Context-driven expectations about focus alternatives

1

3 Context-driven expectations about focus alternatives

4

5

6 Christina S. Kim a,⇑, Christine Gunlogson b, Michael K. Tanenhaus b,c, Jeffrey T. Runner b,c

7 a University of Kent, Department of English Language and Linguistics, Rutherford College, Canterbury, Kent CT2 7NF, UK8 b University of Rochester, Department of Linguistics, 503 Lattimore Hall, Rochester, NY 14627, USA9 c University of Rochester, Department of Brain & Cognitive Sciences, Meliora Hall, Rochester, NY 14627, USA

1011

1 3a r t i c l e i n f o

14 Article history:15 Received 1 July 201216 Revised 7 February 201517 Accepted 28 February 201518 Available online xxxx

19 Keywords:20 Focus21 Discourse processing22 Visual world eye-tracking23 Alternatives24 Context dependence25 Domain restriction26

2 7a b s t r a c t

28What is conveyed by a sentence frequently depends not only on the descriptive content29carried by its words, but also on implicit alternatives determined by the context of use.30Four visual world eye-tracking experiments examined how alternatives are generated31based on aspects of the discourse context and used in interpreting sentences containing32the focus operators only and also. Experiment 1 builds on previous reading time studies33showing that the interpretations of only sentences are constrained by alternatives explic-34itly mentioned in the preceding discourse, providing fine-grained time course information35about the expectations triggered by only. Experiments 2 and 3 show that, in the absence of36explicitly mentioned alternatives, lexical and situation-based categories evoked by the37context are possible sources of alternatives. While Experiments 1–3 all demonstrate the38discourse dependence of alternatives, only explicit mention triggered expectations about39alternatives that were specific to sentences with only. By comparing only with also,40Experiment 4 begins to disentangle expectations linked to the meanings of specific opera-41tors from those generalizable to the class of focus-sensitive operators. Together, these find-42ings show that the interpretation of sentences with focus operators draws on both43dedicated mechanisms for introducing alternatives into the discourse context and general44mechanisms associated with discourse processing.45! 2015 Elsevier B.V. All rights reserved.46

47

48

49 1. Introduction

50 Language processing involves not only continuously51 integrating multiple information sources but also generat-52 ing expectations about the upcoming discourse (Altmann53 & Kamide, 1999; Marslen-Wilson, 1973, 1975; Tanenhaus,54 Spivey-Knowlton, Eberhard, & Sedivy, 1995). In this paper,55 we investigate the role of context-based expectations in56 the processing of sentences with alternative-triggering

57focus operators (or focus particles). Increasingly, the role58of context-relevant alternatives has emerged as a central59factor in both semantics and pragmatic interpretation.60Consider the example in (1).

(1) Jane only has some apples.65

66In addition to conveying what Jane does have, (1) cru-67cially conveys that Jane does not have any other element68from some contextually-determined set of alternatives to69apples. Examining discourse expectations associated with70alternative-triggering focus operators like only makes it71possible to separate expectations about contextually72-determined alternatives, from expectations about explicit73descriptive content, which have been the primary focus of74previous research. Studying how listeners use focus

http://dx.doi.org/10.1016/j.cognition.2015.02.0090010-0277/! 2015 Elsevier B.V. All rights reserved.

⇑ Corresponding author at: University of Kent, English Language andLinguistics, School of European Culture and Languages, Cornwallis NorthWest, Canterbury, Kent CT2 7NF, UK. Tel.: +44 (0)1227 827539;fax: +44 (0)1227 823641.

E-mail addresses: [email protected] (C.S. Kim), [email protected] (C. Gunlogson), [email protected] (M.K. Tanenhaus),[email protected] (J.T. Runner).

Q1

Q2

Q3

Cognition xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Cognition

journal homepage: www.elsevier .com/locate /COGNIT

COGNIT 2961 No. of Pages 23, Model 3G

12 March 2015

Please cite this article in press as: Kim, C. S., et al. Context-driven expectations about focus alternatives. Cognition (2015), http://dx.doi.org/10.1016/j.cognition.2015.02.009

75 operators to access and construct alternatives thus pro-76 vides a window into one of the fundamental puzzles in77 real-time language processing, namely how listeners con-78 struct the contextual domains that support rich incremen-79 tal interpretation.

80 1.1. Incremental interpretation based on semantically81 indeterminate input

82 Linguistic input is highly structured: every element is83 interpreted relative to other elements. For example, rela-84 tions hold between a main verb and its arguments,85 between a wh- element and its gap, between functional86 elements and their arguments, and between arguments87 and their modifiers. Linguistic dependencies give rise to88 expectations about their completions (Gibson, 1998,89 2000; Hale, 2003; Levy, 2008; Lewis, 1993; Lewis &90 Vasishth, 2005). In expectation-based experimental para-91 digms like visual world eye-tracking (Cooper, 1974;92 Tanenhaus et al., 1995), eye movements not only respond93 in a time-locked manner to linguistic events in the audito-94 ry input, but are also anticipatory, reflecting expectations95 generated by the listener about how the sentence or dis-96 course will resolve (Altmann & Kamide, 1999, 2007;97 Chambers, Tanenhaus, Eberhard, Filip, & Carlson, 2002).98 In typical visual world studies, the target sentence con-99 tains a definite description of the target referent (e.g. the

100 green triangle) and the display of visually-presented objects101 establishes the initial referential domain. Given the102 uniqueness presupposition carried by the, listeners use103 the linguistic input (along with other cues) to determine104 the unique candidate matching the description, proceeding105 incrementally as descriptive content becomes available.106 We refer to reference resolution that is triggered by explic-107 it descriptive input as description-matching.108 In sentences like (2) description-matching and contex-109 tual restriction of the referential domain triggered by the110 functional element the are conflated. The identification111 task is to pick out the object that is green and triangular,112 excluding non-green and non-triangular candidates.

(2) Click on the green triangle.117

118 Semantic and pragmatic constraints generate strong119 expectations about the likelihood of reference to objects120 in a scene, influenced by both linguistic factors (e.g. old/121 new status in the discourse) and non-linguistic informa-122 tion (e.g. knowledge about contingencies and likelihoods123 of eventualities in real world situations; Altmann &124 Kamide, 1999; Chambers, Tanenhaus, & Magnuson, 2004;125 Chambers et al., 2002). Moreover, pragmatically-driven126 expectations can lead to rapid reference resolution even127 before disambiguating descriptive content is encountered.128 For example, Sedivy, Tanenhaus, Chambers, and Carlson129 (1999) examined processing of noun phrases with gradable130 prenominal adjectives as in (3) using visual displays where131 more than one item qualified as tall in the context (e.g. a132 tall cup and a tall pitcher).

(3) Click on the tall cup.

137

138When tall is encountered, the property (tall) in the descrip-139tion could be mapped onto both the cup and the pitcher.140However, if there is also a smaller cup in the display, then141given a contrastive interpretation, only the taller cup is a142plausible referent. Sedivy et al. (1999) found that contrast143leads listeners to construe the adjective as a disambiguat-144ing modifier, effectively restricting the domain of reference145to the contrastive referents (the two cups). This result146demonstrates rapid use of contextually-coded contrast to147achieve highly incremental semantic interpretation148(Sedivy, 2003; Wolter, Gorman, & Tanenhaus, 2011).149Description matching based on linguistic content and150context-based domain restriction normally work together151to enable highly incremental interpretation. However, in152most visual world studies showing this, interpretive effects153associated with contextual inference are obscured because154these two sources of information are conflated. The current155study uses the focus particle only, which allows these two156interpretive components to be pulled apart.

1571.2. Generating expectations without descriptive content

158In contrast to gradable adjectives, focus particles do159not have intrinsic descriptive content: a focus particle160such as only does not circumscribe reference the way161property-denoting elements such as green and tall do. In162their lack of descriptive content, focus particles resemble163function words like the. Unlike the, however, focus parti-164cles are generally not confined to the noun phrase either165syntactically or semantically. Focus particles can typically166co-occur with a variety of phrasal types, depending on167what is in focus, and when a focus operator does associ-168ate with a noun phrase, such as only with some apples169in (1), the operator need not be part of or even adjacent170to that noun phrase.171We will be concerned with the focus operators only172and also, both of which specify the relation of a focused173element (object noun phrases in this study) to a set of174alternatives (including the focused element) of the same175semantic type. For example, the interpretation of (1),176repeated below as (4a), is standardly assumed to have177two meaning components (4b–c), where (4c) depends178on a set of contextually-determined alternatives such as179(4d) (Rooth, 1992).

(4) a. Jane only has some apples.b. Jane has some apples.c. Jane does not have anything in A besides

apples.d. A = {apples, oranges, tangerines, mangos,

bananas. . .}196

197While both (4a) and its counterpart without only both198say something about apples, (4a) also conveys something199about non-apples—specifically, that Jane has nothing in200the set of non-apples. But not all non-apples are equally201relevant. If they were, Jane would only need to have one202non-apple in her possession to rule out the applicability203of only—a book, say, or a toothpick, or in the more general204domain of what an individual can ‘‘have,’’ a body, a sense of

2 C.S. Kim et al. / Cognition xxx (2015) xxx–xxx


12 March 2015


205 humor, or the trivial property of being self-identical.206 Instead, (4a) conveys that with respect to some constrained207 set of alternatives of which apples is one member, Jane208 has apples but lacks all of the other alternatives. The iden-209 tity of these alternatives remains implicit in the only sen-210 tence. In (4a), description-matching suffices to identify211 what Jane has (apples), but no part of (4a) explicitly212 describes the non-apple alternatives she lacks. Thus, while213 the accuracy of (4a) as a whole rides on an appropriate214 determination of the alternative set, there is no clear point215 in the sentence where the explicit linguistic content is216 decisive with respect to alternatives.217 Some prior studies have shown that the alternatives218 associated with only are available in online sentence com-219 prehension. Ni, Crain, and Shankweiler (1996) compared220 sentences with a temporary main verb/reduced relative221 clause ambiguity like (5), which contained either the222 determiner the or the particle only.

(5) {The,Only} businessmen loaned money at lowinterest were told to record their expenses.

228

229 The alternative set evoked by only led comprehenders230 to expect modification on businessmen to distinguish the231 businessmen that the sentence was about from other busi-232 nessmen. This encouraged comprehenders to construe233 loaned as introducing a reduced relative clause, weakening234 the garden path that normally results when the sentence235 begins with a definite description (but cf. Paterson,236 Liversedge, & Underwood, 1999). Sedivy (2002) replicated237 Ni et al.’s only effect, and also showed that providing238 explicit alternatives in the prior discourse, which, by239 hypothesis, eliminates the expectation for modification,240 indeed strengthened the garden path. These results strong-241 ly suggest that the alternatives depend on the discourse242 context.243 The examples in (6) suggest that the alternatives used244 to interpret a sentence like (4a) are dependent on elements245 made salient in some way by the prior discourse, whether246 this is by explicit mention or some other more indirect247 means. Context (6a) explicitly mentions two elements248 (apples and pears) which can then be understood as the249 alternatives to apples in (4a); this is comparable to the250 explicitly provided alternatives in Sedivy (2002).

(6) a. Beth is asking Jane if she has any apples orpears.

b. Beth wants to make a pie. She’s asking herroommates if they have any ingredients tomake one.

c. Jane is alone in the kitchen at night when shehears a burglar trying to break in.She looks around for something she can pelthim with as he comes through the window.

(4) a. Jane only has some apples.274

275 Context (6b) is similar in content to (6a), but does not276 rely on explicit content to evoke the appropriate alterna-277 tives for (4a), namely kinds of fruit that might go into a

278pie. (6c) suggests that abstract aspects of the prior context279such as goals can suggest as alternatives the more hetero-280geneous collection of objects suitable for hurling at intrud-281ers as in (7).

(7) A = {apple, baseball, soup can, orange, kitchentimer. . .} 287

288While these examples show that alternatives can be289inferred from the context indirectly, it is unclear whether290alternatives associated with such potentially open-ended291categories have the same restrictive power as explicitly292mentioned alternatives.293The current paper addresses two questions about the294generation and use of expectations triggered by focus-sen-295sitive operators:

2961. What is the nature of the expectations triggered by297focus-sensitive operators? Specifically, to what extent298are alternatives calculated incrementally in the (initial)299absence of descriptive content? Do alternatives evoked300by indirect means have the same status as those intro-301duced explicitly into the discourse, and do focus opera-302tors preferentially distinguish between direct and303indirect mention?3042. Which parts of the expectations associated with the305lexical items only and also can be generalized to the306class of focus-sensitive operators, and which parts are307attributable to particular operators?308

309We can frame possible answers to the first question in310terms of two hypotheses, which we illustrate using sen-311tence (4a) and its hypothetical alternative set (4d).

3121.2.1. Lexical associates hypothesis313Comprehenders generate expectations about the314upcoming descriptive content of the sentence based on lex-315ical or conceptual similarity with prior discourse content.316For (4a), this means that comprehenders generate expecta-317tions about the focus value apples based on prior discourse318information, as in an analogous sentence without only. This319is illustrated by the context in (6b), repeated below.

(6) a. Beth wants to make a pie. She’s asking herroommates if they have any ingredients tomake one.

(4) a. Jane only has some apples.d. A = {apples, oranges, tangerines, mangos,

bananas, . . .} 334

335On this hypothesis, existing lexical representations and336their conceptual relatedness do the work of determining337which continuations are expected: there need not be any338moment to moment calculation of the relative likelihoods339of possible alternatives. The alternatives should always340be linked to the lexical content of the focus value, and they341should be predictable only to the extent that prior lexical342content increases the availability of conceptually similar343lexical material.

Q4

C.S. Kim et al. / Cognition xxx (2015) xxx–xxx 3


12 March 2015


344 1.2.2. Situation-driven alternatives hypothesis345 Comprehenders generate expectations about alterna-346 tives based on the specific properties or goals of the situa-347 tion being described. In a context like (6b), comprehenders348 will infer something like (4d) as the alternatives because349 the situation described in the prior context makes350 them salient as ingredients for making a pie; the same351 alternatives are expected under the Lexical Associates352 Hypothesis due to facilitation from prior lexical material.353 However, in a context like (6c), only the Situation-driven354 Alternatives Hypothesis predicts that comprehenders will355 infer that alternatives in (7) are more likely than other pos-356 sible alternative sets such as (4d). In the absence of direct357 conceptual similarity between prior lexical content and358 focus alternatives, the alternatives in (7) would only be359 predicted if comprehenders anticipate alternatives that360 are salient to the particular situation described.

(6) c. Jane is alone in the kitchen at night when shehears a burglar trying to break in.She looks around for something she can pelthim with as he comes through the window.

(7) A = {apple, baseball, soup can, orange, kitchentimer. . .}

375

376 If aspects of the discourse situation influence the per-377 ceived likelihood of alternatives, existing models of sen-378 tence processing would have to be enriched; for example,379 expectation-based models would need to track probability380 distributions over candidate alternatives. We address these381 issues regarding time course and contextual cues in382 Experiments 2 and 3.383 We can separate out the meaning contributions of dif-384 ferent lexical items from general properties shared by all385 members of the class by comparing only with another386 alternative-triggering lexical item, also. Like only, also has387 two meaning components—when the direct object is388 focused, a sentence like (8a) conveys both of the meanings389 in (8b) and (8c). The value of the alternative set, A, is deter-390 mined by the context, as with only. Given the particular391 alternative set in (8d) (=(4d)), (8a) ends up meaning that392 Jane has some apples, and that she has one or more of393 the other items in A.

(8) a. Jane also has some apples.b. Jane has some apples.c. Jane has something in A besides apples.d. A = {apples, oranges, tangerines, mangos,

bananas. . .}410

411 The difference in meaning between (8a) and its counter-412 part with only (4a) reflects how the focus value (apples)413 relates to the focus alternatives in A. However, the mechan-414 ism that generates and predicts likely alternatives might be415 common to both lexical items, and more generally to the416 broader class of focus sensitive operators. Characterizing417 the meaning differences associated with these operators418 would allow us to identify the broader processes involved419 in predicting alternatives. To this end, Experiment 4 com-420 pares the expectations triggered by only and also.

421Before addressing the two questions just described in422Experiments 2–4, we replicate the basic result of Sedivy423(2002) in Experiment 1 using the visual world paradigm.424In addition to showing that the interpretations of sen-425tences with only are influenced by prior discourse content,426the current methodology allows us to more precisely427examine the time course of this effect.

4282. Experiment 1: focus alternatives and discourse429mention

430Experiment 1 examines the focus particle only using431sentences like (4a), where only precedes the focused mate-432rial it associates with. Many examples from the semantics433literature illustrate how focus alternatives can be restrict-434ed by recent mention. For example, the second clause of435(9) is most naturally interpreted with Tom, Bill, and436Harry as the alternatives to Bill—that is, it conveys that that437John introduced Bill but not Tom or Harry to Sue.

(9) John brought Tom, Bill, and Harry to the party,but he only introduced [Bill]F to Sue. (Rooth,1996, example 24) 444

445In a self-paced reading study, Sedivy (2002) used446reduced relative clauses to demonstrate that mentioned447contrastive sets are construed as focus alternatives when448followed by a sentence with only, as in (10).

(10) a. All of the secretaries and accountants weremade to take a tough computing course.Only the secretaries prepared for theexam. . .

b. . . .and earned significant pay raises.c. . . .passed and earned pay raises. 465

466When the context did not establish contrast, the expecta-467tion for alternatives triggered by only made the verb pre-468pared likely to be interpreted as introducing a reduced469relative clause (10b), resulting in increased reading time470when the sentence resolved with a main clause parse471(10c). However, when the requirement for alternatives472was obviated by contrast recently established in the dis-473course, main clause continuations were more expected,474and reduced relative continuations were associated with475longer reading times.476In Experiment 1, we compare processing of sentences477with only (4a) to their counterparts lacking a focus opera-478tor (11) (target sentences).

(4) a. Jane only has some apples.(11) Jane has some apples. 488

489Listeners viewed a four-item display as they heard the tar-490get sentence. They were instructed to click on the item Jane491had (apples). Target sentences appeared in discourses that492either did or did not mention the eventual focus (the target493word, e.g. apples).494Based on Sedivy’s findings, we expect recently men-495tioned items to be naturally interpretable as focus alterna-496tives in sentences like (4a), with participants preferring



12 March 2015


497 mentioned items in the display. Secondly, comparison with498 sentences lacking only like (11) will show whether any such499 bias is a general preference, or a bias triggered specifically500 by the presence of the operator, as suggested by Sedivy501 (2002). Finally, if the expectation for recently mentioned502 items is triggered by the operator, the bias in favor of men-503 tioned items should emerge once only has been encoun-504 tered, even before any auditory information about the505 target word is available. Such time course information506 would provide additional support for the discourse-depen-507 dence of alternative-triggering focus operators like only.

508 2.1. Method

509 2.1.1. Participants510 Twenty-four undergraduate students from the511 University of Rochester, recruited from introductory512 Linguistics courses and flyers posted on campus, were paid513 $7.50. Participants were native speakers of American514 English, with normal or corrected-to-normal vision.

515 2.1.2. Materials and design516 Experimental materials consisted of twenty discourses.517 In each two-sentence discourse, the first (context) sentence518 mentioned two items, and the second (target) sentence519 mentioned a single item (the target word). Four versions520 of each item were constructed by crossing two factors:521 Mention (whether the target word had appeared in the522 context sentence) and Only (whether the target sentence523 contained only) (see Table 1). In Mention trials, the target524 word was equally likely to occur first or second in the con-525 text sentence.526 Twenty displays corresponded to the twenty527 experimental items. Each display contained four 200 !528 200 pixel images located at the corners of the529 1024 ! 768 pixel computer screen (with images flush with530 screen edges): the target referent, a competitor, and two531 unrelated distractors. The competitor was a picture whose532 name begins with the same syllable as the target word,533 making the target and competitor names phonological534 cohorts (Marslen-Wilson, 1987)—for example, a target535 apples might have the competitor axes. The reasoning536 behind using phonological cohort competitors is as fol-537 lows: if listeners were identifying the target based solely538 on the auditory input, with the target and competitor in539 the same cohort, then they would have to hear enough of

540the target word to distinguish it from the competitor541before being able to identify the target. If, however, we542observe an increase in target looks relative to competitor543looks before the auditory input would have provided suffi-544cient information to distinguish between them, then we545can infer that some other cue (here, previous mention of546the target word) is being used to identify the target refer-547ent. The target and competitor were not semantically relat-548ed. The distractor items were neither cohorts of the target549and competitor, nor of each other. The visual context550restricts the space of possible responses. Because only551one of the previously mentioned items appeared in the dis-552play on Mention trials, if participants track discourse men-553tion, they should be able to uniquely identify that item in554the display. An example display for the discourses in555Table 1 is shown in Fig. 1.556Four lists were created using a Latin square design, with557Mention and Only counterbalanced across lists. Each par-558ticipant was assigned to one list and saw one version of559each item. Experimental trials were interspersed with 60560filler trials. The position of objects in each display was ran-561domized, and a random trial order was generated for each562run of the experiment.563For all experiments, filler trials had the same format as564experimental trials (here, one context sentence followed565by the target sentence). Certain properties were distribut-566ed across filler items to minimize statistical regularities567across the materials. First, trials featuring displays with568cohort pairs were equally frequent across the entire set569of materials as those without cohort pairs; in displays with570cohort pairs, the eventual target was equally likely to be a571cohort member or one of the other two referents. Fillers572were also constructed to minimize the salience of alterna-573tive readings of sentences with only, which could have574interfered with the intended exclusive reading. For exam-575ple, only has a scalar reading under which the example in576Table 1 can be paraphrased as ‘‘Neil has a valuable item,577while Jane has a less valuable item,’’ or ‘‘Neil has two items,578while Jane has one item.’’1 To make the latter reading less579salient, target sentences were equally likely to contain two580target words (‘‘Jane has some apples and some crayons’’)581as they were to contain one target word. Care was taken582to select competitor and distractor referents for the visual583displays which did not obviously differ in inherent value.584No filler items contained only.585Sentences were recorded by a native speaker of586American English. In all experiments, the same context587sentence recordings were used across trials sharing the588same context sentence. The sentences were pronounced589with prominence on the final noun of the object noun590phrase in both only and non-only sentences, a pattern con-591sistent with either direct object focus or unmarked whole-592sentence focus (Ladd, 1996). Readers might be concerned

Table 1Design and example stimuli for Experiment 1.

Only No only

Mention Neil has someapples and somecards

Neil has someapples and somecards

Jane only hassome apples

Jane has someapples

No mention Neil has somelanterns andsome cards

Neil has somelanterns andsome cards

Jane only hassome apples

Jane has someapples

1 The scalar reading of only is particularly salient since sentence-finalprominence is compatible with both scalar only and the (intended)exclusive reading. A potential concern is that prominence on only mayhighlight the exclusive reading, while final prominence highlights thescalar reading. While this may be the case, our target sentences all havefinal prominence, therefore any differences observed cannot be attributedto the relative availability of the exclusive and scalar readings.



12 March 2015


593 that any differences in timing could be due to prosodic594 characteristics of the Only and No only stimuli. We include595 files with the stimuli as Supplemental materials for inter-596 ested readers. In Appendix A we present a discussion of597 the prosody issues, including details of acoustic analyses598 that we conducted, and a discussion of how these analyses599 address concerns about the interpretation of the eye-600 movement data that we report in the results section.601 Trials were presented in a random order generated on602 each run of the experiment. Four practice trials—none con-603 taining only or an image that appeared as a target referent604 in an experimental trial—preceded the 80 trials.

605 2.1.3. Procedure606 Each trial began with the participant fixating and click-607 ing on a crosshair in the center of the screen. Participants608 listened to the context and target sentences over head-609 phones. The display appeared on the computer screen at610 the onset of the target sentence; there was no preview.611 Participants were instructed to click on the items that612 the second mentioned character had, corresponding to613 the focused element in the target sentence. The trial ended614 when the participant clicked on a picture. Eye movements615 were recorded from the onset of the target sentence to the616 end of the trial, using a head-mounted SR EyeLink II eye-617 tracking system sampling at 250 Hz.

618 2.1.4. Modeling619 The data were analyzed using mixed-effect logistic620 regression models with Participant and Item as random621 effects (Barr, 2008; Jaeger, 2008).2 The models predicted622 fixations to the target referent, and included the following623 fixed effects: (1) the presence/absence of only, (2) whether624 or not the target was mentioned in the preceding sentence,625 and (3) time (in seconds, with one data point sampled every626 4 ms). Data was unaggregated in all the regression models627 reported. In addition, we included the state of the previous628 fixation (on or off the target) as a predictor to deal with

629the oversampling problem that arises in analyses of visual630world fixation data.3

631All analyses began with the full model, which included632all interactions among Mention, Only and Time. The State633term was left in the model, regardless of significance. All634predictors were centered. Redundant terms were removed635by eliminating one predictor at a time for all terms corre-636lated with one or more other terms in the model, starting637with the highest order term. Model comparison using the638likelihood ratio test determined whether the model includ-639ing the predictor increased the likelihood of the data rela-640tive to the model excluding that term. All analyses641followed the procedure in Barr, Levy, Scheepers, and Tily642(2013) to determine the maximal random effects structure643supported by the data. For each model with a given fixed644effects structure, we started with the model containing645the maximal random effects structure for Participant and646Item, and iteratively removed random effects from the647model if the model failed to converge. The final model is648given in the table of model coefficients. When interaction649terms in the regression models were significant, we650assessed the simple slopes associated with each level of651the relevant predictor by fitting the regression model652separately for each level (Aiken & West, 1991).653Two analysis windows were delimited by salient lin-654guistic events in the stimuli. The early window started at655the onset of the particle only for Only conditions and the656onset of the main verb for No only conditions, and ended657at the onset of the target word (spanning on average658573 ms). The late window started at the onset of the target659word, and ended 500 ms after the onset of the target word.

6602.2. Results

661We present graphs of proportions of fixations to poten-662tial referents (Fig. 2), planned comparisons of target and663competitor fixations over 100 ms intervals, and logistic664regression models predicting target fixations in two time665windows (Tables 2–5).666Fig. 2 shows the mean proportion of fixations to the tar-667get and competitor, and the averaged fixations to the two668distractors, for the four experimental conditions (aggregat-669ed over 100 ms bins in proportion of fixation plots). Data670are aligned to the onset of the target word. In both No671Mention conditions and the Mention-No Only condition,672fixations to the target did not exceed competitor and dis-673tractor fixations until well after 200 ms after target word674onset. In contrast, target fixations in the Mention-Only con-675dition began to increase relative to fixations to the other676scene referents approximately 200 ms after word onset.

Fig. 1. Example display for Experiment 1 (labels for illustration only).

2 See Brown-Schmidt (2009), Brouwer, Mitterer, and Huettig (2012),Mack, Ji, and Thompson (2013) for similar logistic regression analyses ofvisual world eye-tracking data.

3 Looks to objects in a display arise as a result to saccadic eye movementswhich shift the listeners gaze to attended regions; however, eye move-ments take about 150–200 ms to plan and execute. Thus, at the moment intime sampled by the eye-tracker, the best predictor of whether or not aparticipant is looking at the target is whether she was looking at the targeton the previous sample. In the absence of using event-based analyses,which predict the onset of a saccade to a location, using State as a predictoris a principled, hypothesis-neutral way of dealing with the oversamplingproblem. We do not include interactions with State. There is no cleartheoretical reason for State to interact with the variables of interest, andany such interactions would not be interpretable.



12 March 2015


677 Signal-driven eye movements in action-based visual world678 experiments with four-picture displays occur with a mini-679 mal lag of about 200 ms (Allopenna, Magnuson, &680 Tanenhaus, 1998; Salverda, Kleinschmidt, & Tanenhaus,681 2014; but cf. Altmann, 2011). Therefore, fixations begin682 to converge on the target in the context of only before683 the point in time when fixations could reflect a change684 due to the auditory information of the target word.

685To assess the extent to which the competitor competed686with the target referent, we determined when target and687competitor fixations diverged in each condition by conduct-688ing planned comparisons over 100 ms intervals beginning at689the onset of the target word. All planned comparisons were690performed on log-odds transformed proportions of691fixations. Target fixations reliably exceeded competitor

Fig. 2. Mean proportion of fixations to display items, Experiment 1. Top left = No Mention-No only condition; top right = No Mention-Only; bottomleft = Mention-No only; bottom right = Mention-Only. Vertical lines represent (left to right): average focus particle onset; target word onset. Error barsrepresent standard error.

Table 2Estimates of fixed effects, Experiment 1—early window.

TargetFix " Only + Mention + Time + State +(1|Participant) + (1|Item)

Estimate SE z p

Intercept #6.90 0.37 #23.32 <0.0001Only #0.099 0.23 #0.44 n.s.Mention 0.48 0.18 2.64 <0.01Time 3.63 0.34 10.83 <0.0001State 10.41 0.29 35.85 <0.0001

Table 3Correlations of fixed effects, Experiment 1—early window.

Intercept Only Mention Time

Only 0.26Mention 0.13 0.073Time #0.54 0.014 0.064State #0.61 #0.041 0.026 0.55

Table 4Estimates of fixed effects, Experiment 1—late window.

TargetFix " Only + Mention + Time + State + Only:Mention +(1 + Mention|Participant) + (1|Item)

Estimate SE z p

Intercept #6.49 0.23 #28.14 <0.0001Only #0.18 0.22 #0.83 n.s.Mention 0.29 0.21 1.36 n.s.Time 2.18 0.49 4.41 <0.0001State 10.80 0.17 65.34 <0.0001Only:Mention 0.66 0.29 2.28 <0.05

Table 5Correlations of fixed effects, Experiment 1—late window.

Intercept Only Mention Time State

Only #0.43Mention #0.52 0.47Time #0.60 #0.012 0.013State #0.34 0.010 0.013 0.14Only:Mention 0.31 #0.76 #0.66 0.012 0.071



12 March 2015


692 fixations in the 200–300 ms window for the Only-Mention693 condition, t(46) = 2.27, p < 0.05 (100–200 ms window,694 t(46) = 0.85, n.s.), but not until the 400–500 ms window for695 the No only-Mention condition, t(46) = 3.05, p < 0.005696 (300–400 ms window, t(46) = 0.76, n.s.), and the 600–697 700 ms window for both Only-No mention, t(46) = 3.50,698 p < 0.005, and No only-No mention, t(46) = 2.75, p < 0.01699 (500–600 ms window, t(46) = 0.34, n.s. and t(46) = 1.08,700 n.s., respectively).701 Estimates of the coefficients corresponding to fixed702 effects in the regression models, and the correlations703 among fixed effects, are given in Tables 2 and 3 for the ear-704 ly analysis window, and in Tables 4 and 5 for the late win-705 dow. In the early window, there were main effects of Time706 and State. Participants were more likely to fixate the target707 later in the trial and fixation on the target at the previous708 time point was highly predictive of fixation on the target709 at the current time point (these effects will be significant710 in general for all the models presented, so we will not dis-711 cuss them further). There was a main effect of Mention:712 when the target word had been mentioned in the previous713 sentence, participants were more likely to fixate that refer-714 ent. There was no significant effect of only, and none of the715 interactions among Only, Mention, and Time survived716 model comparison.717 In the late window, the main effect of Mention is no718 longer reliable. There was also a two-way interaction719 between presence of only and Mention. Analysis of simple720 slopes revealed a significant facilitative effect due to721 Mention when only was present (b = 0.95, z = 4.32,722 p < 0.0001), but not in sentences without only (b = 0.29,723 z = 1.36, p > 0.1).4

724 2.3. Discussion

725 The main effect of Mention in the early window raises726 the possibility that listeners generally have a higher expec-727 tation for previously mentioned material in upcoming dis-728 course than for discourse-new material. We note, however,729 that the materials used in Experiment 1 leave open the

730possibility that the Mention effect might be subsumed by731a more general bias for discourses to remain on topic. We732will address this possibility in Experiments 2 and 3 by733using materials that distinguish between alternatives734introduced by explicit mention and those introduced more735indirectly. This will also allow us to evaluate the hypoth-736esis that only might have a specific preference for explicitly737mentioned alternatives.738The Mention-Only interaction that emerges in the late739window replicates Sedivy’s (2002) finding that, specifically740in sentences with only, discourse-given alternatives affect741how subsequent sentences are interpreted. Since primary742prominence is on the direct object for both Only and No743only sentences, and both sentence types are consistent744with direct object focus, the finding that only modulated745the Mention bias suggests that the presence of an overt746operator makes the crucial difference in restricting alterna-747tives for subsequent interpretation.748When the operator was present, the effects of the749Mention bias preceded acoustic phonetic input about the750target word. The visual context is a powerful constraint751on the space of possible interpretations (for review see752Altmann & Mirkovic, 2009). Given the already restricted753referential domain specified by the visual context, par-754ticipants were able to identify a probable target referent755based on the presence of only alone, even before any infor-756mation about the target word itself was available. Taken757together, the results show that interpretation is incremen-758tal even in the face of radical indeterminacy with respect to759descriptive content. The sensitivity of focus-sensitive760operators to explicit mention is consistent with previous761studies that have shown that listeners keep track of old/762new status (e.g., Arnold, Altmann, Fagnano, & Tanenhaus,7632004; Arnold, Fagnano, & Tanenhaus, 2003; Kaiser &764Trueswell, 2004; Wolter et al., 2011).

7653. Experiments 2 and 3: sources of focus alternatives

766Experiments 2 and 3 examine how alternatives are con-767structed when they are not introduced explicitly, focusing768on predictions made by the Lexical Associates and769Situation-driven Alternatives Hypotheses. In addition, use770of potential alternatives that are not explicitly mentioned771allows us to evaluate whether or not there is a general772mention bias and whether only is biased toward focus773alternatives that have been explicitly mentioned.774According to the Lexical Associates Hypothesis, the775ongoing retrieval and activation of lexical–conceptual rep-776resentations during discourse processing supplies focus777alternatives when required by a focus operator.778Mechanistically, retrieving the meaning of a concept ‘‘acti-779vates’’ semantic neighbors in a gradient manner (Collins &780Loftus, 1975; Cree, McRae, & McNorgan, 1999). When alter-781natives are not explicitly supplied by the discourse context,782the processor may use this existing mechanism as a source783of alternatives. According to this hypothesis, alternatives784inferred from prior discourse are always linked to specific785lexical items; they should be more predictable when they786are closer lexical neighbors to mentioned elements.787Such a system, with focus operators piggybacking on788existing lexical retrieval mechanisms, is compatible with

4 We also conducted ANOVAs over log-odds transformed proportions oftarget fixations (Jaeger, 2008), using the same two analysis windows as forthe regression models. The results were comparable to the regressionanalyses. In the early window, the advantage due to Mention wassignificant in the by-Subjects (F1(1,20) = 16.2, MSE = 1243.5, p < .0001)and by-Items (F2(1,16) = 4.3, MSE = 250.5, p < .05) analyses; there was anadvantage for No only sentences which was marginal by Subjects(F1(1, 20) = 3.3, MSE = 255.8, p = .07) and significant by Items(F2(1,16) = 4.2, MSE = 247.4, p < .05); the Mention-Only interaction wasonly marginal by Items (F1(1,20) = .85, MSE = 64.8, n.s.; F2(1,16) = 3.3,MSE = 192.8, p = .07). In the late window, the Mention advantage remainedreliable (F1(1, 20) = 72.6, MSE = 5332.0, p < .0001; F2(1, 16) = 86.5,MSE = 4718.0, p < .0001); there was an Only advantage, reliable only bySubjects (F1(1, 20) = 7.4, MSE = 543, p < .01; F2(1, 16) = 2.9, MSE = 157,p = .09); importantly, the Mention-Only interaction was significant(F1(1,20) = 6.8, MSE = 498, p < .01; F2(1,16) = 8.9, MSE = 487, p < .005). Posthoc comparisons revealed that the Mention advantage was highly reliableby Subjects and Items when the target sentence contained only,t1(766) = 5.1, p < .00001; t2(638) = 5.3, p < .00001, but reliable only bySubjects when it did not, t1(443) = 1.9, p = .05; t2(335) = 1.4, n.s. Theprimary effects of interest are the same in both regression and ANOVAanalyses: (i) there is facilitation due to Mention in the early window, and(ii) the Mention-Only interaction is reliable only in late window.



12 March 2015


789 the formal account of alternative generation proposed by790 Blok and Eberle (1999), where alternatives are identified791 with elements in a sort hierarchy that combines syntactic792 and conceptual information. Crucially, alternatives are793 determined relative to a specific lexical item. Using the794 example in (4a), the focused element apples is immediately795 dominated in the sort hierarchy by the sort APPLES, which is796 classified both in terms of its syntactic properties (e.g. plu-797 ral, count noun) and its conceptual properties (e.g. subtype798 of the conceptual class FRUIT). The alternatives to apples are799 dependents of the category FRUIT for which the language800 has a lexical item, in other words, something approximated801 by (4d).

(4) d. A = {apples, oranges, tangerines, mangos,bananas. . .}808

809 Blok and Eberle’s proposal assumes that the focused810 element is the input to computing alternatives, but this811 could be adapted to take as input lexical items appearing812 in the preceding discourse. A context mentioning oranges813 and mangos would generate the alternatives in (4d), and814 explain why members of the category FRUIT are readily815 interpreted as alternatives to apples. However, because816 the conceptual component of Blok and Eberle’s hierarchy817 is meant to be constant across languages and speakers,818 there is no room for individual experience to alter category819 structure in that framework.820 By contrast, the Situation-driven Alternatives821 Hypothesis states that focus alternatives are restricted822 based on a composite representation of the situation823 described, rather than associations with individual lexical824 items. According to this hypothesis, understanding dis-825 course involves constructing a mental representation of826 situations or events (see Zwaan & Radvansky, 1998 for a827 review); concepts are selectively activated according to828 their relevance to a specific situation. Often, both829 Situation-driven and Lexical Hypotheses will make the830 same predictions: a context like (6b) (repeated below)831 could make apples more expected as an alternative because832 it is lexically related to words like pie, or because it is rele-833 vant to a situation where someone is looking for potential834 pie ingredients. However, only the Situation-driven835 Hypothesis predicts that alternatives will be expected836 based on their situation relevance even when they lack837 inherent similarity with prior lexical content, as in (6c).

(6) b. Beth wants to make a pie. She’s asking herroommates if they have any ingredients tomake one.

c. Jane is alone in the kitchen at night when shehears a burglar trying to break in.She looks around for something she can pelthim with as he comes through the window.854

855 Experiments 2 and 3 ask whether sharing a conceptual856 category with a mentioned item facilitates target identifica-857 tion, independent of discourse old–new status. Experiment858 2 uses lexically-based categories, whose members share a859 set of properties (e.g. apple, orange, mango are members of860 the category fruit). Experiment 3 uses ad hoc categories

861associated with specific situations (e.g. hot dogs, nachos,862Coke are items one might buy at a baseball game). The863Lexical Associates Hypothesis predicts the lexically-an-864chored alternatives in Experiment 2, but not the situation-865specific alternatives in Experiment 3. The Situation-driven866Alternatives Hypothesis predicts expectations about situa-867tion-relevant alternatives in both Experiments 2 and 3868(including lexically associated ones). In both cases, we also869ask whether category effects are limited to sentences with870only, like the mention effect from Experiment 1, or whether871it is observed across the board, as would be expected if such872effects reflect general-purpose processes that occur regard-873less of the presence of a focus operator.

8743.1. Experiment 2: lexically-anchored alternatives

8753.1.1. Method8763.1.1.1. Participants. Twenty-four undergraduate students877from the University of Rochester were recruited from878introductory Linguistics courses and flyers posted on the879university campus, and were paid $7.50. All participants880were native speakers of American English, with normal881or corrected-to-normal vision. For all experiments present-882ed, participants were not excluded for previous participa-883tion in another of the experiments as long as the sessions884were separated by at least one academic semester.

8853.1.1.2. Materials and design. Experimental materials were88624 discourses, each with a context sentence followed by a887target sentence. Six lists were created according to a Latin888square design, counterbalancing ContextType (whether889the context sentence explicitly mentioned the target word,890mentioned members of the same conceptual category as891the target word, or mentioned members of a different con-892ceptual category) and Only (whether the target sentence893contained only) (see Table 6). We refer to the two condi-894tions without explicit mention as Same Category Novel895and Different Category Novel, respectively. The Explicit896mention and Different category conditions correspond to897the Mention and No mention condition from Experiment 1.898The display for each test item had a target referent cor-899responding to the target word, a competitor in the same900phonological cohort as the target, and two unrelated901distractors.902Participants heard each experimental item in one of six903conditions (Table 6). Experimental trials were interspersed904with 48 filler trials designed to eliminate statistical regula-905rities in the materials. Trials featuring displays with cohort906pairs were equally frequent as those without cohort pairs;907in displays with cohort pairs, the target was equally likely908to be a cohort member or one of the other referents. Across909the entire set of materials, 42% of target sentences con-910tained two target words, and 29% of trials had displays911with cohort pairs. A native speaker of American English912recorded the discourses. The experiment began with four913practice trials. The procedure was the same as in914Experiment 1.

9153.1.1.3. Modeling. Three time windows were analyzed: (i)916the pre-particle window, the 500 ms window ending at

Q5



12 March 2015


917 the particle onset for Only conditions and at the onset of918 the main verb for No-Only conditions (the earliest segment919 of the trial, when the visual display and the initial word in920 the auditory stimulus are available), (ii) the pre-target win-921 dow, starting at particle or main verb onset and ending at922 target word onset (spanning on average 551 ms), and (iii)923 the post-target window, starting at target word onset and924 ending 500 ms after target onset. To more directly compare925 looks to same category targets relative to mentioned and926 different category targets, we show the mean proportion927 of fixations to only the target referents across conditions.928 These are shown in Fig. 3 for the three No-Only conditions929 and the three Only conditions. Time is aligned to the onset930 of the target word.931 Results were assessed by fitting target fixations from932 the experimental conditions with mixed-effect logistic933 regression models, using the three analysis windows.934 Models included fixed effects of Only, Target Type, Time935 and State. Target Type was Helmert-coded (the first con-936 trast compares Same category Novel with Different catego-937 ry Novel; the second contrast compares Mentioned with938 both Same category and Different category Novel). Only939 and State were contrast-coded. Fixed and random effects940 structures were determined as described above.

9413.1.2. Results942We present regression models predicting target fixa-943tions in the pre-particle, pre-target and post-target analy-944sis windows (Tables 7–9). Graphs of the proportions of945target fixations are given in Fig. 3.946In the pre-particle window (Table 7), there is already a947main effect of Category status—when only visual informa-948tion about the display referents is available early in the tar-949get sentence, listeners prefer referents sharing a conceptual950category with recently mentioned elements (this is indicat-951ed by the significance of the second Helmert contrast for952the TargetType variable, which compares Same vs.953Different Category targets). This effect increases over954Time, as Different category referents become increasingly


Only No only

Explicit mention Neil has some pearsand some apples

Neil has some pearsand some apples

Alex only has someapples

Alex has some apples

Same categoryNovel

Neil has some pearsand some oranges

Neil has some pearsand some oranges



DifferentcategoryNovel

Neil has some sandalsand some boots

Neil has some sandalsand some boots



Fig. 3. Mean proportions of target fixations, Experiment 2. Left = No Only condition; right = Only. Vertical lines indicate (left to right): average targetsentence onset, average particle onset, target word onset.

Table 7Estimates of fixed effects, Experiment 2—pre-particle window.

TargetFix " TargetType + Only + Time + State +TargetType:Only + TargetType:Time + Only:Time +TargetType:Only:Time + (1 + TargetType|Participant) +(1|Item)

Estimate SE z p

Intercept #6.04 0.51 #11.90 <0.0001TargetType[Same v.

Different Category]1.72 0.61 2.80 <0.01

TargetType[Mentioned v.Same/Diff Cat]

0.72 0.37 1.98 <0.05

Only #1.68 0.64 #2.64 <0.01Time #0.35 0.59 #0.59 n.s.State 10.25 0.13 80.22 <0.0001TargetType[Same v. Diff

Cat]:Only#1.59 0.84 #1.90 0.06

TargetType[Mentioned v.Same/Diff Cat]:Only

#1.47 0.42 #3.47 <0.001

TargetType[Same v. DiffCat]:Time

2.28 0.73 3.12 <0.005

TargetType[Mentioned v.Same/Diff Cat]:Time

0.44 0.40 1.10 n.s.

Only:Time #2.06 0.78 #2.65 <0.01TargetType[Same v. Diff

Cat]:Only:Time#2.35 1.02 #2.31 <0.05

TargetType[Mentioned v.Same/DiffCat]:Only:Time

#1.75 0.51 #3.40 <0.0001



12 March 2015


955 dispreferred (b = #3.07, z = #3.06, p < 0.005). There is also a956 main effect of Mention: fixations to explicitly mentioned957 targets are facilitated relative to (Same or Different catego-958 ry) novel target fixations. Interestingly, there is a negative959 effect of Only, which decreases over Time, as indicated by960 the negative Only:Time interaction in the model. To the961 extent that participants were able to pick up on prosodic962 cues to the presence of only this briefly decreased fixations963 to the target referent. There is also a negative interaction964 between Mention and the presence of Only, in the opposite965 direction from the Mention-Only interaction in the late966 window in Experiment 1: Mention had a facilitative effect967 in sentences without only (b = 0.72, z = 1.97, p < 0.05), and968 decreased target looks in sentences with only (b = #0.74,969 z = #2.30, p < 0.01).970 The Same category advantage persists in the pre-target971 analysis window (particle onset to target word onset), sug-972 gesting that listeners expect continuations that share a973 conceptual category with recently mentioned content.974 None of the remaining fixed effects except for State were975 reliable predictors of target fixations (Table 8).976 The Same category effect remains reliable in the post-977 target window (beginning at the onset of the target word).978 In addition, a Mention-Only interaction emerges: previous979 mention facilitates target identification relative to no-

980mention in discourses with only (b = 0.31, z = 3.05,981p < 0.005), but not in those without only (b = 0.014,982z = 0.12, p > 0.1). This replicates the Mention-Only interac-983tion in the late window in Experiment 1. The negative984Same category-Time and Mention-Only-Time interactions985indicate that the strength of the Category and Mention-986Only effects decrease with Time, as fixations converge on987the target across conditions late in the trial (Table 9).

9883.1.3. Discussion989The results of Experiment 2 suggest that conceptual990associates of recently retrieved lexical items are interpret-991ed as focus alternatives when required by the presence of a992focus operator. This result is predicted by the Lexical993Associates Hypothesis. However, as discussed earlier, these994lexical effects would also be predicted by the Situation-dri-995ven Alternatives Hypothesis.996The Mention and Same category effects have clearly dif-997ferent profiles. Explicit mention (Experiments 1 and 2)998selectively facilitates alternatives when only is present,999whereas category status influences expectations about1000the target referent irrespective of the presence of only. In1001fact, listeners prefer same category referents in the earliest1002window, when visual context and discourse context from1003the preceding sentence are available, but it is not yet1004known whether the target sentence will contain a focus1005particle. This suggests that the mechanism underlying1006the category effect is not dedicated to generating focus1007alternatives, but is more general. For example, the category1008effect may reflect listeners’ expectations that a discourse1009will generally stay ‘‘on topic’’ irrespective of discourse-1010old/new status (Ochs, 1979; Segal, Duchan, & Scott,10111991), in the absence of a clear indication of topic shift.1012We remain agnostic about whether the facilitation for1013same category items can be explained by a general1014phenomenon such as lexically-based semantic priming1015(Yee & Sedivy, 2006)—this would be consistent with the1016Lexical Associates Hypothesis. However, an explanation1017that only appeals to priming would not explain the interac-1018tions with only in the case of explicit mention. In1019Experiment 3, we turn to ad hoc categories, which are1020not expected to be facilitated under the Lexical1021Hypothesis, and cannot be explained by semantic priming.

Table 8Estimates of fixed effects, Experiment 2—pre-target window.

TargetFix " TargetType + Only + Time + State +TargetType:Only + (1+Only|Participant) + (1|Item)

Estimate SE z p

Intercept #5.46 0.13 #41.99 <0.0001TargetType[Same v. Diff

Category]0.21 0.089 2.37 <0.05

TargetType[Mentioned v.Same/Diff Cat]

0.041 0.049 0.84 n.s.

Only 0.014 0.14 0.10 n.s.Time 0.14 0.31 0.45 n.s.State 9.91 0.10 98.04 <0.0001TargetType[Same v. Diff

Cat]:Only#0.17 0.12 #1.37 0.17

TargetType[Mentioned v.Same/Diff Cat]:Only

0.034 0.070 0.49 n.s.

Table 9Estimates of fixed effects, Experiment 2—post-target window.

TargetFix " TargetType + Only + Time + State + TargetType:Only + TargetType:Time + Only:Time + TargetType:Only:Time +(1|Participant) + (1|Item)

Estimate SE z p

Intercept #5.78 0.18 #32.58 <0.0001TargetType[Same Category] 0.39 0.19 2.04 <0.05TargetType[Mentioned] 0.014 0.11 0.12 n.s.Only 0.40 0.21 1.85 0.06Time 3.03 0.53 5.78 <0.0001State 10.36 0.12 84.24 <0.0001TargetType[SameCat]:Only #0.44 0.26 #1.66 0.10TargetType[Mentioned]:Only 0.30 0.15 1.99 <0.05TargetType[SameCat]:Time #1.34 0.65 #2.08 <0.05TargetType[Mentioned]:Time 0.53 0.37 1.46 0.14Only:Time #0.99 0.71 #1.39 0.16TargetType[SameCat]:Only:Time 2.97 0.87 3.42 <0.001TargetType[Mentioned]:Only:Time #2.65 0.51 #5.22 <0.001



12 March 2015


1022 3.2. Experiment 3: ad hoc categories and situation-relevant1023 alternatives

1024 Experiment 3 addresses effects of prior mention of1025 instances of a conceptual category on subsequent process-1026 ing of same category material, using ad hoc categories1027 evoked by the goals or expectations associated with a1028 situation in the world. For instance, items like hot dogs1029 or nachos could become salient alternatives in a situation1030 where something is being purchased at a baseball game—1031 not because they inherently share features with baseball,1032 but because our cumulative experience with baseball1033 games tells us that hot dogs and nachos are likely to be1034 sold there. The current experiment asks whether, in a base-1035 ball game scenario, hot dogs and nachos become salient1036 alternatives, despite having little inherent conceptual simi-1037 larity with mentioned items. If category effects resembling1038 those in Experiment 2 are observed, their explanation must1039 involve experience-based inference about likely situations1040 in the world—as expected on the Situation-driven1041 Alternatives Hypothesis—in addition to any lexically-based1042 semantic priming.1043 We created pairs of scenarios, where one scenario was1044 compatible with a narrow set of situations (Biasing con-1045 text), and the other with a wider range of situations1046 (Neutral context). The biasing contexts are analogous to1047 the Same category conditions in Experiment 2—here,1048 rather than evoking lexically-anchored categories, contex-1049 tual bias evoked highly constrained situations which1050 should support formation of goal-related categories.

1051 3.2.1. Method1052 3.2.1.1. Participants. Thirty-eight undergraduates from the1053 University of Rochester, recruited from introductory1054 Linguistics courses and flyers posted on campus, were paid1055 $7.50. All participants were native speakers of American1056 English with normal or corrected-to-normal vision.

1057 3.2.1.2. Materials and design. Experimental materials were1058 32 three-sentence discourses. Each discourse had an initial1059 context sentence describing a setting, a second context1060 sentence, and a target sentence describing what two indi-1061 viduals want to buy. The design crosses Mention (whether1062 the target word appeared in the second context sentence),1063 Bias (whether the initial sentence described a biasing sce-1064 nario consistent with a narrow set of outcomes, or a1065 relatively neutral scenario), and Only (whether only1066 appeared in the target sentence). Eight lists were created1067 using a Latin square design; Table 10 shows an example1068 item.1069 Displays were constructed for each test item. Each dis-1070 play had a referent corresponding to the target word, a1071 competitor in the same phonological cohort as the target,1072 and two distractors. Displays for neutral conditions had1073 four referents compatible with the scenario; in biasing1074 contexts only one referent was compatible with the1075 scenario.1076 Participants heard each experimental item in one of the1077 eight conditions shown in Table 10. Experimental trials1078 were interspersed with 53 filler trials to minimize statisti-1079 cal regularities in the materials. Trials featuring displays

1080with cohort pairs were equally frequent as those without1081cohort pairs; in displays with cohort pairs, the target was1082equally likely to be a cohort member or one of the other1083referents. In addition, half of the target sentences con-1084tained two target words, and half contained one. A native1085speaker of American English recorded the discourses. The1086experiment began with four practice trials. The procedure1087was as in Experiment 1.

10883.2.1.3. Modeling. We analyzed fixations in three time win-1089dows: (i) the pre-particle window, spanning the 500 ms1090before particle onset for Only conditions and before main1091verb onset for No-Only conditions, (ii) the pre-target win-1092dow, starting at particle or main verb onset and ending at1093target word onset (spanning on average 918 ms), and (iii)1094the post-target window, the 500 ms window starting at1095the target word onset. For each analysis window, results1096were assessed by fitting target fixations from the eight1097experimental conditions using mixed-effect logistic regres-1098sion models. Models included fixed effects of Only,1099Mention, Bias, Time and State. All predictors were con-1100trast-coded.

11013.2.2. Results1102We present regression models predicting target fixa-1103tions in the three analysis windows (Tables 11–13).1104Graphs of the proportions of target fixations are shown in1105Fig. 4. The top two panels of Fig. 4 show the proportion1106of target fixations for the Neutral context conditions. The1107bottom panels show the corresponding Biasing context1108conditions. Time is aligned to the onset of the target word.1109In the pre-particle window (Table 11), none of the pre-1110dictors reach significance (except State). Effects of Mention1111and Bias emerge in the pre-target window (Table 12; par-1112ticle or main verb onset to target word onset). As in1113Experiments 1 and 2, there is a Mention-Only interaction:1114prior mention facilitates target identification to a greater1115extent when only is present (b = 0.59, z = 8.25,1116p < 0.0001), and has the reverse effect, decreasing the like-1117lihood of target fixations, when it is not (b = #0.26,1118z = #3.58, p < 0.001). The Mention-Only interaction1119strengthens over the early window, as indicated by the1120Mention-Only-Time interaction. In addition, there is a main


Biasingcontext

Mention Neil and Alex are at the baseball gameAlex wants to buy some hot dogs andsome nachosNeil (only) wants to buy some hot dogs

Nomention

Neil and Alex are at the baseball gameAlex wants to buy some Coke and somenachosNeil (only) wants to buy some hot dogs

Neutralcontext

Mention Neil and Alex are at the supermarketAlex wants to buy some hot dogs andsome cherriesNeil (only) wants to buy some hot dogs

Nomention

Neil and Alex are at the supermarketAlex wants to buy some bell peppers andsome cherriesNeil (only) wants to buy some hot dogs



12 March 2015


1121 effect of Bias. Fixations converge on the target earlier in1122 biasing contexts than in neutral contexts. Like lexical1123 category in Experiment 2, the contextual bias effect does1124 not interact with the presence of only.1125 The main effect of Bias persists in the post-target win-1126 dow (Table 13; beginning at the onset of the target word).1127 None of the Mention or Only terms from the early window1128 remain significant as all fixations converge on the target1129 referent toward the end of the trial.

1130 3.2.3. Discussion1131 Ad hoc categories evoked by particular situation types1132 (represented by contextual bias) pattern like the lexical

1133categories in Experiment 2: category information influ-1134ences expectations about upcoming discourse across the1135board, rather than specifically when required by particles1136like only. However, since ad hoc categories are made up1137of elements that lack inherent conceptual similarity, the1138bias effect in Experiment 3 cannot be explained entirely1139by lexically-based priming, or as a by-product of lexical1140access (the Lexical Associates Hypothesis). Rather, expecta-1141tions based on these categories require situation-specific1142reasoning about likely outcomes, as predicted by the1143Situation-driven Alternatives Hypothesis.1144Together, the results of Experiments 2 and 3 show that1145listeners generate expectations about upcoming discourse1146content based in part on conceptual properties of recently1147mentioned material. Although the (lexical and ad hoc) cate-1148gories evoked by prior discourse can serve as a source of1149alternatives when this is required by the presence of a1150focus operator, the generality of the category effects1151suggests they reflect mechanisms involved in general1152discourse processing. Indeed, since Same Category1153(Experiment 2) and Biasing Context (Experiment 3) target1154words were always discourse-new, the bias toward same1155category referents cannot be a result of listeners keeping1156track of previously mentioned discourse referents. By con-1157trast, the mention bias is modulated by the presence of1158only, suggesting that introducing alternatives into a dis-1159course by explicit mention is qualitatively different from1160increasing the salience of certain elements by other, indi-1161rect means such as naming a superordinate category.

11624. Experiment 4: the lexical contribution of the focus1163operator

1164We now turn to the second question introduced in1165Section 1: Which parts of the expectations associated with1166the lexical items only and also can be generalized to the1167class of focus-sensitive operators, and which parts are1168attributable to particular operators? By looking at other1169alternative-triggering lexical items, we can begin to sepa-1170rate out the meaning contributions of different focus1171operators from general properties shared by all members1172of the class.1173Based on Experiment 1, listeners seem to readily inter-1174pret the set of recently mentioned items as a restriction on1175the set of potential alternatives. The restricted alternatives1176together with the meaning of only gave rise to the expecta-1177tion that the upcoming focus would be one of the1178mentioned items—that is, a subset of the restricted alterna-1179tives. In order to tease apart the contributions of general1180focus processing and lexical content, Experiment 4 com-1181pares only with discourses like (12), featuring also, another1182alternative-triggering focus operator.

(12) Jane also has some apples. 1187

1188We restrict ourselves to cases where also associates1189with the direct object of the sentence, as opposed to the1190subject. For a sentence with also, VP or direct object focus1191is indicated by prominence on the focus value, and lack of1192prominence on the particle. Based on Experiment 1, we1193assume that the set of items mentioned in the context

Table 11Estimates of fixed effects, Experiment 3—pre-particle window.

TargetFix " Bias + Mention + Only + Time + State +Mention:Only + Only:Time + (1|Participant) + (1|Item)

Estimate SE z p

Intercept #11.16 4.50 #2.48 <0.05Bias 0.51 0.75 0.68 n.s.Mention #0.87 0.99 #0.88 n.s.Only #9.76 7.58 #1.29 n.s.Time 1.43 2.91 0.49 n.s.State 15.13 1.28 11.86 <0.0001Mention:Only 1.61 1.48 1.10 n.s.Only:Time #5.65 4.72 #1.20 n.s.

Table 12Estimates of fixed effects, Experiment 3—pre-target window.

TargetFix " Bias + Mention + Only + Time + State +Bias:Mention + Bias:Only + Bias:Time + Mention:Only +Mention:Time + Only:Time + Mention:Only:Time +(1|Participant) + (1|Item)

Estimate SE z p

Intercept #6.05 0.16 #38.81 <0.0001Bias 0.31 0.16 2.02 <0.05Mention #0.36 0.19 #1.88 0.06Only #0.24 0.19 #1.23 n.s.Time #0.25 0.27 #0.92 n.s.State 10.71 0.066 162.73 <0.0001Bias:Mention #0.068 0.13 #0.53 n.s.Bias:Only #0.10 0.13 #0.78 n.s.Bias:Time 0.12 0.24 0.50 n.s.Mention:Only 0.67 0.25 2.62 <0.01Mention:Time #0.61 0.34 #1.80 0.07Only:Time #0.061 0.34 #0.18 n.s.Mention:Only:Time 1.03 0.48 2.12 <0.05

Table 13Estimates of fixed effects, Experiment 3—post-target window.

TargetFix " Bias + Mention + Only + Time + State +Bias:Only + (1|Participant) + (1|Item)

Estimate SE z p

Intercept #6.08 0.10 #59.20 <0.0001Bias 0.22 0.089 2.52 <0.05Mention #0.015 0.064 #0.24 n.s.Only #0.11 0.092 #1.18 n.s.Time #0.25 0.12 #2.07 <0.05State 10.70 0.066 163.40 <0.0001Bias:Only #0.10 0.13 #0.80 n.s.



12 March 2015


1194 sentence is interpreted as the alternative set for interpret-1195 ing the target sentence. The presupposition of also together1196 with the focus on the direct object should then lead listen-1197 ers to expect the upcoming focus to be a discourse new1198 item, where the proposition expressed by the target sen-1199 tence is true for the focus value and the set of recently1200 mentioned alternatives. Given a context sentence like1201 (13a), we would expect (13b) (=12) to convey something1202 like (13c).

(13) a. Neil has some pears and some oranges.b. Jane also has some apples.c. Jane has some pears, some oranges, and

some apples.1215

1216 By contrast, the counterpart to (12) with only (4a) gives1217 rise to an expectation that the focus will be a subset of1218 the mentioned items. Thus, we predict that discourses with1219 only and also will lead to opposite expectations about the1220 discourse-old/new status of upcoming focused referents.1221 Because this information is carried by the focus operator,1222 we predict that patterns of fixations for only and also trials1223 will diverge early in the target sentence. Finally, conver-1224 gence on a display referent should be facilitated when1225 the target sentence is consistent with expectations based1226 on the focus operator: fixations on the referent should con-1227 verge earlier in Mention than Novel conditions for Only

1228trials, and earlier in Novel than Mention conditions for1229Also trials.

12304.1. Method

12314.1.1. Participants1232Twenty-six undergraduate students recruited from1233introductory Linguistics courses and flyers posted on the1234University of Rochester campus, were paid $7.50.1235Participants were native speakers of American English,1236with normal or corrected-to-normal vision.

12374.1.2. Materials and design1238Experimental materials were 24 discourses with a con-1239text sentence followed by a target sentence. In half of the1240test items, the target word was explicitly mentioned in1241the context sentence; in the other half the target word1242had not been mentioned. Half of the target sentences1243included only, and half, also. Each participant saw six1244tokens of each condition (see Table 14).1245Displays were constructed for each test item. Each dis-1246play had a referent corresponding to the target word.1247However, due to the presupposition of also (here, that1248Jane has something in the alternative set other than1249apples), some of the target sentences required targets that1250included multiple objects. For example, given the context1251sentence ‘‘Neil has some pears and some oranges,’’ the

Fig. 4. Mean proportion target fixations, Experiment 3. Top left = Neutral context, Mention condition; top right = Neutral context, No mention; bottomleft = Biasing context, Mention; bottom right = Biasing context, No mention. Vertical lines indicate (left to right): average target sentence onset, averageparticle onset, target word onset.



12 March 2015


1252 target sentence ‘‘Jane also has some apples’’ conveys that1253 Jane has not only apples, but pears and oranges as well.1254 To minimize differences among conditions, and maintain1255 roughly equal visual complexity among the four display1256 quadrants, we constructed displays with four sets of refer-1257 ents, as illustrated in Fig. 5. The status of the display refer-1258 ents differed by condition. The four sets were: a subset of1259 the mentioned items (apples in the Only condition example1260 in Fig. 5), a superset of the mentioned items (apples, pears,1261 oranges), the mentioned items (apples, pears), and a novel1262 item without either mentioned item (oranges).1263 Participants heard each experimental item in one of the1264 four conditions shown in Table 14. Experimental trials1265 were interspersed with 84 filler trials designed to mini-1266 mize statistical regularities in the materials. Half of the tar-1267 get sentences contained two target words, and half1268 contained one. The discourses were recorded by a native1269 speaker of American English. Trials were presented in1270 random order, after four practice trials.

1271 4.1.3. Procedure1272 As in Experiments 1–3, participants were instructed to1273 click on the items that Jane had, rather than items men-1274 tioned in the target sentence. Sentences with the same tar-1275 get words could correspond to different referents1276 depending on the focus particle: for Only trials, the target1277 corresponded to either the subset or the novel referent,1278 while for Also trials, the target was either the superset or1279 the same set referent.

1280 4.1.4. Modeling1281 Three analysis windows were designated: (i) the pre-1282 particle window, the 500 ms window ending at the onset

1283of the focus particle, (ii) the pre-target window, starting at1284particle onset and ending at target word onset, and (iii)1285the post-target window, the 500 ms starting at target onset.1286Onsets used to delimit analysis windows were determined1287on a trial-by-trial basis.1288To assess the influence of the focus operator, we fit fixa-1289tions to the Novel/Superset referent using a mixed-effect1290logistic regression model in the pre-particle analysis win-1291dow. The model predicted fixations to the novel referent1292for Only conditions, and fixations to the superset referent1293for Also. In order to compare the two focus particles despite1294their presuppositional differences, we created a binary1295variable whose value was 1 if either the Novel or the1296Superset referent had a value of 1, and 0 otherwise. This1297was used as the dependent variable. The random effects1298structure was determined as previously described. State1299was included as a fixed effect.1300Because the time courses of effects associated with the1301focus particle diverged in the later analysis windows,1302Mention effects were analyzed separately for Only and1303Also. For each particle, we asked whether prior mention1304influenced identification of the referent. We fit fixations1305to the target referent (defined by trial as the referent1306clicked on in that trial) using logistic regression models,1307for the pre-target and post-target analysis windows.1308Mention, Time, and State were included as fixed effects,1309and Participant and Item as random effects.

13104.2. Results

1311The predicted target referent varies by condition as1312follows: subset for Only-Mention, novel for Only-Novel,1313superset for Also-Novel, and same set for Also-Mention.1314Two types of analyses were performed. First, we asked1315whether the choice of focus operator predicted fixations1316to novel scene referents. This analysis assesses differences1317in novelty biases based on the operator in a region where1318information about the target is not yet available. Next, we1319analyzed looks to the target referent separately for Only1320and Also conditions, assessing differences in how quickly1321fixations converged on the target referent for mentioned1322and novel targets. Proportions of fixations to scene refer-1323ents are presented in Fig. 6 for Also conditions, and1324Fig. 7 for Only.


Only Also

Mention Neil has some pears andsome apples

Neil has some pears andsome apples

Jane only has some apples Jane also has some apples

Novel Neil has some pears andsome oranges

Neil has some pears andsome oranges

Jane only has some apples Jane also has some apples

Fig. 5. Example displays for Experiment 4 (labels for illustration only). Left panel: Only conditions; right panel: Also conditions.



12 March 2015


1325 For the Also-Mention condition, the direct-object focus1326 leads to a conflict between the presupposition of also and1327 the expectation of a novel object. Therefore, none of the1328 displayed referents is a perfectly felicitous interpretation1329 of the target sentence. This infelicity is reflected in both1330 the variability in responses and later response times in this1331 condition (Fig. 6). There were two main response types in1332 the Also-Mention condition: in the majority of trials, par-1333 ticipants chose the Subset referent, which is consistent1334 with the form of the target word, but violates the presup-1335 position associated with also (that Jane has items other1336 than the referent corresponding to the target word). In

1337most of the remaining trials, participants chose the Same1338set referent, which satisfies the presupposition of also1339(but, like the Subset referent, violates the expectation that1340the target will be discourse-novel).5 Fig. 8 shows the distri-1341bution of response types by condition.

Fig. 6. Mean proportions of target fixations, Experiment 4—Also conditions. Top left = Also-Novel condition; top right = Also-Mention (subset responses);bottom right = Also- = Mention (same set responses). Vertical lines indicate (left to right): average target sentence onset, focus particle onset, average targetword onset, average response time (mouse click).

Fig. 7. Mean proportions of target fixations, Experiment 4—Only conditions. Left = Only-Novel; right = Only-Mention. Vertical lines indicate (left to right):average target sentence onset, focus particle onset, average target word onset, average response time (mouse click).

5 Twenty of the 26 participants responded consistently throughout.Within the 6 inconsistent responders, 2 converged on Subset responses inlate trials, 1 converged on Same set responses, and 3 responded variablythroughout the experiment. The convergence on response types in the Also-Mention condition did not affect responses in the other conditions, whichreceived consistent responses throughout the experiment.



12 March 2015


1342 Responses from the Also-Novel condition are plotted in1343 the top left panel of Fig. 6, and the two main response1344 types in the Also-Mention condition are shown in the two1345 right panels.

1346 4.2.1. Novelty bias predicted by focus operator1347 We predicted that also would lead to expectations for a1348 superset referent, while only would lead to expectations1349 for a subset referent. Because this difference is tied to the1350 information carried by the focus operator, we expect it to1351 emerge early in the utterance, in principle as soon as the1352 focus particle has been processed.1353 Table 15 shows estimated coefficients from the regres-1354 sion model for the pre-particle window. The pre-particle1355 window shows a main effect of focus operator, with fewer1356 fixations to the Novel/Superset referent when the target1357 sentence contained only than when it contained also. This1358 is apparent in the early increase in fixations to the1359 Superset referent in Also conditions (Fig. 7), compared to1360 Only conditions (Fig. 8). The effect is carried by a novelty1361 bias associated with also: planned comparisons over1362 100 ms intervals showed no advantage due to Mention in1363 Only conditions in the pre-particle window, t(50) = 1.08,1364 n.s. (100 ms preceding target word onset). However, for1365 sentences with also, fixations converged on the superset1366 referent in the 100 ms window preceding the onset of1367 the particle, t(50) = 2.83, p < 0.01 (in the 200–100 ms pre-1368 ceding particle onset, t(50) = 1.22, n.s.). The effect of focus1369 operator interacts with Time: the advantage for Novel/

1370Superset fixations associated with also weakens over the1371pre-particle analysis window. We return to this point1372shortly. While the focus operator effect is in the direction1373we predicted, it is surprising that it is already present1374before the onset of the focus particle.1375What might account for these early effects? We1376hypothesized that participants were sensitive to acoustic1377differences between the initial segments of sentences con-1378taining only and also—differences that reflect the prosody1379associated with their respective focus structures. To test1380this hypothesis, we first measured minimum and maxi-1381mum F0, and duration for the subject argument in only1382and also target sentences (Table 16). Indeed the subject in1383also sentences had a higher maximum F0, t(23) = 2.2,1384p < 0.05, and larger pitch excursion, t(23) = 2.97, p < 0.001,1385than subjects in only sentences, likely serving as cues to1386the identity of the focus operator before it became available1387in the auditory input. (Note that the magnitude of the Only/1388Also differences is considerably larger than the differences1389between Only and No only sentences in Experiment 1.)1390To examine whether prosody was responsible for the1391particle effects in the pre-particle window, we re-ran the1392analyses with four additional fixed effects included in the1393model: (i) the maximum F0 in the pre-particle window,1394(ii) the F0 range in the pre-particle window, (iii) the rate1395of F0 change between the minimum and maximum F0 val-1396ues, and (iv) the duration of the segment between the1397minimum and maximum F0 points. Duration and F0 rate1398remained in the model after model comparison. The main1399effect of Particle is no longer significant in the resulting1400model (Table 17), suggesting that prosodic cues in the1401pre-particle window were responsible for the early diver-1402gence between Only and Also sentences. We return below1403to the differences in contextual support required by only1404and also, and their consequences for prosody.

Fig. 8. Distribution of responses by condition, Experiment 4.

Table 15Estimates of fixed effects, Experiment 4: Effect of focus particle, pre-particlewindow.

Novel/SupersetFix " Mention + Particle + Time + State +Particle:Time + (1 + Mention|Participant) + (1 +Mention|Item)

Estimate SE z p

Intercept #3.98 0.26 #15.28 <0.0001Particle[Only] #0.71 0.18 #3.90 <0.0001Mention #0.19 0.20 #0.95 n.s.Time 6.09 0.53 11.48 <0.0001State 9.39 0.10 91.29 <0.0001Particle[Only]:Time #2.69 0.71 #3.80 <0.001

Table 16F0, duration means (standard deviation) for Experiment 4, target sentencesubject.

Min F0 (Hz) Max F0 (Hz) DF0 (Hz) Duration (ms)

Only 170.3 (24.1) 196.2 (8.2) 26.0 (27.3) 307.0 (13.3)Also 154.7 (35.7) 202.6 (13.6) 47.9 (36.6) 316.7 (13.9)

Table 17Estimates of fixed effects, Experiment 4: Effects of focus particle andprosodic cues, pre-particle window.

Novel/Superset " Particle + Mention + Time + Duration +F0Rate + State + Particle:Duration + Particle:Time +(1|Participant) + (1|Item)

Estimate SE z p

Intercept #3.71 1.81 #2.05 <0.0001Particle[Only] #4.57 2.70 #1.69 0.09Mention #0.038 0.11 #0.35 n.s.Time 0.73 0.47 1.58 n.s.Duration #5.64 5.77 #0.98 n.s.F0 Rate #0.11 0.16 #0.65 n.s.State 10.70 0.12 92.66 <0.0001Particle[Only]:Duration 15.89 8.66 1.83 0.07Particle[Only]:Time #1.94 0.69 #2.83 <0.01



12 March 2015


1405 The negative interaction between focus operator and1406 Time appears in both models for the pre-particle window.1407 The decrease in the focus operator effect after an early1408 advantage for also suggests that Novel/Superset fixations1409 do not increase monotonically, as is typically found for tar-1410 get fixations. In addition, Figs. 7 and 8 show that the time1411 course of target fixations differs substantially by focus1412 operator, with effects emerging later for only than for also.1413 We therefore performed separate analyses for each particle1414 for the pre-target and post-target analysis windows.

1415 4.2.2. Time course of mention/novelty biases1416 Estimated model coefficients are shown in Tables 181417 and 19 for the Only conditions, and in Tables 20 and 211418 for Also.1419 For the Only conditions, there is no significant effect of1420 Mention in the pre-target window (Table 18). In the1421 post-target window, a Mention advantage emerges1422 (Table 19). Thus, the Only conditions showed more target1423 fixations when the target had been mentioned than when1424 it was discourse-new, replicating the Mention-Only effect1425 from Experiment 1. Planned comparisons over 100 ms1426 intervals revealed that fixations converged on the target1427 referent in the 0–100 ms window after target word onset1428 for the Only-Mention condition, t(50) = 2.47, p < 0.051429 (#100 to 0 ms window, t(50) = 1.08, n.s.) and in the1430 400–500 ms window for the Only-No Mention condition,1431 t(50) = 3.08, p < 0.01 (300–400 ms window, t(50) = 0.86,1432 n.s.).1433 The Also conditions (Tables 20 and 21) show a different1434 time course. In the Also-Mention condition, the novelty1435 bias that emerged in the pre-particle window persists in1436 the pre-target window until the 200–100 ms window pre-1437 ceding the onset of the target word, t(50) = 2.31, p < 0.05.1438 The novelty bias reemerges in the pre-target window in1439 the 200–300 ms following target word onset, t(50) = 3.12,1440 p < 0.01. The early emergence of this effect, even before1441 the onset of the particle, may suggest a confluence of early1442 prosodic cues consistent with also rather than only, and the

1443resulting novelty bias being boosted by a general prefer-1444ence for discourse-new material. In the Also-Mention con-1445dition, fixations do not converge on the target referent1446even 800–900 ms after target onset, t(50) = 0.13, n.s., well1447after the target word.

14484.3. Discussion

1449Based on the meaning difference between only and also,1450we expected also to give rise to expectations for a superset1451referent (mentioned referents plus a novel target), and1452only, to expectations for a subset referent (one of two men-1453tioned referents). This is indeed what we observed in1454Experiment 4. In addition, the patterns of eye movements1455for only and also sentences diverged early, as expected if1456the difference in expectations was linked to information1457carried by the focus operator. This pattern of time courses1458is consistent with Romoli, Khan, Sudo, and Snedeker1459(2015), who also found earlier target fixations with also1460relative to only conditions.1461The different time courses associated with the mention1462and novelty biases might be explained in part by a general1463novelty preference. The novelty bias associated with also1464would be boosted by such an underlying preference, while1465the mention bias associated with only would be working1466against it, explaining the early novelty preference in also1467sentences. Such a bias would seem to be at odds with the1468main effect of Mention observed in Experiment 1.1469However, Experiment 1 did not include a Same/Different1470category manipulation, and as such, the general Mention1471preference (in contrast to the Mention-Only interaction)1472might actually reflect a Same category bias, as in1473Experiments 2–3. To substantiate this idea, we conducted1474a sentence completion study using the online crowd-1475sourcing platform Amazon Mechanical Turk, asking1476whether discourse-new continuations are preferred after

Table 18Estimates of fixed effects, Experiment 4: Only, pre-target window.

TargetFix " Mention + Time + State + (1 +Mention|Participant) + (1 + Mention|Item)

Estimate SE z p

Intercept #4.86 0.81 #5.97 <0.0001Mention 1.20 0.94 1.29 n.s.Time 0.92 0.11 8.45 <0.0001State 0.31 0.04 7.49 <0.0001

Table 19Estimates of fixed effects, Experiment 4: Only, post-target window.

TargetFix " Mention + Time + State + (1 +Mention|Participant) + (1 + Mention|Item)

Estimate SE z p

Intercept #4.49 0.46 #9.77 <0.0001Mention 1.08 0.53 2.03 <0.05Time 3.15 0.10 32.61 <0.0001State 0.32 0.03 10.01 <0.0001

Table 20Estimates of fixed effects, Experiment 4: Also, pre-target window.

TargetFix " Mention + Time + State + Mention:Time +(1 + Mention|Participant) + (1 + Mention|Item)

Estimate SE z p

Intercept #2.20 0.31 #7.01 <0.0001Mention #1.21 0.61 #1.99 <0.05Time 1.14 0.12 9.75 <0.0001State 1.13 0.03 34.42 <0.0001Mention:Time #2.22 0.18 #12.35 <0.0001

Table 21Estimates of fixed effects, Experiment 4: Also, post-target window.

TargetFix " Mention + Time + State + Mention:Time + (1 +Mention|Participant) + (1|Item)

Estimate SE z p

Intercept #2.45 0.38 #6.49 <0.0001Mention #0.95 0.59 #1.62 0.11Time 1.37 0.13 10.27 <0.0001State 1.17 0.04 32.96 <0.0001Mention:Time #1.09 0.21 #5.17 <0.0001



12 March 2015


1477 the diverging mention biases associated with only and also1478 have been taken into account. We presented 24 par-1479 ticipants with 24 pairs of sentences like (14)–(15). The first1480 sentence mentioned two same category referents; the sec-1481 ond sentence either contained only, also, or no particle. The1482 sentence pairs were based on the test items used in1483 Experiment 2.

(14) Neil has some apples and some pears.(15) Jane {only,also,Ø} has _____.1490

1491 98.8% of all responses were same category comple-1492 tions.6 Consistent with Experiment 4, only and also gave rise1493 to different tendencies for completions containing explicitly1494 mentioned content (Fig. 9). However, all conditions showed1495 a preference for continuations that contained novel lexical1496 material which nevertheless stayed within the conceptual1497 category introduced in the prior discourse, consistent with1498 the Same category bias observed in Experiment 2.1499 Returning to the results of Experiment 4, the novelty1500 bias associated with also is superimposed on the underly-1501 ing general novelty bias: together with the early prosodic1502 cues, this yields an early, very prominent novelty bias in1503 also sentences. However, the same general novelty bias1504 works against the mention bias associated with only. A1505 consequence of these conflicting pressures may be the1506 relatively late emergence of the mention bias in only1507 sentences.1508 The early divergence of the patterns of eye movements1509 in only and also sentences suggests listeners were able to1510 discern the difference between these sentence types based1511 on prosodic cues in the pre-particle window. Why would1512 the target sentences with only and also have perceptible

1513prosodic differences? The sentences have identical syntac-1514tic structures, with the focus particle syntactically attach-1515ing to and taking scope over the verb phrase. We suspect1516the difference lies in how differences in focus structure1517are encoded for each sentence type. In Experiment 2, we1518restricted ourselves to cases of VP-focus, setting aside sub-1519ject-focus sentences like the ones exemplified in (16) and1520(17); paraphrases in terms of alternative sets are given in1521(16c) and (17c). (18) and (19) give VP-focus sentences for1522comparison. Italics represent the focused argument, and1523capitals represent (primary) prosodic prominence.

(16) a. Neil and Jane have some pears.b. Only JANE has some apples.c. No one in the alternative set has apples

except Jane.(17) a. Neil and Jane have some apples.

b. Jane ALSO has some apples.c. There is a set of alternative individuals who

have apples; Jane has apples.(18) Jane only has some APPLES.(19) Jane also has some APPLES. 1552

1553For only, the focal difference between (16b) and (18) is1554encoded by word order: the particle precedes the subject1555when the subject argument is in focus. Because focus1556structure is signaled unambiguously by word order,1557prosodic prominence is, by comparison, a relatively unreli-1558able cue to focus structure.1559By contrast, the subject-focus and VP-focus also sen-1560tences in (17b) and (19) have identical word orders.1561Instead, the focus structure difference is encoded1562prosodically (Beaver & Clark, 2008; Kripke, 2009; König,15631991; Rullman, 2003). In both cases, the subject argument1564bears (secondary) prominence; however, while also is non-1565prominent when focus is in the VP as in (19), it bears pri-1566mary prominence when the subject is focused, as in1567(17b). Since we wanted to signal VP-focus readings rather1568than subject-focus, we avoided prominence on also in our1569materials. This may have had the side effect of rendering1570the subject more prominent, thereby distinguishing it from1571subjects in the analogous only sentences. The early effects1572in Experiment 2 suggest that comprehenders are sensitive1573to these prosodic cues and can use them to anticipate the1574upcoming focus structure. Detecting a tendency for more1575prominence on subjects in the also sentences would have1576allowed subjects to infer the type of particle (also), with1577eye movements reflecting the novelty bias associated with1578that particle.1579In summary, only and also lead to different expectations1580with respect to prior mention: while only facilitated iden-1581tification of mentioned targets (corresponding to a subset1582of the mentioned items), also gave rise to an expectation1583for discourse-new targets (corresponding to a superset of1584the mentioned items). Listeners were also sensitive to dif-1585ferences in prosodic prominence between only and also1586sentences. Prosodic differences before the onset of the1587focus operator shaped expectations about the identity of1588the operator itself. These results, together with the results1589of Experiments 2–3, suggest that the interpretation of sen-1590tences with focus operators like only and also draw on both

Fig. 9. Proportion of completion types, sentence completion experiment.

6 We assigned each discourse a basic level category prior to analyzing theresponses; where there was a salient subordinate category (e.g. writingimplement instead of office supplies for pen and pencil), we used the moregeneral category. We classified responses as Mentioned if the exact form ofa mentioned item was used, not counting differences in determiners orplurality. For the remainder of responses, responses falling in the pre-assigned categories were counted as Same category. In most cases, Samecategory responses were consistent with the more specific category (e.g.the completion marker after pen and pencil). Most Different categoryresponses involved a salient relation between the completion and men-tioned items, despite not falling into the pre-assigned categories (e.g. thecompletion flowers after candy canes and chocolate).



12 March 2015


1591 general mechanisms that are ongoing during discourse1592 processing (giving rise to lexical and ad hoc category1593 effects), and dedicated mechanisms for introducing alter-1594 natives into the discourse context (explicit mention),1595 which are triggered when a focus operator requires that1596 the context supply a set of alternatives.

1597 5. General discussion

1598 The experiments presented allow us to at least partially1599 answer the questions posed in the Introduction about the1600 processing of context-relevant alternatives triggered by1601 focus particles. The first question was about how to charac-1602 terize the contextual information that shapes expectations1603 about focus alternatives. The second question was about1604 the division of labor between the lexical content of indi-1605 vidual focus operators and the class of alternative-sensitive1606 operators. We revisit these questions, spelling out the1607 implications of our findings for existing models of lan-1608 guage processing, and discussing areas for future research.

1609 5.1. Incremental inference about abstract alternatives

1610 We demonstrated that interpretation proceeds1611 incrementally even when it involves restricting abstract1612 alternatives. Because focus particles like only or also do1613 not provide information about descriptive content, we1614 can infer that the expectations listeners generate when1615 they encounter such particles are shaped by the prior dis-1616 course context. In the absence of explicitly given alterna-1617 tives, the processor uses alternatives made salient by1618 general-purpose mechanisms involved in discourse pro-1619 cessing. Moreover, the calculation of context-relevant1620 alternatives appears to be separable from the requirements1621 of specific lexical items; in the case of only and also, the1622 choice of focus operator determined how the focus related1623 to its alternatives, thereby generating expectations about1624 whether the focus was given or new.1625 Issues related to the processing of abstract alternatives1626 have not been directly investigated by much previous real-1627 time processing research, which has tended to focus on1628 how listeners use context to resolve local indeterminacy1629 rather than on how listeners create context. Existing mod-1630 els of sentence comprehension are not explicit about1631 whether and how expectations about implicit alternatives1632 influence incremental sentence processing. For instance,1633 cue-based retrieval models (Lewis, 2000; Lewis &1634 Vasishth, 2005; Van Dyke & Lewis, 2003) only generate1635 predictions about syntactic category. However, to accom-1636 modate the Situation-driven Alternatives Hypothesis,1637 existing theories need to be scaled up to account for how1638 abstract domain representations are constructed.1639 Referential Theory (Altmann & Steedman, 1988) and the1640 broader family of constraint satisfaction models1641 (MacDonald, Pearlmutter, & Seidenberg, 1994; McRae,1642 Spivey-Knowlton, & Tanenhaus, 1998; Spivey &1643 Tanenhaus, 1998) claim that discourse factors—for example,1644 the presence of contrast—influence expectations about like-1645 ly parses; but as with other models originally designed to

1646account for syntactic ambiguity resolution, the expectations1647generated are about likely parses, not about likely represen-1648tations of the context. Predictions of surprisal-based models1649(Hale, 2001; Levy, 2008) are similarly restricted to syntactic1650parses. Explaining the findings presented here will require1651scaling up such models to include tracking probability dis-1652tributions over sets of alternatives—ultimately over com-1653mon ground representations.

16545.2. The nature of expectations about alternatives

1655In addition to tracking discourse mention, listeners gen-1656erate expectations about candidate alternatives based on1657properties such as category membership. The category1658effects observed in Experiment 2 may be explained in part1659by general processes of lexical retrieval and conceptual1660activation required in processing a sentence (Collins &1661Loftus, 1975; Cree et al., 1999; Srinivas & Roediger, 1990;1662Tversky & Hemenway, 1983), or by semantic priming1663(Yee & Sedivy, 2006). However, the situation-driven effects1664observed in Experiment 3 suggest that comprehenders also1665generate alternatives on the fly, which may involve novel1666ad hoc categories that are determined by the constraints1667or goals associated with a particular discourse (see1668Ballard & Hayhoe, 2009 for a discussion of task effects on1669gaze control; also Salverda, Brown, & Tanenhaus, 2011).1670Arguments for communicative goals or relevance as orga-1671nizing principles of linguistic behavior have been present-1672ed by theorists from different perspectives (Clark, 1992,16731996; Roberts, 1996; Sperber & Wilson, 1987; among1674others; see also the discussion of conceptual covers in1675Aloni, 2000). In future research it will be important to1676explore how listeners generate goal-oriented alternatives.1677It will also be important to examine how expectations1678about alternatives interact with structural properties of1679the discourse, such as discourse coherence relations1680(Kehler, 2002), which have independently been shown to1681influence discourse-level dependencies.1682To conclude, the experiments presented support a pro-1683cessor that generates incremental predictions with respect1684to both upcoming discourse content and implicit alterna-1685tives required for focus interpretation. For the two focus1686operators investigated here (only and also), it does so by1687tracking mention within the local discourse. In addition,1688alternatives are generated on the basis of ad hoc categories,1689and therefore must be generated in a manner that is sensi-1690tive to features of particular contexts of use. Most general-1691ly, then, the current results provide an empirical1692foundation for future research that uses insights from1693linguistic operators such as focus particles to explore the1694representations and processes that underlie discourse1695comprehension, and for enriching current processing1696models to include the machinery necessary to explain the1697behavior of constructions that require contextually1698-determined alternatives.

16996. Uncited references

1700Grodner et al. (2010) and Huang and Snedeker (2009).

Q6

Q7



12 March 2015


1701 Appendix A. Experiment 1: analysis of prosodic cues

1702 One of the objectives of Experiment 1 was to assess the1703 time course of interpretive effects associated with the1704 focus operator. Specifically, we were interested in whether1705 listeners anticipate the identity of the target referent based1706 on contextual information (the content of the preceding1707 sentence) and the presence of the operator alone, at a point1708 when auditory input from the target word is not yet avail-1709 able. We found that in the early window (starting at the1710 onset of the particle in Only conditions and the main verb1711 in No only conditions, and ending at the onset of the target1712 word), there was a main effect of Mention which did not1713 interact with the presence of only. In the late window start-1714 ing at the onset of the target word, a Mention-Only interac-1715 tion emerged: in sentences with only, mentioned referents1716 were more likely to be considered as targets. Since the1717 Mention-Only advantage emerges 200–300 ms after target1718 word onset, and signal-driven eye movements in this type1719 of experiment occur at a 200 ms delay, we argue that this1720 advantage preceded any acoustic phonetic input about1721 the target word. This would show that the presence of only1722 restricted the set of candidate alternatives to just the men-1723 tioned ones.1724 Such a conclusion assumes that prosodic information1725 from earlier in the sentence was not responsible for the1726 Mention-Only effect. In creating the materials, we ensured1727 that primary prominence was on the direct object for both1728 only and non-only sentences, since this pattern is consis-1729 tent with either direct object focus or unmarked whole-1730 sentence focus. Because all target sentences (including fil-1731 lers) had sentence-final prominence, with secondary1732 prominence on the subject, participants were unlikely to1733 expect alternative prosodic realizations of the target sen-1734 tence over the session. However, the possibility remains1735 that small prosodic differences early in the sentence were1736 used by participants to distinguish between conditions.1737 In particular, we were concerned that systematic increased1738 prominence on the subject argument might signal an only1739 target sentence to participants, since such a pattern may1740 suggest that the sentence will resolve with contrastive1741 prominence on the direct object. This would provide an1742 alternative explanation for our the Mention-Only effect,1743 in terms of prosodic patterns rather than the presence or1744 absence of the focus particle.1745 To assess whether participants could have used promi-1746 nence on the subject to anticipate whether they were in an1747 only sentence, we measured minimum, maximum and1748 mean F0, pitch excursion, and duration for the subject1749 argument in the target sentence (Table A.1).

1750Only sentences had a lower maximum F0 (t(38) = #0.33,1751n.s.), a smaller pitch excursion (t(38) = #3.65, p < 0.001),1752lower mean F0 (t(38) = #2.14, p < 0.05) and shorter dura-1753tion (t(38) = #2.40, p < 0.05) relative to sentences without1754only. If prosodic prominence is associated with larger pitch1755excursions and longer duration, then subjects in only sen-1756tences are actually less prominent than are subjects in sen-1757tences without only, and consequently less likely to signal1758contrastive prominence on the direct object. We think that1759early anticipation of direct object focus is therefore unlike-1760ly to be associated with increased prominence on the sub-1761ject argument.

1762Appendix B. Supplementary material

1763Supplementary data associated with this article can be1764found, in the online version, at http://dx.doi.org/10.1016/1765j.cognition.2015.02.009.

1766References

1767Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and1768interpreting interactions. Newbury Park, CA: Sage.1769Allopenna, P. D., Magnuson, J. S., & Tanenhaus, M. K. (1998). Tracking the1770time course of spoken word recognition: Evidence for continuous1771mapping models. Journal of Memory and Language, 38, 419–439.1772Aloni, M. (2000). Conceptual covers in dynamic semantics. In L. Cavedon,1773P. Blackburn, N. Braisby, & A. Shimojima (Eds.). Logic, language and1774computation (Vol. III). Stanford, CA: CSLI.1775Altmann, G. T. M. (2011). Language can mediate eye movement control1776within 100 milliseconds, regardless of whether there is anything to1777move the eyes to. Acta Psychologica, 137(2), 190–200.1778Altmann, G. T. M., & Kamide, Y. (1999). Incremental interpretation at1779verbs: Restricting the domain of subsequent reference. Cognition,178073(3), 247–264.1781Altmann, G. T. M., & Kamide, Y. (2007). The real-time mediation of visual1782attention by language and world knowledge: Linking anticipatory1783(and other) eye movements to linguistic processing. Journal of1784Memory and Language, 57, 502–518.1785Altmann, G. T. M., & Mirkovic, J. (2009). Incrementality and prediction in1786human sentence processing. Cognitive Science, 33, 583–609.1787Altmann, G. T. M., & Steedman, M. J. (1988). Interaction with context1788during human sentence processing. Cognition, 30(3), 191–238.1789Arnold, J. E., Altmann, R., Fagnano, M., & Tanenhaus, M. K. (2004). The old1790and thee, uh, new. Psychological Science, 578–582.1791Arnold, J. E., Fagnano, M., & Tanenhaus, M. K. (2003). Disfluencies signal1792theee, um, new information. Journal of Psycholinguistic Research, 32(1),179325–36.1794Ballard, D. H., & Hayhoe, M. (2009). Modelling the role of task in the1795control of gaze. Visual Cognition, 17(6–7), 1185–1204.1796Barr, D. J. (2008). Analyzing ‘visual world’ eyetracking data using1797multilevel logistic regression. Journal of Memory and Language, 59,1798457–474.1799Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects1800structure for confirmatory hypothesis testing: Keep it maximal.1801Journal of Memory and Language, 68(3), 255–278.1802Beaver, D. I., & Clark, B. Z. (2008). Sense and sensitivity: How focus1803determines meaning. Wiley.1804Brouwer, S., Mitterer, H., & Huettig, F. (2012). Speech reductions change1805the dynamics of competition during spoken word recognition.1806Language and Cognitive Processes, 27(4), 539–571.1807Brown-Schmidt, S. (2009). The role of executive function in perspective1808taking during online language comprehension. Psychonomic Bulletin &1809Review, 16(5), 893–900.1810Chambers, C. G., Tanenhaus, M. K., Eberhard, K. M., Filip, H., & Carlson,1811G. N. (2002). Circumscribing referential domains in real-time1812sentence comprehension. Journal of Memory and Language, 47,181330–49.1814Chambers, C. G., Tanenhaus, M. K., & Magnuson, J. S. (2004). Action-based1815affordances and syntactic ambiguity resolution. Journal of1816Experimental Psychology: Learning, Memory, and Cognition, 30,1817687–696.

Table A.1F0, duration means (standard deviation) for target sentence subject,Experiment 1.

Min F0

(Hz)Max F0

(Hz)DF0

(Hz)Mean F0

(Hz)Duration(ms)

Only 216.6(7.4)

245.4(8.2)

28.8(6.1)

232.2(1.6)

233.7(12.5)

No only 217.3(5.4)

253.5(8.5)

36.2 (6.7) 236.6 (1.4)

243.2 (12.5)



12 March 2015


1818 Clark, H. H. (1992). Arenas of language use. Chicago: University of Chicago1819 Press.1820 Clark, H. H. (1996). Using language. Cambridge: Cambridge University1821 Press.1822 Collins, A. M., & Loftus, E. F. (1975). A spreading activation of semantic1823 processing. Psychological Review, 82, 407–428.1824 Cooper, R. M. (1974). The control of eye fixation by the meaning of spoken1825 language: A new methodology for the real-time investigation of1826 speech perception, memory, and language processing. Cognitive1827 Psychology, 6(1), 84–107.1828 Cree, G. S., McRae, K., & McNorgan, C. (1999). An attractor model of lexical1829 conceptual processing: Simulating semantic priming. Cognitive1830 Science, 23, 371–414.1831 Gibson, E. (1998). Linguistic complexity: Locality of syntactic1832 dependencies. Cognition, 68, 1–76.1833 Gibson, E. (2000). The dependency locality theory: A distance-based1834 theory of linguistic complexity. In Y. Miyashita, A. Marantz, & W.1835 O’Neil (Eds.), Image, language, brain (pp. 95–126). Cambridge, MA:1836 MIT Press.1837 Grodner, D. J., Klein, N. M., Carbary, K. M., & Tanenhaus, M. K. (2010).1838 ‘‘Some’’, and possibly all, scalar inferences are not delayed: Evidence1839 for immediate pragmatic enrichment. Cognition, 116, 42–55.1840 Hale, J. T. (2001). A probabilistic Earley parser as a psycholinguistic model.1841 In Proceedings of the second meeting of the North American chapter of1842 the association for computational linguistics.1843 Hale, J. T. (2003). The information conveyed by words in sentences.1844 Journal of Psycholinguistic Research, 32(2), 101–123.1845 Huang, Y. T., & Snedeker, J. (2009). Online interpretation of scalar1846 quantifiers: Insight into the semantics–pragmatics interface.1847 Cognitive Psychology, 58, 376–415.1848 Jaeger, T. F. (2008). Categorical data analysis: Away from ANOVAs1849 (transformation or not) and towards logit mixed models. Journal of1850 Memory and Language, 59, 434–446.1851 Kaiser, E., & Trueswell, J. C. (2004). The role of discourse context in the1852 processing of a flexible word-order language. Cognition, 94(2),1853 113–147.1854 Kehler, A. (2002). Coherence, reference, and the theory of grammar. CSLI1855 Publications.1856 König, E. (1991). The meaning of focus particles: A comparative perspective.1857 London: Routledge.1858 Kripke, S. A. (2009). Presupposition and anaphora: Remarks on the1859 formulation of the projection problem. Linguistic Inquiry, 40(3),1860 367–386.1861 Ladd, D. R. (1996). Intonational phonology. Cambridge: Cambridge1862 University Press.1863 Levy, R. (2008). Expectation-based syntactic comprehension. Cognition,1864 106(3), 1126–1177.1865 Lewis, R. L. (1993). An architecturally-based theory of human sentence1866 comprehension. Unpublished doctoral dissertation, Carnegie Mellon1867 University.1868 Lewis, R. L. (2000). Specifying architectures for language processing:1869 Process, control, and memory in parsing and interpretation. In M. W.1870 Crocker, M. Pickering, & C. Clifton, Jr. (Eds.), Architectures and1871 mechanisms for language processing. Cambridge: Cambridge1872 University Press.1873 Lewis, R. L., & Vasishth, S. (2005). An activation-based model of sentence1874 processing as skilled memory retrieval. Cognitive Science, 29, 375–419.1875 MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994). The1876 lexical nature of syntactic ambiguity resolution. Psychological Review,1877 101, 676–703.1878 Mack, J. E., Ji, W., & Thompson, C. K. (2013). Effects of verb meaning on1879 lexical integration in agrammatic aphasia: Evidence from eyetracking.1880 Journal of Neurolinguistics, 26(6), 619–636.1881 Marslen-Wilson, W. D. (1973). Linguistic structure and speech shadowing1882 at very short latencies. Nature, 244, 522–523.1883 Marslen-Wilson, W. D. (1975). Sentence perception as an interactive1884 parallel process. Science, 189, 226–228.1885 Marslen-Wilson, W. D. (1987). Functional parallelism in spoken word1886 recognition. Cognition, 25, 71–102.1887 McRae, K., Spivey-Knowlton, M. J., & Tanenhaus, M. K. (1998). Modeling the1888 influence of thematic fit (and other constraints) in on-line sentence1889 comprehension. Journal of Memory and Language, 38, 283–312.

1890Ni, W., Crain, S., & Shankweiler, D. P. (1996). Sidestepping garden paths:1891The contribution of syntax, semantics and plausibility in resolving1892ambiguities. Language and Cognitive Processes, 11, 283–334.1893Ochs, E. (1979). Planned and unplanned discourse. In T. Givón (Ed.),1894Syntax and semantics (XII): Discourse and syntax (pp. 51–80). New1895York: Academic Press.1896Paterson, K. B., Liversedge, S. P., & Underwood, G. (1999). The influence of1897focus operators on syntactic processing of short relative clause1898sentences. The Quarterly Journal of Experimental Psychology, 52A(3),1899717–737.1900Roberts, C. (1996). Information structure in discourse: Towards an1901integrated formal theory of pragmatics. In J. H. Yoon & A. Kathol1902(Eds.), Ohio State University working papers in linguistics (Vol. 49).1903Romoli, J., Khan, M., Sudo, Y., & Snedeker, J. (2015). Resolving temporary1904referential ambiguity using presupposed content. In F. Schwarz (Ed.).1905Experimental approaches to presupposition. Studies in theoretical1906psycholinguistics (Vol. 45, pp. 67–87). Cham, Switzerland: Springer1907International Publishing.1908Rooth, M. (1992). A theory of focus interpretation. Natural Language1909Semantics, 1(1), 75–116.1910Rooth, M. (1996). On the interface principles for intonational focus. In T.1911Galloway & J. Spence (Eds.), Proceedings of SALT VI (pp. 202–226).1912Ithaca, NY: Cornell University.1913Rullman, H. (2003). Additive particles and polarity. Journal of Semantics,191420, 329–401.1915Salverda, A. P., Brown, M., & Tanenhaus, M. K. (2011). A goal-based1916perspective on eye movements in visual-world studies. Acta1917Psychologica, 137(2), 172–180.1918Salverda, A. P., Kleinschmidt, D., & Tanenhaus, M. K. (2014). Immediate1919effects of anticipatory coarticulatory information on lexical access in1920spoken-word recognition. Journal of Memory and Language, 71, 145–163.1921Sedivy, J. C. (2002). Invoking discourse-based contrast sets and resolving1922syntactic ambiguities. Journal of Memory and Language, 46, 341–370.1923Sedivy, J. C. (2003). Pragmatic versus form-based accounts of referential1924contrast: Evidence for effects of informativity expectations. Journal of1925Psycholinguistic Research, 32(1), 3–23.1926Sedivy, J. C., Tanenhaus, M. K., Chambers, C. G., & Carlson, G. N. (1999).1927Achieving incremental interpretation through contextual1928representation: Evidence from the processing of adjectives. Cognition,192971, 109–147.1930Segal, E. M., Duchan, J. F., & Scott, P. J. (1991). The role of interclausal1931connectives in narrative structuring: Evidence from adults’1932interpretations of simple stories. Discourse Processes, 14, 27–54.1933Sperber, D., & Wilson, D. (1987). Precis of relevance: Communication and1934cognition. Behavioral and Brain Sciences, 10, 697–754.1935Spivey, M. J., & Tanenhaus, M. K. (1998). Syntactic ambiguity resolution in1936discourse: Modeling the effects of referential context and lexical1937frequency. Journal of Experimental Psychology: Learning, Memory, and1938Cognition, 24, 1521–1543.1939Srinivas, K., & Roediger, H. L. (1990). Classifying implicit memory tests:1940Category association and anagram solution. Journal of Memory and1941Language, 29, 389–412.1942Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., & Sedivy, J. E.1943(1995). Integration of visual and linguistic information in spoken1944language comprehension. Science, 268, 1632–1634.1945Tversky, B., & Hemenway, K. (1983). Categories of scenes. Cognitive1946Psychology, 15, 121–149.1947Van Dyke, J. A., & Lewis, R. L. (2003). Distinguishing effects of structure1948and decay on attachment and repair: A retrieval interference theory1949of recovery from misanalyzed ambiguities. Journal of Memory and1950Language, 49(3), 285–316.1951Wolter, L., Gorman, K. S., & Tanenhaus, M. K. (2011). Scalar reference,1952contrast and discourse: Separating effects of linguistic discourse from1953availability of the referent. Journal of Memory and Language, 65(3),1954299–317.1955Yee, E., & Sedivy, J. C. (2006). Eye movements to pictures reveal transient1956semantic activation during spoken word recognition. Journal of1957Experimental Psychology: Learning, Memory, and Cognition, 32(1), 1–14.1958Zwaan, R. A., & Radvansky, G. A. (1998). Situation models in language1959comprehension and memory. Psychological Bulletin, 123(2), 162–185.

1960



12 March 2015


Context-driven expectations about focus alternatives

Documents