children and adults concerning thedrawing of scalar implicatures. Moreover, Garrett and Harnish (2007)’s experimen-tal results also favour a standardisation view of GCIs, although they are cautiousthat this may be due to their particular interpretation of Levinson’s I-heuristic. Re-call also that Levinson himself characterises the level of utterance-type meaning assubsuming such phenomena as standardised inferences, although he suggests thatGCIs are distinct from those. Overall, I would suggest, the experimental resultsconcerning the processing of putative GCIs support the idea that there might be aspecial level of meaning at which sentences are paired with preferred interpretationsand to which GCIs, amongst other phenomena, belong. However, the fact that re-sults of experiments testing SIs and so-called I-based implicatures differ from oneanother, does suggest the possibility that the phenomena collectively called GCImight in fact be distinct and thus exhibit different properties.49

4.1.4 Speech Acts

The notion of speech act captures the fact that usually in using language we carryout specific actions. Thus, the speaker of (153) does not only utter a complexexpression and thus expresses a specific meaning, but, more importantly, undercertain conditions he also enters into a specific kind of commitment, namely to dowhat he said he would do.

(153) I promise to come.

Speech acts in general are assumed to be a phenomenon that can be identified atthe level of what is meant/communicative sense; at a level, that is, where criteriaare taken into account that are specific for general human interpersonal interaction.Thus, criteria are considered that may influence the hearer’s interpretation of whatthe speaker meant – and thus, what his intentions were in making a particular ut-terance – which are not necessary for figuring out what the utterance as such means.Recall the example from Bierwisch, repeated below as (154), which, depending onthe context of utterance, may be used to make a promise, statement or as a threat.Regardless of which speech act this is taken to be a token of, the hearer is able tounderstand what the utterance as such means.

Moreover, as Bierwisch observes, although (153) involves a performative verb(or illocutionary force indicating device), the sentence may be felicitously used ina situation in which the intended speech act is not one of promising. Thus, thekind of speech act that a speaker intends to carry out with his utterance, very muchdepends on the utterance situation. Although there are linguistic indicators that areconventionally associated with particular speech acts, it is questionable whether thispotential should be considered as being part of the semantics of such expressions.Moreover, in order to carry out a speech act of promising, it is not even necessaryto use the expression promise at all. Thus, as we have already seen, an utterance of(154) made in the appropriate circumstances may also count as a promise.50

49This point will be taken up in section 5.2.1.50This does not, of course, hold true for all types of speech act. Thus, for the action of naming to

be successfully carried out, an utterance has to be made by the appropriate person in appropriate

(154) I will come tomorrow evening.

The situation is similar as regards the correlation of particular sentence types withspeech acts, giving rise to the notion of indirect speech act. Thus, the three basicsentence types declaratives, interrogatives and imperatives are usually associatedwith the speech acts of asserting or stating, asking or questioning and ordering orrequesting, respectively. If the sentence type of an utterance matches the intendedillocutionary force of that utterance, the speech act carried out is said to be a directspeech act. If, however, there is no such match, the speech act is said to be indirect.In (155), for instance, the speaker overtly asks a question, but covertly performs aspeech act of requesting.

(155) A: Can you pass the salt?

B: Yes. [B passes A the salt]

Thus, the utterance meaning of (155) can be described as asking for the truth orfalsity of the related affirmative proposition ‘You can reach the salt’. However, thespeaker’s intentions would not be adhered to, if in answer to (155), B merely saidyes and did not also actually pass the salt. Thus, A most likely intended to makea request by his utterance. The content of his actual utterance guides the hearerto what it is specifically, that A is requesting. This is so because what the speakerseems to be questioning is one preparatory condition for requesting of the hearerthat he pass the salt.

One possible analysis of indirect speech acts, proposed by Searle (1975), assumesthat such speech acts have two illocutionary forces associated with them. Whereasone of them is characterised as the literal, but in this case secondary illocutionaryforce (in the example above: questioning), the other is the non-literal, but primaryillocutionary force (here: requesting). Thus, we also find the notion of literal andnon-literal meaning aspects applied at the level of what is meant/communicativesense, with respect to the notion of indirect speech act (e.g. Capone 2004, Huang2007).

Searle furthermore assumes that in the interpretation of indirect speech acts, someinferencing is involved. He suggested that the latter, similarly to conversationalimplicatures, are based on the Gricean principle of co-operation and the maximsof conversation. Interestingly, quite a number of what are called indirect speechacts have become conventionalised to such an extent that naive speakers, on a firstblush, may not even be aware that they actually are not direct. The degree ofconventionality becomes apparent, if one compares, e.g. (155) with the sentencesbelow, which, although they express the same meaning as (155), cannot be used torequest of the addressee that he pass the salt.51

(156) a. Are you able to pass the salt?

b. Do you have the ability to pass the salt?

circumstances, where that utterance has to include the expression christen or name.51Thus, note that whereas in (155) it would be possible to add please, thereby overtly indicating

the intended speech act, this is not permissible in the examples in (156).

To account for such facts, Morgan (1978) introduced the idea of short-circuited im-plicature: although the implicature from the literal illocutionary force of an utteranceto the non-literal, intended illocutionary force is in principle calculable, in cases ofconventionalised indirect speech act, the implicature is not in fact (fully) calculated.Thus, Morgan, as Busse later on, uses the notion of Gricean implicature to accountfor the process of conventionalisation (see also Konig 1988, Rolf 1995).

So, traditionally, indirect speech acts are treated as conversational implicatures,albeit as to some extent conventionalised ones. Nevertheless, similarly to the tradi-tional treatment of metaphor and irony in terms of conversational implicature, theassumption is that the direct speech act is determined first and the indirect speech actafterwards and on the basis of the direct speech act if that does not fit the situationat hand. However, as for metaphor, it has been shown also for indirect speech actsthat they are not necessarily determined after potential direct speech acts. Thus,Shapiro and Murphy (1993) showed that processing of indirect speech acts comparedto direct speech acts is not necessarily secondary and optional. Similarly, in a studyinvestigating the comprehension of indirect speech acts in aphasic subjects, Hirstet al. (1984) found that subjects with anterior left-hemisphere damage did compre-hend the indirect speech act, although they were not able to determine the directspeech act on the basis of the syntactic structure of the utterance.

In their first experiment, Shapiro and Murphy (1993) tested the traditional viewconcerning the interpretation of indirect speech acts, which they call the serial model,as well as an alternative model, which they call the parallel model. As we saw before,according to the serial model, indirect speech acts are derived after the direct speechact has been computed and judged as not fitting the context of utterance. Theformer, thus, depends on the latter and is only optionally derived. In contrast,the parallel model assumes that direct and indirect meanings of an utterance areprocessed in parallel. That is, in terms of processing, indirect meanings are notbased on direct meaning in the sense that the former are derived only after thelatter has been. Moreover, this model assumes that accessing a potential indirectmeaning is not optional, but rather obligatory.

Subjects were presented sentences of which they were asked to decide whetherthey had a plausible direct meaning. There were four stimulus conditions, namely,plausible direct (+D) and implausible direct (-D) meaning and plausible indirect(+I) and implausible indirect (-I) meaning. The items were designed such that eachcould be understood as having a direct and/or an indirect meaning or neither (e.g.,+D-I, +D+I, -D+I and -D-I).

(157) +D-I: Where is Sandy this weekend?

+D+I: Don’t you think this is exciting?

-D+I: Can you stop whistling?

-D-I: Can text books tell time?

The serial model predicts that the availability of a plausible indirect meaning shouldplay no role in subjects’ decision on the plausibility of the direct meaning. Thatis, reaction times in conditions (+D-I and +D+I) should be equally fast. Theparallel model, however, leads to the following predictions. Although subjects werespecifically asked to judge whether critical items had a plausible direct meaning only,

if the critical item has a plausible indirect meaning, this should be automaticallyprocessed, leading to longer reaction times in the (+D+I) condition as compared tothe (+D-I) condition, as the subject will have to decide which of the two availableinterpretations is the direct one.

The results of the experiment support the parallel model. Thus, subjects tooklonger to judge whether a target sentence had a plausible direct meaning if thatsentence also had a plausible indirect meaning, showing that the indirect mean-ing interfered. Thus, even though in the experiment there was no context givento bias an indirect meaning interpretation and subjects were specifically asked toconcentrate on direct meanings only, they nevertheless processed plausible indirectmeanings and actually could not ignore them. This shows that indirect meanings arenot secondary to direct meanings and they are not optional but rather obligatory.

Hirst et al. (1984) investigated further the question of whether processing of thedirect speech act is necessary for the understanding of the indirect speech act. Theirsubjects were anterior aphasics, who generally seem not to be able to use knowledgeof syntax in their interpretation of utterances. Thus, ‘. . . they may not be able toparse a “Can you . . .” utterance as a question since it requires sophisticated syntacticcompetence to distinguish a question from a declarative sentence’ (Hirst et al. 1984,p. 28). However, anterior aphasics might make use of contextual information aswell as of their world knowledge and knowledge of the individual words in order tointerpret a particular utterance.

Recall once more that the serial model holds that the processing of an indirectspeech act depends on the prior processing of the direct speech act, which is thenjudged as not fitting. This model predicts, then, that anterior aphasics should notbe able to interpet an utterance as an indirect request, when they are not able toprocess the direct question. In contrast, the parallel model predicts that anterioraphasics should be able to interpret an utterance as an indirect request, employingcontext, even if they are not able to interpret the direct question, employing context,because the two interpretations are independent of one another.

Subjects watched colour-videotaped episodes, with two characters involved indifferent kinds of activity in different situations. In each episode, one of the charac-ters asked a ‘Can you X?’-question, to which the other character responded eitherby starting to carry out some action or giving a simple verbal yes/no answer. Allin all, four different responses were possible: appropriate or inappropriate question-answering and appropriate or inappropriate action carried out. Subjects were askedto judge whether the response to the ‘Can you X?’-question was appropriate.

The results show that the anterior aphasics did understand indirect requests,since in those cases they indicated at a rate above chance that an action would beappropriate in response. However, in those cases where ‘Can you X?’ was intendedas a direct question, anterior aphasics judged the appropriate ‘yes’ answer as actuallyappropriate in only 34% of the cases, indicating that they did not understand the‘Can you X?’-question as a genuine question. Thus, Hirst et al. (1984)’s resultssuggest once again that the prior processing of the direct speech act is not necessaryfor the understanding of an indirect speech act.

To summarise, it seems that the interpretation of indirect speech acts neitherrequires the prior determination of the direct speech act made by the speaker of a

particular utterance, nor is it optional. That is, if an indirect meaning is available,it will be processed even in situations in which there is no context which could besaid to have triggered it (cf. Shapiro and Murphy 1993). In a way, then, indirectspeech acts can be compared to familiar metaphors, although in contrast to theformer, the latter are not based on utterance meaning. That is, it seems that likefamiliar metaphor, indirect speech acts of the type investigated by the studies citedhere, might be assumed to be conventionalised to such an extent that they areprocessed in parallel to the potential direct speech act expressed.52 In fact, theparticular characteristics of such, in a sense, conventionalised indirect speech actscan be captured by assuming that they, like GCIs are part of Levinson (2000)’s levelof utterance-type meaning. Thus, like GCIs, conventionalised indirect speech actsare based on utterance meaning, they arise independently of the particular contextof utterance and they have to be cancelled if they are not intended.

4.2 Differentiating What is Said from What is

As we already saw, the different approaches to utterance interpretation that assumeat least two separate context-dependent meaning levels (e.g., Grice, Bierwisch) as-sume that the one level (i.e., what is meant/communicative sense) is, in some sense,based on the other (i.e., what is said/utterance meaning). However, it is not clearhow exactly this relation is to be explicated. Thus consider again the example (67).

EX: I had a very large breakfast today.

CI: B doesn’t want lunch.

It seems obvious that what B actually wants to convey in uttering I had a very largebreakfast is that he does not want to have lunch. However, this is not explicitlyexpressed in his utterance of I had a very large breakfast. Rather, it is the hearer’stask to arrive at the speaker’s intended meaning and that is only possible via anumber of pragmatic processes applying to the meaning of the complex expressionB uttered. Thus, although in this example, B can be taken to have meant whathe said, what he meant is not exhausted by what he said. Moreover, in order todetermine the sum, as it were, of what B meant, the hearer first has to determine themeaning of the utterance that B made, in order to derive potential conversationalimplicatures that the speaker might have wanted to convey.

Note that examples such as the one above contrast with examples, where it isnot totally clear which role the speaker’s actual utterance plays in the process ofdetermining the meaning she intended to convey. Thus, consider another, well-known example of conversational implicature (taken from Levinson 1983).

52However, the studies cited did not test whether indirect speech acts are also initially processedeven when there is contextual information available that biases the interpretation towards a directone, a behaviour characteristic for familiar metaphors.

Differentiating What is Said from What is Meant 185

(158) A. I do think Mrs Jenkins is an old windbag, don’t you?

B. Huh, lovely weather for March, isn’t it?

B’s utterance in (158), at first blush, seems totally irrelevant to the topic that Aintroduced with his utterance. However, this example can be analysed along thelines that B’s utterance might in fact be relevant in that he may have conveyed theproposition that he does not like to talk about people behind their back or that theperson A just talked about, is standing right behind A. However, the question, forboth these potential interpretations, is which role, if any, the actual content of B’sutterance plays in the determination of what B actually meant. What can be saidabout this example, regardless of how to answer that question, is that the meaningof B’s utterance can be determined, in a sense, independently of what he intendedto convey by it. That is, arguably, A will be able to conclude what B intendedto communicate by his utterance regardless of whether his utterance appropriatelydescribes the state of the weather at the time of utterance. More specifically, evenif B says what he does on an uncomfortably rainy and windy day, A will be able tofigure out what B primarily intended to convey was something other than an irony(as this is not relevant in the situation). Thus, B’s utterance may or may not betaken to be part of what he meant by making it.

Lastly, consider an example of what might be assumed to be an idiom of German.

(159) Wir stehen hinter dir

‘We stand behind you’

The sentence in (159) has a very salient non-literal meaning out of context amountingto ‘we support you’. A likely context in which the sentence in (159) can be felicitouslyuttered is one in which it qualifies as a promise on the part of the speaker. It is ratherunlikely that (159) with the reading ‘we support you’ would be used by a speakerto, e.g., warn the addressee. Having said this, the sentence in (159) can of coursebe used to issue a warning, however, the meaning associated with it would probablybe something along the lines of ‘We are watching what you’re doing’, another non-literal meaning of the sentence. Thus, in a case like (159), recognising the speaker’sintention in making the utterance, can actually contribute to understanding theutterance meaning itself.

The fact that it is not clear how what is said/utterance meaning relates to what ismeant/communicative sense has been argued to show that actually no differentiationbetween these two putative levels of meaning is possible. Thus, Borg (2004b) andCappelen and Lepore (2005) argue against the claim put forward by Bach (1994a)that indirect speech reports capture what is said by a particular utterance. Rather,indirect speech reports seem to pick up on all sorts of aspects associated with aparticular utterance, thus, again constituting evidence that what is said cannot bestrictly differentiated from what is meant. In contrast, Recanati (2004) assumes thatwhat is said should be differentiated from what is meant and he proposes that thedifference be captured by distinguishing two general types of pragmatic processes,where the one type leads to what is said by an utterance and the other to what ismeant by the speaker of that utterance. Similarly, Bierwisch (1980) claims that adifferentiation of utterance meaning from communicative sense is necessary, although

for different reasons, namely the fact that it is possible both to understand what aspeaker said without necessarily knowing what he meant by his utterance and viceversa.

4.2.1 What is Said/What is Meant and Indirect Speech Re-ports

Thus, Borg (2004b) as well as Cappelen and Lepore (2005) agree on the claim thatindirect speech reports cannot, across the board, be taken as reflecting what wassaid (in the Gricean sense) by the reported utterance/speaker. It should be notedthat Carston (2002c) concurs in this view. In fact, the idea that starting froman utterance’s meaning, or what is said, one can recover the semantic meaning ofthe sentence used to make the utterance, is not an idea that is generally held bycontextualists.53 Borg (2004b) shows that amongst the possible set of such indirectspeech reports of the utterance of a particular sentence are such which clearly are notsemantically related in content to the semantic meaning of the sentence originallyused. Thus, consider the sentence in (160), where the sentences given in (161)exemplify possible indirect speech reports for a particular utterance of (160) by someperson Jim in different possible contexts54.

(160) Blair lives at No. 10.

(161) a. Jim said that Tony Blair lives at No. 10 Downing St, London, UK.

b. Jim said that the current Prime Minister lives at No. 10.

c. Jim said that that man lives there.

d. Jim said that Baby Leo’s father lives in No. 10.

e. Jim said that the most right-wing Labour leader to date secured powerwhere other, more left-wing, predecessors had failed.

f. Jim said that he knows about British politics.

There are several points to note here. First, although the reports all include the ex-pression x said that . . ., this does not ensure that what is reported here is anythingclose to what the theoretic notion of what is said tries to capture. In this con-nection, recall the experiments mentioned above on the difference between minimalproposition (involving something close to Borg’s liberal truth conditions), propositionexpressed (involving actual truth-conditions) and purely implicated meaning, wherethe results showed that naive speaker-hearers do not understand and use the termsay in the way Grice intended it in his theoretical considerations. Moreover, con-trary to what is predicted by Grice’s characterisation of the levels of what is saidand what is meant and their interaction, the results of the experiments mentionedsuggest that interpreters do not generally differentiate between what a speaker said

53As mentioned, e.g. Bach (1994a) holds this view. It should be noted, however, that in orderfor the intended correlation between what is said by an utterance and an indirect speech reportthereof to hold, certain restrictions on the form of the indirect speech report apply. The critiqueremains, though, that such restrictions obscur the fact that in reality, indirect speech reports maytake a multitude of forms.

54Examples are taken from Borg (2004b) and, due to their subject matter, are somewhat dated.

and what he implied with his utterance. In my opinion, this is clearly to be seen byexamples such as (161e) and (161f), which would seem odd at the least, if they wereintended as indirect speech reports of what Jim ‘said’ in the theoretical sense. Theyseem to involve uses of the term say that might, e.g., be paraphrased as follows.

(162) a. Jim meant that the most right-wing Labour leader to date secured powerwhere other, more left-wing, predecessors had failed.

b. Jim intended to show that he knows about British politics.

This, however, would indicate that what we have here, if anything, is a report onwhat the ‘reporter’ thinks the speaker meant or intended by making the particularutterance he did. Thus, it does not capture what is said but rather what wasmeant. The fact that for Borg all the different mentioned indirect speech reports arepossible/appropriate lead her to assume that there is no clear-cut boundary betweenthe levels of what is said and what is meant. Thus, whether some particular meaningaspect counts as part of what is said or what is meant may differ from occasion tooccasion.

To repeat, Borg, takes what is said to be part of speaker meaning and as such,it is totally pragmatically determined and therefore one cannot expect that fromknowing what was said by a speaker one can deduce the meaning of the sentencethe speaker uttered. Thus, there is only a one-way relationship between sentencemeaning and what is said, namely such that the latter is based on the former. Thequestion, again, is what exactly it means for what is said to be ‘based’ on sentencemeaning (cf. RT’s ‘development’). This question is especially interesting here, asBorg argues that the minimal semantic content of a sentence, its sentence meaning,need not be part of what the speech act means, that sentence is used to carry out.That is, semantic content need not be a part of speech act content. However, if whatis said is part of speaker meaning, where the latter can be assumed to correspond tospeech act content, and sentence meaning is not part of speaker meaning, then whatexactly is the relation that sentence meaning and what is said are in? Moreover,from Borg’s discussion it rather seems as if it were not at all clear whether what issaid actually can be taken to be a level of meaning distinct from what is meant andif that is the case, it is actually questionable whether we can state how we get fromsentence meaning to what is said.

Now, in Cappelen and Lepore (2005)’s approach, the speaker generally is notmuch of an authority when it comes to what it was she said when making a particularutterance. As already noted, Cappelen and Lepore (2005) state that indirect speechreports cannot capture the technical notion of what is said, however, they seem tothink that they do capture what the speaker said in a wider sense. Thus, theylist a few examples of reports on the so-called ‘Smoking Gun’ utterance55 made by

55‘When you get in these people, when you get these people in, say: “Look, the problem is thatthis will open the whole Bay of Pigs thing, and the president just feels that,” ah, without goinginto the details . . . don’t, don’t lie to them to the extent to say there is no involvement, but justsay this is sort of a comedy of errors, bizarre, without getting into it, “the president believes thatit is going to open the whole Bay of Pigs thing up again, and ah because these people are pluggingfor, for keeps and that they should call the FBI in and say that we wish for the country, don’t gofurther into this case.” Period. That’s the way to put it, do it straight.’

President Nixon, all of which they claim are appropriate and report on what thespeaker said. Here are just three of them.

(163) a. Nixon told Haldeman to tell the CIA to tell the FBI not to pursue theirinvestigation into the Watergate burglary.

b. Nixon wanted the CIA Director, Richard Helms, to thwart the FBI’s probeof the Watergate burglary by saying it was a CIA operation.

c. Nixon told Haldeman to break the law.

These reports exemplify Cappelen and Lepore’s wide understanding of what is said.In fact, we already noted above that possible indirect reports of a speaker’s utterancemay pick up on what is said by a speaker but also on what was meant. Cappelen andLepore do not distinguish these aspects within speech act content, since they assumethat there is no possible way of doing so. From the discussions so far and withindifferent frameworks, it is clear that it is indeed difficult to differentiate between thetwo levels. Nevertheless, I find the argument they give to defend their particularview of indirect reports stating what a speaker said not entirely convincing.

Here are a couple of objections we have encountered . . .. Did Nixon reallysay what we attribute to him? Did he, strictly speaking say it? Did heliterally say it? Let’s see: We know that this particular utterance wasthe cornerstone of the impeachment case against Nixon. [. . .] Imaginethe absurdity of a defense of Nixon that he didn’t strictly speaking, askthe CIA to block the FBI investigation. (Cappelen and Lepore 2005,p. 196-7)

It goes without saying that it would have been absurd to defend Nixon on the linesthat he did not strictly speaking request the CIA to block the FBI investigation,but this does not get rid of the discrepancy there is between what a speaker mayconvey or mean by making a particular utterance and the meaning the utterance assuch can be attributed.

Note also that this is not where the problems with indirect speech reports end.Thus, apart from possibly capturing what is said and what is meant by (the speakerof) an utterance, they may pick up on aspects of an act carried out by a speakerhe himself may not be aware of. Thus, (163c) picks up on the fact that the CIA’sinterfering with an FBI investigation constitutes an illegal act. In this particularcontext, it is reasonable to assume that Nixon was aware of that fact, but still,neither did he tell Haldeman (in the sense of ‘order’, which seems to be intendedhere) to break the law, nor is it clear whether it is appropriate in this scenario toasumme that (163c) constitutes part of what Nixon meant by his utterance. Anotherexample may make this point more obvious: a German hiker meets a British hiker ontour through Scotland. Towards evening they get to a bigger town, go to a grocerystore to get some food, etc. and then, since the weather is nice, decide to have theirdinner on a bench in a picturesque public place with nice views. While they unpacktheir proviant, the German says (164) to the Briton, handing him a can of Germanbeer.

(164) Here. Have a beer.

What the German does not know is that in many public places in Britain consumingalcohol is against the law. In this context, would it be appropriate to report theGerman has having invited or urged the Briton to break the law? Probably not.However, one might object that this scenario does not compare to the Nixon scenariosince we can assume of Nixon that he did know that if Haldeman did what Nixontold him to, he would be breaking the law. But even if we change the context in ourexample such that the German actually knows about this particular law, I wouldcontent it still is unappropriate to report what he meant, let alone ‘said’, as ‘Heinvited/urged the Briton to break the law.’ Of course, if the Briton excepts the beerand drinks it, it follows that he does break the law, due to the logical relation holdingbetween carrying out this particular action in Britain (i.e., consuming alcohol in apublic place) and breaking the law in Britain. However, it is questionable, whetherthis is part of what the speaker intended with his utterance.56

Generally, the problem with indirect speech reports can be traced back to the ap-parent vagueness involved in the expression say. Thus, recall once again the resultsof Nicolle and Clark (1999)’s first experiment, which had three conditions, askingsubjects to choose a paraphrase of the speaker’s utterance that best reflected eitherwhat the sentence meant the speaker used, what the words meant the speaker usedin his utterance, or what the speaker communicated by his utterance. The resultsacross the different conditions did not significantly differ, suggesting that subjectsdid not differentiate between utterance level meaning and speech act content. How-ever, these results do not show that in interpreting natural language utterances,people do not build up a level of context-sensitive meaning distinct from speech actcontent. Rather, what they show is that the expression say is vague as it can beused to refer to both these levels. Moreover, as the use of say in some of the indirectspeech reports cited above shows, it can even be used to pick up on an entailment ofwhat the speaker actually said/meant, although this entailment may not actually bewhat the speaker intended to convey or what was foremost in his mind when makingthe reported utterance. In fact, the original speaker of some reported utterance maynot even have been aware of the reported entailment (cf. Gross 2006).

Thus, indirect speech acts neither exclusively capture the technical what is saidby an utterance, nor do they simply capture what the speaker intended to con-vey or meant by his utterance. However, that fact as such is no evidence that nodifferentiation between the levels of what is said and what is meant is possible ornecessary.

4.2.2 Primary vs. Secondary Pragmatic Processes

Recanati (2004) assumes that a differentiation of types of pragmatic processes ispossible and that the distinction allows to differentiate between the levels of whatis said/utterance meaning and what is meant/communicative sense. Thus, he dif-ferentiates between so-called primary and secondary pragmatic processes. Primarypragmatic processes are characterised by the fact that they take place at a sub-personal level57 and determine what is said by an utterance. Such processes take

56Cp. Gross (2006) for similar worries.57The terms sub-personal and personal level are taken from Dennett (1969).

place unconsciously; we are not aware of them and thus cannot reflect upon them.They comprise such processes as the resolving of reference, fixing indexicals, dis-ambiguation and free enrichment (and possibly other pragmatic processes). Theimportant characteristic primary pragmatic processes have in common is that theydo not require a fully propositional form to apply to. Rather, they (may) applyto the sub-propositional output of the linguistic process of semantic composition,which yields the semantic form of a sentence. The final and conscious output ofthese unconsciously proceeding processes then, is what is said, which is of a fullypropositional form and equatable with the truth-evaluable content of an utterance.Note, however, that Recanati’s notion of what is said differs from Grice’s traditionalcharacterisation in that he assumes processes like free enrichment to contribute towhat is said. Moreover, the primary processes are such that they do not take intoaccount speaker intentions.

In contrast, secondary pragmatic processes take place on a personal level; theyare accessible by our consciousness and we can reflect upon them. Thus, Recanaticlaims, normal interpreters are usually both aware of what has been said and ofwhat has been implied. However, while they are consciously able to recapitulatehow they arrived at what was implied by some utterance on the basis of what hasbeen said, they cannot ‘calculate’ how they arrived at what was said (or the utterancemeaning) on the basis of the surface sentence used to make that utterance. Thus,they are not aware of the primary pragmatic processes applying to the semantic formof the sentence uttered; it is only the final result of the application of those processesthat is consciously available to them. Moreover, they are able to recapitulate theinferential connection between what is said and what is implied.

The common characteristic of secondary pragmatic processes is that they doneed a fully propositional form to apply to. That is because from the fact that p,they proceed to inferentially derive further propositions (implicatures). Moreover,secondary processes do take speaker intentions into account. Note that Recanati’scharacterisation of primary and secondary pragmatic processes as being derived onthe basis of sub-propositional and fully propositional forms, respectively, impliesthat there is a temporal ordering to their application – hence their names – primaryprocesses take place first, whereas secondary processes take place afterwards.

Recanati’s view is similar to that held by Ruhl (1989), who differentiates betweenconscious and unconscious cognitive processes. As the names suggest, while theformer are consciously available to us, the latter are not. Thus, Ruhl for examplesuggests that the reason why we assume of a particular expression to have a certainliteral meaning is that we are not conscious of the fact that there are processes in ourminds, which abstract over uses and thus create default interpretations. However,an expression might be used with a ‘non-default’ (non-salient) but literal readingnonetheless.

With respect to Recanati’s differentiation between primary and secondary prag-matic processes, it is important to recall that he rejects what he calls the ‘minimalist’view (following the Gricean view) of what is said, where what is said is the minimalproposition derivable form the semantic form of an utterance. Thus, as already men-tioned, he argues that the minimalist what is said is not even consciously availableto a normal hearer-speaker. Thus, for B’s utterance in (165)

Recanati predicts that a normal hearer-speaker if asked to state what B actuallysaid with his utterance, would come up with what Carston called the explicature ofthe utterance, namely ‘The speaker had a very large breakfast today’. Thus, theassumed minimal proposition ‘The speaker had a very large breakfast prior to thetime of utterance’ is not available to the hearer, as it is the result of sub-consciouspragmatic processes, enriching the semantic form, but not the final result of allsuch processes. In fact, it is questionable whether this is a level of meaning thatinterpreters derive at all during interpretation. Recanati’s notion of ‘what is said’is that it is the actual conscious output of the application of primary pragmaticprocesses to the semantic form of an utterance and the only consciously available‘proposition expressed’ by the utterance of a particular sentence58.

sentence meaning

what is said

Pragmatics

primary pragmaticprocesses

what is meant

secondarypragmaticprocesses

sub-propositional

fully propositional

+ narrow context

+ broad context

Figure 4.2: Recanati’s view

We already saw that this view is supported by the results of experiments carriedout by Gibbs and Moise (1997). Thus, the results show that apparently subjects didnot have anything like a minimal proposition available, rather, when asked what aspeaker actually said with his utterance, they would choose the enriched version ofthe traditional, Gricean what is said, namely one corresponding to RT’s explicature.Note that, in the last experiment in Gibbs and Moise (1997), which we have notmentioned yet, it was tested whether in a relevant context, subjects would actuallychoose a rather minimal proposition as being ‘said’ by a speaker. The results showedthat they did, however, in this case what they chose still was what was said by therespective utterance in Recanati’s enriched sense. It just happened to coincide withwhat minimalists would refer to as what is said, that is, the minimal proposition, dueto the very specific contexts of utterance. Recall, however, that Gibbs and Moise(1997)’s results were criticised on the grounds that they presupposed of the subjectsthat they make use of the differentiation between what is said in the enriched sense

58This is why Recanati’s approach is called the availability based approach. See above for theformulation of the availability principle.

and what is implicated, when, in fact, it might be that they are guided by sometotally different criterion. Thus, Nicolle and Clark (1999)’s study was geared to showthat naive speakers do not normally differentiate between an utterance’s explicatureand its communicative sense in the technical understanding. Rather, what seemsto matter to them, albeit not consciously so, is the way in which an utteranceachieves relevance in a particular context. Thus, if what is predominantly relevantin a given context is a communicated implicature, naive interpreters might pick theimplicature as what has been said by the speaker. That is, they do not share thetechnical understanding of such notions as ‘say’ and ‘communicate’ presupposed bytheorists.

Thus, while Recanati’s differentiation between primary pragmatic processes andsecondary pragmatic processes was supposed to capture the different levels of mean-ing explicature and what is meant, respectively, it seems that such a differentiationactually is not possible, at least if speakers’ intuitions are assumed to be a guidewhen characterising these different levels of meaning. However, it is quite possiblethat, in a particular context of utterance, the naive interpreters are not aware of aparticular proposition on which they consciously base further inferences they draw.Moreover, even though speakers may not be consciously aware of a level of meaningthat corresponds to the minimal proposition of an utterance that does not meanthat this level may not play a role during interpretation after all.

Note also that Recanati’s characterisation of primary and secondary pragmaticprocesses actually cannot be applied to differentiate between explicatures and impli-catures (cf. e.g. Carston 2002a). This is because RT claims that implicatures neednot be drawn on the basis of a full proposition. Rather, the fact that RT assumesmutual parallel adjustment of explicatures and implicatures rules out the claim thatprocesses leading to the latter type of pragmatic inference need a full propositionto apply to. Furthermore, RT assumes that speaker intentions already play a rolefor the determination of an utterance’s explicature (i.e., the proposition the speakerintended to express). Also, due to mutual parallel adjustment, there cannot beany inherent temporal ordering to the application of pragmatic processes leadingto explicatures or implicatures. Moreover, RT does not assume that the processesretrieving implicatures from what the speaker said necessarily are part of a personallevel, which is characterised by allowing for conscious reflection. The assumption isthat there is one pragmatic system that has the function of recovering the speaker’sintention in making a particular utterance, however, this system is taken to be a sub-personal system, not necessarily requiring conscious reflection on what the speaker’sintention(s) might have been.59

Carston (2002a, p. 145) gives the following example to make clear the differencein assumptions about its interpretation according to Recanati’s view in contrast toRT.

(166) A: Do you want something?

B: Yes. I’ve run out of paper.

B answers A’s question in the affirmative. Thus, it is likely that A, having

59For a thorough discussion of Recanati’s differentiation of primary and secondary pragmaticprocesses see Carston (2007).

processed this first part of B’s reply, at this point forms the ‘. . . anticipatory as-sumption schema [B wants ]’ Carston (2002a, p. 145). The second part of B’sreply, then, provides a completion for that schema – ‘B wants some paper’. Now,according to RT this completed proposition happens to be an implicature of B’ssecond utterance, as it is not the result of a ‘development’ of the semantic form ofB’s second utterance. Note, however, that if B’s second utterance had been ‘Somepaper’ or ‘I want some paper’, then A’s completed assumption schema would ratherhave counted as an explicature of that utterance. This fact is unproblematic for anapproach to verbal interpretation as RT, since it assumes that there is only a singlepragmatic system responsible for the derivation of both explicatures as well as impli-catures. However, according to Recanati’s differentiation between different types ofpragmatic processes, the completed assumption schema [B wants some paper] wouldhave to be viewed as the result of one distinct type of pragmatic process in the casewhere it turns out to be an explicature, and of another distinct type of pragmaticprocess in the cases where it is an implicature of B’s utterance.

‘According to Recanati’s view . . ., the understood answer, that is thecompleted schema, ‘B wants some paper’, must be achieved by funda-mentally different kinds and levels of processes in the two cases . . ..However, this is not supported by intuitions about the two possibilities,and, in the absence of compelling arguments, it is difficult to see why weshould adopt this view.’ (Carston 2002a, p. 145)

A similar critique is provided by Borg (2006)’s review of Recanati (2004). Shepoints out that Recanati’s category of primary pragmatic processes also includessuch processes as free enrichment, which are characterised by the fact that they areoptional and actually do apply to full propositions. She argues that this fact blursthe usefulness of a distinction between primary and secondary pragmatic processes.Thus, consider the following example. Jack asks Jill whether she wants to go andhave something to eat and she replies:

(167) I’ve had breakfast.

Amongst other things, according to Recanati, it is a primary pragmatic processthat constrains the timescale within which Jack will interpret Jill to have had break-fast. The point here is that the process that constrains the timescale might deliverdifferent results. It could lead to Jack interpreting as what is said by Jill that ‘Jillhas had breakfast today’. If that is the case, then a further implicature, derivedvia a secondary pragmatic process will be necessary to provide Jack with a relevantanswer to his question: ‘Jill does not want to go and eat’. However, nothing speaksagainst assuming that what the process constraining the timescale results in leadsto the following what is said by Jill: ‘Jill has had breakfast recently enough to makeeating currently undesirable’, which would be relevant enough to provide an answerto Jack’s question. Thus, in this latter case, no further inferences would have to bederived.

‘Yet, if both interpretations are possible, then it looks as if somethingwhich counts as an implicature to one hearer might count simply as what

is said to another, in which case we don’t seem to have anything like adeterminate notion of the literal meaning of an utterance.’ (Borg 2006,p. 3)60

As we saw above, Recanati’s characterisation of primary pragmatic processes asbeing sub-personal and secondary pragmatic processes as being personal is not helpfulhere either, as speakers, contrary to Recanati’s assumption do not seem to haveconcrete and static intuitions about the difference between the theoretical notionsof what is said and what is implicated. That is, the experiments carried out on thedistinction did not verify Recanati’s availability principle – speakers in interpretingan utterance do not necessarily and consciously differentiate between what a speakersaid and what he merely implicated.61

4.2.3 What is Said/What is Meant and Distinct KnowledgeSystems

Recall that Bierwisch differentiates between three distinct levels of meaning, ex-pression meaning, utterance meaning and communicative sense. This differentiationreflects his view that there are three distinct knowledge systems involved in pro-ducing what he calls ‘communicative linguistic behaviour’, namely, knowledge oflanguage, everyday knowledge and knowledge about social interaction. That is, thebasic meaning SEM (E) of an expression E, be it simple or complex, is determinedby the language system alone. The meaning of an utterance exemplar t of E, M (t)is determined by what is contributed to it by SEM (E) on the one hand, and thecontext of utterance C, which in turn is determined by our everyday knowledge, onthe other. Both context and utterance meaning are mental representations that aredetermined by the conceptual system. The third level of meaning, the communica-tive sense CS (t) of an utterance exemplar t of E is constituted by the contextuallydetermined meaning M (t) of that utterance exemplar, plus knowledge about socialinteraction in general which determines the specific interactional situation IAS inwhich the utterance takes place.

As mentioned already, in contrast to Grice, Bierwisch does not assume that ut-terance meaning necessarily forms part of communicative sense. Generally, he claimsthat the differentiation between the two levels is required since it is possible for ahearer to interpret correctly what the speaker meant by what he said, without know-ing what was actually said. Bierwisch cites the example of the American prisonerof war in Italy, who recites the following line of a German poem with the intentionof showing that he is German:

(168) Kennst Du das Land wo die Zitronen bluhen.

‘Do you know the country where the lemons bloom.’

It can be argued that in this case, if neither speaker nor hearer are speakers ofGerman, they might not understand the meaning of the utterance of (168), but the

60Where the phrase literal meaning here should be interpreted as the utterance meaning or whatis said by an utterance.

61See also Garcia-Carpintero (2006) for another critique of Recanati’s availability principle.

does notdependenton

Semantics (simple) expression meaning

(complex) expression meaning

utterance meaning

Pragmatics

reference resolutionfixing indexicalsdisambiguation. . .

communicative sense

ConceptualSystem

sub-propositional

fully propositional

+ narrow context

+ broad context

underspecified

Figure 4.3: Bierwisch’s view on the relation of utterance meaning and communicativesense

hearer of (168) might nevertheless discover what the speaker’s intention might havebeen in uttering it. Note that the reverse can be the case as well, namely thatit is possible for a hearer to understand an utterance without necessarily graspingwhat the speaker of that utterance meant by it. Thus, imagine a situation whereA is helping B to prepare for an exam. They have both learned the subject mattertogether and additionally, A has asked B exam-like questions for B to check how farhe has progressed in learning. As A takes leave of B he utters the sentence in (169).

(169) I will come tomorrow evening.

Irrespective of B’s understanding of what A has said, he might nevertheless beunsure whether in uttering (169) A wanted to promise that he will learn togetherwith B tomorrow, or whether he threatened that he will get angry if by tomorrowevening B has not advanced in memorising the subject matter.

It should be noted, however, that although an argument can be made that inprinciple the interpretation of the two levels of meaning what is said and what ismeant is independent of one another, more often than not, determining the level ofwhat is meant does depend on knowing the proposition expressed by the utterancemade or is at least influenced by what the hearer assumes was expressed by theutterance made. In fact, this can be taken to be the normal communication situation.

Bierwisch criticises further that aspects of meaning at the level of communicativesense are usually viewed as directly related to linguistically determined meaning. Forinstance, speech act theory is taken to be ‘. . . an extension of the theory of meaningin natural language’ (Wunderlich 1977, p. 243). However, as Bierwisch points out,this view obscures the fact that there is a basic distinction to be made betweenlanguage and communication. Thus, Bierwisch shows that in order to explain sucha phenomenon as performative utterances, it is not sufficient to assume that theyhave to be analysed in terms of felicity conditions and that the notion of truth-conditions is not applicable here. As he puts it: ‘. . . under certain conditions asentence having an explicit performative formula as its topmost clause specifies a

truth condition that can be satisfied by the speech acts in which the sentence isused.’ (Bierwisch 1980, p. 12).

(170) I promise to come.

The truth condition specified by (170) can be paraphrased very loosely as ‘Thespeaker promises to come’. This condition is fulfilled only if the speaker actuallyperforms the speech act of promising in using the sentence in (170), that is, if theinteractional setting of the utterance of (170) is such that it qualifies as the settingnecessary for the performance of an act of promising. That is, whether an utteranceof an explicit performative such as (170) constitutes the speech act of promising,depends on the interactional setting in which it is uttered. This view also explainswhy it is possible to use a sentence that includes an explicit performative formulawithout actually performing a speech act.

(171) A: Why are all my parties such a success?

a. B: I promise to come.

In this exchange, the interactional setting for B’s utterance can be assumed to besuch that it does not qualify as a setting necessary for the felicitous performance ofan act of promising. What is crucial for examples such as the above is that thereis a difference to be made between meaning as it is determined by linguistic andcontextual factors and the interpretation of a linguistic utterance as a specific typeof social interaction. Therefore, ‘Speech act theory is a branch of the theory ofcommunication, viz. that involving linguistic utterances, rather than a part of thetheory of language.’ (Bierwisch 1980, p. 3).

Turning back to the idea that speaker intentions play a particular role at thelevel of communicative sense, it should be noted that in figuring out why a speakermade a particular utterance, hearers make use of information resulting from theirconceptualisation and evaluation of the social relationship between themselves andthe speaker. Thus, although determining a speaker’s intention in making a certainutterance involves making assumptions about what the speaker had in mind, theassumptions made also crucially depend on how the hearer, possibly unconsciously,judges his relationship to the speaker.

This idea is captured in Schulz von Thun (1982)’s communication-psychologicalapproach by assuming that there actually are four sides to a message (utterance)that have to be taken into account: the level of content (‘Sachebene’), the aspectof relation (‘Beziehungsebene’), the aspect of self-revelation (‘Selbstkundgabe’) andthe aspect of appeal (‘Appellseite’). Thus, consider the example in (172) (adaptedfrom Schulz von Thun 1982) of how meaning is determined by the social settingin which communication takes place. The context for the small verbal exchange in(172) is the following: A man (A) and a woman (B) are sitting in a car, which thewoman is driving.

(172) A: Du, da vorne ist grun.

‘Listen, it’s green ahead.’

B: Fahrst Du oder fahre ich?!

‘Are you driving or am I?!’

The four sides to A’s utterance, as B interprets them, might be (crudely) para-phrased as follows:

(173) Level of content: the traffic lights are greenAspect of relation: You need my help.Aspect of self-revelation: I’m in a hurry.Aspect of appeal: Drive faster!

Schulz von Thun (1982) points out that it is the interpreter’s free choice to reactto any of the four sides of the message. In the example in (172), the side that B isapparently verbally reacting to is the aspect of relation. However, in the situationstated, she might also react to the aspect of appeal of A’s message non-verballyat the same time by actually driving faster. Her verbal reaction and the way it isshaped might only turn against the aspect of relation that in her opinion is conveyedby A’s message. In the case where A actually did not intend his utterance to beinterpreted on the aspect-of-relation side as ‘You need my help’, the question is ifwe would like to say that the communication between A and B was unsuccessful,although B actually did react as intended to one side of the message by driving faster.That is, assuming that successful interpretation of a speaker’s meaning involvesdetermining the speaker’s intention in making an utterance, does not capture thecomplexity of such a task, as we have to assume that there are various sides to verbalinteraction, highlighting different aspects of the overall intention of the speaker.More specifically, we might want to say that in making an utterance, the speakerdoes not necessarily have only one intention, but rather several, possibly on differentlevels and with different prominence.

Be that as it may, the important point to note here, once again, is that the typeof information apparently necessary for determining what a speaker might haveintended with his utterance seems to be distinctly different from the knowledge nec-cessary to determine what the speaker’s utterance as such means. This hypothesisis also corroborated by findings concerning autistic childrens’ understanding of thespeaker’s intention in making a particular utterance in contrast to their understand-ing of what the utterance as such means. Thus, MacKay and Shaw (2004) testedautistic children’s understanding of various types of figurative language as well asthe likely intentions with which they were employed by the speakers. Althoughthe autistic children did understand what the utterances meant more or less well,they had considerably more problems in determining why a speaker should use aparticular type of figurative language in the first place.

In this connection, recall RT’s assumption that the determination of explicaturesand implicatures actually is subject to mutual parallel adjustment, where assump-tions about what the speaker meant by his utterance seem to influence how thatutterance is interpreted in the first place. Although viewing the evolving of the in-terpretation process in these terms is not unproblematic – as it is not clear on whichbasis, then, to differentiate explicatures from implicatures and why to assume a richerlevel of utterance meaning than originally proposed by Grice in the first place – whatit does capture is the fact that people do seem to understand utterances differentlyif they have a grasp on what the speaker might possibly have intended in makingthat utterance, than when they do not have a definite grasp on those intentions. For

instance, autistic people tend to interpret language literally and a possible reasonfor that might be that they do not understand why a speaker said what he said. Inother words, it might be that autistic people have problems considering a speaker’spossible intentions and the fact that the speaker actually persued a particular goalin making a particular utterance.

However, while there may be differences between autistic and ‘normal’ people indetermining what is meant by a speaker of an utterance, this fact alone does nottell us whether there really is a difference in what the two groups take to be theutterance meaning of that utterance. As we saw, one cannot rely on intuitions whenit comes to spelling out what the speaker said vs. what he meant by his utterance.In MacKay and Shaw (2004)’s study, subjects were asked both for aspects of thebackground against which a particular utterance took place, as well as the reasonthe speaker might have had to make this particular utterance. For example, in thehyperbole condition, subjects where asked what they thought how many CDs thespeaker of ‘I’ve got thousands of CDs’ actually has. In other words, they were askedto infer from the utterance the situation it was supposed to describe. Since thesubjects did not take into account the possibility that the speaker might want toboast in making such an utterance, most of them assumed that the speaker doeshave a very large number of CDs. This is different for the control group of normallydeveloped children, who judge the situation differently.62 However, that does nottell us how the two groups in fact interpreted the utterance as such. It might bethat that interpretation does not differ, but that the difference in judgements simplyis due to how these children differ in their evaluation of the situation in which suchan utterance is made and which purpose it has.

Having said this, I do not want to deny that hearers come to the interpretationtask with more or less specific assumptions about what the speaker might want toconvey in a particular discourse situation. It is just that I am reluctant to call theseassumptions implicatures, since it seems to me that these kinds of inferences aredifferent in kind from what is usually understood by the term implicature. Thus,in contrast to ‘real’ PCIs, such assumptions are not (necessarily) intended by thespeaker to be inferred by the hearer and they are not (necessarily) the product ofan inferencing process that is based on the meaning of a prior utterance on the partof the speaker. In fact, it seems to me that generally what a hearer’s assumptionsin a particular utterance situation influence is the final interpretation of what hetakes the speaker to have meant by his utterance (cf. the differences in judgementof ’normal’ and autistic children above).

Recall once again Schulz von Thun (1982)’s idea that a message has several sidesto it and the interpretation of one of them is based on the assumptions a hearer hasas to the speaker’s opinion of him. For example, if B thinks that A thinks B is alooser, then it is possible that in a situation in which A, B and several others areplanning a cycling tour they want to go on together and A says (174) to B aboutthe potential tour they are just discussing, B might understand him to have meant

62Thus, of the answers given by children in the control group, only two involved numbers higherthan 50 (namely 70 and ‘hundreds’). In contrast, in the subject group, eight children gave numericalresponses over 50. In fact, their answers involved such unlikely quantities as ‘million’, ‘zillion’ andeven ‘infinity’.

something like ‘You won’t be able to make it’ or ‘This is too difficult for you’.

(174) This is a very difficult tour.

In other words, particular assumptions hearers bring to the interpretation processmay influence how they understand why a speaker said what he said, but it is notgoing to change their understanding of what it in fact was the speaker said. Thus,if in this particular situation what A wanted to convey – apart from the fact thathe considers the contextually relevant tour as difficult – was in fact the idea thathe thinks doing this tour will involve very careful planning (and that he specificallyaddressed himself to B because he actually thinks B is good at organising things),one might argue that communication between A and B was only partly successfulin that B did understand what A said but was wrong about his reasons for sayingwhat he did.

Thus, as we saw above, assumptions the interlocutors have concerning otherinterlocutors’ opinions about one another may influence the interpretation of whata speaker meant by his utterance. Moreover, assumptions concerning the socialsetting in which a particular utterance takes place influence what the speaker willbe taken to have meant with his utterance. Thus, Capone (2004) mentions theexample of a teacher using (175) in addressing his favourite pupil Michelangelo whois prompting a classmate.

(175) I saw you.

Capone (2004) says of this example that even though the teacher might actuallyadmire the fact that Michelangelo tries to help his fellow pupil, in the particularsocial setting in which his utterance occurs, it is clear that part of what the teachermeans in saying (175) is that he wants Michelangelo to stop prompting.63

Note that the assumptions the hearer forms in a particular discourse situationare also crucial in explaining irony. That is, on a first blush, in cases of irony, theutterance made by a speaker does not seem to fit with the situation in which itis made. This is not to say that the hearer has clear-cut expectations of what aspeaker is going to say, however, whatever the speaker does in fact say, the hearerwill try to integrate with his conception of the discourse situation. This can be seenespecially well in such cases of irony in which what the speaker said might actuallyalso be taken to be (part of what is) meant by her, but where the meaning expressedby her utterance nevertheless in some way contrasts with the situation it is used in.

(176) I love children who keep their rooms clean.[Said by a mother having just looked into the messy rooms of her kids.]

Thus, the speaker of (176) can be taken to use the sentence she does descriptively,but at the same time her utterance in the situation it occurs in will be interpreted as

63In fact, Capone (2004, p. 7) goes further in that he assumes that the utterance as such is goingto be understood differently from its use in another context: ‘. . . this example nicely instantiatesthe view that the context is the total social setting in which the speech event takes place, themeaning of an utterance being determined by its place in an interactional sequence.’ It should beclear that I disagree.

ironic, since she does not voice her negative attitude to the messiness of the roomsdirectly.

In sum, the important point to note is that interpreting a particular utteranceas ironic, as used to carrying out a specific speech act or as giving rise to particularPCIs involves contextual information beyond that needed to determine the utterancemeaning as such. Moreover, in the majority of utterance interpretation situations,the proposition taken to be expressed by the utterance of a speaker is involved in de-termining what the speaker meant in making that utterance. However, assumptionsabout the intentions behind or the reasons for making the utterance the speaker did,do not influence a hearer’s interpretation of the meaning of an utterance as such, itonly influences her interpretation of what she takes the speaker to have meant.

4.3 Summary

One of the aims of the present chapter was to review the phenomena traditionallyassumed to arise at the level of what is meant/communicative sense to see whetherfor their characterisation it is necessary to assume that they are based on a fullypropositional utterance meaning. As we saw, for metaphor at least, this does notseem to be the case. Rather, arguments were given for assuming that metaphor actu-ally belongs to the level of utterance meaning. Thus, on the one hand, metaphoricalmeaning usually is associated not with entire utterances, rather, it seems that it issimple expressions (or at least expressions below sentence level) that may receivea metaphorical interpretation. On the other hand, it was argued that metaphorcan be treated along the same lines as metonymy (e.g., Dolling 2001) and othertypes of figurative language use assumed to arise at the level of utterance meaning(e.g., Sperber and Wilson 2008). Moreover, it seems that metaphor interpretationproceeds independently of the intentions the speaker had in making the utterancethat includes the metaphor (e.g., Borg 2001). In addition, metaphor seems to differin some important aspects from irony, which type of figurative language use wasacknowledged as belonging to the level of what is meant/communicative sense.

Thus, in contrast to metaphor, the interpretation of an utterance as ironic doesseem to require a prior determination of a kind of ‘basic’ utterance meaning (e.g.Giora et al. 2007). Furthermore, interpreting an ironic utterance is cognitively moreexacting than interpreting an utterance involving metaphor (cf. Happe 1993, Colstonand Gibbs 2002, MacKay and Shaw 2004). What is similar in non-familiar metaphorand irony interpretation is that the ‘basic’ meaning the respective figurative meaningis based on stays activated to some degree and for some time during interpretation.

What differentiates both metaphor and irony from conversational implicatureis that in cases of the former the meaning understood differs from some ‘basic’meaning, whereas in cases of the latter the inferred meaning aspects are addedto what is understood. Thus, Grice assumed that conversational implicatures arebased on some ‘basic’ utterance meaning and added to it. Generally, CIs originallywere assumed to be both speaker intended and cancellable. However, as we saw,not all CIs exhibit these two features to the same degree. Thus, whereas PCIsonly arise in (broad) contexts in which they are clearly speaker intended, GCIs areinitially computed regardless of whether they are speaker intended or not. As a

Summary 201

consequence of this (in)dependence on speaker intentions, PCIs actually are notcancellable, whereas GCIs are. What is special about GCIs is that if they are notcancelled, they consequently are assumed to have been intended by the speaker afterall. Thus, what makes CIs and irony similar and differentiates them from metaphoris that both are based on some full utterance meaning and both are (ultimately)speaker intended.64

Another phenomenon undisputably taken to arise at the level of what is meantare speech acts. Thus, determining the speech act intended by a speaker in making aparticular utterance allows the hearer to interpret how what the speaker uttered is tobe taken. Of particular interest for our purposes are indirect speech acts, since theyare traditionally characterised as involving a basic, literal illocutionary force anda derived, non-literal one. Arguments were given for not treating indirect speechacts in terms of conversational implicatures, since their determination – contraryto what was traditionally assumed – does not seem to necessarily involve a priordetermination of a potential but non-fitting direct speech act.

The second aim of this chapter was to find an answer to the question of whattypes of information have to be available to the interpretation process to determinephenomena taken to belong to the level of what is said and such taken to belong tothe level of what is meant. To this end, we looked at some of the arguments givenfor and against assuming two distinct levels of context-sensitive meaning. That is,even though particular phenomena were characterised as being based on some levelof what is said/utterance meaning, there are arguments that there actually are noclear criteria which allow us to differentiate between meaning aspects that are part ofwhat is said/utterance meaning and such as are part of what is meant/communicativesense. Thus, Borg (2004b) as well as Cappelen and Lepore (2005) argue against theaccuracy of indirect speech reports as a tool to capture what is said or the utterancemeaning of the reported utterance. However, while their reservations are absolutelywarranted, the fact that indirect speech reports cannot capture the technical notionof what is said by an utterance is not sufficient to show that a differentiation betweenwhat is said and what is meant is not in principle possible. Incidentally, the samepoint holds true concerning the results of experiments aimed at testing whether peo-ple differentiate between the respective two levels of meaning. As already mentioned,what the experiments in question showed is that naive speaker-hearers understandthe expression say differently from its technical characterisation. Moreover, theyshow that people do not necessarily consciously reflect on the supposed difference.However, again, these facts are not by themselves sufficient to show that there ac-tually are no two levels of context-sensitive meaning that can be differentiated fromone another.

Recanati (2004) proposes a differentiation between two general types of prag-matic processes, which is supposed to capture the difference between the two levelsof what is said and what is meant. Whereas the sub-personal and thus, unconsciouslyoperating primary pragmatic processes result in what is said by an utterance, thepersonal and thus consciously operating secondary pragmatic processes lead to whatis meant by an utterance. Recanati (2004)’s differentiation of pragmatic processes in

64Note, however, that the discussion of GCIs will be taken up once more in chapter 5, leadingto a somewhat different characterisation.

terms of whether they are consciously available predicts that naive speaker-hearersshould be aware of the difference between what is said and what is meant (e.g., hisavailability principle). However, the results of experiments aimed at verifying thisprediction show that it is inaccurate. Thus, a differentation of what is said fromwhat is meant cannot rely on speaker intuitions.

Bierwisch (1980) also assumes that a differentiation between what is said andwhat is meant is necessary and he makes reference to different knowledge systemsthat seem to be involved in determining the different levels of meaning. Thus, whilethe processes leading to utterance meaning make use of information provided bygeneral background knowledge, the processes leading to communicative sense relyon information drawn from knowledge about social interaction. Generally, the twoknowledge systems are independent of one another and so are, strictly speaking,the two meaning levels for the determination of which they are employed. How-ever, more often than not the utterance meaning of what some speaker said plays adistinctive role in figuring out what the speaker meant in saying what he did, i.e.,the communicative sense of his utterance. Nevertheless, there still is a differencebetween what an utterance means and what a speaker means or intends in making aparticular utterance and experimental studies with autistic subjects corroborate theview that it is quite possible to understand the former without grasping the latter(cf. MacKay and Shaw 2004).

Thus, generally, and following Bierwisch, the argument can be made that thecontextual information the processes leading to what is said draw on is taken fromgeneral background knowledge, whereas the contextual information the processesleading to what is meant draw on is taken from a knowledge base about social inter-action. However, this again raises the question as to the role of speaker intentions ininterpretation, as reasoning about these is usually taken to draw on the latter typeof knowledge system, i.e., that on social interaction. This issue will be taken up inchapter 5 (section 5.2).

Chapter 5

Varieties of Meaning, Context andthe Semantics/PragmaticsDistinction

Recall that this thesis started out from the observation that although the termsliteral meaning and non-literal meaning are extensively used in the semantics andpragmatics literature, usually there is no clear indication of what kinds of mean-ing aspects these two terms are actually assumed to pick out. Moreover, althoughthere seem to be some sort of standard characterisations for the two terms, theyare not used consistently with those characterisations and – what is worse – thosecharacterisations can be shown to be inappropriate for the description of the phe-nomena the two terms are intuitively taken to pick out. Since, traditionally, thetwo terms were used to differentiate semantics and pragmatics from one another,this is a very unsatisfactory state of affairs. Thus, in this chapter, I will turn to al-ternative characterisations of literal meaning and non-literal meaning, before finallyformulating my own (section 5.1). However, while my alternative characterisation ofliteral meaning and non-literal meaning captures the kinds of meaning aspects thesenotions are usually taken to refer to, it does not allow the two terms to figure inthe characterisation of the differentiation between semantics and pragmatics either.Thus, in section 5.2, I will turn back to another notion traditionally used in thatdifferentiation, namely that of context-(in)dependence. While in section (5.2.1), Iwill offer a proposal concerning the particular type of contextual information madeuse of by the process of free enrichment, in section (5.2.2), I will defend a view ofthe semantics/pragmatics distinction that actually does not make crucial referenceto the notion of context-(in)dependence, but rather to the nature of the processesintuitively taken to belong to the individual systems.

5.1 Towards an Alternative Characterisation of

(Non-)Literal Meaning

When thinking about literal (vs. non-literal meaning) in natural language, it isinteresting to look at an expression, which, supposedly and intuitively, expresses

204 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction

just that ‘meaning’ as its literal meaning. Thus, Israel (2002) gives a short butrevealing survey of the development of the contribution made by the expressionliterally to the meaning of utterances in which it occurs. Whereas in its earliestusages in English it did refer to the meaning or sense of expressions taken to be,in some sense, ‘basic’ (cf. 177), the conditions of usage for literally where extendedover time.

(177) All those passages are not to be literally understood.

Thus, Israel (2002) cites later examples, where literally seems to be used to indicatethe quality of a speaker’s commitment to his utterance. This applies to examplesas in (178), where literally is used to indicate that the speaker is committed to thetruth of a literal interpretation of his utterance.

(178) Most of these kids cannot think, they literally cannot come in out of the rain.

However, it applies also to examples as in (179), where literally is used to emphasisea fitting choice of words, although the intended interpretation clearly has a derivedcharacter and thus cannot be taken to be literal in the sense of ‘basic’.

(179) [In his music videos he] literally brings words to life; one of his favouritetechniques is to superimpose song lyrics on a background image.

Whereas in examples (178) and (179) literally is used to distinguish different possibleinterpretations of an utterance, there are also uses of literally, where it is used todifferentiate between speech acts a speaker might carry out in making the utterancehe does.

(180) You wouldn’t understand. I don’t mean that as an insult, I mean it literally.

Thus, the expression literally can be used to refer to the meaning or sense of expres-sions in their literal form; it can also be used to indicate the speaker’s commitmentto the truth of a literal interpretation of his utterance and it can be used to pickout, arguably, what is in some sense considered to be the basic speech act amongstthe different possible speech acts which can plausibly be performed by one and thesame utterance. In addition, the expression literally can also be used to indicate theappropriateness of the use of expressions that will receive a figurative interpretationin the respective utterance.

Turning back to the terms literal and non-literal meaning; in chapter 1 I claimedthat these terms are used to refer to types of meaning arising at different levels ofmeaning as well. In fact, in chapter 2 I argued that their traditional characteri-sations are not consistent and that the notions actually do not capture the datathey were intented to. A crucial point made in that chapter was that lexical mean-ing should actually be viewed as underspecified and that if there is some meaningaspect that might usefully be termed literal, this should not be looked for at thelevel of expression meaning. Furthermore, in chapters 3 and 4, a range of meaningaspects were discussed, which do not easily fit into the traditional literal/non-literalmeaning dichotomy, such as the meaning aspects contributed to utterance meaningby processes of ad-hoc concept formation or free enrichment. Such meaning aspects

Towards an Alternative Characterisation of (Non-)Literal Meaning 205

differ from what we traditionally and intuitively understand as non-literal meaning.At the same time, we would not necessarily want to characterise them as literal.A similar point can be made concerning the notion of sentence non-literality iden-tified by Bach. That is, traditionally, we take non-literal meaning to capture suchfigurative language use as found in metaphor or metonymy, where it seems thatthe non-literalness is associated with a particular expression in the utterance made.Generally, the question arises, how these different meaning aspects are related toone another and in particular, how they relate – at least intuitively – to the notionsof literal meaning and non-literal meaning.

Thus, in this section I will first discuss two proposals for alternative characteri-sations of literal meaning and non-literal meaning, one of which specifically tries tocapture the relations to the various meaning aspects identified in chapters 3 and 4(i.e., Recanati’s). However, what both proposals have in common is that they as-sume the notion of literal meaning to be applicable at the level of expression meaningand also that the lexically given meaning of an expression is a full-fledged concept.As such, it is clear that they will not provide the characterisation of literal meaningI am after (as essentially context-dependent), but they are a starting point. In thefinal part of this section I will offer an alternative characterisation of the notions ofliteral and non-literal meaning, based on the data accrued in the preceding chapters.

5.1.1 Literal Meaning and Types of Non-literal Meaning

Recanati (2001, 2004) offers a detailed differentiation of kinds of meaning in order toaccount for differences between semantic or context-free meaning, minimally prag-matically enriched meaning and what we intuitively take to be non-literal meaning.For the characterisation of different types or aspects of meaning, he assumes thatthe pre-theoretic intuitions of naive speakers should be preserved.

Thus, the t-literal meaning1 of an expression is its literal meaning understood asits conventional meaning. The occurrence of an expression inherits the expression’st-literal meaning, but in most cases the expression’s actual meaning also dependson contextual features. In the cases where the departure from the t-literal meaningof an expression is only minimal, Recanati uses the term m-literal meaning. M-literal meaning is not a case of non-literal meaning as traditionally understood,rather, Recanati argues, it is the non-minimal departures from literal meaning (inRecanati’s terms m-nonliteralness) that can constitute non-literal meaning.2 Thus,Recanati’s differentiation between minimal and non-minimal departures from the t-literal meaning of an expression basically captures the difference between saturationprocesses and processes such as free enrichment. Thus, minimal departures fromthe t-literal meaning of an expression must be linguistically licensed, that is, the t-literal meaning of the expression under consideration must provide a variable which,when that expression is actually used, is provided a contextually determined value.Thus, in a sentence that involves indexical expressions, such as (181) below, the

1t stands for (expression) type2In this respect Recanati’s view differs significantly from Borg’s, since the latter considers all

meaning aspects that go beyond the m-literal meaning as non-literal. She does not differentiatebetween non-minimal meanings that are perceived by the hearer as deviating from some underlyingprimary meaning and such that are not.

semantics or t-literal meaning of the indexical can be said to license ‘. . . the searchfor a contextual value’ (Recanati 2004, p. 69).

(181) He is thirsty.

The situation is different in case of the process of free enrichment, which addsmeaning aspects to the minimal proposition without there being any variable in thesemantic form of the expression. Thus, cases of free enrichment constitute cases ofnon-minimal departures from t-literal meaning. However, non-minimal departurefrom the t-literal meaning of an expression, while being a necessary condition fornon-literalness in the ordinary sense, is not a sufficient condition.

Thus, Recanati makes a further distinction namely that between primary andsecondary meaning. Secondary meaning (or in Recanati’s terms p-nonliteral mean-ing) is meaning that is derived from some more basic, primary meaning (e.g., p-literalmeaning). The meaning of an utterance is non-literal in the ordinary sense, if it hasa secondary character, that is, if it is derived from some prior, primary meaning (p-literal meaning). Examples of such non-literal meaning include conversational impli-catures, indirect speech acts and, arguably cases of irony. Recanati defines primarymeaning as not being inferentially derived from some earlier determined meaning.For example, an utterance of (181) can be used to make a statement about someparticular person. For instance, in a situation where the speaker wants to make astatement about some person Paul, he can do so by uttering (181). The propositionexpressed by that sentence in that utterance situation is ‘Paul is thirsty.’, but thatis due to the specific contextual factors holding. It is a possible m-literal meaning ofthe sentence in (181), since the only departure from the sentence’s t-literal meaningis a saturation of the variable provided by the semantics of the indexical he. Thespeaker uttering (181) may, however, intend more than just to make a statementconcerning Paul. He might make the utterance in order to convey that Paul shouldbe offered a drink. The proposition that Paul should be offered a drink is a proposi-tion in its own right and can therefore be said to be independent of the propositionthat Paul is thirsty. However, its interpretation depends on a prior recovery of them-literal meaning of (181). Thus, the m-literal meaning in this case is also a p-literalmeaning as it forms the basis for the interpretation of the secondary or non-literalmeaning that Paul should be offered a drink. Moreover, it is primary because thereis no inference involved in moving from the t-literal to the m-literal meaning, or atleast none the interpreters themselves are aware of. Rather, the m-literal meaningis directly determined.

Recanati goes on to argue that there are meanings that involve non-minimaldepartures from t-literal meaning, yet are considered to be p-literal meanings nev-ertheless. For example we tend to interpret an utterance like (182) in a way wherethe marriage took place prior to the children being born.

(182) They got married and had many children.

However, this temporal ordering is not part of the t-literal meaning of (182). Neitheris it part of its m-literal meaning, since m-literal meaning is defined as departuresfrom t-literal meaning that are licensed by the conventions of language. Thus,although (182) is a case of a non-minimal departure from t-literal meaning, it is

not a case of non-literal meaning in the ordinary sense. It is still considered to havep-literal meaning, since its meaning is not derived from some earlier determinedmeaning.

Recanati makes similar points regarding meaning aspects that are the result offree enrichment. As we saw above, such meaning aspects are considered to be non-minimal departures from the t-literal meaning of the expression concerned, however,they do not constitute non-literal meaning in the traditional sense, since they do nothave a secondary character, that is, they are not determined on the basis of someunderlying meaning that has been determined first. Thus, looking once more at ex-ample (78), the idea is that what is contributed by the process of free enrichment (orexpansion in Bach’s terminology) is not non-literal in the traditional understanding,as it does not have a secondary nature. That is, ‘from this cut’ is not the result ofa reanalysis of some previously determined primary meaning.

b. You’re not going to die [from this cut]

Secondariness, then, is a sufficient criterion for some meaning to be viewed as non-literal in the traditional sense. However, Recanati argues that it is not a necessarycondition. What is characteristic for non-literal meaning as traditionally understoodis that it is perceived by the language user as being special or deviant. This is dueto the fact that non-literal meaning is an aspect of language use, thus, it comesabout during the interpretation of some utterance and therefore is available to theinterpreter. Recanati calls this the transparency condition and this is a definingproperty of any case of non-literal meaning.

‘A use of words counts as non-literal in the ordinary sense only if thereis something special about that use that is, or can be, perceived by thelanguage users themselves.’ (Recanati 2004, p. 75, emphasis in the orig-inal)

Thus, there are other types of meaning that qualify as non-literal althoughthey lack the property of secondariness. For example, in the case of metonymyor metaphor, their non-literal nature differs from the kind of secondariness identi-fied by Recanati for conversational implicature or indirect speech acts. As mentionedabove, conversational implicature are based on the utterance meaning and added toit. In contrast, in instances of metaphor or metonymy, the non-literal meaning is notadded to an earlier derived utterance meaning, but rather, its computation formspart of the derivation of the utterance meaning itself. Thus, an utterance such as(18a) is non-literal although it does not have secondary meaning in the sense definedabove.

(18a) The ham sandwich in the corner wants some more coffee.

Rather, the non-literal meaning comes about due to the deviance of the readingof ham sandwich employed in this particular utterance situation, from the readingsthat are conventionally associated with the expression ham sandwich. However, thisnon-literal reading of ham sandwich is not secondary in that it is not based on some

earlier computed and literal utterance meaning. Thus, it is rather more similar toother cases of what Recanati calls meaning adjustment, e.g., free enrichment.

The difference between metonymy or metaphor and meaning aspects due tofree enrichment is that the latter elaborates the meaning conventionally associatedwith some expression (thus, Recanati’s term sense elaboration), whereas the formerloosens it (i.e., sense extension). Recanati uses the notion of schema to differentiatebetween the two types of processes. Thus, in free enrichment the expression thatis enriched is used in its ordinary meaning, that is, the schema provided by theexpression fits the situation at hand and the process of free enrichment only addsmore specific details. However, in the case of metaphor or metonymy, the schemaprovided by the expression used does not conventionally fit the situation at hand.It therefore has to be adjusted in order to fit. That is, those meaning aspects whichconflict with the aspects of the situation at hand, are filtered out. However, themeaning aspects that have thus been filtered out still stay activated to some extent.It is the number of meaning aspects filtered out and the extent to which they re-main activated which is responsible for whether a hearer is consciously aware of adiscrepancy between conventional word meaning and the reading an expression isused in in a particular situation.

To summarise, in his (2004) book, Recanati distinguishes between different kindsof meaning. Not only is there a difference between t-literal and m-literal meaning,but also a difference between p-literal and p-non-literal meaning. Moreover, notall p-literal meanings need be m-literal meanings. That is, there are some primary,underived meanings that, at the same time, constitute non-minimal departures fromthe t-literal meaning of the expression concerned. Thus, Recanati’s notions of m-literal meaning and p-literal meaning can be used to describe cases of what Bachcalls sentence non-literality more specifically; cases, that is, in which there is anon-minimal departure from t-literal meaning without that departure constitutinggenuine non-literal meaning.

From the examples Recanati gives in explaining this difference, it seems thatthe pairs of terms apply to different levels of meaning. That is, he identifies m-literal meaning as Grice’s level of what is said, thus, a type of meaning holding atthe level of utterance meaning. In contrast, he takes conversational implicature andindirect speech acts to be good examples of what he calls secondary meaning (or p-non-literal). As indicated above, these phenomena are usually taken to occur at thelevel of communicative sense. However, Recanati also identifies cases of non-minimaldepartures from t-literal meaning, which, on the one hand, differ from Grice’s notionof what is said in that they include contextually provided meaning aspects that arenot linguistically licensed, and on the other hand are not secondary in that they arenot based on some earlier determined meaning. This type of meaning he calls p-literal. Interestingly, this type of meaning involves different meaning aspects, namelysuch as result from processes such as free enrichment as well as cases of figurativeuses of language as metonymy and metaphor. Thus, Recanati differentiates betweendifferent types of non-literal meaning, where some are secondary in nature but othersare not.

There are a number of problems with Recanati’s characterisation of the varioustypes of (non-)literal meaning. First, he assumes that t-literal meaning – that is,

t-literal

��

m-literal(saturation processes)

m-non-literal

��

p-literal

��

sense elaboration(enrichment)

sense extension

��

below threshold above threshold(figurative uses)

p-non-literal(indirect speech acts,

CIs, . . .)

Figure 5.1: Non-literal meaning in Recanati’s classification

the linguistically coded meaning – is conventional.

‘. . . start with a sense of the phrase ‘literal meaning’ which is reasonablyclear and raises no particular problem. In that sense, the literal meaningof a linguistic expression is its conventional meaning: the meaning ithas in virtue of the conventions which are constitutive of the language.’(Recanati 2004, p. 68)

As we saw in the preceding chapters, however, it is not at all clear that a character-isation of semantic/linguistic meaning in terms of conventionality is unproblematic,especially since it implies that conventionality is an all-or-nothing property, which,as I argued above, it is not.

Second, it is somewhat irritating that Recanati takes conversational implicaturesand indirect speech acts to be good examples of what is usually understood as non-literal meaning. As Stern (2006, p. 248, footnote 2) comments: ‘[o]ne might balkat classifying such inferred meanings as either literal or non-literal’. Rather, it isphenomena like metaphor or metonymy that are considered prototypical examplesof non-literal meaning. What makes conversational implicatures and indirect speechacts good examples of non-literal meaning for Recanati seems to be the fact that inhis classification these two meanings do not only fulfil the transparency condition,but at the same time have a secondary character. The latter cannot be said ofmetaphor and metonymy, whose non-literalness Recanati’s system can only captureby applying to the transparency condition. However, recall that, traditionally, non-literal meaning is understood as being the result of a reinterpretation of an utterance.Thus, the resulting meaning, in a sense, replaces the original meaning.3 Although wesaw that the traditional view is problematic, the intuition that in cases of non-literalmeaning the basic meaning is in some sense replaced by the non-literal meaning,remains. In case of conversational implicatures, however, the situation seems to bedifferent. Here, the meaning contributed by the CIs is added to the basic meaning

3Even though, as we saw, at least some aspects of the underlying meaning remain active andthus play a role during the interpretation of a non-literal meaning.

from which they are derived. That is, the CIs are not part of the utterance meaningof the utterance made, but rather they are part of the communicative sense or whatis meant by the speaker of that utterance. Thus, in the cases in which CIs aredrawn with respect to a particular utterance, the meaning of the utterance itself isnot changed or becomes non-literal as seems to be suggested by Recanati.4

Note further that the transparency condition is problematic itself. It imputesto speakers the ability to reliably differentiate between cases of literal meaning andcases of non-literal meaning as classified by Recanati. However, at least for thephenomenon of conversational implicature we already saw that this is not the case.Thus, although CIs are originally assumed to be derived on the basis of some priorilydetermined utterance meaning, speakers are not able to reliably differentiate betweenthe ‘said’ and the ‘implied’. Moreover, Recanati’s characterisation of how the non-literalness of metonymic or metaphoric readings for some expression comes aboutis rather vague. Thus, especially in the case of metonymy, one might wonder whatmeaning aspects of the literal meaning of the respective expression are filtered outand thus lead to it to be understood as metonymic. This characterisation worksbetter with metaphors but even in that case, especially where the metaphor vehicleand the metaphor target are of different sorts, deriving a metaphoric interpretationdoes not seem to simply be a case of ‘feature-dropping’. Having said all that,Recanati’s characterisation can also not capture the fact that there are cases ofnon-literal meaning which seem to be as easily/fast understood as literal meaningis. That is, arguably, in the case of what Giora calls familiar metaphors, there isno strenuous process of metaphoric meaning-generation, with non-fitting meaningaspects being filtered out but staying activated to a sufficient extent. Nevertheless,the respective meanings are or at least can be perceived as non-literal.

With respect to the process of free enrichment, the question arises why it is notcharacterised as secondary, although it adds meaning aspects to minimal proposi-tions and why its results are not perceived as deviating, although they are addedto minimal propositions. Recanati’s argument is that free enrichment is a primarypragmatic process and as such not consciously available to hearers. Thus, they can-not know that it is based on some priorily determined meaning. However, as wesaw, assumptions relying on hearers’ awareness of some process/meaning aspect isproblematic.

Note also that Recanati’s differentiation between free enrichment on the onehand and metaphor on the other in terms of sense elaboration and sense extension,respectively is seen as problematic by proponents of RT. Thus, Carston (1997) hasargued for treating narrowing (Recanati’s sense elaboration) and loosening (Reca-nati’s sense extension) as basically two opposite directions in which one and the sameprocess of ad-hoc concept formation might turn, especially since there are cases ofconcept adjustment in which both narrowing and loosening are involved. Thus, inRT free enrichment and the process leading to a metaphoric interpretation are not

4‘The interpretation of both conversational implicatures and indirect speech acts involves aninference from the utterance’s primary meaning to its derived meaning’ (Recanati 2001, p. 266).

viewed as distinct.5

5.1.2 Literal Meaning as ‘Minimal Meaning’

Ariel (2002) proposes to replace the traditional notion of literal meaning by three dis-tinct types of minimal meaning. Thus, she differentiates between linguistic meaning,salient meaning and privileged interactional meaning. The reasons for abandoningthe traditional notion of literal meaning are similar to the ones put forward here,most importantly the fact that the traditional characterisation of literal meaning isinconsistent. Her differentiation of meanings is based on the motivations behind thetraditional concept of literal meaning, which she takes to be linguistic, psycholin-guistic and interactional in nature. Thus, Ariel suggests a three-fold distinction ofdifferent types of what she terms minimal meaning on the basis of the kinds of func-tions that these meanings serve. Ariel claims that what we take to be the semantic,that is, linguistic meaning of an expression may not coincide with the reading thatis most salient from a psychological point of view. Similarly, the interpretation ofsome utterance that we take a speaker to be bound by may be different both fromthe result of the composition of the linguistic meanings of the expressions involved aswell as from the readings that are most rapidly activated during the interpretationof that utterance. Let us look at these different types of literal meaning, or ratherminimal meaning in more detail.

The first type of minimal meaning that Ariel distinguishes is the linguisticmeaning of expressions, that is, their encoded meaning. Linguistic meaning is notwholly truth-conditional, that is, it includes meaning aspects that are not truth-conditionally relevant and the truth-conditional aspects of meaning that it doesprovide may not exhaust the entire set of truth-conditions necessary to determinethe truth or falsity of the meaning of some utterance of the linguistic expressions.Moreover, although linguistic meaning generally is characterised as being composi-tional, non-compositional meanings (such as for idioms) may also grammaticise andthus become linguistic meaning.

From how Ariel characterises linguistic meaning, it seems that she does not take itto be characterised by underdeterminacy. On the contrary, like Recanati, she seemsto assume that linguistic meanings are full-fledged readings for expressions. Thus,she argues that although linguistic meanings rarely figure in actual conversationunenriched, interpreters are able to opt for this ‘bare’ linguistic meaning. Arielgives an example to illustrate this possibility.

(183) a. S: Tire, ani hitxalti ita lifney xamesh shanim‘Look, I started with/made a pass at her five years ago’

b. M: Ata hitxalta ita o she . . .‘You made a pass at her or . . .’

c. S: Lo . . .‘No . . .’

5See, however, Burton-Roberts (2007b)’ critique of the symmetric treatment of what do seemto be two distinct processes. Also, see below for a general critique of the way ad-hoc conceptformation is characterised by RT.

In this example, the construction hitxalti ita is used, which is ambiguous as it maymean ‘started working with her’ or ‘made a pass at her’. Ariel claims that the contextin which this little conversation took place, made it quite clear that S intended theformer reading. M, however, picks up on the second possible reading. Ariel arguesthat M’s interpretation of S’s utterance cannot be dismissed in the way a completelyirrelevant meaning might have been. She assumes that this is the case becauseM’s interpretation of S’s utterance is linguistically justified, since he picks up on a‘bare’ meaning of the construction hitxalti ita. This example suggests two things.First, that Ariel takes linguistic meanings to be full-fledged readings. Moreover,she explicitly states that the various meanings of homonymous6 and polysemousexpressions are ‘. . . no doubt . . .’ (Ariel 2002, p. 394) instances of linguistic meaning.As should have become obvious from what has been said sofar, the assumption thatthe meanings of polysemous or homonymous expressions generally are linguisticallycoded is not at all uncontroversial. Second, what identifies such ‘bare’ readings asinstances of linguistic meaning is the fact that in the situations they are opted for,they are not contextually appropriate. In other words, they arise or are chosen as theintended meaning by interpreters, although they do not fit the respective context.Thus, Ariel assumes they cannot be the result of interpretation ‘in context’, as itwere. However, I do not find this evidence convincing, because what Ariel calls‘bare’ linguistic meaning does not have to be taken to be a linguistic meaning at all.It might belong to the second level of minimal meaning identified by Ariel, namelysalient meaning, which notion she directly borrows from Giora.

Thus, as mentioned before, this type of meaning can be seen as a psycholin-guistically motivated minimal meaning in that it is the meaning that is activatedfirst, automatically and obligatorily. In the sense that this meaning is the easiestto access – or the meaning the activation of which is least costly – it may be re-garded as a kind of literal meaning.7 Ariel argues that not all salient meanings arenecessarily identical to linguistic meanings and vice versa. This is due to the factthat the criterion for saliency of meaning is the speed with which we process therespective meaning, rather than whether the meaning under consideration is con-ventional. However, recall Giora’s definition of salient meanings as being a functionof several aspects, one of them actually being conventionality. Moreover, Ariel doesnot give a detailed explanation of what she takes the notion of conventionality tocapture. She does mention, in a footnote, that her view of conventionality differsfrom Giora’s. Thus, whereas Giora takes conventionality of a meaning to be oneproperty that influences the status of that meaning as to whether it is (less-)salientor not and thus (co-)determines whether that meaning will be lexically coded, Arieldifferentiates between conventional uses of an expression with a particular readingand the lexicalised meanings of an expression. Conventional uses determine thesalience of a meaning, but not its linguistic status. A particular reading is only lin-guistically conventionalised if it is lexicalised. But at this point the question arises

6Note that Ariel (2002) uses the term ambiguous to refer to homonymous expressions. However,since the notion of ambiguity usually is understood to encompass both homonymy and polysemy, toavoid confusion, I will consistently use the term homonymous where Ariel (2002) has ambiguous.

7However, Giora emphasises that the notion of salient meaning cuts across the traditionalliteral/non-literal distinction.

which aspects lead to the lexicalisation of a particular reading of some expression.Note that Ariel’s view on conventionality is similar to Busse’s, (discussed in chapter2) who also differentiates between conventions of language and conventions of lan-guage use. Recall, however, that Busse takes the process of conventionalisation toinclude the process of lexicalisation proper, that is, the process that finally changesthe status of some reading from being pragmatically derived to being semanticallycoded. As was suggested above, conventionality seems to be closely connected tothe frequency of use of an expression with a particular reading. Thus, arguably, it isincreased frequency of use that leads to a shortening of the inference chain endingin a pragmatic inference. The point where the inference chain becomes so obscuredthat we are no longer able to consciously reconstruct it, can be taken to be the pointat which the reading under consideration lexicalises. Ignoring for the moment thequestion whether one wants to assume that semantic meanings are full-fledged read-ings or not, the crucial point about these assumptions concerning the lexicalisationprocess is that it does rely on the conventionality of a particular reading, which inturn can be taken to be determined by the frequency of use of the reading underconsideration. If that is the case, then I see no reason why it should be necessary(in Ariel’s sense) to differentiate a level of linguistic meaning from a level of salientmeaning.

To come back to the example mentioned above, the reading ‘made a pass at her’of the construction hitxalti ita of which Ariel claims that it cannot be dismissed inthe way a completely irrelevant meaning might have been due to the fact that it islinguistically justified: this conclusion is not necessary as the respective reading maybe taken to be a salient meaning of the construction. Thus, Ariel mentions studiesthe results of which show that during the interpretation of an ambiguous utterance,all salient meanings are initially activated, but most of them are supressed ratherquickly, as soon as it becomes clear that they are not intended. However, theexperimental situation is, I think, an idealised one, in that it is very likely thatthe most prominent of the intentions of the subjects is to comply with what theyare asked to do by the experimenter. In other words, in such an experimentalenvironment there is no other, personal reason to be less than fully co-operative, or,to put it differently, there is no particular reason why the subject should have otherpersonal intentions that determine his interpretation of a given utterance.

The idea is that if, as Ariel suggests, M in this scenario is not being fully co-operative, that is, if his relation to S, or to the topic discussed, is such that, say,he enjoys teasing S and in this situation does so by willfully misunderstanding him,then that attitude towards S and the intention that grows out of it will influencethe way he interprets S’s utterance. The interpretation process will not necessarilyunfold in the way that the plausibly not intended reading of the construction hitxaltiita is heavily suppressed. Recall Schulz von Thun’s characterisation of a messagehaving more than just two sides, where this characterisation takes into account thata hearer’s interpretation of a speaker’s utterance is influenced by how the formerassesses the relation holding between himself and the speaker. Moreover, speakersand hearers may have certain attitudes towards one another, which may be affectedby the interactional context. So, in the example above, it seems that M has a certainattitude toward S, which he does not want to express directly, but which is expressed

by his apparently willfully misunderstanding S. Or it might be that he simply wantsto humorously tease S and does so by choosing to willfully misunderstand him. Thepoint is, that if M actually has any one of the intentions mentioned, they will in-fluence the way he interprets S’s utterance and thus may lead to the interpretationprocess resulting in more than one interpretation.

Another problematic aspect about Ariel’s differentiation of linguistic meaningfrom salient meaning concerns the role the two are assumed to play in the inter-pretation process. Looking once more at the example above, Ariel states that theinteractionally implausible interpretation of S’s utterance by M is linguistically jus-tified, because the basis for it is a linguistic meaning of the construction hitxalti ita.Does that mean that in interpretation linguistic meanings and salient meanings areaccessed independently of one another? If the main reason for M’s arriving at thatparticular interpretation of S’s utterance is the fact that the reading is linguisticallyjustified rather than a salient meaning of the construction in question, then it shouldbe possible for that reading to be chosen as the interpretation of the construction,although it is not a salient meaning. Thus, linguistic meaning and salient meaningshave to be taken to play different roles in the interpretation process. I find such anassumption highly implausible. First, the interpretation process is a cognitive taskthat can be measured by appropriate methods. As has been shown in numerousexperiments, manipulating the input to that process may affect it and the results itreturns. I do not see how linguistic meaning (in Ariel’s sense) can figure in such aprocess without being picked up by the instruments used to measure the unfoldingof that process. Second, it is not clear to me on which basis Ariel distinguishesbetween linguistic meaning and (less-)salient meaning. She claims that the formeris a significant level of meaning as it ‘characterises the native speaker’s competencein her language.’ (Ariel 2002, p. 362). However, I am not quite sure what it meansfor linguistic meaning to characterise a speaker’s competence in his language. More-over, such a characterisation of the role of linguistic meaning only makes sense if oneassumes that linguistic meaning is constituted by full-fledged readings. However, aswe saw, there are a number of arguments for not doing so. Thus, if linguistic mean-ing is taken to be characterised by underdeterminacy, then it cannot easily be takenas being reflected in a native speaker’s competence in his language. Rather, whentalking about a speaker’s semantic competence, what is actually meant is her abilityto use a particular expression appropriately in situations in which that expressionis, more or less, standardly or conventionally used.

Finally, let us turn to the third type of minimal meaning Ariel suggests: theprivileged interactional meaning. This type of meaning is minimal in the sensethat it may be taken to capture that content of an utterance, which the speakeris taken to be bound to. It thus constitutes the most basic level of communicatedmeaning. Ariel concedes, however, that there is no unique meaning representationthat may be taken to invariably function as the privileged interactional meaning.Rather, what is taken as the privileged interactional meaning of some speech act mayactually differ from interlocutor to interlocutor. Recall that this was also suggestedby the experiments in which subjects were asked to choose among a number ofalternative paraphrases the one they thought best captured what a speaker saidwith his utterance. Depending on the nature of the testing material and situation,

subjects did not always choose the paraphrase that captured the theoretical notionof what is said. Rather, they sometimes chose paraphrases which included whattheoretically would be identified as clear cases of implicature (e.g. Nicolle and Clark1999, Bezuidenhout and Cutting 2002).

Ariel argues that this type of meaning is both distinct from linguistic meaningand salient meaning. In contrast to the two latter types of meaning, which are of-ten subconscious, privileged interactional meaning is fully conscious. Moreover, itclearly differs from linguistic meaning in that it is a cognitive hybrid, including bothcoded and inferred meaning aspects, whereas linguistic meaning is clearly coded.As evidence that salient meaning differs from privileged interactional meaning Arielsuggests the differences in processing that have been found for the interpretation ofhomonymous and polysemous expressions. Thus, for both types of expression, earlyon during their interpretation, all their meanings which are equally salient are ac-cessed. However, for homonymous expressions, their unintended meanings are thenrather quickly suppressed, whereas the unintended readings of a polysemous expres-sion remain activated for much longer. In a particular speech situation, though, theinterpreter of a polysemous or homonymous expression, having determined the read-ing of that expression in that particular situation, may not consciously be aware ofother possible readings, although they were unconsciously accessed during the inter-pretation process. Moreover, the interpreter is not consciously aware of the fact thatthe unfolding of the interpretation process differs for homonymous and polysemousexpressions.

Thus, privileged interactional meaning is supposed to capture a level of meaningsimilar to explicature. That is, it consists both of coded and inferred meaningaspects and it is a meaning level which can be evaluated regarding its truth orfalsity. As already stated, Ariel admits that there is not necessarily a way of uniquelydetermining this particular level of meaning. Nonetheless, it is characterised as thatminimal level of meaning of an utterance the speaker is taken to be committed to.This view of privileged interactional meaning as a type of literal meaning seems to bebased on the traditional idea according to which what is said by an utterance is bothfully propositional as well as literal in that it only consists of the lexical meanings ofthe expressions that are part of the utterance and of meaning aspects contributed bythe traditionally assumed saturation processes. As such, it is traditionally assumedto be that level of meaning to the assertion of which the speaker of the respectiveutterance is taken to be bound to. However, as we saw especially in chapter 3,actually the level of meaning a speaker is taken to be bound to, is richer thanGrice’s level of what is said and, crucially, it may involve meaning aspects thattraditionally are classified as non-literal (such as metonymy or metaphor).

5.1.3 Nature of the Processes Determining (Non)-LiteralMeaning

Before turning to my own characterisation of literal meaning and non-literal mean-ing, I would like to summarise and discuss the various processes identified in chapter3 and 4 involved in determining the meaning of an utterance. The point I want tomake is that there seem to be no processes specifically aimed at determining what

might be called the literal meaning of an expression.Thus, recall the processes discussed in connection with the level of utterance

meaning in chapter 3. We saw that the processes of disambiguation, fixing of indexi-cals and reference resolution lead to what Grice called what is said. However, we alsosaw that this traditional level of meaning may not actually be of fully propositionalform. Since in chapter 2 I suggested that literal meaning is to be found at the levelof fully propositional meaning, the saturation and disambiguation processes leadingto Grice’s traditional what is said cannot determine this type of meaning. However,recall that Bierwisch proposes the two processes conceptual shift and conceptualdifferentiation which, arguably, do result in literal meaning. These two processessupply meaning aspects that fix the particular reading the expression they apply tohas in a particular (narrow) utterance context. Thus, they are saturation processesin that they provide the values for variables in the semantic forms of the expressionsconcerned. Note that, according to Bierwisch, in both cases what the processes con-tribute are primary or non-derived meaning aspects. That being so, conceptual shiftand conceptual differentiation can be seen as processes that result in literal meaningaspects being integrated into the propositional content expressed by an utterance.However, note that similarly to Grice’s view, such a model of (non-)literal meaninginterpretation would again amount to a literal first or serial model of interpreta-tion. In fact, as the results of experiments on the interpretation of novel cases ofnon-literal meaning suggest, such a model may be appropriate when it comes tomodelling how unfamiliar cases of non-literal meaning are interpreted. However, forcases of familiar non-literal meaning, it makes the wrong predictions.

An alternative is to assume that saturation processes may already integratemeaning aspects that are intuitively non-literal, but which have a high frequencyand are thus familiar and might be taken to be conventionalised to the degree thatthey have become part of the conceptual family from which the processes of con-ceptual shift or conceptual differentiation choose a member. Thus, some types ofmetonymic shift – such as e.g., the producer-for-product metonymy – seem to behighly productive (at least for English), allowing for unproblematic extension tonew cases (cp. Frisson and Pickering 2007). That is, identifying some person as anauthor seems to automatically allow the use of that person’s name to refer to booksproduced by that person. That being so, one could assume that such highly produc-tive processes apply virtually automatically, thus, leading to no perceptible readingtime or semantic integration effects. Another possibility is that for highly frequentmetonymies, the metonymic reading in fact has become part of the concept familyon which the process of conceptual shift operates. For example, as soon as some per-son is characterised as an author, the possibility of using that person’s name for hisbooks is made available by the general relation holding between authors and theircreations. Similarly, in the case of institution-for-person metonymy, the metonymicreading seems fairly conventional, due to the general knowledge that institutions areassociated with people.

The fact that such readings may still be perceived as non-literal might be due tothe particular semantic relations taken to hold between those readings and the so-called literal ones or it might simply be due to the fact that processes exist that createthese kinds of meaning relations, allowing for the use of a particular expression in a

particular way in the first place. Thus, the process of metonymic shift may apply toa concept that is part of the set of primary readings of a particular expression andresult in a metonymic interpretation for that expression in that particular situation.If the relevant expression is frequently used in that metonymic reading, the resultof the process might be integrated into the concept family of which the primaryconcept the expression is used to express is part. In that case it is conventionalisedand the interpretation process no longer has to infer it.

Note, however, that frequent use of an expression with a particular metonymicreading is not a necessary condition for it to be integrated into the set of primaryreadings of the expression. Thus, recall Frisson and Pickering (2007)’s experiment inwhich subjects showed no difficulty interpreting a ‘producer-for-product’-metonymicuse of an expression they had only encountered once before and not in a metonymicreading. That is, the ‘producer-for-product’-metonymy seems so regular and pro-ductive that it is enough for speaker-hearers to identify some person as an authorto be able to use that person’s name to refer to his works. Thus, the particularmetonymic reading is added to the set of primary readings for the expression auto-matically, as it were. That is the point at which a metonymic reading is accessed asquickly as a primary one. However, due to the existence of the metonymic processand the fact that people can consciously reconstruct how the process works, they arealso consciously able to realise that some reading is metonymic, even if that readingis highly familiar and in that sense conventional.

This idea suggests that something is left behind by the process of metonymicshift that allows the system to recognise that a particular metonymic reading for aparticular expression has been generated before because it seems that this is neededto account for how such a reading can be conventionalised in the first place. If it isonly created on the fly, as it where, and no trace of it is left behind, it is unclearhow the system would be able to recognise that it is used often enough for it to bemore efficient to get permanently stored.8

Thus, the set of potential primary meanings for an expression from which suchprocesses as conceptual shift and conceptual differentiation choose, includes not onlynon-derived, basic and in that sense literal meanings, but also such as are clearlyderived, non-basic and in that sense non-literal. Thus, membership in the set ofprimary meanings is not determined on the basis of ‘literalness’, rather, it is basedon salience of meaning. That is why the set of potential readings for an expressionthat a pointer points to should rather be called the set of salient readings for anexpression.

In fact, as suggested already in chapter 2, actually what a pointer points to isnot simply the set of salient readings for an expression – where that set, as we saw,has to be characterised as being dynamic – rather, it is a conceptual region, or whatRT calls the encyclopaedic entry of a concept, which may be taken to be organisedby salience and thus subsumes the set of salient readings. That is, the encyclopaedicentry includes ready-made readings for an expression but also material from which

8This idea might be modelled by assuming that the particular reading determined by the in-terpretation process and due to a particular use of an expression does leave a trace in memory oneach occasion it is determined by the interpretation process. The more such traces exist, the easierit will be to activate this particular reading. (Cp. Hintzman 1986).

a contextually fitting reading may be ‘build’.In contrast to saturation processes in general, free enrichment (or expansion) is

characterised as a process that operates on the fully propositional semantic formof an utterance adding meaning aspects that can be taken to have been intendedto be expressed by the speaker and that are necessary in order for the resulting,enriched proposition to form the basis for further pragmatic inferences. Thus, incontrast to conceptual shift, conceptual differentiation, metonymical shift or ad-hocconcept-formation, free enrichment does not operate on meaning below the levelof full propositionality. Rather, according to Bach (1994a), what is characteristicfor cases where free enrichment applies is that each of the expressions involved inthe respective utterance is used literally (or with primary, non-deviating meaning),however, the proposition taken to be expressed by the speaker of the utteranceis ‘. . . conceptually more elaborate . . .’ (Bach 1994a, p. 135) then the unenrichedproposition. That is why Bach (1994a, p. 135) calls this phenomenon sentence-nonliterality: ‘. . . a sentence is used nonliterally without any of its constituents beingso used’. This is in contrast to metonymy and metaphor, where the non-literality isusually associated with simple expressions.

Thus, recall the examples Dolling (2001) gave for metonymic shift and metaphor-ical interpretation, respectively of an expression. As we saw, these processes inte-grate conceptual material into the inflected SF of the respective expression, usingthe structure provided by the met-operator. What is important for the resultinginterpretation to be coherent is that the SF-parameter within the SEM of the expres-sion concerned be fixed when the process of metonymic shift or that of metaphoricalinterpretation applies.9

For example, Dolling (2001) gives the respective parameter-fixed SFs (i.e. 121)for the individual metonymic uses of Lamm given in (120), where the SF-parameterwithin the basic SEM of Lamm has been fixed by �.

(120) a. Hans mochte heute Lamm essen.

‘Hans wants to eat lamb today.’

b. Maria tragt seit gestern Lamm.

‘Since yesterday, Maria wears lamb.’

c. Anna weigert sich, Lamm zu essen oder zu tragen.

‘Anna refuses to eat or wear lamb.’

(121) a. λx. FLEISCH(x) & ∃y [MAT(y)(x) & (y) � (lamm)]

b. λx. FELL(x) & ∃y [MAT(y)(x) & (y) � (lamm)]

c. λx. FELL∨FLEISCH(x) & ∃y [MAT(y)(x) & (y) � (lamm)]

Similarly for potential metaphoric uses of Lamm as in (123) below. Here the param-eter in the SEM of Lamm has been fixed to =.

(123) Annas Mann ist ein Lamm.

‘Anna’s husband is a lamb.’

9Or at least that they get fixed in the course of the application of such processes.

(124) λx. PERSON(x) & ∃y [INST(y)(x) & ART(y) & ∃z [AHNLM(z)(y) & (z) =(lamm)]]

Another process that operates on the semantic form of simple expressions is ad-hoc concept formation. This process can be taken to determine non-literal meaningsfor the expression it operates on, since it is characterised as contextually adjustingthe encoded meaning of an expression. However, if one assumes that semanticmeaning actually is highly underspecified, it does not seem useful to characterise ad-hoc concept formation as a process that applies to the encoded meaning and adjustsit. That is, the encoded meaning under the assumption of underspecification is noteven a full meaning; it does not constitute a full concept. The intuition underlyingthe assumption of a process like ad-hoc concept formation is, however, that it doesapply to a full reading or concept and returns a concept that is based on that originalconcept. Recall also that in RT it is assumed that ad-hoc concept formation is theunderlying process for meaning aspects differing in their degree of perceived non-literalness (e.g., loose uses, hyperbole, metaphor). One possibility of capturing thisintuition is to assume that ad-hoc concept formation operates in those cases, where,based on the consideration of the context, the process that determines the value ofthe SEM-parameter in that particular context, does not (easily) find a reading thatis sufficiently general/specific enough to fit in that context. Thus, the reading ad-hocconcept formation operates on is that member of the value set of the SEM-pointerwhich is best compatible with the context at hand, but not general/specific enough.Recall example (66) from chapter 3, repeated below.

(66) Kato (of O. J. Simpson, at his trial):He was upset, but he wasn’t upset.[He was upset’ but he wasn’t upset”]

The interpretation of the two occurrences of upset in this example can be adequatelydescribed employing the idea of ad-hoc concept formation. Thus, the first occurrenceof the expression upset is assigned one of the expression’s salient readings, namelythe one that fits best the contextual circumstances. The second occurrence of upset,however, seems to be used by the speaker to convey a very specific concept, given theparticular circumstances of utterance. To repeat, a rough paraphrase of what Katowanted to express might be ‘O.J. Simpson was sad and angry to a certain degree,but he was not sad and angry to a degree that led him to killing his wife.’ Thus,it seems rather unlikely that the speaker entertained this particular concept priorto the particular situation he found himself in and in which he made the utterancein (66). It is even more unlikely that this concept is stored as part of the salientreadings for upset for either speaker or hearer. Thus, the concept is created online,during the interpretation of the respective utterance and by specifying the concepttaken to be expressed by the first occurrence of upset in the utterance.

While the idea of ad-hoc concept formation works well for such examples as theone given above, it is actually questionable whether this process also is involved inwhat is called loose use in RT, examples of which are repeated below.

(103) a. France is hexagonal.

b. Holland is flat.

c. I must run to the bank before it closes.

In fact, at least for the examples given, it is questionable whether a specific processis necessary to explain this type of phenomenon at all (cf. Burton-Roberts 2007a).Recall that, in the above example, RT claims, the expressions hexagonal, flat andrun are used to express concepts that are less strict or specific than the ones they aretaken to encode. The first thing to note is that if what expressions encode actuallyare not full concepts, then what is encoded by hexagonal, flat and run respectively,cannot said to be a strict or specific reading, as it is no full reading at all. Second,and as noted by Burton-Roberts (2007a) for the assumed encoded meaning for theseexpressions, assuming that the strict or specific concepts are the only concepts thepointer in the SEM of the respective expressions points to, seems a rather arbitrarystipulation. On the contrary, the fact that the uses of the relevant expressionsin the examples above only seem ‘special’ on second thoughts and only under theassumption that their salient meanings are strict, makes it even more questionablewhether this assumption is in fact appropriate.

In this connection, recall also Ruhl (1989)’s cautionary note not to attribute‘non-literalness’ to a particular reading for an expression only on the basis thatthis reading deviates from some other reading that expression might be used toexpress. Note also that, at least for the expression flat, the assumed ad-hoc conceptin fact does not actually seem to be so ad-hoc. That is, using flat to describe thegeneral geographical circumstances of a country or some region seems to be a ratherconventional means of making such a description. In fact, for the correspondingGerman expression flach, the respective reading figures in the compound Flachland(i.e., lowland, plain), indicating that the relevant concept expressed by flach is notat all ad-hoc.

Thus, I would tentatively suggest that the concepts expressed by so-called looseuses of expressions in fact are nothing more then possible salient meanings, and thatin order for them to be expressed on a particular occasion of use of the respectiveexpression, nothing else happens than that the interpretation process chooses thatreading from the set of salient meanings the expression might be used to express.This suggestion is only tentative, as the sole intuition that such uses are not felt toinvolve some effortful process of concept-formation might be misleading. Thus, theidea that cases of what is called loose use actually involve nothing more than salientreadings of the respective expression needs to be empirically tested.

Having said that, one might argue that even though such loose readings might bemembers of the set of salient meanings today, diachronically speaking, they mightstill be the result of some loosening process – or as argued by RT, of ad-hoc conceptformation – which applied to the strict meaning and created the loose reading fromthat. Thus, recall that the set of salient meanings as I have characterised it alsoincludes non-basic, derived readings. As mentioned already, the problem is thatthere does not seem to be a way of objectively ascertaining whether a particularmeaning is literal or not. Be that as it may, the point is that for the process ofad-hoc concept formation, as it is characterised by RT, one would assume that itsoperation in a particular interpretation situation would involve greater cognitiveeffort than simply choosing a ready-made reading from a given set. My assumption,then, is that in the interpretation of the examples of loose use that RT provide,

this process does not operate. Rather, I would suggest that the respective readingsinvolved in so-called loose uses stand in the same relation to the so-called strictreadings as the salient readings of a polysemous expression generally do.

Recall that in RT the process of ad-hoc concept formation is generally assumedto be at play for all kinds of more or less non-literal meaning aspects, amongst them,of course, metaphor. However, even if ad-hoc concept formation is taken to involveloosening of the denotation of an existing concept, it is questionable whether thisappropriately characterises what is going on in metaphor interpretation. That is,while the idea, as we saw, works for cases such as in (106), it is questionable whetherad-hoc concept formation can be applied to cases, where the salient concept of themetaphor vehicle could not even in principle be attributed of the concept of themetaphor target, as is the case in (184).10

(106) John is a soldier.

(184) Peter is a bulldozer.

That is, the interpretation of bulldozer as denoting a category to which Peter is saidto belong, does not seem to involve simply a cancellation of particular features ofthe original concept bulldozer is taken to express ‘literally’.11 Thus, Dolling (2000)’streatment of metaphor as involving a similarity function AHNL that relies on a setof features M with respect to which two concepts are considered to be similar bettercaptures what seems to be going on in metaphor, although how those features aredetermined remains an open question.

To summarise, the nature of the processes contributing to utterance meaningthat we looked at suggests that it is differences in the information used by thoseprocesses that leads to the resulting utterance meaning being perceived as eitherliteral or non-literal. That is, there does not seem to be a distinctive set of processesresulting only in literal meaning. Taking into account the fact that potential read-ings for an expression differ in their grade of salience, there are processes that assignvalues to parameters in the semantic form of an expression – so-called saturationprocesses – and the readings that might potentially be that value are characterisedby being highly salient, but not necessarily primary in the sense of ‘non-derived’.That is, already saturation processes might provide as a value for some variable inthe semantic form of an utterance what might be called a non-literal meaning. Incontrast, there do seem to be processes whose function it is to derive a particu-lar meaning from some underlying, basic, full meaning, such as metonymic shift,metaphorical interpretation, ad-hoc concept formation and – if one accepts Bach’sargumentation – free enrichment.

5.1.4 (Non-)Literal Meaning as (Non-)Basic Meaning

In this section, I want to propose an alternative characterisation of the notions literalmeaning and non-literal meaning, based on the assumption that linguistically coded

10Note that Glucksberg and McGlone (1999)’s attributive category approach faces the sameproblem.

11This is fully acknowledged by Carston (2002b), who suggests a few directions in which to turnfor a possible solution to the problem, but does not clearly offer one herself.

meaning actually is highly underspecified and the fact that these are very intuitivenotions. I start by recapitulating the different uses of the two terms in the literature.

Recall that the traditional view assumes of literal meaning that it is to be found atthe level of expression meaning, stored in the lexicon as the lexical meaning of simpleexpressions and as such entering into the process of semantic composition, which,therefore, also results in larger structures that have literal meaning. Thus, in its useas characterising lexical meaning, literal meaning refers to some basic, primary, non-deviating, full meaning. Recall further that traditionally, it was assumed that toreach the level of utterance meaning or what is said, only such processes operate thatresolve reference, assign values to indexicals and disambiguate the potential semanticstructures generated by semantic composition. There is a sense in which the thusresulting meaning is literal in that it does not include any derived meaning.12 Thus,this particular contextually informed meaning might be called literal in the senseof non-derived, because the processes it results from only assign values to variablesprovided by the semantic form of the respective expression. Therefore, in its seconduse as characterising a particular pragmatically enriched meaning, literal is usedto refer to some disambiguated meaning that is constituted of basic, primary andnon-deviating meanings plus values assigned to variables in the semantic form of theutterance by the processes of fixing of indexicals and reference resolution. Note thatthis amounts to the assumption that what is contributed by saturation processesin general contributes to literal utterance meaning.13 Finally, a particular speechact determined by processes leading to the level of communicative sense might becalled literal, because it is taken to be the basis for the hearer’s drawing of furtherinferences in his interpretation of what the speaker meant by his utterance, leadingto an indirect speech act.

Thus, similarly to the various uses of the expression literally cited at the begin-ning of section 5.1, three different uses of the term literal can be differentiated, one,where it refers to a particular type of meaning associated with simple expressions,one, where it refers to some contextually enriched meaning of an utterance and one,where it refers to the direct speech act attributed to the speaker. What all threeuses share is the general aspect that the meaning described as literal is in somesense basic and underived, although the meaning levels the term is applied to differin their complexity.

Similar points can be made concerning the term non-literal meaning. That is,particular meanings may be associated with a simple expression, where those mean-ings are derived from some underlying basic full meaning and in that sense arenon-literal. However, what is crucial is that traditionally, such non-literal readingsfor simple expressions were assumed to only arise due to a particular use of an ex-pression by a particular speaker in a specific context. That is, in contrast to literalmeaning, the term non-literal meaning traditionally is taken as not applicable atthe level of expression meaning. In addition, the term is used to refer to utterances

12This is so because traditionally, the lexical meanings that are composed into the sentencemeaning of the complex expression uttered are necessarily literal.

13However, as I will suggest below, under the assumption that lexical meaning actually is un-derspecified, what is contributed to utterance meaning by such processes as conceptual shift andconceptual differentiation actually may not necessarily considered to only involve primary or literalreadings of a particular expression.

where what the utterance explicitly/implicitly means is not exhausted by what thesentence used to make the utterance means (i.e., Bach’s sentence nonliterality re-sulting from the process of expansion). Finally, the term non-literal is also usedto refer to meaning aspects arising at the level of communicative sense, namely forcases in which what the utterance meaning describes, deviates in a fairly special wayfrom the situation at hand or, generally, from what one might reasonably expectthe speaker to intend to express (irony). Moreover, when it comes to the speech acta speaker is taken to have carried out with his utterance, indirect speech acts areusually characterised as non-literal, that is, as deviating from some underlying basicor primary direct speech act.

Thus, as for literal meaning, for the term non-literal meaning (at least) three dif-ferent uses can be differentiated, one, where it refers to a type of meaning associatedwith the particular use of a simple expression, one, where it refers to some contextu-ally enriched meaning of an utterance and one, where it refers to the indirect speechact attributed to the speaker. In addition, the term non-literal is further used tocharacterise the meaning arising from a particular use of a complex expression (i.e.,irony). Here, what the four uses of non-literal share is the general aspect that themeaning described is in some sense non-basic or derived and deviating from somerelevant alternative meaning. However, as with literal meaning the meaning levelsthe term is applied to differ in their complexity.

To summarise, both the terms literal meaning and non-literal meaning are used todescribe particular meaning aspects at different levels of meaning. The only charac-teristic that the thus described meaning aspects have in common is that at the levelof meaning at which they occur, they have a basic (thus being literal) or a deviating(thus being non-literal) status. However, note that this common characteristic doesnot allow for the inference that the kinds of meaning aspects described as literal ornon-literal are in any way comparable or similar. This is because what is describedas basic or deviating is basic or deviating with respect to the level of meaning atwhich the respective meaning aspect occurs. In other words, the fact that particularmeaning aspects occurring at different levels of meaning are collectively described aseither literal or non-literal does not make the thus denoted categories homogeneous.Thus, on the one hand, literal meaning at the level of expression meaning refersto context-free, stored basic meanings of expressions. Literal meaning at the levelof utterance meaning refers to context-dependent, complex meanings, where thosemeanings do not include any non-literal meaning aspects. At the level of commu-nicative sense, the term literal refers to yet a more complex meaning representation,including inferences concerning the speech act carried out by the speaker of the ut-terance in question. On the other hand, non-literal meaning used with respect toparticular readings an expression might have, refers to a meaning of some more orless complex expression that traditionally is taken to be determined in the particularcontext in which that expression is used. When Bach speaks of sentence nonliteral-ity, he refers to a meaning that is fully propositional but not solely the product ofsaturation processes. Irony is characterised as involving non-literal meaning becauseof the particular discrepancy between what the utterance made by a speaker meansand what the speaker, given the particular utterance situation, can reasonably betaken to actually mean. Finally, indirect speech acts are characterised as non-literal

due to the assumption that there is a conventional fit between sentence form usedand speech act carried out, which is not adhered to in cases of indirect speech acts.

Thus, it seems that if one wants a characterisation of literal meaning and non-literal meaning that captures all the different phenomena at the different meaninglevels these terms are used to refer to, that characterisation is going to have to be verygeneral and vage. However, as indicated already in chapter 2 and afterwards, thereare reasons for not counting some of the meaning aspects traditionally characterisedas literal or non-literal, respectively, amongst the phenomena these terms in fact referto. Let me repeat the claims I made concerning the nature of literal meaning andnon-literal meaning in chapter 2. Neither literal meaning nor non-literal meaningcan be found at the level of expression meaning, if – as I argued it should – themeaning available at this level is assumed to be highly underspecified, abstract andnot actually constituting any full reading at all. Note that due to the assumptionof underspecification of lexical meaning in general, I do not differentiate betweenlinguistic meaning and salient meaning in Ariel’s sense. In other words, there areno full readings for expressions stored in the lexicon and even if there were, suchmeanings would not be unaffected by the features that determine the salience ona particular occasion of particular readings potentially expressed by a phonologicalform.

Contrary to the traditional view, I argued in chapter 2 that both literal meaningas well as non-literal meaning can be found at the level of utterance meaning. Thus,what we have at this level of meaning are both literal as well as non-literal interpreta-tions or readings for expressions below the sentence level. And although non-literalmeaning can intuitively be characterised as ‘deviating’ from some underlying or ba-sic meaning, in terms of the time-course of the interpretation process, a non-literalmeaning need not necessarily be derived from some underlying meaning online, ifthat non-literal meaning is familiar, sufficiently frequent and conventionalised to acertain extent. Thus, the type of non-literal meaning characterised so far is whatBach calls constituent non-literality. However, recall that Bach argues for anothertype of non-literal meaning to be identified at the level of utterance meaning, namelysentence non-literality. He uses this term for a particular type of meaning resultingfrom the process of expansion operating on a level of meaning that is characterisedas fully propositional (in Bach’s terms: what is said plus the results of the pro-cess of completion), but not the proposition the speaker is intuitively taken to haveintended to express with his utterance. Note, however, that, intuitively at least,the relation between the type of meaning resulting from expansion and the minimalproposition of the utterance that process is based on is different from that between,say, the metaphorical or metonymic meaning of a given expression and the meaningthese non-literal readings are based on. Whereas in cases of metaphor or metonymy,we start out from whatever is the basic meaning and end up with a meaning thatdiffers in a principled way from that basic meaning, in case of expansions of min-imal propositions, what we have as a result is not some different meaning, but ameaning that is more precise (hence Recanati’s term sense elaboration). In otherwords, in case of expansion it seems that meaning aspects are added to the mini-mal proposition to make it more precise, whereas in case of metaphor or metonymymeaning is ‘changed’. In this respect expansion is more similar to (particularised)

conversational implicature than to cases of metaphor or metonymy. That is, (P)CIsare drawn on the basis of the proposition expressed or impliciture and are added toit rather than replacing the utterance meaning. As such, they are not cases of whatis normally intuitively understood as non-literal meaning. However, expansion andthe processes leading to PCIs differ in that the former integrate sub-propositionalmeaning aspects into the minimal proposition, resulting in the proposition expressed,whereas PCIs are themselves full propositions inferred on the basis of the propositionexpressed.

Recall that traditionally irony is also considered a good example of non-literalmeaning. That is because originally, irony was understood as involving two mean-ings, as it were, namely the one ‘literally’ expressed by the utterance made anda second, opposite to the meaning expressed by the utterance made but actuallyintended to be conveyed by the speaker. However, neither of the two theories ofirony interpretation discussed in chapter 4 adhere to this traditional view of whatis involved in irony. More specifically, recall that RT’s approach to irony does noteven make reference to a non-literal meaning being involved in producing or under-standing this ‘figure of speech’. However, what licenses the idea of the non-literalin cases of irony is the fact that there seems to be some contrast involved betweenwhat the speaker says and the specifics of the contextual circumstances in whichshe makes her utterance. Having said this, if one wants to characterise non-literalmeaning as a meaning that (is derived and thus) deviates from some underlying,basic meaning and replaces it, irony actually should not be classified as a type ofnon-literal meaning.

It seems that similar points have to be made with respect to indirect speech acts.Thus, recall that the idea is that indirect speech acts are non-literal in that theyare derived from some priorily determined, underlying direct speech act, which inturn is determined based on the sentence form of the utterance made by a speaker.However, recall also the studies cited above which showed that indirect speech actsdo not take longer to interpret than direct speech acts and that indirect speech actsmay be understood although the assumed underlying or basic direct speech act is not.This suggests that, actually, what are called indirect speech acts are not derived fromsome priorily determined direct speech act after all (or at least not online). Note alsothat the idea of indirect speech acts relies on the assumption of the so-called literalforce hypothesis, according to which particular sentence types are conventionallyassociated with particular speech act types or illocutionary forces. If one does notaccept this hypothesis, then there actually is no such thing as an indirect speechact, since all speech acts have to be determined in the social context in which anutterance is made and the sentence type used in a particular utterance is no failsaveguide towards the intended speech act. Having said this, as we saw in the studieson so-called indirect speech acts, there are, of course, conventionalised speech actssuch that the interpretation process that determines the speech act carried out on aparticular occasion of utterance may be short-circuited, as it were.

After the whole discussion concerning the standard notions of literal meaningand non-literal meaning, characterising the former simply as basic and the latteras derived or deviating from basic meaning and, in a sense, replacing it might beviewed as a rather meagre and disappointing result. However, there are a few points

in favour of this rather general characterisation. First, recall that in the standardcharacterisations of literal and non-literal meaning specific reference was made totheir role in processing. As we saw, however, the distinction between literal andnon-literal meaning does not necessarily play a role in the interpretation process.That is, although one might argue that there must have been a point in time atwhich a non-literal meaning for an expression was derived from its literal meaning,this fact does not of itself lead to perceivable differences in the interpretation ofthe expression in its non-literal meaning at some later point. Having said this, aswas shown by Frisson and Pickering (2007) for the producer-product metonymy,under certain circumstances even interpreting a totally unfamiliar non-literal use ofa particular expression may involve relatively little effort.

Second the characterisation of non-literal as derived or deviating from some basicmeaning allows the term to be applied to idioms. That is, the meaning associatedwith an idiomatic expression is conceived of as deviating from some basic mean-ing in the sense that speaker-hearers are able to consciously recognise that eitherthe idiomatic expression does not have a compositional meaning and also no ‘ho-mophonous’ counterpart with compositional meaning, or, for cases where there doexist compositional counterparts, that the idiom in question has a meaning that isnot compositional and in that sense deviating.

Third, note that non-literal meaning is not secondary in Recanati’s sense. Thatis, for an expression to be interpreted non-literally, the meaning of the utterance it ispart of does not have to be determined first. Thus, I do not consider conversationalimplicatures and so-called indirect speech acts as good examples of non-literal mean-ing. Rather, both of these phenomena seem to me to have much more in commonwith free enrichment/expansion, which processes Recanati also does not characteriseas resulting in non-literal meaning.

Incidentally, the idea of above and below threshold activation Recanati resorts toin order to explain the non-literalness without secondariness found e.g., in metaphorand metonymy is interesting with respect to less or unfamiliar non-literal uses ofexpressions, which, it might be argued, are easier to identify as non-literal than moreconventional non-literal uses. However, such latter uses can be identified as non-literal by speakers on second thoughts, thus, non-literalness cannot be characterisedentirely in terms of above/below threshold activation. Having said that, recall thatsolely relying on speaker’s intuitions is problematic, as there is the danger that, ifasked, a speaker-hearer will try to establish the status of a particular reading ofan expression as non-literal with respect to some other reading when in fact thatpost hoc established connection may not actually have had any role to play in thespeaker’s understanding of the reading in question before.14 More generally, theproblem is that, with the notion of salience cutting across the notion of literalness,there does not seem to be a principled way of establishing whether a particularreading of an expression is basic or not (since it is not necessarily the meaning thatcomes to mind first on encountering a particular expression).

Be that as it may, the important point to note is that with the very generalcharacterisations given for literal and non-literal meaning the two notions still cannot

14This point will be taken up below in the discussion of loose use and RT’s characterisation ofit in terms of ad-hoc concept formation.

The Nature of Context in Utterance Interpretation 227

be used in a differentiation of semantics from pragmatics. Moreover, the fact that aparticular meaning aspect is classified as literal or non-literal alone does not allowthe drawing of any conclusions concerning the time-course of its interpretation. Inlight of these facts, it seems that the characterisation of literal meaning and non-literal meaning given above actually is not very useful, beyond, perhaps, capturingintuitions behind the notions of literal meaning and non-literal meaning and beingconsistent. The major problem is that there seems to be no objective criterion bywhich to differentiate literal from non-literal meaning. That is, it is not clear onwhich basis – other then pure intuition – we judge a particular meaning to be basic.In particular, as we saw above, cases of literal and non-literal meaning cannot bedifferentiated on the basis of the processes that bring them about.

5.2 The Nature of Context in Utterance Interpre-

tation

As we saw throughout this thesis, in the discussion about the nature of the twolevels of meaning utterance meaning and what is meant and whether they actuallycan and should be differentiated, assumptions about the nature of the contextualinformation necessary to determine the two respective levels of meaning play a cru-cial role. Thus, one aspect in which the various views on the nature of the level ofutterance meaning differ is whether the processes determining this level are takento be sensitive to considerations about the speaker’s potential intentions in makingthe utterance he did. The notions of narrow vs. broad context capture this differencein that although both consist of the information about who the speaker and ad-dressee(s) of an utterance are as well as the time and place at which it takes place,only the latter also includes speaker intentions. In addition to the concrete utter-ance situation – involving or not involving assumptions about speaker intentions –general background assumptions are important when it comes to interpreting ut-terances. Thus, utterances are intuitively judged as (involving) literal or non-literal(meaning) against a particular set of background assumptions, where this, in turn,is taken to be sensitive to the particular utterance situation. More generally, es-pecially in approaches that assume underspecification of lexical meaning, a furtherimportant pool of conceptual knowledge for interpretation processes to draw fromis the overall world knowledge, which provides the actual readings for linguistic ex-pressions. As argued above, knowledge concerning social interaction should be takenas distinct from overall world knowledge, since the former only contributes to thelevel of what is meant, whereas the latter is already important for the level of whatis said/utterance meaning.

It should be noted that when it comes to the interpretation of ongoing discourse,the notions of both narrow as well as broad context are not sufficient. Rather, whathas been called a common ground is needed, that is, the context that is constitutedand updated by the ongoing discourse the participants are engaged in. Interestingly,in the approaches considered above, this particular type of context has not beenexplicitly mentioned, as the focus rather is on the interpretation of single utterancesand not so much on the interpretation of ongoing discourse. However, as I will argue

below, the principles taken to hold for the interpretation of discourse to be coherentmay also play a role when looking at the interpretation of individual utterances.What is more, particular assumptions made concerning the resources used in theinterpretation of discourse may also be fruitfully exploited for the interpretation ofindividual utterances.

Thus, in section 5.2.1, I will first introduce some important aspects concerningthe interpretation of ongoing discourse. Moreover, Irmer (2009)’s suggestion will beintroduced that in the interpretation of discourse, one source of information thatis made use of is conceptual frames that are evoked by the speaker’s use of theexpressions that make up his utterance. Taking up this idea, I will offer a proposalof how such implicit meaning aspects as taken by RT to be contributed by the processof free enrichment are determined without the need to take recourse to assumptionsabout the speaker’s intentions in making a particular utterance.

Recall further that the different approaches to the nature of utterance meaning orwhat is said also differ in their conceptions of whether or not the semantics compo-nent of the language faculty should be taken to be concerned with context-dependentmeaning and if so, to which extend. An interesting question to ask is why particularsemantically oriented approaches to utterance interpretation should assume of thesemantics component that it be sensitive to specific contextual information in thefirst place. I will take up this question in section 5.2.2 and argue that it has to dowith assumptions about what types of operations semantics generally is taken to beinvolved in.

5.2.1 Context and the Interpretation of Implicit MeaningAspects

Free Enrichment and Implicit Meaning Aspects

Let us begin with the idea within RT that free enrichment and, say, metaphoricalinterpretation are simply two different directions taken by one and the same processof ad-hoc concept formation. Actually, this view is glossing over some importantdifferences between the two processes and the resulting meaning aspects. Thus,whereas in cases of metaphor what is being operated on is the concept associatedwith a simple expression, in case of free enrichment what we have is the addition ofa conceptual element to the semantic form of the utterance, without there being alinguistically mandated slot for it. More specifically, the process of free enrichmentis usually understood to operate on fully propositional forms.15

In fact, it seems that the conceptual elements contributed by free enrichment areactually standardised interpretations for the particular form the utterance takes. Assuch, they may be considered as belonging to the level of meaning proposed by Levin-son for GCIs: utterance-type meaning. Garrett and Harnish (2008)’s experimenton particular types of what they call (following Bach) impliciture meaning aspects

15Recall that Recanati assumes of free enrichment that it is a primary pragmatic process, whereone of the latter’s characteristic is that it operates on sub-propositional forms. However, Recanati,in a sense, is forced to this treatment of free enrichment, since he takes the process to be sub-personal and thus not available to consciousness. See, however, Carston (2007) for a criticism ofthis view.

corroborate this assumption. Thus, it is not only the case that speakers in saying‘X broke a finger’ very often actually intend to express ‘X broke one of X’s fingers’,rather, even if they make use of this particular sentence in a context where it isclear that it is not a finger of X that is broken, this particular interpretation arisesnevertheless. Thus, it seems to have a special status in that it arises despite notbeing intended.

Recall that one of the arguments that lead RT to assume that the meaningaspects contributed by free enrichment are part of the explicitly expressed utterancemeaning is the fact that they may play a truth-conditional role. However, as alreadymentioned, this fact may be captured in a way other than assuming that theyhave to be speaker-intended to begin with. Thus, although the term GCI suggeststhat what these inference are is some special type of CI, in fact, what they are issomething that has been abstracted away from genuine CIs.16 In Bach’s terms, theyhave become standardised and as such arise independently of speaker intentions.However, because they are standardised, they have the ability to contribute to thetruth conditions of the utterance they are part of in those cases where they clearlyare intended. In other words, speakers (unconsciously) can make use of the factthat a particular utterance-type is standardly associated with a particular meaningaspect when they explicitly want to communicate that meaning aspect. They canrely on the fact that hearers will access the meaning aspect in question, once therespective utterance-type with which it is standardly associated is used/instantiatedand that hearers will retain it as indeed intended, unless evidence to the contraryarises.

As we saw, Burton-Roberts (2006) suggests that it is a defining property of GCIsonly that they are cancellable. If this is so then it would supply a theoretical testfor whether a particular meaning aspect taken to be contributed by free enrichmentmight be treated in terms of GCI: it has to be cancellable in principle. Unfortunately,the cancellation test only works for a subset of what are usually considered GCIs(e.g., by Levinson). Crucially, the test does not yield the desired results for theimplicatures based on what Levinson calls the I-heuristic and which Garrett andHarnish (2008) tested. In fact, Burton-Roberts (2006) cites the following exampleto support the view that explicatures are not cancellable.

(185) a. I haven’t had breakfast.

b. I haven’t had breakfast today.

c. ??I haven’t had breakfast, but I did today.

Thus, it seems he does not count such I-based implicatures among GCIs. However,as Garrett and Harnish (2008)’s results indicate, these phenomena do show themain characteristics of GCIs as stated by Burton-Roberts (2006), namely arisingeven in cancelling contexts and arising when the utterances that give rise to themare decontextualised. Note also that for the positive variant of the above example,the cancellation of the implicit meaning aspect (IMA) does seem to be possible.

(186) a. I’ve had breakfast.

b. I’ve had breakfast today.

16Cp. Borg (2001).

c. I’ve had breakfast, but not today.

Similar points can be made for the other types of the – in Levinson’s terms – I-basedimplicatures tested by Garrett and Harnish (2008). Thus, in the positive variants,cancellation of the IMA seems possible, but not so for the negative variants.

(187) a. It is(n’t) raining.

b. It is(n’t) raining here.

c. It is raining, but not here.

d. ??It isn’t raining, but it is raining here.

(188) a. John broke/didn’t break a finger.

b. John broke/didn’t break his own finger.

c. John broke a finger, but not one of his own.

d. ??John didn’t break a finger, but he did break one of his own.

Thus, both the positive as well as the negative variants given in the (a) examples goalong with the respective IMA explicitly spelled out in the (b) examples. The (c)examples show that it is possible to suspend the specific IMA standardly associatedwith the respective utterance-types in case of the positive variants. Nevertheless,note that intuitively, the (c) examples are understood as still expressing some IMA,although that meaning aspect is much more general than the one standardly associ-ated with the respective utterance-type.17 Thus, looking back at (186c) the speakeris understood to have expressed that he has had breakfast before the day of the ut-terance, in (187c) the speaker is understood to express that it is raining somewhereother than the location where he makes his utterance and in (188c) the speaker isunderstood to express that John did break a finger, but that it wasn’t one of hisown.

Note, however, that whether or not a particular IMA is cancelled seems to dependvery much on the terms in which the putative cancellation is couched. Thus, considerKorta (1997)’s example of an impliciture cancellation, of which he assumes that itis quite acceptable.

(189) I haven’t eaten breakfast, but I do not mean that I haven’t eaten breakfastthis morning. In fact, I have never eaten breakfast.

Thus, what seems to be going on in this example is similar to what we saw for (186c),namely, the IMA today is cancelled, resulting in an understanding of the speaker’sutterance as expressing that he has not had breakfast prior the time of utterance,in other words, he’s never had breakfast. That is why the IMA in this particularcase cannot be cancelled by the sort of continuation offered by Burton-Roberts(2006) (i.e., but I did today), as this would be contradictory to how the sentence isunderstood without the IMA. The same holds for the other examples stated above.That is, (187c) and (188c) are odd, because if the respective IMAs are cancelled,then continuing the respective sentences in the way it is done by Burton-Roberts

17This fact might actually be taken to show that what we have here is in fact not genuinecancellation. See, however, below for a general critique of the cancellation test.

(2006) leads to contradiction in both cases. Thus, if the IMA here is cancelled in(187c), what we get is an interpretation where it is not raining anywhere. If theIMA john’s own is cancelled in (188c), the interpretation would be that John didnot break any finger.

Moreover, trying to apply the cancellation test to scalar implicatures as in (190)and (191) below shows that even with those examples, that is, the ones that Burton-Roberts (2006) wants to establish are cancellable and thus genuine GCIs, the testturns out to be problematic.

(190) a. Some of the students went to the party.

b. Some but not all of the students went to the party.

c. Some of the students went to the party, in fact all of them did.

d. None of the students went to the party.

e. ??Some of the students didn’t go to the party, in fact all of them did.

f. ?? It is not the case that some of the students went to the party, in fact allof them did.

Thus, with the scalar expression some, it isn’t actually possible to have a negativevariant of the sentence in (190a), in which the implicature-triggering scalar someoccurs. Therefore, the negation (190d) of (190a) cannot be said to cancel an implica-ture, because no scalar implicature is triggered in the first place. Trying to formulatea negative variant of (190a), as is done in (190e), which potentially does give rise toa scalar implicature shows that in such a case a continuation which is supposed tocancel the implicature makes this sentence contradictory as well. More specifically,(190e) is acceptable only if considered an example of metalinguistic negation.

Consider now an example involving a numeral.

(191) a. John has got three children.

b. John has got exactly three children.

c. John has got three children, in fact, he has four.

d. ?John doesn’t have three children, he has four.

Here it is easier to formulate a negative variant of the positive sentence that gaverise to the implicature. However, as in the example before, the negative sentenceactually does not give rise to the scalar implicature, so, again, there can’t be anycancellation here. Nevertheless, in the case of (191d), if this is understood as involv-ing genuine truth-functional negation rather than metalinguistic negation, it againis contradictory.

Generally, what is problematic about the idea of cancellation of GCIs is thatthe phenomena categorised as GCIs differ in some crucial aspects and this factactually makes it difficult to apply one overall theoretical test to decide whether aparticular phenomenon should be treated as a GCI or not. Thus, it seems that ifone wants to verify whether a particular putative IMA is characterised by the factthat it arises even when it is not intended by the speaker and that it arises out of aparticular context of utterance, one has to resort to psycholinguistic (and possiblyto neurolinguistic) experimental methods. Having said this, one would still like tohave a theory of how such standardised IMAs arise and what exactly is going on in

those situations in which it is clear that they are not intended by the speaker. Inother words, if one wants to assume a level of utterance-type meaning, the questionis how this figures in the actual interpretation process and on which basis suchstandardised meaning aspects are integrated into (and, as the case may be, deletedfrom) the semantic form of the utterance at hand.

Discourse Interpretation and Information from Conceptual Frames

One possibility is to assume that in interpreting a particular utterance, what thehearer makes use of is information stored in frames that are evoked by the expres-sions the speaker used in his utterance (e.g. Baker et al. 1998, Nissim et al. 1999,Baker et al. 2003, Fillmore et al. 2003, Ruppenhofer et al. 2010). Irmer (2009)suggests this idea in his analysis of bridging inferences18 within the framework ofSegmented Discourse Representation Theory (SDRT) (e.g., Asher and Lascarides2003, Lascarides and Asher 2007). I will briefly describe Irmer (2009)’s approachbelow. Thus, consider the example in (192).

(192) a. John was murdered yesterday.

b. The knife lay nearby.

On a first blush, there is nothing in these two sentences that makes their sequenceovertly coherent. Nevertheless, they are understood as forming a text. More specif-ically, the knife in (192b) is understood as potentially referring to the instrumentused by the unmentioned murderer in his act of murdering. Understanding knife in(192b) as referring to the potential instrument used in the murdering event men-tioned in (192a) is the result of a bridging inference. The question is how sucha bridging inference is drawn. In particular, what is crucial here is the questionon which kinds of information such inferences may be based. One likely source ofinformation is the so-called common ground against which verbal communicationtakes place. This consists of knowledge about the situational context, the discoursecontext as well as (relevant) world knowledge. The situational context involves thepeople participating in the discourse, the time and place at which the discoursetakes place and the objects present (as they may be potential referents for indexicalexpressions). The discourse context includes the propositions expressed or conveyedby preceding utterances. Depending on the general topic of the discourse, aspectsof world knowledge relevant to that topic are also taken to form part of the commonground.

This last point is especially interesting in the present discussion, as the questionarises how what is taken to be relevant information is identified and becomes partof the common ground. Irmer (2009)’s suggestion, at least with respect to bridginginferences, is to assume that during the interpretation of an utterance, specific piecesof conceptual knowledge drawn from frames are integrated into the semantic form ofthe utterance at hand. Before looking at how this works in more detail, let us firstturn to how such frames are characterised in the framework Irmer (2009) makes useof, namely Frame Semantics.

18Note incidentally, that in Levinson’s classification of implicatures, bridging inferences actuallyare treated as a type of I-implicature.

Definition: A Killer or Cause causes the death of the Victim.

Core Frame Elements:

FE description inherited FE sort

Killer The person or sentient entity that Agent sentientcauses the death of the Victim

Victim The living entity that dies as a Patient sentientresult of the killing

Instrument The device used by the Killer to Instrument physicalbring about the death of the Victim entity

Cause An inanimate entity or process that Causecauses the death of the Victim

Means The method or action that the Killer Means state ofor Cause performs resulting in the affairsdeath of the Victim

Non-core Frame Elements: Beneficiary, Manner, Place, Purpose, Time . . .

Lexical Units: annihilate.v, annihilation.n,. . . murder.v, murderer.n, . . .

Figure 5.2: The Killing frame

As the name suggests the leading idea of Frame Semantics is that world knowl-edge is organised in frames. More specifically, it is stereotypical situations thatframes are mental representations of. The ideas of Frame Semantics are imple-mented in the on-line lexical resource FrameNet, which currently consists of about11 6000 lexical units, 960 semantic frames and 150 000 annotated sentences.19. InFrameNet, each semantic frame consists of a number of elements, the frame ele-ments (FEs), some of which are taken to be crucial to the identification of thatparticular frame (i.e., conceptually necessary core frame elements); some of whichare peripheral or extra-thematic (i.e., not conceptually necessary non-core frameelements).

Peripheral FEs mark such notions as Time, Place, Manner, Means,Degree, and the like. They do not uniquely characterize a frame, andcan be instantiated in any semantically appropriate frame. [. . .] Extra-thematic frame elements situate an event against a backdrop of anotherstate of affairs, either an actual event or state of the same type, . . . orby evoking a larger frame within which the reported state of affairs isembedded . . .. (Ruppenhofer et al. 2010, p. 27)

19For an illustration of a semantic frame, consider the Killing frame in (5.2) evoked by thelexical unit murder used in (192.)

Each frame has a definition, which is an informal description of the concept theframe represents. Moreover, the lexical units associated with the particular frame arelisted. A lexical unit is an unambiguous sound-meaning pairing, thus, polysemousexpressions consist of a number of lexical units, each of which might potentially beassociated with a different frame. If a particular lexical unit is used in an utterance,it is said to evoke the frame it is associated with. In addition to the informationlisted above, the relation of a given frame to other frames is indicated.20 Thus, aparticular frame might inherit some or all of the elements of a more abstract frame;a particular frame might be a subframe of a more complex frame, etc. Note also thatthe more abstract frames might actually not have any lexical units associated withthem. Their function is rather to relate the more concrete frames to one another.

The core FEs roughly correspond to the semantic roles involved in the eventualityconceptually represented in the frame. However, as can be seen from example (192a),not all core FEs need necessarily be expressed in a sentence about an instanceof the particular eventuality. On the one hand, this might happen for structuralreasons (e.g., 192a is a passive sentence, thus, the subject of the correspondingactive sentence is not syntactically obligatory), on the other hand, the unexpressedcore FEs might be implicitly understood or inferrable from the context.

To explain how such brigding inferences as in (192) are drawn, Irmer (2009)proposes to integrate the core FEs of the frame evoked by the verb used into thesemantic form build up for the respective utterance. Thus, whereas in SDRT thebasic semantic form for the first sentence in (192) looks as in (193), with the inte-gration of the core FEs from the Killing frame evoked by murder, it looks as in(194).

named(j, john)murder(e)patient(e, j)

e, j | x1, x3

e : Killingkiller(e, x1), x1 =?victim(e, j), named(j, john)instrument(e, x3), x3 =?

For reasons of space, I do not want to introduce the formalism of SDRT in too muchdetail. However, a few notes on the graphical representation used in that theory: inthe upper part of the horizontally divided boxes the discourse referents introduced bythe respective utterance are given, whereas in the lower part conditions holding forthose referents are stated. The conditions might involve properties of discourse ref-erents and their relations to other referents. The vertical line used in the upper part

20For reasons of space, these relations are omitted in 5.2.

of the box in (194) is introduced by Irmer (2009) in order to differentiate betweenwhat he calls regular (i.e., linguistically introduced) and weak discourse referents(i.e., introduced through evoked frames). In this particular case, the Victim dis-course referent introduced by the Killing frame has already been identified with thelinguistically given discourse referent j, as this is taken to have the property of beinga Patient and the thematic role Victim actually is a specific type of Patient.21

The Killer and Instrument roles, in contrast, remain unspecified (indicated by thenotation x1 =?, x3 =?). However, in (192), the second sentence linguistically intro-duces a discourse referent that may be identified with the Instrument role. Thisis because the expression knife evokes the Weapon frame with the core-FE Weapon.However, and as Irmer (2009) points out, although knives can be used as weapons,they are not necessarily designed as such. ‘The only knowledge we can use is thatthere is no clash of sorts: both knives and killing instruments are physical entities’(Irmer 2009, p. 203). Nevertheless, identifying the knife introduced in the secondutterance with the weak discourse referent x3 that is a murder-instrument from thefirst utterance is plausible and results in a consistent interpretation. Having saidthis, generally, what is even more important for the identification of discourse refer-ents to be at all possible is that a particular discourse referent introduced earlier isactually accessible for the newly introduced discourse referent. Without going intothe details here, accessibility is established subject to the discourse relation assumedto hold between the utterances under consideration.22

Free Enrichment and Information from Conceptual Frames

As already said, Irmer (2009) restricts the elements used from the frames to thosethat are characterised as core FEs. In fact, he only concentrates on the core FEs offrames evoked by the main verb in the respective utterances. However, all contentwords evoke frames and generally it can be assumed that the information in theindividual evoked frames in some way or other enters into the interpretation of theutterance at hand. In addition, there is a case to be made for allowing core as wellas non-core elements of evoked frames to figure during the interpretation process.In that case, the information provided by the evoked frames may be considered toform the minimal background, as it were, against which the utterance in questionis interpreted. In this connection, recall that Irmer (2009) introduces a two-waydistinction between regular and weak discourse referents. However, as he himselfsuggests, a more fine-grained differentiation between discourse referents with respectto their salience might be useful. A differentiation of discourse referents based ontheir salience in a particular context would allow for an assumption concerning theoperation of the processes resolving underspecification of discourse referents: onlysuch weak discourse referents down to a particular degree of salience will be triedto get resolved, identified with others, etc.

Turning now to the IMAs taken to be contributed to an utterance’s meaning

21In this particular case, this relation is due to the inheritance relation between the Killing

frame and its parent frame Transitive action, which, amongst others, has the two core-FEsAgent and Patient, which are inherited by the Killing frame as the core-FEs Killer and Victim,respectively.

22In the case of (192), this being Background.

by the process of free enrichment: recall once again that the meaning aspects con-tributed are not in any way linguistically mandated; they are unarticulated con-stituents. Recall further the examples Wilson and Sperber (2000) use to argueagainst Stanley (2000)’s hidden indexical approach.

(93) I’ve often been to their parties, but I’ve never eaten anything [there].

(94) I must wash my hands. I’ve eaten [using my hands].

Wilson and Sperber (2000) use the above examples to argue that if one assumesa hidden indexical approach, one would be forced to posit such variables in thesemantic forms of each of the second sentences in the respective examples above. Ialready indicated in chapter 3 that I do not think this argumentation goes through.This is because in these examples what we have is actually small texts and theIMAs assumed to be expressed by the speaker in the respective second sentences ofthese texts can be taken to ‘arise’ due to the particular discourse context in whichthe respective utterance takes place. The question, of course, is, what is meantby ‘arise’. Here, I think, we can follow Irmer (2009)’s suggestion that during theinterpretation of utterances, information from evoked frames is integrated into thesemantic form build up for that utterance.

Thus, let us concentrate on (93) first, starting with the first utterance. Unfortu-nately, FrameNet does not provide information on the frame(s) associated with thecopula, however, intuitively the sense expressed in (93) could be paraphrased withgo to, which does have an associated frame in FrameNet, namely the Attending

frame, with the core-FEs Agent and Event. The noun party, in turn evokes theSocial Event frame, with the core-FEs Attendee and Social event. Assumingthat these latter two roles can be traced back to the roles Agent and Event23, re-spectively, by some inheritance relation, the more specific roles can be identifiedwith the more general roles and integrated in the representation of the utterance’smeaning. Note that in addition to its core-FEs, the frame Social Event has quitea number of non-core FEs, amongst them Refreshment as well as Place.

Turning now to the second sentence in (93), the verb to eat, in turn, evokes theIngestion frame (see figure 5.3), with its core-FEs Ingestor and Ingestibles,as well as, amongst others, the non-core FEs Time, Place and Manner. Intuitively,the place at which the eating-event expressed in the second utterance takes place isinterpreted as identical to the place of the respective party. This intuition can becaptured by assuming that the two non-core FEs place of the two respective framesare identified, although not further specified.24 Moreover, it seems that the informa-tion that a party stereotypically involves refreshments is necessary to understand thediscourse in (93) as coherent in the first place. Thus, the core-FE Ingestibles fromthe Ingestion frame evoked by the second utterance is related to the non-core-FE

23Unfortunately, FrameNet as yet has not specified the inheritance relations of the Attending

frame to other frames.24This identification is based on the principle introduced by Cohen (2006) Equality by Default:

Two discourse referents are assumed to be equal, unless it can be proven that they are distinct.As this is a very general principle and as such prone to overgeneralisation, it should be mentionedthat there exist quite a number of more specific principles that constrain the applicability of thisparticular principle. (See, e.g., Irmer (2009) for more information on the interaction of this principleand more specific ones.)

Definition: An Ingestor consumes food or drink (Ingestibles), which entailsputting Ingestibles in the mouth for delivery to the digestive system. This mayinclude the use of an Instrument. Sentences that describe the provision of food toothers are NOT included in this frame.

Core Frame Elements:

FE description inherited FE sort

Ingestible The Ingestibles are the Substance n. s.entities that are beingconsumed by the Ingestor

Ingestor The Ingestor is the person Agent Sentienteating or drinking

Non-core Frame Elements: Degree, Duration, Instrument, Manner, Means,Place, Purpose, Source, Time

Lexical Units: breakfast.v, consume.v,. . . drink.v, eat.v, . . .

Figure 5.3: The Ingestion frame

Refreshments of the Social Event frame evoked by the first utterance.25 Thus, itis the background information that a party might involve refreshments that makesit appropriate to express some sort of contrast by using the expression but, althoughovertly what is talked about in the individual sentences (parties in the first, eatingin the second) is not necessarily related. This can be seen when considering, e.g.,Asher and Lascarides (2003)’s characterisation of the discourse relation Contrast.Generally, the authors argue that the relation of Contrast between two utterancesα and β holds, if β involves one of the expressions but, however or in contrast. Thiscondition is trivially fulfilled by (93). In addition, Asher and Lascarides (2003) arguefurther that the two utterances be structurally similar while semantically dissimilaror contrasting. Arguably, the condition that the semantic content of the two utter-ances be contrasting is only fulfilled by assuming that attending a party normallyinvolves partaking of refreshments.

Let us now turn briefly to the interpretation of (94). Intuitively, the relationbetween the two utterances is one of Explanation. Thus, the second utterance inthis small discourse is understood as providing an explanation for the propositionexpressed in the first. If the speaker’s hands introduced by the first utterance areidentified as the instrument used for eating, the explanation becomes plausible.In fact, the Ingestion frame evoked by the verb to eat does include Instrument

25Although this is not specified in FrameNet, the idea would be that Ingestibles andRefreshments are related with respect to their semantic sorts. The problem here is that thetwo frames the respective FEs are part of are not in any inheritance relation, which actually makesit difficult in FrameNet to relate them to one another other then by intuition.

amongst its non-core FEs. Needless to say, as in example (192), identifying theaforementioned speaker’s hands as the instrument used in the eating-event requiresrecourse to world knowledge that it is possible for a person to use his hands as theinstrument in an eating event.

Let us now turn to examples involving IMAs that consist of only single sentences.For example, the speaker’s use of breakfast in (186a) arguably again evokes the In-

gestion frame, where one of the non-core frame elements (FEs) is the time at whichingestibles are ingested.26

(186a) I’ve had breakfast.

In fact, according to FrameNet, in this particular case, the expression to have alsoevokes the Ingestion frame. The fact that the sentence is in the past tense maylead to a partial specification of the non-core FE Time as having taken place at somepoint before the time of utterance. By taking into account knowledge about usualpractices in the culture at hand, namely that it is quite usual for people to havebreakfast once every day, the relevant time span may be restricted to the morninghours of the day of utterance.

Consider now the interpretation of an utterance of (188a) – another example ofa meaning aspect, according to RT, due to free enrichment.

(188a) John broke/didn’t break a finger.

In an utterance of (188a), using the expression break evokes a frame Experience

bodily harm with the core FEs Body part and Experiencer, while finger evokesa frame Observable Bodypart with the core FEs Body part and Possessor. Ob-viously, the referent of finger can be identified as playing the role of body part,whereas John is an accessible discourse referent and an appropriate value for thePossessor and the Experiencer-variable, where the former is identified with thelatter, the latter being retained in the representation of the utterance’s meaning asit is more specific.27

Consider next example (78), repeated below.

b. You’re not going to die from this cut.

Here, the expression to die evokes the death frame, which has as one of its non-coreFEs the cause of death. From the situation in which this utterance is said to bemade, the cut the addressee has suffered might potentially be the cause of death, ifit is severe. Now, it seems that what the speaker expresses is that the addressee’scut is not going to be the cause of his death.

Incidentally, this remainds me of another example, mentioned by Bach (2000)and involving the positive variant of (78a). The situation is one where an ontologist

26Note that FrameNet only lists the verb to breakfast, which, however, does evoke the Ingestionframe.

27Once again it should be noted that the two frames Experience bodily harm and Observable

Bodypart are not in any inheritance relation. However, the informal characterisation of the FEExperiencer in the former frame entails that the referent filling this role be a possessor of thereferent filling the FE body part.

is talking to one of his patients about that patients’ cancer. The patient wants afrank prognosis and the ontologist utters (195a).

(195) a. You’re going to die, but not from this cancer.

b. You’re going to die.

c. You’re going to die from this cancer.

Here it seems that the speaker is alluding to the common sense knowledge that allliving beings must die, the knowledge that death might have an identifiable causeas well as the common sense knowledge that having cancer is a potential cause fordeath. It is likely that in the particular utterance situation assumed the addresseewill interpret the first part of (195a) as expressing (195c). That is, the verb todie again will evoke the death frame and, due to the particular situation in whichthis utterance takes place, the non-core FE element cause will get specified as theaddressee’s cancer. However, this is taken back in the second part of the ontologists’utterance. In other words, the fact that a death might have a cause remains, butin this situation the cause is not specified, or rather, what is expressed is that thepatient’s cancer is not going to be the cause for his death.

Finally, an utterance of (187a) is normally understood as (187b), because theverb to rain evokes the Precipitation frame amongst whose core-FEs are time

and place.

(187a) It is(n’t) raining.

(187b) It is(n’t) raining here.

Since the sentence used is in the present tense and progressive aspect, the time

variable of the Precipitation frame can be filled by the time of the context ofutterance, which can be assumed to be given anyway as it is needed for the resolutionof deictic expressions. The same goes for the place variable. Since the context ofutterance provides a value for that place variable, this will be used, unless thereis some information in the discourse background that makes such an identificationinappropriate.

A further interesting point to note concerns a potential interaction of the expres-sions used in an utterance when it comes to deciding which lexical unit to assumein those cases where an expression is polysemous. Thus, recall once more example(186a), in which the expression to have is used. Actually, in the FrameNet database,there are several entries for this verb, each of which is associated with a differentframe.

(186a) I’ve had breakfast.

have.v Have associated

have.v Inclusion

have.v Birth

have.v Ingestion

have.v Possession

However, as already indicated, I think it is likely that the noun breakfast is asso-ciated with the Ingestion frame as well and this fact will influence the choice ofthe appropriate lexical unit for the expression to have (i.e., that also evoking theIngestion frame). The same can be said for Ruhl (1989)’s example (52).

(52) The thief took the jewels.

The verb take is highly polysemous and, in addition, occurs as part of quite a numberof phrasal verbs.

take.v Has as requirement

take.v Taking time

take.v Capacity

take.v Taking

take.v Ride vehicle

take.v Ingest substance

take.v Removing

take.v Bringing

take.v Conquering

In one of its numerous readings, take is associated with the Taking frame. Theco-occurrence of the two expressions take and thief, where the latter evokes a frameTheft, which actually inherits from the Taking frame, might lead the interpretationprocess to assume that the lexical unit the expression take in this particular utteranceis used to express, actually is the one that is associated with the Taking frame.

Having analysed the explicature/impliciture examples above with the help of(more or less) general frames and their elements, the question arises where thenotion of a level of utterance-type meaning comes in. Note that the examples Ihave analysed were mostly such that there was no evidence that the IMAs were notappropriate. As a last point, I want to make a suggestion as to what happens ifa sentence like (188a) is interpreted in a context in which it is clear that it is notJohn’s finger that is broken.

(188a) John broke/didn’t break a finger.

As we saw from the experimental results obtained by Garrett and Harnish (2008),it seems that the specific IMA ‘his own finger’ is initially interpreted, even thoughthe forgoing context does not support it and it, thus, has to get cancelled. A firstthing to note is that actually, the dispreferred but appropriate reading in which Johnbreaks someone else’s finger involves a sense of break that differs from that used inthe preferred reading, thus evoking another frame, namely the frame cause harm

with the core-FEs Agent, Body part, Cause and, crucially, Victim. Whereas inthe Experience bodily harm frame it is the Experiencer that is identified as thePossessor of the Body part under discussion, in the case of the cause harm frame,it is the Victim that is identified as the Possessor of the Body part. The ques-tion is how the preferred reading of (188a) can be explained. Note that in thisparticular case, co-occurence of the expression to break with finger is not of anyhelp, as the frame Observable bodypart evoked by the latter is plausible with

both the relevant readings of break. However, an important fact that I have notelaborated on is the fact that discourse interpretation is taken to be subject to cer-tain constraints or principles.28 One of these constraints is Asher and Lascarides(2003)’ meta-rule ‘maximise discourse coherence’, which consists of four parts, onebeing ‘resolve underspecifications’ and the one of special interest in the present dis-cussion. Thus, note that if to break is assumed to evoke the frame cause harm,the Possessor core-FE of the Observable bodypart frame evoked by finger wouldhave to get identified with the Victim core-FE of the cause harm frame, where thisactually remains underspecified in an utterance of (188a). In contrast, if the frameExperience bodily harm is used, the Observable bodypart’s Possessor core-FEwill be identified with the former frame’s core-FE Experiencer which is actuallyassigned the value john. Thus, if the frame Experience bodily harm is used ratherthan the frame cause harm, more underspecifications will get resolved, which makesthis interpretation preferrable.29

In a sense, then, the level of utterance-type meaning – at least as far as the kindsof meaning aspects are concerned we looked at here – boils down to the preferredresolutions of unspecified potential discourse referents and their relations. A con-sequence of this view is that GCIs – at least those traditionally taken to be basedon Levinson’s I-principle – actually are not implicatures at all but rather IMAs,integrated into the semantic form of an utterance during its interpretation. Thischaracterisation, of course, only holds if the primary characteristic of implicaturesgenerally is taken to be that they arise on the basis of propositions. If the primarycharacteristic is rather taken to be the idea that implicatures arise on the basis ofparticular conversational maxims or principles, then the IMAs discussed above mightstill be characterised as implicatures, as their integration into the semantic repre-sentation of an utterance, as we saw, is guided by overarching principles on generaldiscourse interpretation. Be that as it may, note that the approach sketched herecaptures Recanati’s intuition that free enrichment is a primary pragmatic process,although one characteristic of these processes is that they do not need a propositionas a basis and free enrichment normally is taken to be based on propositions. Infact, with an approach that assumes that such IMAs are the result of the integrationinto the semantic form of an utterance of information from frames evoked by indi-vidual expressions used in an utterance, the process of free enrichment is no longercharacterised as applying to propositions.

Consequences

What makes an approach to utterance interpretation using frames interesting is thefact that it allows to keep the semantics of the expressions used in an utteranceminimal, while still allowing particular meaning aspects to figure in the interpreted

28as, e.g., the principle Equality by Default mentioned below.29Note also that the preferred reading involves a model of the world that is more minimal

than the model for the dispreferred reading, as the former involves a smaller number of discoursereferents than the latter (i.e., the FE victim). As Irmer (2009, secion 5.2) points out, the notionof minimality of models plays an important role in discourse interpretation. Thus, the preferrencefor a particular reading of an ambiguous expression may be traced back to the fact that the modelfor that reading is more minimal than the model for the alternative reading(s).

utterance, where those meaning aspects may or may not get specified and wherethey are defeasible. Moreover, assuming that expressions evoke particular framesthat provide a restricted number and type of elements as potential discourse referentsrestricts the search space for values for newly introduced discourse referents in anenfolding discourse. In addition, and as is also argued by Irmer (2009), introducingsuch variables for potential discourse referents does not mean that they have toget specified in the course of utterance or further discourse interpretation. Thatis, it is quite possible that such variables will remain unspecified. Conversly, as wesaw above, it is always possible to clarify the exact value assigned to some frame-related element to the point where the element may in fact stay underspecifiedafter all.30 Crucially, such an approach allows to characterise the process called freeenrichment as contributing meaning aspects to the meaning of an utterance, wherethose meaning aspects are not the values of semantically provided slots. However,free enrichment can still be seen as a type of saturation process and thus, in asense restricted, in that it provides values to variables introduced into the utterancecontext by frames that are evoked by the expressions used in a particular utterance.Moreover, free enrichment can be assumed to actually not rely on a considerationof the speaker’s potential intentions in making the utterance he did.

In a sense, then, assuming that the interpretation process makes use of informa-tion from frames is to eliminate hidden indexicals from the semantics of expressionsto reintroduce them into the conceptual information used by the interpretation pro-cess. However, this seems to be indicated anyway by the fact that backgroundinformation is generally assumed to play an important role in determining evenwhat is traditionally called the literal meaning of an utterance. Moreover, sinceframe-related elements are integrated into the semantic form of an utterance whilethis is built up by the interpretation process, one would actually expect those ele-ments to have a truth-conditional effect. However, the exact effect they will haveon a particular occasion depends on the specificity of the actual value assigned tothem.

5.2.2 Context, Semantic Interpretation and the Semantics/Pragmatics Distinction

As we saw, especially in chapter 3, approaches to utterance interpretation differ intheir conception of the semantics component in the language faculty. In particular,one cause for dissonance is the question of whether what the semantics compo-nent delivers should be taken to be propositions or only sub-propositional semanticforms. Approaches assuming that semantics only delivers sub-propositional forms doso based on the ideas that, generally, semantics only deals with context-independent

30Thus, recall the example sentences such as repeated below, where the preferred interpretationof the respective IMA is taken back but the general element this IMA is taken to be the value ofis still felt to be present.

(i) It is raining, but not here.(ii) I had breakfast, but not today.(iii) John broke a finger, but not one of his own.

meaning aspects of expressions and that those context-independent meaning aspectsare not sufficient for the determination of the proposition actually expressed by therespective sentence in a particular utterance situation. Approaches assuming thatsemantics does in fact determine the proposition semantically expressed by somesentence do so on the basis of the assumption that – contrary to its traditional char-acterisation – semantics in fact is not concerned with context-independent meaningonly. Rather, semantic processes of meaning composition take into account partic-ular types of contextual information. However, the approaches differ in what theytake the nature of these pieces of contextual information to be (i.e., objectively at-tainable contextual information for Borg; contextual information partly influencedby assumptions about the speaker’s intentions in making his utterance in Cappe-len and Lepore’s approach). In any case, contextual information is assumed to beintegrated into the semantic form of an utterance, subject to the presence of somelinguistically given element that licenses such an integration. That is, Borg as wellas Cappelen and Lepore do not assume that unarticulated constituents should betaken to be part of a sentence’s meaning relative to a context of utterance.

Having said this, it should be noted that the information provided by linguisticsemantics alone may not be sufficient to capture what a reader or hearer understandswhen she is presented with a sentence out of context. Thus, even if I do not knowwho John is, when reading a sentence such as John broke a finger, I will understandwhoever the name John refers to to have broken one of his own fingers. It is clear thatthe relevant IMA is not semantically provided and as such is defeasible. However,it arises without a particular context being necessary. Thus, even if applying thetraditional criterion for semantic meaning – that it is the meaning understood whenpresented a sentence without any contextual information (Katz’s anonymous lettersitatuation) – this is not strict enough to exclude so-called unarticulated constituents.Thus, the IMA ‘his own’, in Levinson’s terms, is a preferred interpretation that isdetermined without any consideration of the potential speaker’s intentions or a likelycontext of utterance being necessary.

Be that as it may, the important point to note is that if it is propositions thatthe semantics component of the language faculty delivers, then that component canalso be assumed to be involved in what is called the semantic interpretation of thosepropositions. If, however, the semantics component only delivers sub-propositionalsemantic forms, it cannot be involved in the semantic interpretation, since thesesemantic forms do not yet allow for an evaluation concerning their truth or falsity.In a sense, then, the question of the semantics/pragmatics distinction seems to boildown to whether the process of semantic interpretation should be assumed to bepart of the semantics component.

Recall that both in Borg’s as well as Cappelen and Lepore’s approaches semanticinterpretation is taken to be part of the semantics component. However, the truthconditions delivered and semantically interpreted by that component are only liberaltruth conditions rather than the actual truth conditions assumed to be involved inthe particular utterance situation. However, in both approaches it is assumed thatthese liberal truth conditions have an important role to play. Thus, Borg assumesthat they establish the inferential relations between the expressions of a language,whereas Cappelen and Lepore assume that they are the minimal level of meaning a

hearer can fall back on in cases where for some reason or other it is not quite clearwhat the actual truth conditions should be taken to be.

Turning back to the approaches that assume of the semantics component thatit ‘only’ delivers sub-propositional forms: recall that RT in particular does not evenrecognise a level of meaning corresponding to Borg or Cappelen and Lepore’s lib-eral truth conditions as playing any role whatsoever in the interpretation process.Rather, RT assumes that it is explicatures, that is, pragmatically enriched semanticforms that play an important part in the interpretation process, as they can be takento constitute the level of meaning that is the basis for further pragmatic inferencesas well as that level of meaning to the expression of which the speaker is bound.Thus, even if one were to assume that what the semantics component of the lan-guage faculty delivers are propositions, rather than propositional radicals, semanticinterpretation would still have to be assumed to take place again once the actualtruth conditions have been determined. In other words, even if an evaluation as tothe truth or falsity of liberal truth conditions might be possible within the semanticscomponent of the language faculty, according to all approaches discussed above, itis not possible for the actual truth conditions, thus, those truth conditions whichcan be considered as the ones really mattering in actual communication situations.

Note that Levinson’s as well as RT’s conceptions of semantics seem like at-tempts at preserving the intuition that semantic interpretation should be part ofsome sort of semantics system. Thus, whereas Levinson assumes that the semanticscomponent in the language faculty is intruded upon by a restricted number of prag-matic processes, leading him to differentiate between two semantic sub-systems, RTassumes that there actually are two different semantic systems, namely linguisticvs. real semantics, where the latter actually is the ‘language of thought’ and thusthe locus of semantic interpretation. What Levinson’s conception better captures isthe fact that the information contributed by the pragmatic processes intruding intosemantics is essentially defeasible, non-monotonically determined conceptual infor-mation. As such these processes are crucially distinct from the kinds of processesinvolved in semantic interpretation, which essentially are monotonic. That is, thedrawn conclusions as such are not defeasible; only the premises, as it were, on whichthey are based might turn out to be false (as the premises may involve pragmaticallydetermined IMAs).

For example, suppose Mary knows that (196d) holds and in her conversationwith Jane concerning how to spend their time Jane answers Mary’s question (196a)as in (196b) and Mary interprets this as expressing (196c). On the basis of thisinformation, she will be able to draw the conclusion (196e).

(196) a. Mary: How about playing tennis?

b. Jane: It is raining.

c. It is raining in Leipzig at the moment.

d. If it is raining at location li at time tj, it is not possible to play tennis atlocation li at time tj

e. It is not possible to play tennis in Leipzig at the moment.

This conclusion follows monotonically from the two premises (196d) and (196c).However, the determination of the proposition (196c) functioning as one of the

premises actually involves pragmatically determined meaning aspects. Nevertheless,once the interpretation process has decided on the particular representation of themeaning of the utterance at hand, semantic interpretation will straightforwardlyapply to that proposition, following the standard rules. If the thus determinedproposition is interpreted as true, it can function as a premise in an act of deductivereasoning as exemplified above. However, it might turn out that the pragmaticallydetermined meaning aspect featuring in the supposed proposition expressed by Janefor some reason or other is inappropriate and consequently has to be changed. Inthat case the premise (196c) assumed in the reasoning process would change also,leading to a different conclusion. However, it would not affect the nature of thereasoning employed as such.

Viewed from the perspective of the nature of processes involved in the determi-nation of the proposition expressed by a sentence and its semantic interpretation,Levinson’s view (repeated below in figure 5.4) of the semantics component as al-lowing the intrusion of pragmatic processes in the determination of an utterance’smeaning is sensible.

SEM 1: sentence meaning

Pragmatics

what is meant

PCIspeech acts. . .

sub-propositional

fully propositional

+ narrow context

+ broad context

SEM 2: what is said

pragmaticintrusion

Figure 5.4: Levinson’s view

However, recall Bierwisch’s suggestion that an evaluation of speech acts withrespect to their truth or falsity is possible and useful as well (cf. section 4.2). Thus,consider again example (174), repeated here as (197a), where this is said by A to Band B generally thinks that A has a low opinion of B.

(197) a. This is a very difficult tour.

b. The Coast to Coast Cycling Path is a very difficult tour to do.

Suppose further that what B takes A to have expressed (and what A did in factintend to express) with his utterance is (197b). On the basis of B’s beliefs concerningA’s opinion of B, B might take A to have intended to convey that B won’t be able tomake the relevant tour or that it is too difficult for B, etc. More specifically, B mightdetermine on the following proposition: ‘A believes that the Coast to Coast CyclingPath is too difficult for B’. If B assumes this proposition to be true – and based onhow he arrived at it, he trivially will – this will have consequences on his knowledge

base of social interaction. Thus, on the one hand, it will strengthen his assumptionsconcerning A’s opinion of him as well as possibly lead him to take active steps tochange what he believes is A’s opinion of him, etc.

Be that as it may, the crucial point is that what is called semantic interpretationactually has to be assumed to take place also at the level of meaning what is meant.Note that the nature of the process that determines the truth or falsity of a givenproposition is the same regardless of whether that proposition corresponds to thelevel of sentence meaning, utterance meaning or communicative sense/what is meant.This fact might be captured by expanding the semantics component to also includewhat is meant (possibly as Semantics 3 as indicated in figure 5.5), with pragmaticprocesses – although arguably of a different nature – also intervening or intrudingbetween what is said and what is meant.

SEM 1: sentence meaningpragmaticintrusion

pragmaticintrusion

SEM 3: what is meant

PCIsspeech acts. . .

sub-propositional

fully propositional

+ narrow context

+ broad context

SEM 2: what is said

Figure 5.5: Semantics throughout

However, this view raises the question whether semantics as conceived here gen-erally should be viewed as belonging in its entirety to the language faculty. It isunlikely that anyone would want to assume that the processes needed to determinewhat is meant should be viewed as intruding into a linguistic semantics component.Thus, after all, RT’s conception of there being two seperate, as it were, semanticssystems seems to better capture the similarities and differences between the pro-cesses assumed by Levinson to belong to one or the other of two subsystems ofsemantics.

Thus, whereas linguistic semantics deals with the lexical meanings of linguisticexpressions and their composition, it is in real semantics that propositions generallyare semantically interpreted. Linguistic semantics and real semantics are related,however, in that the former is assumed to ‘inherit’ its properties from the latter.

In fact, an approach is possible that denies that there is something like linguisticsemantics separate from real semantics at all. Thus, Burton-Roberts (2005, 2007a)argues that linguistic semantics is in fact empty. This follows from his Represen-tational Hypothesis (RH) (Burton-Roberts 1994, 2000, Burton-Roberts and Poole2006, Chng 1999), which is the assumption that phonological forms only represent –rather than encode – the concepts or thoughts a speaker-hearer might want to com-

LinguisticSemantics

lexical meaning

sentence meaning

Real Semantics explicature

Real Semantics

Pragmatics

saturation processesfree enrichmentad-hoc concept form.. . .

what is meant

sub-propositional

+ broad context

Figure 5.6: RT’s differentiation of linguistic vs. real semantics

municate. In other words, there are no such objects, traditionally called e.g., words,morphemes, where those objects ‘consist’ of phonological and semantic material.Crucially, linguistic expressions are not considered to be sound-meaning pairings.Rather, phonological forms represent particular concepts by convention. That is,there is a set of convention rules that govern the particular range of the representa-tion function the individual phonological forms in a language have.

Note that under the RH, the view of what a language is also differs cruciallyfrom the view taken by approaches that presuppose the existence of Saussuriansigns. Thus, Burton-Roberts (2007b) argues that languages are and differ fromone another in their phonologies. In other words, different languages are simplydifferent phonologies.31 In contrast, it is the conceptual structures phonologicalforms represent that are taken to have semantics and syntax. These conceptualstructures are part of the Language of Thought.

If one accepts the view of phonological forms as merely representing concepts,then it no longer makes sense to speak of ‘encoded meaning’. Note that such a viewis in line with the assumption that the semantic form of an expression constitutesa pointer to the set of possible readings of that expression. As such, the semanticform is empty of meaning; its sole function is to link phonological forms and theconcepts they represent. However, if that is how pointers are characterised thenthere actually is no reason why they should be called semantic forms at all. Rather,hearing a particular string of sounds and identifying that string as an instance/arealisation of a particular phonological form is sufficient to trigger the activation ofassociated concepts.

However, if phonological forms, in a sense, are directly linked to and activateassociated meanings, several questions arise. One question is how to explain thedifferences found in the interpretation of homonymous vs. polysemous expressions.Recall that in the case of homonymous expressions, frequency of occurrence of a

31It is important to stress, however, that under the RH, phonology has to be taken to be muchricher than traditionally assumed. Thus, for instance, it also includes the convention rules governingthe representation of concepts by phonological forms.

Phonology phonological forms

conceptsLanguage of Thought

point to

semantic composition. . .

what is meant

Figure 5.7: Only real semantics in Burton-Roberts’ approach

given meaning as well as the particular contextual circumstances influence the acti-vation of that meaning, whereas in the case of polysemous expressions, these factorsdo not seem to play any role. One possibility to explain this difference in a RH-framework is to assume that hearing the realisation of a particular phonological formthat is used to represent concepts from distinct families of concepts actually leads toan activation of both these distinct concept families. The assumption then, I guess,would have to be that initiating and retaining the activation of two distinct conceptfamilies is effortful and cognitively taxing and therefore the interpretation processhas to ‘decide’ which of the activation paths, as it were, to retain and which toinhibit, leading to the kind of competition effect noted by Klepousniotou (2002).32

In the case of phonological forms pointing to concepts that are semantically relatedand traditionally were associated with one expression characterised as polysemous,in contrast, only one general concept family is activated and thus, there is no com-petition.

Another question is what happens to the notion of semantic composition and,relatedly to the idea of compositionality of meaning. Generally, assuming that lin-guistic semantics is in fact empty has as one consequence that the process of semanticcomposition can no longer be viewed as essentially linguistic. That is, if pointersonly point, then there is in fact no lexical meaning. If pointers do not have logicalproperties, they cannot be combined by operations taking logical properties intoaccount. Thus, semantic composition cannot be linguistic; it has to be purely con-ceptual. A hearer interpreting an utterance identifies the phonological forms thespeaker realised and those phonological forms in turn activate the concepts theymay be used to represent by convention. Pragmatic processes will be necessary todetermine on the basis of the particular utterance situation exactly which conceptsthe speaker intended the phonological forms he used on that occasion to represent.

32Recall that Klepousniotou (2002) found that subjects were faster in their lexical decisions forpolysemous as compared to homonymous target words. She explains this result by assuming thatin case of homonymy, two distinct lexical entries are initially activated and compete with oneanother.

The concepts themselves do have logical properties and those properties will deter-mine how they are combined, finally terminating in a full proposition. Thus, undersuch a conception, the main task of pragmatic processes is, in a sense, disambigua-tion. That is, from the phonological forms the speaker used, the convention rulesgoverning their use as well as the situation in which he used them, the hearer hasto infer which concepts the speaker intended those phonological forms to represent.To be sure, in those cases where a phonological form is used to represent a conceptwhere this representation is not governed by a convention rule, pragmatic processesare involved also in either determining which concept is in fact represented or increating such a concept.

Note that – as far as I am aware – at present, there exists no clear exposi-tion of how exactly the interpretation of utterances works under the RH.33 As aconsequence, there are quite a number of unresolved issues concerning utteranceinterpretation in this framework. For instance, it is not quite clear to me at whichstage during interpretation considerations of speaker intentions come into play and,thus, whether a differentiation is made between the concepts phonological formsare taken to represent in a given utterance situation and the concepts the speakeris taken to have intended to communicate in that situation. In other words, it isnot clear whether a differentiation of what is said/utterance meaning and what ismeant/communicative sense is assumed, as argued for in chapter 4. Another worry iswhether the assumption behind the RH that phonological forms represent conceptsby convention can capture the fact that convention actually is a gradual property, asargued in chapter 2. Finally, it seems that the empirical results supporting the viewof underspecification of linguistic meaning reported in chapter 2 are not compatiblewith the RH hypothesis.

Generally, looking at different approaches to the differentiation between seman-tics and pragmatics, it seems that it is not so much a question of whether contextplays a role and if so which for that differentiation. Rather, it is the difference in thenature of the processes employed that is to be captured by the differentiation. How-ever, since, at the same time, the differentiation of semantics from pragmatics goesalong with assumptions concerning the relations of these systems to the languagefaculty in general, the problem arises of how much of the semantic or pragmaticprocesses to allow to figure in the language faculty. In any case, it actually seemsthat there is no way of differentiating between two systems involved in the interpre-tation of utterances, if these two systems have to be assumed to strictly follow oneanother. Rather, the different types of processes employed seem to interleave.

Thus, pending a clearer exposition of the RH position, I will assume the followingversion of the semantics/pragmatics distinction. Taking into account the principleof economy concerning the size and nature of a semantics component within the lan-guage faculty, I will assume, following Bierwisch, Dolling and RT, that it providesunderspecified semantic forms as the lexical meaning of expressions as well as therules for their combination, following the principle of compositionality. Semanticinterpretation, then, does not take place within the linguistic semantic component,but rather in what RT calls real semantics (Bierwisch’s conceptual system). How-ever, in the spirit of Dolling’s multi-level model of meaning, there are various levels

33Thus, the only fragmentary graphic illustration of that approach in figure 5.7.

LinguisticSemantics

lexical meaning

sentence meaning

utterance meaning

Real Semantics

saturation processesfree enrichment. . .

communicative sense

sub-propositional

Pragmatics

interprets

+ narrow context

+ broad context

Figure 5.8: My view on the semantics/pragmatics distinction

of meaning at which semantic interpretation might take place. And these levels ofmeaning are attained by pragmatic processes enriching the sub-propositional seman-tic form determined by the linguistic semantic component in a number of steps. Forreasons given above, I assume that a differentiation between something like the levelof utterance meaning and the level of communicative sense is useful and neccessary.As for the question of which criterion should be assumed to differentiate between theprocesses involved in determining the utterance meaning of an utterance and thosedetermining the communicative sense of the speaker making that utterance: I aminclined to follow Bierwisch’s suggestion that it is the consideration of the speaker’sintentions in making the utterance she did. Thus, even if assuming that in additionto the traditionally assumed processes something like free enrichment needs to beassumed to determine the actual truth conditions of an utterance, in section 5.2.1 Isuggested a way of modelling that process such that it need not make reference topotential speaker’s intentions in making the utterance she did.

Let me adapt a metaphor used by Levinson (2000) in his characterisation ofDiscourse Representation Theory, were he writes the following.

. . . there is a common slate, a level of propositional representation, uponwhich both semantics and pragmatics can write – the contributions maybe distinguished, let’s suppose, by the color of the ink: semantics inblack, pragmatics in red. Semantics and pragmatics remain modular“pens” as it were: they are separate devices making distinctively differ-ent contributions to a common level of representation. (Levinson 2000,p. 193)

When Levinson writes semantics, he of course refers to his conception of the lin-guistic semantics component, consisting of two subsystems. However, I think themetaphor works equally well for a conception of semantics as actually two separate,though related systems that interact with a pragmatics system. Thus, it is the con-tributions of linguistic semantics and pragmatics that are interpreted by real seman-tics at various steps during utterance interpretation. While the linguistic semantic

Summary 251

contribution is stable, the pragmatic contributions might change, since they are de-feasible. Moreover, different pragmatic processes may be seen as adding successivelymore material, while at particular points during that process the representation issemantically interpreted by real semantics.

As concerns the objection that already for the resolving of reference of (true)demonstratives appeal to speaker intentions might be necessary: I do not have anexplicit treatment to offer, however, there are two points that I would like to make.First, it seems to me that a speaker is restricted in his use of a true demonstrativeto making reference to entities that the hearer can objectively identify in or accom-modate to the discourse context, precisely because she cannot rely on the hearerto make use of reasoning about the speaker’s intentions in making an utterance infixing those referents. Thus, in a situation were A and B are at the zoo, have beenlooking at the apes and have been talking about what funny things apes are proneto do, A’s utterance of Look at that, accompanied by a pointing gesture towardsa particular ape will probably be interpreted by B as meaning ‘Look at what thatape I’m pointing at is doing’. However, if A and B have been talking about somecompletely different subject and A suddenly stops, points towards some ape andsays Look at that, it will not be so clear what he refers to and I would suppose thatB will fix on a referent that he assumes is mutually accessible to A and B (e.g., theape itself).

Second, even if in resolving the reference of true demonstratives the hearer em-ploys reasoning as to what it most likely is that the speaker intended to refer to,I would content that this kind of reasoning still is different from reasoning aboutthe speaker’s intention in making the utterance she did as such. Thus, while in thelatter case, the hearer will employ more or less specific assumptions drawn fromhis knowledge base of social interaction, as, for instance, assumptions concerningthe particular social relationship between speaker and hearer, I do not think thatthis type of information is used when it comes to determining the referent of a truedemonstrative. Thus, simply making reference to speaker intentions in the differ-entiation between the pragmatic processes leading to utterance meaning and thoseleading to what is meant actually might not be fine-grained enough. Whereas inthe determination of the former, reasoning concerning what the speaker intended torefer to by his use of a demonstrative is involved, for the determination of the latter,reasoning concerning why the speaker said what he did is necessary and only thelatter kind of reasoning requires access to knowledge concerning social interaction.

5.3 Summary

In this chapter, we looked at alternative characterisations of literal meaning andnon-literal meaning and (broadly) at the role context might be taken to play inthe overall interpretation process as well as in the differentiation of semantics andpragmatics. Let me summarise the points that were made.

Recanati offers a typology of meanings that is intended to differentiate betweenwhat are traditionally taken to be literal meaning and non-literal meaning as wellas further meaning aspects that can be identified but may not be taken to fit oneor the other description. Thus, he identifies an expression‘s t-literal meaning as the

meaning conventionally associated with it and as such the meaning traditionallycalled literal. He differentiates from t-literal meaning a number of more or lessdeviating meanings, only some of which are non-literal in the ordinary understandingof the term. As we saw, there are a number of problems with Recanati’s classificationof the different types of meaning. Thus, his characterisation of t-literal meaningas both being conventional as well as literal in the traditional sense is debatable,since, as we saw, conventionality does not necessarily have to be seen as goingalong with literalness of meaning. Also, according to his characterisation of non-literal meaning, conversational implicatures and indirect speech acts should be bestexemplars of that category, as they are both secondary and – according to Recanati –fulfil the transparency condition. However, intuitively it is such figurative meaningsas metaphor and metonymy that are regarded good examples of non-literal meaning.Moreover, it seems that a notion such as the transparency condition, which relies onspeaker-hearers’ intuitions does not produce the desired results.

Ariel’s differentiation of three types of what she calls minimal meanings is guidedby the assumption that the notion of literal meaning has been used to refer to thesethree distinct types of meaning because the notion captures the fact that all threemeanings are minimal or basic. However, she points out, they are only minimal orbasic with respect to the different domains they belong to. Thus, she differentiateslinguistic meaning both from salient meaning and privileged interactional meaning.The main problem with Ariel’s classification concerns her characterisation of lin-guistic meaning and salient meaning as being totally independent of one another.In fact, from how she characterises the two meanings, it seems that both are takento be independently accessed during the interpretation of a particular expression.Reasons were given for why this seems highly implausible.

In contrast to both Recanati and Ariel who assume of lexical meanings that theyare full-fledged readings, I argued for viewing lexical meaning as underspecified. Asa consequence – and if one wants to keep the intuition that the notion applies tofull readings – literal meaning can no longer be regarded as semantic. Thus, literalmeaning, like non-literal meaning, has to be seen as pragmatically determined. Todifferentiate the one type of meaning from the other, I characterised the formeras basic and non-derived and the latter as non-basic, derived or deviating and insome sense replacing the meaning from which it derives or deviates. With such acharacterisation, metaphor and metonymy are captured as prototypical examples ofnon-literal meaning, whereas, contrary to Recanati, neither PCIs nor indirect speechacts are considered non-literal. In fact, with this characterisation, the results ofad-hoc concept formation would be considered non-literal, but not the results of freeenrichment. This is because whereas in the former case a basic meaning is changedand replaced, in the latter case a meaning is specified, as it were, and aspects areadded to it. Similar points have to be made with respect to idioms and irony. Thus,while the former may be characterised as non-literal, this is not so clear for thelatter.

Note again that this characterisation of literal meaning and non-literal meaningdoes not distinguish between types of meanings that play distinctive roles in theinterpretation process and it does not allow the two notions to be used in differ-entiating semantics from pragmatics. However, this seems to be simply due to the

Summary 253

fact that these notions are rather intuitive and vague to begin with. Another notiontraditionally used in the differentiation between semantics and pragmatics is thatof context-(in)dependence. In addition, particular types of pragmatic processes aredifferentiated with respect to the nature of the contextual information they takeinto account and more specifically, whether they make reference to the speaker’s in-tention in making a particular utterance. For instance, the process free enrichmentis usually assumed to take recourse to potential speaker intentions. However, thereare experimental results which indicate that such implicit meaning aspects (IMAs)as taken to be contributed by free enrichment actually seem to arise whether thespeaker intended them or not. This effect follows if one assumes that during the in-terpretation of an utterance, information stored in frames evoked by the expressionsused may be made use of. This allows for a contribution of IMAs to the inter-pretation of an utterance without the need to take recourse to speaker intentions.Moreover, although the process of free enrichment is taken to operate independentlyof linguistically mandated slots in the semantic form of an utterance, the processis not entirely free. Thus, it does not add any meaning aspects whatsoever, butrather its contribution may be taken to be restricted to meaning aspects that arethe values of variables contributed by frames.

As concerns the role of the notion of context-(in)dependence in the differentia-tion of semantics and pragmatics, I argued that, actually, the dichotomy context-independent vs. context-dependent may only be used to differentiate what might becalled linguistic semantics from pragmatics. In contrast, the process of semantic in-terpretation actually only applies to meaning representations that have already beenpragmatically enriched, as the output of the context-independent linguistic seman-tics component is only sub-propositional. Thus, the notion of context-(in)dependencecannot be used to differentiate between the meaning aspects pragmatics as comparedto real semantics deals with. Rather, these two systems are differentiated by the na-ture of the processes that constitute them: monotonic reasoning with non-defeasibleoutput in the case of real semantics vs. non-monotonic reasoning with defeasibleoutput in the case of pragmatics.

Turning, lastly, to the differentiation of types of pragmatic processes, we sawthroughout this thesis that amongst others, one important aspect in which pragmaticprocesses are assumed to differ is whether they take into account the speaker’spotential intentions in making her utterance. For the processes that contribute tothe level of utterance meaning, I argued that the information they use is restrictedto narrow context34 and conceptual knowledge including common sense knowledge,possibly in the form of frames evoked by the different expressions used, but also inthe form of meaning postulates which capture stereotypical behaviour and practices.In contrast, the processes leading to what is meant by the speaker of a particularutterance are assumed to make use of information resulting from reasoning about thespeaker’s potential intentions and generally information from the knowledge base ofsocial interaction.

34possibly with the exception of demonstratives, whose resolution may require reasoning aboutwhat the speaker intended to demonstrate

Chapter 6

Summary

256 Summary

List of Figures

2.1 Traditional differentiation of types of meaning . . . . . . . . . . . . . 112.2 Grice’s distinction of what is said and what is meant . . . . . . . . . . 132.3 The typical cat-on-the-mat configuration . . . . . . . . . . . . . . . . 152.4 A rather unusual cat-on-the-mat configuration . . . . . . . . . . . . . 162.5 Revised differentiation of types of meaning . . . . . . . . . . . . . . . 192.6 (Non-)Literal meaning and Grice’s levels of what is said/what is meant 202.7 Busse’s view of degrees of conventionality . . . . . . . . . . . . . . . . 302.8 Idioms in Busse’s framework . . . . . . . . . . . . . . . . . . . . . . . 322.9 Salient meanings according to Busse’s characterisation . . . . . . . . 332.10 Giora’s characterisation within Busse’s framework . . . . . . . . . . . 342.11 Lexical entry for spulen according to Engelberg (2000) . . . . . . . . 422.12 Lexical entry for to bake according to Pustejovsky (1995) . . . . . . . 452.13 Lexical entry for cake according to Pustejovsky (1995) . . . . . . . . . 462.14 Semantic representation for the VP bake a cake according to Puste-

jovsky (1995) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472.15 Literal and non-literal meaning in Relevance Theory . . . . . . . . . . 572.16 Lexical entry according to Giora . . . . . . . . . . . . . . . . . . . . . 582.17 Salient and non-salient meaning in Giora’s approach . . . . . . . . . . 592.18 A lexical entry adapted from Relevance Theory . . . . . . . . . . . . 602.19 Partial syntactic structure for Mittie feeds the dog . . . . . . . . . . . 632.20 e.g. the destruction of the city, the city’s destruction . . . . . . . . . 642.21 e.g. (John)destroyed the city . . . . . . . . . . . . . . . . . . . . . . . 642.22 e.g. (John’s) destroying the city . . . . . . . . . . . . . . . . . . . . . 652.23 The semantic form of Max began to read every novel. . . . . . . . . . 652.24 The semantic form of Max began every novel. . . . . . . . . . . . . . . 652.25 Reaction times from Beretta et al. (2005) . . . . . . . . . . . . . . . . 692.26 M350 latencies Beretta et al. (2005) . . . . . . . . . . . . . . . . . . . 69

3.1 Grice’s distinction of what is said and what is meant more detailed . . 853.2 Grice’s types of meaning and their relation to what is said/what is

meant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873.3 Bierwisch’s three levels of meaning . . . . . . . . . . . . . . . . . . . 893.4 Minimal proposition vs. proposition expressed . . . . . . . . . . . . . 993.5 The Relevance-theoretic View . . . . . . . . . . . . . . . . . . . . . . 1003.6 Bach’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1043.7 Stanley’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1063.8 Borg’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

258 List of Figures

3.9 Cappelen and Lepore’s view . . . . . . . . . . . . . . . . . . . . . . . 122

4.1 Levinson’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1674.2 Recanati’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1914.3 Bierwisch’s view on the relation of utterance meaning and commu-

nicative sense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

5.1 Non-literal meaning in Recanati’s classification . . . . . . . . . . . . . 2095.2 The Killing frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2335.3 The Ingestion frame . . . . . . . . . . . . . . . . . . . . . . . . . . . 2375.4 Levinson’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2455.5 Semantics throughout . . . . . . . . . . . . . . . . . . . . . . . . . . . 2465.6 RT’s differentiation of linguistic vs. real semantics . . . . . . . . . . . 2475.7 Only real semantics in Burton-Roberts’ approach . . . . . . . . . . . 2485.8 My view on the semantics/pragmatics distinction . . . . . . . . . . . 250

Bibliography

Ariel, M.: 2002, The demise of a unique concept of literal meaning., Journal ofPragmatics 34, 361–402.

Asher, N. and Lascarides, A.: 2003, Logics of Conversation., Cambridge: CambridgeUniversity Press.

Assimakopoulos, S.: 2008, Intention, common ground and the availability of seman-tic content: a relevance-theoretic perspective., in I. Kecskes and J. Mey (eds),Intention, common ground and the egocentric speaker-hearer., Mouton de Gruyter.

Bach, K.: 1994a, Conversational Impliciture., Mind and Language 9, 124–162.

Bach, K.: 1994b, Semantic slack: what is said and more., in S. L. Tsohatzidis(ed.), Foundations of Speech Act Theory. Philosophical and linguistic perspectives.,Routledge, pp. 267–291.

Bach, K.: 1997, The Semantics-Pragmatics Distinction: What It Is and Why ItMatters., Linguistische Berichte, Special Issue on Pragmatics 8, 33–50.

Bach, K.: 2000, Quantification, Qualification and Context. A Reply to Stanley andSzabo., Mind and Language 15, 262–283.

Bach, K.: 2001a, Speaking Loosely: Sentence Nonliterality., Midwest Studies inPhilosophy 25, 249–263. Figurative Language.

Bach, K.: 2001b, You don’t say?, Synthese 128, 15–44.

Bach, K.: 2004, Minding the gap., in C. Bianchi (ed.), The Semantics/PragmaticsDistinction., Stanford: CSLI.

Bach, K.: 2005, Context ex Machina., in G. S. Zoltan (ed.), Semantics vs. Pragmat-ics., Oxford: Clarendon, pp. 15–44.

Bach, K.: 2006a, Impliciture vs. Explicature: What’s the difference?,userwww.sfsu.edu/~kbach\Bach.ImplExpl.pdf. Paper given at the ‘Workshopon Explicit Communication’, University of Granada, Spain.

Bach, K.: 2006b, The Excluded Middle: Semantic Minimalism without MinimalPropositions., Philosophy and Phenomenological Research 73, 435–442.

Bach, K.: n.d.a, From the strange to the bizarre: Another Reply to Cappelen andLepore., userwww.sfsu.edu/~kbach\reply2toC&L.pdf.

260 Bibliography

Bach, K.: n.d.b, Minimalism for Dummies: A Reply to Cappelen and Lepore.,userwww.sfsu.edu/~kbach\replytoC&L.pdf.

Baker, C., Fillmore, C. and Cronin, B.: 2003, The structure of the FrameNetdatabase., International Journal of Lexicography 16, 281–296.

Baker, C., Fillmore, C. and Lowe, J.: 1998, The Berkeley FrameNet project., Pro-ceedings of the 36th ACL and 17th COLING, San Francisco, pp. 86–90.

Baron-Cohen, S., Leslie, A. M. and Frith, U.: 1985, Does the autistic child have a”theory of mind”?, Cognition 21, 37–46.

Beretta, A., Fiorentino, R. and Poeppel, D.: 2005, The effects of homonymy andpolysemy on lexical access: an MEG study., Cognitive Brain Research 24, 57–65.

Bezuidenhout, A. L.: 2006, The coherence of contextualism., Mind and Language21, 1–10.

Bezuidenhout, A. L. and Cutting, J. C.: 2002, Literal meaning, minimal proposi-tions, and pragmatic processing., Journal of Pragmatics 34, 433–456.

Bezuidenhout, A. L. and Morris, R. K.: 2004, Implicature, Relevance, and DefaultPragmatic Inference., in D. Sperber and I. Noveck (eds), Experimental Pragmat-ics., Palgrave Macmillan, Basingstoke, pp. 257–282.

Bianchi, C. (ed.): 2004, The Semantics/Pragmatics Distinction., CSLI Publications.

Bierwisch, M.: 1979, Wortliche Bedeutung - eine pragmatische Gretchenfrage.,in G. Grewendorf (ed.), Sprechakttheorie und Semantik., Frankfurt am Main,Suhrkamp Verlag, pp. 119–148.

Bierwisch, M.: 1980, Semantic Structure and Illocutionary Force., Speech act theoryand pragmatics., D. Reidel, Dordrecht.

Bierwisch, M.: 1983, Semantische und konzeptuelle Reprasentation lexikalischerEinheiten., in R. Ruzicka and W. Motsch (eds), Untersuchungen zur Semantik.,Akademie Verlag, Berlin, pp. 61–99.

Bierwisch, M.: 1997, Lexical Information from a Minimalist Point of View., inC. Wilder, H. M. Gerlach and M. Bierwisch (eds), The Role of Economy Principlesin Linguistic Theory., Akademie Verlag, Berlin.

Bierwisch, M. and Schreuder, R.: 1992, From concepts to lexical items., Cognition42, 23–60.

Blasko, D. G. and Connine, C. M.: 1993, Effects of familiarity and aptness inmetaphor processing., Journal of Experimental Psychology: Learning, Memoryand Cognition 19, 295–308.

Borer, H.: 2004, Structuring Sense., Vol. 1 and 2, Oxford: Oxford University Press.

Bibliography 261

Borg, E.: 2001, An Expedition Abroad: Metaphor, Thought and Reporting., inP. French and H. Wettstein (eds), Midwest Studies in Philosophy XXV, Oxford:Blackwell, pp. 227–248.

Borg, E.: 2004a, Formal semantics and intentional states., Analysis 64, 215–23.

Borg, E.: 2004b, Minimal Semantics., Oxford: Oxford University Press.

Borg, E.: 2006, Review of Literal Meaning., Mind 115, 461–5.

Borg, E.: 2007, Minimalism vs. Contextualism in Semantics., in G. Preyer andG. Peter (eds), Context Sensitivity and Semantic Minimalism: Essays on Seman-tics and Pragmatics., Oxford University Press, pp. 546–571.

Borg, E.: 2009, On three theories of implicature: Default Theory, Relevance Theoryand Minimalism., International Review of Pragmatics 1, 63–83.

Braun, D.: 2010, Indexicals., in E. N. Zalta (ed.), The Stanford Encyclopedia ofPhilosophy., summer 2010 edn, Stanford University.

Breheny, R., Katsos, N. and Williams, J.: 2006, Are generalised scalar implicaturesgenerated by default? an on-line investigation into the role of context in generatingpragmatic inferences., Cognition 100.

Brennan, J. and Pylkkanen, L.: 2008, Processing events: Behavioral and neuromag-netic correlates of Aspectual Coercion, Brain and Language 106.

Burton-Roberts, N.: 1994, Ambiguity, sentence and utterance., Transactions of thePhilological Society 92, 179–213.

Burton-Roberts, N.: 2000, Where and what is phonology? A representational per-spective., in N. Burton-Roberts, P. Carr and G. Docherty (eds), Phonologicalknowledge., Oxford University Press, Oxford, pp. 39–66.

Burton-Roberts, N.: 2005, Robyn Carston on semantics, pragmatics and ‘encoding’.,Journal of Linguistics 41, 389–407.

Burton-Roberts, N.: 2006, Cancellation and Intention. Paper given at the ‘Workshopon Explicit Communication’, University of Granada, Spain.

Burton-Roberts, N.: 2007a, Varieties of semantics and encoding: negation, narrow-ing/loosening and numerals., in N. Burton-Roberts (ed.), Pragmatics., Palgrave,Basingstoke.

Burton-Roberts, N. (ed.): 2007b, Pragmatics., Palgrave, Basingstoke.

Burton-Roberts, N. and Poole, G.: 2006, ‘Virtual conceptual necessity’, feature-dissociation and the Saussurian legacy in generative grammar., Journal of Lin-guistics 42, 575–628.

Busse, D.: 1991, Diachrone Semantik und Pragmatik. Untersuchungen zur Erklarungund Beschreibung des Sprachwandels., Tubingen: Niemeyer.

262 Bibliography

Cacciari, C. and Tabossi, P.: 1988, The Comprehension of Idioms., Journal of Mem-ory and Language 27, 668–683.

Capone, A.: 2004, Speech acts: literal and non-literal. internet resource:http://semanticsarchive.net.

Capone, A.: 2006, On Grice’s Circle (a theory-internal problem in linguistic theoriesof the Gricean type.), Journal of Pragmatics 38, 645–669.

Capone, A.: 2009, Are explicatures cancellable? Toward a theory of the speaker’sintentionality., Intercultural Pragmatics 6–1, 55–83.

Cappelen, H. and Lepore, E.: 2005, Insensitive Semantics: a defense of semanticminimalism and speech act pluralism., Blackwell.

Cappelen, H. and Lepore, E.: n.d.a, Kent Bach on Minimalism for Dummies,userwww.folk.uio.no/hermanc/docs/KBM.pdf.

Cappelen, H. and Lepore, E.: n.d.b, Reply to Bach,userwww.folk.uio.no/hermanc/docs/BachPPR.pdf.

Carston, R.: 1997, Enrichment and loosening: complementary processes in derivingthe proposition expressed?, Linguistische Berichte, Special Issue on Pragmatics8, 103–27.

Carston, R.: 1999, The semantics/pragmatics distinction: A view from RelevanceTheory., in K. Turner (ed.), The Semantics/Pragmatics Interface from Differ-ent Points of View., Current research in the semantics - pragmatics interface 1,Elsevier Science, Oxford.

Carston, R.: 2002a, Linguistic meaning, communicated meaning and cognitive prag-matics., Mind and Language 17(1/2), 127–48. Special Issue on Pragmatics andCognitive Science.

Carston, R.: 2002b, Metaphor, ad hoc concepts and word meaning - more questionsthan answers., UCL Working Papers in Linguistics 14, University College London.

Carston, R.: 2002c, Thoughts and Utterances. The Pragmatics of Explicit Commu-nication., Blackwell Publishing, Oxford.

Carston, R.: 2004a, Explicature and semantics., in S. Davis and B. S. Gillon (eds),Semantics: A Reader., Oxford University Press, pp. 817–845.

Carston, R.: 2004b, Relevance Theory and the saying/implicating distinction., inL. Horn and G. Ward (eds), Handbook of Pragmatics., Oxford: Blackwell.

Carston, R.: 2004c, Truth-conditional content and conversational implicature., inC. Bianchi (ed.), The Semantics/Pragmatics Distinction., CSLI Publications.

Carston, R.: 2005, Relevance Theory, Grice and the neo-Griceans: a response toLaurence Horn’s ‘Current issues in neo-Gricean pragmatics’., Intercultural Prag-matics 2(3), 303–319.

Bibliography 263

Carston, R.: 2007, How many pragmatic systems are there?, in M.-J. Frapolli (ed.),Saying, Meaning, Referring: Essays on the Philosophy of Francois Recanati., Pal-grave, pp. 18–48.

Chierchia, G.: 2004, Scalar Implicatures, Polarity Phenomena, and the Syn-tax/Pragmatics Interace., in A. Belletti (ed.), Structures and Beyond., OxfordUniversity Press, pp. 39–103.

Chierchia, G., Fox, D. and Spector, B.: 2010, The Grammatical View of ScalarImplicatures and the Relationship between Semantics and Pragmatics., www.

semanticsarchive.net/Archive/WMzY2ZmY/CFS_EmbeddedSIs.pdf.

Chierchia, G. and McConnell-Ginet, S.: 2000, Meaning and grammar: an introduc-tion to semantics, 2. edn, Cambridge, Mass.: MIT.

Chng, S. W.: 1999, Language, thought and literal meaning., Phd thesis, Universityof Newcastle upon Tyne.

Clark, H. H.: 1973, Space, time, semantics and the child., in T. E. Moore (ed.),Cognitive development and the acquisition of language., New York: AcademicPress, pp. 27–63.

Cohen, A.: 2006, Anaphora resolution and minimal models., in J. Bos and A. Koller(eds), ICoS-5, Buxton, England.

Cole, P. (ed.): 1981, Radical pragmatics, Academic, New York.

Colston, H. L.: 2000, On necessary conditions for verbal irony comprehension.,Pragmatics and Cognition 8, 277–324.

Colston, H. L.: 2002, Contrast and assimilation in verbal irony., Journal of Prag-matics 34, 111–142.

Colston, H. L. and Gibbs, R. W.: 2002, Are Irony and Metaphor Understood Dif-ferently?, Metaphor and Symbol 17, 57–80.

Curco, C.: 2000, Irony: Negation, echo and metarepresentation., Lingua 110, 257–280.

Dascal, M.: 1987, Defending literal meaning., Cognitive Science 11, 259–281.

Dascal, M.: 1989, On the roles of context and literal meaning in understanding.,Cognitive Science 13, 253–257.

de Villiers, J., Myers, B. and Stainton, R.: 2010, Differential Pragmatic Abilitiesand Autism Spectrum Disorders: The Case of Pragmatic Determinants of LiteralContent., in M. Macaulay and P. Garces-Blitvich (eds), Pragmatics and Context.,Toronto: Antares.

de Villiers, J., Stainton, R. and Szatmari, P.: 2007, Pragmatic abilities in autismspectrum disorder: a case study in philosophy and the empirical., Midwest Studiesin Philosophy 31, 292–317.

264 Bibliography

Dennett, D. C.: 1969, Content and Consciousness., Routledge and Kegan Paul,London.

Dews, S. and Winner, E.: 1999, Obligatory processing of literal and nonliteral mean-ings in verbal irony., Journal of Pragmatics 31, 1579–1599.

Dolling, J.: 1997, Semantic Form and Abductive Fixation of Parameters., inJ. Dolling (ed.), Systematische Bedeutungsvariationen: Semantische Form undkontextuelle Interpretation, University of Leipzig.

Dolling, J.: 2000, Formale Analyse von Metonymie und Metapher., in R. Eckhardtand K. von Heusinger (eds), Meaning Change – Meaning Variation., Universityof Konstanz, pp. 31–53.

Dolling, J.: 2001, Systematische Bedeutungsvariationen: Semantische Formund kontextuelle Interpretation., Linguistische Arbeitsberichte 78, University ofLeipzig, Germany.

Dolling, J.: 2005, Semantische Form und pragmatische Anreicherung: Situations-ausdrucke in der Außerungsinterpretation., Zeitschrift fur Sprachwissenschaft24, 159–225.

Dolling, J. and Zybatow, T.: 2007, Verb Meaning: How much Semantics is in theLexicon?, in A. Spath (ed.), Interfaces and Interface Conditions., Mouton – deGruyter, pp. 33–75.

Egg, M.: 1994, Zur Reprasentation extrem polysemer Lexeme., in M. Schwarz (ed.),Kognitive Semantik., Narr, Tubingen.

Egg, M.: 2002, Semantic construction for reinterpretation phenomena, Linguistics40, 579–609.

Egg, M.: 2003, Beginning Novels and Finishing Hamburgers: Remarks on the Se-mantics of to begin, Journal of Semantics 20, 163–191.

Egg, M.: 2005, Flexible Semantics for Reinterpretation Phenomena., CSLI, Stan-ford.

Eviatar, Z. and Just, M. A.: 2006, Brain correlates of discourse processing: An fMRIinvestigation of irony and conventional metaphor comprehension., Neuropsycholo-gia 44, 2348–2359.

Fillmore, C., Johnson, C. and Petruck, M.: 2003, Background to FrameNet., Inter-national Journal of Lexicography 16, 235–250.

Fodor, J. A.: 1983, Modularity of Mind: An Essay on Faculty Psychology., Cam-bridge, Mass.: MIT Press.

Fodor, J. A.: 1998, Concepts. Where cognitive science went wrong., Oxford: Claren-don.

Bibliography 265

Fodor, J. A. and Lepore, E.: 1998, The Emptiness of the Lexicon: Reflections onJames Pustejovsky’s The Generative Lexicon., Linguistic Inquiry 29, 269–288.

Frisson, S.: 2009, Semantic Underspecification in Language Processing., Languageand Linguistics Compass 3(1), 111–27.

Frisson, S. and Pickering, M. J.: 1999, The Processing of Metonymy: Evidence fromEye Movements., Journal of Experimental Psychology 25(6), 1366–1383.

Frisson, S. and Pickering, M. J.: 2007, The processing of familiar and novel senses ofa word: why reading Dickens is easy but reading Needham can be hard., Languageand Cognitive Processes 22, 595–613.

Garcia-Carpintero, M.: 2006, Recanati on the Semantics/Pragmatics distinction.,Critica 38(112), 35–68.

Garrett, M. and Harnish, R. M.: 2007, Experimental Pragmatics: Testing for im-pliciture., Pragmatics and Cognition 16(1), 65–90.

Garrett, M. and Harnish, R. M.: 2008, Skating along the syntactic verge., in W. D.Lewis, S. Karimi, H. Harley and S. O. Farrar (eds), Time and Again. TheoreticalPerspectives on Formal Linguistics., Benjamins, Amsterdam.

Garrett, M. and Harnish, R. M.: 2009, Q-Phenomena, I-Phenomena and Impliciture:Some Experimental Pragmatics., International Review of Pragmatics 1, 84–117.

Gazdar, G.: 1979, Pragmatics: implicature, presupposition and logical form., Aca-demic Press.

Gernsbacher, M. A., Keysar, B., Robertson, R. R. W. and Werner, N. K.: 2001, TheRole of Suppression and Enhancement in Understanding Metaphors., Journal ofMemory and Language 45, 433–450.

Gernsbacher, M. A. and Robertson, R. R. W.: 1999, The role of suppression infigurative language comprehension., Journal of Pragmatics 31, 1619–1630.

Gibbs, R. W.: 1994, The poetics of mind: Figurative thought, language, and under-standing., Cambridge University Press, Cambridge.

Gibbs, R. W.: 2002, A new look at literal meaning in understanding what is saidand implicated., Journal of Pragmatics 34, 457–486.

Gibbs, R. W. and Moise, J. F.: 1997, Pragmatics in understanding what is said.,Cognition 62, 51–74.

Giora, R.: 1995, On irony and negation., Discourse Processes 19, 239–264.

Giora, R.: 1997, Understanding figurative and literal language: The graded saliencehypothesis., Cognitive Linguistics 8, 183–206.

Giora, R.: 1999, On the priority of salient meanings: Studies of literal and figurativelanguage., Journal of Pragmatics 31, 919–929.

266 Bibliography

Giora, R.: 2002, Literal vs. figurative language: Different or equal?, Journal ofPragmatics 34, 487–506.

Giora, R. and Fein, O.: 1999, On understanding familiar and less-familiar figurativelanguage., Journal of Pragmatics 31, 1601–1618.

Giora, R., Fein, O., Kaufman, R., Eisenberg, D. and Erez, S.: 2009, Does an ”ironicsituation” favour an ironic interpretation?, in G. Brone and J. Vandaele (eds),Cognitive poetics. Goals, gains and gaps., Berlin/New York: Mouton de Gruyter,pp. 383–399.

Giora, R., Fein, O., Laadan, D., Wolfson, J., Zeituny, M., Kidron, R., Kaufman, R.and Shaham, R.: 2007, Expecting Irony: Context Versus Salience-Based Effects.,Metaphor and Symbol 22, 119–146.

Giora, R., Fein, O. and Schwartz, T.: 1997, On the time course of understandingmetaphor: The graded salience hypothesis., In preparation .

Glucksberg, S. and McGlone, M. S.: 1999, When love is not a journey: Whatmetaphors mean., Journal of Pragmatics 31, 1541–1558.

Glucksberg, S., P., G. and Bookin, H.: 1982, On understanding non-literal speech:Can people ignore metaphors?, Journal of Verbal Learning and Verbal Behavior1, 85–96.

Gold, R., Faust, M. and Goldstein, A.: 2010, Semantic integration dur-ing metaphor comprehension in asperger syndrome., Brain and Language .doi:10.1016/j.bandl.2010.03.002.

Grice, H. P.: 1969, Utterer’s Meaning and Intentions., Philosophical Review pp. 147–177. Reprinted in Grice 1989, pp. 86–116.

Grice, H. P.: 1975, Logic and Conversation., in P. Cole and J. L. Morgan (eds),Syntax and Semantics 3: Speech Acts, Academic Press, New York. Reprinted inGrice 1989, pp. 22–40.

Grice, H. P.: 1989, Studies in the ways of words., Harvard University Press, Cam-bridge and London.

Groefsema, M.: 2007, Concepts and word meaning in relevance theory., inN. Burton-Roberts (ed.), Pragmatics., Palgrave, Basingstoke, pp. 136–157.

Gross, S.: 2006, Can One Sincerely Say What One Doesn’t Believe?, Mind andLanguage 21, 11–20.

Happe, F. G. E.: 1993, Communicative compentence and theory of mind in autism:A test of relevance theory., Cognition 48, 101–119.

Hintzman, D. L.: 1986, ”Schema Abstraction” in a Multiple-Trace Memory Model.,Psychological Review 93, 411–428.

Bibliography 267

Hirst, W., LeDoux, J. and Stein, S.: 1984, Constraints on the Processing of IndirectSpeech Acts: Evidence from Aphasiology., Brain and Language 23, 26–33.

Hobbs, J. R., Stickel, M., Appelt, D. and Martin, P.: 1993, Interpretation as Ab-duction., Artificial Intelligence 63, 69–142.

Huang, Y.: 2007, Pragmatics, Oxford: Oxford University Press.

Huang, Y. T. and Snedeker, J.: 2009, Online interpretation of scalar quantifiers:Insight into the semantic-pragmatic interface., Cognitive Psychology 58, 376–415.

Huang, Y. T. and Snedeker, J.: in press, Logic and Conversation revisited: Evidencefor a division between semantic and pragmatic content in real time languagecomprehension., Language and Cognitive Processes .

Irmer, M.: 2009, Bridging Inferences in Discourse Interpretation. Ms., doctoralthesis.

Israel, M.: 2002, Literally Speaking., Journal of Pragmatics 34.

Kadmon, N.: 2001, Formal Semantics. Semantics, pragmatics, presupposition, andfocus., Blackwell.

Kaplan, D.: 1989a, Afterthoughts., in J. Almog, J. Perry and H. Wettstein (eds),Themes from Kaplan, Oxford.

Kaplan, D.: 1989b, Demonstratives., in J. Almog, J. Perry and H. Wettstein (eds),Themes from Kaplan, Oxford.

Katz, J.: 1977, Propositional Structure and Illocutionary Force., New York: Crowell.

Keysar, B.: 1989, On the Functional Equivalence of Literal and Metaphorical Inter-pretations in Discourse., Journal of Memory and Language 28, 375–385.

Klepousniotou, E.: 2002, The Processing of Lexical Ambiguity: Homonymy andPolysemy in the Mental Lexicon., Brain and Language 81, 205–223.

Konig, E. u. E. C. T.: 1988, Pragmatic Strengthening and Semantic Change: TheConventionalizing of Conversational Implicature., in Hullen and Schulze (eds),Understanding the Lexicon., Tubingen: Niemeyer, pp. 110–124.

Korta, K.: 1997, Implicitures: Cancelability and Non-detachability., Technical Re-port ILCLI-97-LIC-6, Donostia: ILCLI.

Korta, K. and Perry, J.: 2008, Pragmatics., in E. N. Zalta (ed.), The StanfordEncyclopedia of Philosophy., fall 2008 edn, Stanford University.

Kratzer, A.: 1996, Severing the external argument from its verb., in J. Rooryck andL. Zaring (eds), Phrase Structure and the Lexicon., Dordrecht: Kluwer, pp. 109–137.

268 Bibliography

Lakoff, G. (ed.): 1987, Women, Fire, and Dangerous Things: What CategoriesReveal About the Mind., University of Chicago Press.

Lascarides, A. and Asher, N.: 2007, Sdrt: Dynamic semantics with discourse struc-ture., in H. Bunt and R. Muskens (eds), Computing Meaning: Volume 3, Tilburg:Kluwer.

Laurent, J.-P., Denhieres, G., Passerieux, C., Iakimova, G. and Hardy-Bayle, M.-C.:2006, On understanding idiomatic language: The salience hypothesis assessed byERPs., Brain Research 1068, 151–160.

Leslie, A. M.: 1987, Pretense and representation: The origins of theory of mind.,Psychological Review 94, 412–426.

Levinson, S. C.: 1983, Pragmatics., Cambridge University Press, Cambridge.

Levinson, S. C.: 2000, Presumptive Meanings. The Theory of Generalised Conver-sational Implicature., MIT Press, Cambridge, Mass.

Lewis, D. S.: 1969, Convention., Harvard University Press, Cambridge, Mass.

Lyons, J.: 1987, Semantics., in J. Lyons (ed.), New Horizons in linguistics., Vol. 2,London: Penguin, pp. 152–178.

MacKay, G. and Shaw, A.: 2004, A comparative study of figurative language inchildren with autistic spectrum disorders., Child Language Teaching and Therapy20, 13–32.

Marantz, A.: 1997, No escape from syntax. don’t try morphological analysis inthe privacy of your own lexicon., in A. Dimitriadis (ed.), Proceedings of the 21thAnnual Penn Linguistics Colloquium., Penn Linguistics Club, pp. 201–225.

Mashal, N., Faust, M., Hendler, T. and Jung-Beeman, M.: 2008, Hemispheric dif-ferences in processing the literal interpretation of idioms., Cortex 44, 848–860.

McElree, B. and Frisson, Steven Pickering, M. J.: 2006, Deferred interpretations:Why starting dickens is taxing but reading dickens isn’t., Cognitive Science30, 181–192.

McElree, B., Traxler, M. J., Pickering, M. J., Seely, R. E. and Jackendoff, R.: 2001,Reading time evidence for enriched semantic composition., Cognition 78, B17–B25.

McGlone, M.: 1996, Conceptual Metaphor and Figurative Language Interpretation:Food for Thought?, Journal of Memory and Language 35, 544–565.

Morgan, J. L.: 1978, Two Types of Convention in Indirect Speech Acts., in P. Cole(ed.), Syntax and Semantics. Vol. 9., New York/ San Francisco/ London, pp. 261–280.

Nicolle, S. and Clark, B.: 1999, Experimental pragmatics and what is said: a re-sponse to gibbs and moise., Cognition 69, 337–354.

Bibliography 269

Nissim, M., Sanso, A. and Soria, C.: 1999, Towards a compositional frame seman-tics., Proceedings of the European Conference on Cognitive Science . Siena, Italy.

Noveck, I. A.: 2001, When children are more logical than adults: Experimentalinvestigations of scalar implicature., Cognition 78, 165–188.

Noveck, I. and Posada, A.: 2003, Characterising the time course of an implicature:An evoked potentials study., Brain and Language 85, 203–210.

Nunberg, G.: 1978, The Pragmatics of Reference., Indiana University LinguisticsClub 1978, Bloomington, IN.

Nunberg, G.: 1979, The non-uniqueness of semantic solutions: polysemy., Linguis-tics and Philosophy 3, 143–184.

Nunberg, G.: 1995, Transfers of Meaning., Journal of Semantics 12, 109–132.

Nunberg, G.: 2002, The Pragmatics of Deferred Interpretation., in L. Horn andG. Ward (eds), Blackwell Encyclopedia of Pragmatics., Blackwell.

Nunberg, G., Sag, I. A. and Wasow, T.: 1994, Idioms., Language 70(3), 491–538.

Ortony, A., Schallert, D. L., Reynolds, R. E. and Antos, S. J.: 1978, InterpretingMetaphors and Idioms: Some Effects of Context on Comprehension., Journal ofVerbal Learning and Verbal Behavior. 17, 465–477.

Pagin, P.: 2005, Compositionality and context., in G. Preyer (ed.), Contextualismin Philosophy., Oxford University Press, Oxford, pp. 303–48.

Pagin, P. and Pelletier, J.: 2007, Content, context and composition., in G. Peterand G. Preyer (eds), Content and Context. Essays on Semantics and Pragmatics.,Oxford University Press, Oxford, pp. 25–62.

Papafragou, A.: 1996, Figurative language and the semantics/pragmatics distinc-tion., Language and Literature 5, 179–93.

Perry, J.: 1998, Indexicals, Context and Unarticulated Constituents., Proceedings ofthe 1995 CSLI-Amsterdam Logic, Language and Computation Conference., Stan-ford: CSLI Publications.

Pexman, P. M. and Glenwright, M.: 2007, How do typically developing childrengrasp the meaning of verbal irony?, Journal of Neurolinguistics 20, 178–196.

Pickering, M. J., McElree, B., Frisson, S., Chen, L. and Traxler, M. J.: 2006,Underspecification and aspectual coercion., Discourse Processes 42, 131–155.

Pickering, M. J., McElree, B. and Traxler, M. J.: 2005, The difficulty of coercion:A response to de almeida., Brain and Language 93, 1–9.

Pinango, M. M., Winnick, A., Ullah, R. and Zurif, E.: 2006, Time-course of semanticcomposition: The case of aspectual coercion., Journal of Psycholinguistic Research35, 233–244.

270 Bibliography

Pinango, M. M., Zurif, E. B. and Jackendoff, R.: 1999, Real-time processing impli-cations of enriched composition and the syntax-semantics interface., Journal ofPsycholinguistic Research 28, 395–414.

Pustejovsky, J.: 1991, The generative lexicon., Computational Linguistics 17(4).

Pustejovsky, J.: 1995, The Generative Lexicon., MIT Press, Cambridge, Mass.

Putnam, H.: 1975, The Meaning of ”Meaning”., in K. Gunderson (ed.), Language,mind, and knowledge., University of Minneapolis Press, Minneapolis, pp. 131–193.

Pylkkanen, L., Llinas, R. and McElree, B.: 2004, Distinct effects of semantic plausi-bility and semantic composition in meg, in E. Halgren, S. Ahlfors, M. Hamalainenand D. Cohen (eds), Biomag 2004: Proceedings of the 14th International Confer-ence on Biomagnetism, Biomag: Boston, USA.

Pylkkanen, L., Llinas, R. and Murphy, G. L.: 2006, The Representation of Polysemy:MEG Evidence., Journal of Cognitive Neuroscience 18, 1–13.

Pylkkanen, L., Martin, A. E., McElree, B. and Smart, A.: 2009, The AnteriorMidline Field: Coercion or decision making?, Brain and Language 108, 184–90.

Pylkkanen, L. and McElree, B.: 2006, The syntax-semantic interface: On-line com-position of sentence meaning., in M. Traxler and M. A. Gernsbacher (eds), Hand-book of Psycholoinguistics., Elsevier, pp. 537–77.

Pylkkanen, L., Oliveri, B. and Smart, A. J.: 2009, Semantic vs. world knowledge inprefrontal cortex., Language and cognitive processes 24, 1313–1334.

Reboul, A.: 2004, Conversational Implicatures: Nonce or Generalised?, in I. Noveckand D. Sperber (eds), Experimental Pragmatics., Palgrave Macmillan, Bas-ingstoke, pp. 322–333.

Recanati, F.: 1989, The pragmatics of what is said., Mind and Behaviour 4, 295–329.

Recanati, F.: 1995, The Alleged Priority of Literal Interpretation., Cognitive Science19, 207–232.

Recanati, F.: 2001, Literal/Nonliteral., Midwest Studies in Philosophy 15, 264–274.

Recanati, F.: 2002, Unarticulated constituents., Linguistics and Philosophy25, 299345.

Recanati, F.: 2004, Literal Meaning., Cambridge University Press, Cambridge.

Recanati, F.: n.d., Crazy minimalism. internet resource.

Rodd, J., Gaskell, G. and Marslen-Wilson, W.: 2002, Making sense of semantic am-biguity: semantic competition in lexical access., Journal of Memory and Language46, 245–66.

Bibliography 271

Rolf, E.: 1995, Zur Grammatikalisierung konversationeller Implikaturen., inF. Liedtke (ed.), Implikaturen. Grammatische und pragmatische Analysen.,Tubingen: Niemeyer, pp. 87–102.

Rubio Fernandez, P.: 2007, Suppression in Metaphor Interpretation: Differencesbetween Meaning Selection and Meaning Construction., Journal of Semantics24, 345–71.

Rubio Fernandez, P., Breheny, R. and Lee, M. W.: 2003, Context-independentinformation in concepts: An investigation of the notion of ”core features”., inF. Schmalhofer, R. M. Young and G. Katz (eds), Proceedings of EuroCogSci 03:The European Cognitive Science Conference 2003., London: Lawrence Erlbaum.

Ruhl, C.: 1989, On Monosemy. A Study in Linguistic Semantics., State Universityof New York Press, Albany.

Rundblad, G. and Annaz, D.: 2010, The atypical development of metaphor andmetonymy comprehension in children with autism., Autism 14, 29–46.

Rundblad, G. and Annaz, D.: in press, Metaphor and Metonymy Comprehension:Receptive Vocabulary and Conceptual Knowledge., British Journal of Develop-mental Psychology .

Ruppenhofer, J., Ellsworth, M., Petruck, M. R. L., Johnson, C. R. and Scheffczyk,J.: 2010, FrameNet II: Extended Theory and Practice., http://framenet.icsi.berkeley.edu/book/book.pdf.

Sag, I.: 1981, Formal Semantics and Extralinguistic Context., in P. Cole (ed.),Radical Pragmatics., Academic Press, New York, pp. 273–294.

Saussure, F. d.: 1931, Grundfragen der allgemeinen Sprachwissenschaft, de Gruyter,Berlin. edited by Charles Bally and Albert Sechehaye.

Schmidt, G. L. and Seger, C. A.: 2009, Neural correlates of metaphor process-ing: The roles of figurativeness, familiarity and difficulty., Brain and Cognition71, 375–386.

Schulz von Thun, F.: 1982, Miteinander reden. Storungen und Klarungen: Psy-chologie der zwischenmenschlichen Kommunikation., Vol. 1, Rororo.

Schwoebel, J., Dews, S., Winner, E. and Srinivas, K.: 2000, Obligatory Processingof the Literal Meaning of Ironic Utterances: Further Evidence., Metaphor andSymbol 15, 47–61.

Searle, J. R.: 1975, Indirect Speech Acts., in P. Cole and J. L. Morgan (eds), Syntaxand Semantics 3: Speech Acts., Academic Press, London.

Searle, J. R.: 1978, Literal Meaning., Erkenntnis 13, 207–224.

Searle, J. R.: 1979, Metaphor., in A. Ortony (ed.), Metaphor and Thought., Cam-bridge University Press, Cambridge, Mass.

272 Bibliography

Searle, J. R.: 1980, The background of meaning., in J. R. Searle, F. Kiefer andM. Bierwisch (eds), Speech act theory and pragmatics., Dordrecht.

Shapiro, A. M. and Murphy, G. L.: 1993, Can You Answer a Question for Me?Processing Indirect Speech Acts., Journal of Memory and Language 32, 211–229.

Sperber, D. and Wilson, D.: 1995, Relevance: Communication and Cognition.,Blackwell, Oxford. second edition.

Sperber, D. and Wilson, D.: 2002, Pragmatics, Modularity and Mind-reading., Mindand Language 17, 3–23.

Sperber, D. and Wilson, D.: 2008, A Deflationary Account of Metaphor., in R. Gibbs(ed.), Handbook of Metaphor and Thought., Cambridge University Press, Cam-bridge.

Stainton, R.: 2005, In Defense of Non-Sentential Assertion., in Z. G. Szabo (ed.),Semantics vs. Pragmatics., Oxford: OUP, pp. 383–457.

Stanley, J.: 2000, Context and Logical Form., Linguistics and Philosophy 23, 391–434.

Stanley, J. and Szabo, Z. G.: 2000, On Quantifier Domain Restriction., Mind andLanguage 15, 219–261.

Stern, J.: 2000, Metaphor in Context., MIT Press, Cambridge, Mass.

Stern, J.: 2006, Metaphor, Literal, Literalism., Mind and Language 21, 243–279.

Stern, J.: 2009, Metaphor and Minimalism., Philosophical Studies .doi:10.1007/s11098-009-9486-3.

Storto, G. and Tanenhaus, M.: 2005, Are scalar implicatures computed online?, inE. M. et al. (ed.), Proceedings of Sinn and Bedeutung 9, Nijmegen: NijmegenCenter of Semantics.

Strawson, P. F.: 1950, On referring., Mind 59, 320–44.

Stringaris, A. K., Medford, N. C., Giampietro, V., Brammer, M. J. and David, A. S.:2007, Deriving meaning: Distinct neural mechanisms for metaphoric, literal, andnon-meaningful sentences., Brain and Language 100, 150–162.

Taylor, J. R.: 1994, The Two-Level Approach to Meaning., Linguistische Berichte149, 3–26.

Taylor, J. R.: 1995, Approaches to Word Meaning: The Network Model (Langacker)and the Two-Level-Model (Bierwisch) in comparison., in R. Dirven and J. Van-parys (eds), Current Approaches to the Lexicon., Peter Lang Verlag, Frankfurt.

Todorova, M., Straub, K., Badecker, W. and Frank, R.: 2000, Aspectual coercionand the online computation of sentential aspect., Journal of Memory and Lan-guage 53, 1–25.

Bibliography 273

Traxler, M. J., Pickering, M. J. and McElree, B.: 2002, Coercion in sentence process-ing: Evidence from eye-movements and self-paced reading., Journal of Memoryand Language 47, 530–547.

Wearing, C.: 2010, Autism, Metaphor and Relevance Theory., Mind and Language25, 196–216.

Wharton, T.: 2002, Paul Grice, saying and meaning., UCL Working Papers inLinguistics 14.

Wilson, D. and Sperber, D.: 1992, On verbal irony., Lingua 87, 53–76.

Wilson, D. and Sperber, D.: 2000, Truthfulness and Relevance., UCL WorkingPapers in Linguistics 12, 215–254.

Wunderlich, D.: 1977, On Problems of Speech Act Theory., in R. E. Butts andJ. Hintikka (eds), Basic Problems in Methodology and Linguistics., Dordrecht,Holland, pp. 243–258.

The Notions of Literal and Non-literal Meaning in Semantics and Pragmatics

Documents

pragmatics PPW

Pragmatics tutorial

Pragmatics II

Chapter Seven Pragmatics. What is pragmatics? Pragmatics can...

Syntax, Semantics, and Pragmatics of · PDF filepragmatic.....

CHAPTER II LITERATURE REVIEW A. Pragmatics 1. Definition...

The semantics and pragmatics of belief reports in …...The....

Pragmatics Level Dialogue Analysis - Syracuse...

Pragmatics 2015

'Deixis and Pragmatics' for Handbook of Pragmatics Stephen C...

Pragmatics deixisan1

Spoken Language Pragmatics -...

Embracing pragmatics · Embracing pragmatics? Pragmatics of...

Pragmatics (2)

PRAGMATICS 3

PRAGMATICS -- akw5