Bielefeld Dialogue Colloquim November 18, 2005 Staffan Larsson Göteborg University Accommodation, feedback and semantic plasticity.

Bielefeld Dialogue Colloquim November 18, 2005

Staffan LarssonGöteborg University

Accommodation, feedback and semantic plasticity

Background

• Two important dialogue processes– Feedback

• Signaling perception, understanding, acceptance• Signaling failure to percieve & understand; rejection

– Accommodation: Adapting to the behaviour of other DPs • Adapting to presuppositions (adapting the situation model)• Adapting to language use (adapting the language / register)

• Two kinds of coordination in dialogue– Information sharing

• Grounding (updating common ground)– Adaptation of linguistic resources

• Semantic plasticity (& linguistic plasticity in general)• Adaptation of linguistic resources to activity• Gradual changes in language over time

Overview

• Introduction

• Issue-based Dialogue Management

• Accommodating questions

• Accommodating meanings

• An interpretation of Pickering & Garrod (2005)

Introduction

Accommodation

• Lewis (1979): If someone says something at t which requires X to be in the conversational scoreboard, and X is not in the scoreboard at t, then (under certain conditions) X will become part of the scoreboard at t

• Has been applied to referents and propositions, as parts of the conversational scoreboard / information state

Coordination and accommodation

• “If someone says something at t which requires X to be in the conversational scoreboard, and X is not in the scoreboard at t...”

• This seems to indicate a case of lack of coordination– but may also be used as a strategy for conveying

implicit information

• Accommodation can be used when there is miscoordination – Adjustment to presupposition of what is shared

Feedback in dialogue

• Feedback: signals for achieving coordination (alignment, grounding) on several levels (Clark, Allwood)– Contact / attention: +/-– Perception: +/-– Understanding: +/-/?– Reaction: accepting and rejecting utterances

• The hearer can react to some part of an utterance – A word, a phrase, a grammatical construct, or in

general any linguistic construct– Example: Clarification Requests (Ginzburg)

Grounding

• ”To ground a thing … is to establish it as part of common ground well enough for current purposes.” (Clark)

• Common Ground includes– general facts about the world (commonsense knowledge)– More specific facts about the world (e.g. facts about history)– facts about words (dog can mean ”canine animal”)– and more

• Henceforth, we will use grounding in a more limited sense– The process of adding information to the ”Dialogue

Gameboard” (DGB) or Conversational Scoreboard

Two kinds of coordination in dialogue

• Coordination of DGB– Adding new information (questions, propositions, actions) to DGB– Regulated by

• feedback• question accommodation • existensial presupposition accommodation• ...

• Coordination of semantic system, or “register” – Regulated by

• feedback• meaning accommodation• ...?

• Coordination of DGB presupposes a sufficiently coordinated conceptualisation of domain and situation, vocabulary, etc. (language system)

Issue-based Dialogue Management

Issue-based dialogue management

• A computational theory for improving flexibility of linguistic human-computer interaction, based on general principles inspired by observations of human-human dialogue

• Implemented in the GoDiS dialogue system• Background theory

– Ginzburg’s Dialogue Gameboard (DGB) and – Related DGB update protocols– Feedback and grounding (Ginzburg, Clark, Allwood)– Accommodation (Lewis)

• Dialogue moves: ask, answer, greet, quit• Dialogue is, basically, all about raising and

addressing issues– incl. short answers. e.g.”yes”, ”no”, ”paris”, ”in april”

• Dialogue plans drive dialogue in the absence of user intiative

• Extension: – Handling multiple issues– Question accommodation– Feedback– ...

inputinter-pret

TIS

DATABASE LEXICON DOMAIN

data-base

control

update selectgene-rate

output

lexicon domainknowledge

DME

GoDiS

Basic GoDiS information state record type

• SHARED = DGB / scoreboard• QUD: Questions which have been raised but not resolved,

and which are available for resolution of short answers

PRIVATE :PLAN : stackset( Action )

AGENDA : stack( Action )

SHARED :

BEL : set( Prop )

COM : set( Prop )

QUD : stackset( Question )

LU: SPEAKER: Speaker MOVES: OpenQueue( Move )

Sample dialogue plan

< findout(?x.dest-city(x))

findout(?x.depart-city(x))

findout(?x.transport(x))

findout(?x.dept-month(x))

findout(?x.dept-day(x))

raise({?class(economy), ?class(business)}

consultDB(?x.price(x)) >

Answer integration

RULE: integrateAnswerPRE: in( $/SHARED/LU/MOVES, answer(A) )

fst( $/SHARED/QUD, Q )$DOMAIN :: relevant( A, Q )

EFF: ! $DOMAIN: combine( Q, A, P )add( /SHARED/COM, P )

• Before an answer can be integrated by the system, it must be matched to a question on QUD

Basic dialogue with updates

U: ”price information please”; raises price issue– if user asks Q, push Q on QUD and push respond(Q) on AGENDA– if respond(Q) on AGENDA and PLAN empty, find plan for Q and

load to PLAN– if findout(Q) first on PLAN, ask Q

S: ”where do you want to go?”U: ”Paris”

– if LU/MOVES contains answer(A) and A relevant to Q, add P=Q[A] to SHARED.COM

– if P in SHARED.COM and Q topmost on QUD and P resolves Q, pop QUD

– if P in SHARED.COM and P resolves Q and findout(Q) on PLAN, pop PLAN

Basics cont’d

• …S: ”Do you want economy class or business

class?”U: ”economy class”

– if consultDB(Q) on PLAN, consult database for answer to Q; store result in PRIVATE.BEL

– if Q on QUD and P in PRIVATE.BEL s.t. P resolves Q, answer(P)

S: ”The price is £123”

A problem with QUDA: Who won the world cup?B: Where?A: in ItalyB: Are you sure it wasn’t in France?A: YesB: Where did you say it was again?A: ItalyB: Germany

• If QUD=<q1,q2> and q1 is resolved, q2 is available for resolution of short answers– takes no account of how many turns since q2 was raised– but short answers a long distance away from the question are not as

easily processed as an adjacent answer

ISSUES and QUD

• We extend Ginzburg’s DGB by adding ISSUES of type Stack(Question)

• ISSUES contains all raised but unresolved questions – ISSUES determines relevance of user answers

• QUD used for resolving short answers– questions drop off QUD after N turns– a short answer to a question that’s on ISSUES but not

QUD requires adjusting QUD by copying a question on ISSUES

Grounding and feedback

• Grounding– More limited sense than Clark’s: the process of adding

information to the ”Dialogue Gameboard” (DGB)– making sure that the participants are percieving, understanding,

and accepting each other’s utterances• ICM

– Interactive Communication Management– Feedback, Sequencing, Turntaking

• Feedback moves– (short) utterances which signal grounding status of previous

utterance• Sequencing moves

– utterances which signal switching task (”returning to…”), reraising questions (”so…”) etc.

Grounding moves• We want a typology of feedback moves• Some parameters:

– Action levels in dialogue (Allwood, Clark) • contact (con)• perception (per)• understanding

– Word meanings (sem)– Contextual meaning (und)

• acceptance (acc)

– Polarity• Positive (pos)• Negative (neg)• Checking (chk / int)

Some grounding moves in GoDiS• Formal representation

– icm:Level/Type{*Polarity}{:Content}

• Feedback moves– icm:und*neg – ”I don’t understand”– icm:und*pos:P – ”To Paris.”– icm:acc*neg:Q – ”Sorry, I can’t …”– icm:acc*pos – ”Okay”

• Feedback type selected depending on– Quality of recognised speech– Whether system can find a (relevant) interpretation– Whether system can accept what’s been said

• Sequencing moves– icm:reraise:Q – ”Returning to the issue of Q”– icm:loadplan – ”Let’s see…”

Grounding on multiple levels

S> Lets see. What can I do for you?

U> (mumble mumble)

S> Sorry, I didnt hear what you said. What can I do for you?

U> siridus

S> I heard you say siridus . Sorry, I dont understand. What can I

do for you?

U> channel one

S> channel one. I dont quite understand. Do you want to add a

program or change channel?

(dialog3a_edit)

Negative perception

feeback

Positive perception

Negative understanding

(sem)

Positive understanding

(word meaning)

Negative understanding

(und) Task clarification

U> turn on the t vS> turn on the TV… Sorry, I cannot control the TV set. Do you want to add a

program or change channel?”

U> pardonS> Sorry, I cannot control the TV set. Do you want to add a program or change

channel?U> change channelS> Okay.

(dialog3b_edit)

Rejection(negative

acceptance feedback)

Positiveunderstanding

User negative perception

System repetition

Positiveacceptance

Accommodating questions

Question accommodation

• If questions are part of the information state, they too can be accommodated

• If the latest move was an answer, and there is an action in the plan to ask a matching question, then– put that question on ISSUES – (and QUD if it is a short answer)

• Requires that the number of possible matching questions is not too large – (or can be narrowed down by asking clarification

question)

Issue accommodation PLANISSUES

• If – LM=answer(A) – no Q in ISSUES s.t. relevant(A,Q)

• then– find findout(Q) in PLAN s.t. relevant(A,Q)– push Q on ISSUES

• used when prevously unraised question (available in plan) is answered using a short or full answer

QUD accommodation ISSUESQUD

• If – LM=answer(A)– A is a short answer– no Q in QUD s.t. relevant(A,Q)

• then– find Q in ISSUES s.t. relevant(A,Q)– push Q on QUD– raise Q in ISSUES (make Q topmost)

• used when– previously raised question has dropped off QUD, but is

answered using a short answer– previously unraised question is answered using short answer

[needs PLANISSUES accommodation]

• If – LM=answer(A)– no Q in ISSUES s.t. relevant(A,Q)– no findout(Q) in PLAN s.t. relevant(A,Q)

• then– find Plan for some Q’ in DOMAIN s.t. findout(Q) or

raise(Q) in Plan and relevant(A, Q)– push Q’ on ISSUES– set PLAN to Plan

• used when previously unraised question, unavailable in PLAN, is answered using full or short answer (AKA ”task accomodation”)

Dependent issue accommodationDOMAINISSUES (+PLAN)

S: Welcome to the travel agency.U: From London to Paris in April

– not relevant to any question that has been raised, or to any current plan– look in domain knowledge for a plan (for dealing with some question Q)

with matching questions– load this plan, push Q on ISSUES– find in the plan the question(s) matching the user’s answer, and push them

onto ISSUES– integrate answer (requres matching question on ISSUES)

S: Alright, you want to know about price. (…)

– proceed to next plan item

S: How do you want to travel?– ISSUES=<?x.how(x), ?x.price(x)>

Sample dialogue: accommodation

Task clarification (VCR application)

S> What can I do for you? U> channel fiveS> channel five. I dont quite understand. Do you want to

add a program or change channel?U> change channelS> Okay. change channel.S> Lets see. channel five?U> yesS> The channel was changed

Status of question accommodation mechanisms

• QAMs are primarily a set of techniques (based on a common principle) for making dialogue systems more flexible in adapting to the user’s behaviour

• Are QAMs cognitively plausible? – Remains to be investigated

Accommodating meanings

Structuralism

• The sign relation, i.e. the connection between words (linguistic form) and concepts is arbitrary

• The way that linguistic material is divided into words is arbitrary

• The way that the world is divided into concepts is arbitrary

• Focus on study of language as a structure (langue); the concrete use of language (parole) assumed too unruly for scientific study

Poststructuralism

• Langue is continuously being affected by parole• Words change their meanings over time as a

result of language use• If or concepts determine how we understand the

world...• Concrete language use changes our

understanding of the world• Communication is not (just) transmission of

information

Phonetic plasticity

• Cutler, McQueen, Norris (2005)– ACL paper [find it]

• Experiment:– Ambiguous phoneme /?/ between /f/ and /s/– Group A hears words where /?/ replaces /f/,

e.g. ”carafe”– Group B hears words where /?/ replaces /s/

• Result– For group A, the /f/ category became more inclusive

(tested by phoneme categorisation)– For group B, the /s/ category became more inclusive– Exposure to /?/ in non-word context had no effect– Effect generalised to new words, and thus facilitates

word recognition

Semantic plasticity

• As with phonemes, semantic categories can gradually expand, contract, and shift

Kinds of semantic plasticity

• Semantic systems exist on several levels– National– Regional– Domain, activity, language game– Personal (idiosyncratic)– Particular interactions

• Semantic system can be adapted– to a new activity or domain– to a certain individual (who has an idiosyncractic way of using

some concept)– to a certain interaction (ad-hoc)

• Semantic systems can change– over time (diachronic semantics)– over space

• Meaning emerges from a multitude of interactions where the DPs of a linguistic community shape each other’s usage dispositions

• Cf. simulation work in this direction by Steels et al.• A language-user A observes some linguistic construct c

being used in a set of situations Sc

• A generalises over Sc; this generalisation we call the usage disposition [c]

• The way [c] is updated after a new use depends on the feedback given by other DPs

Sketch of a formal general account of semantic plasticity

Indeterminacy

• Upon hearing c in new situation s, A’s reaction (the kind of feedback A gives) partly depends on [c]– But A’s behaviour is not determined by [c]– This means that A can understand and accept uses of

c that deviate from [c]

• A’s own future uses of c are partly determined by [c]– Again, A’s own use of c is not determined by [c]– A can use c in ways that deviate from [c]

Disposition updates

• If follows from the definition of meaning that whenever a construct c is used, Sc will be extended, and so the usage-disposition [c] may change

• This is a disposition update– Disposition reinforcement

• This use of c is consistent with usage disposition, i.e., c is appropriate in s

• No drastic change; previous disposition is reinforced– Disposition revision

• This use of c is non consistent with usage disposition• More or less drastic change of meaning; previous meaning is

revised

Disposition reinforcement

• Assume c was used in situation s• Disposition reinforcement

– This use of c is consistent with usage disposition, i.e., c is appropriate in s

• Appr(c, s)A = True

– No drastic change; previous disposition is reinforced– Monotonic disposition update: [c]A := [c]A + s– Any uses previously deemed appropriate are still

appropriate

Disposition revision

• Disposition revision– This use of c is non consistent with usage disposition– Appr(c,s)A = false– More or less drastic change of meaning; previous

meaning is revised– Nonmonotonic meaning update: [c]A := [c]A * s– some uses previously deemed appropriate would now

be deemed non-appropriate• Exactly how the function is updated depends on

the learning mechanism– This distinction does not make sense for memory-

based learning models

The usage equation

• Use(c, s)A = fuse( fappr(fdisp(ScA, s)), X )

• What does this mean?• Whether A uses c in s depends on

– s: The current situation

– ScA: Situational collocation for c - previous situations

where c has been used, in A’s experience

– fdisp: The way A generalises over these

– fappr: The way A uses this generalisation do determine

the appropriateness of c in s– Any additional factors X

Possible outcomes of using c in s

Speaker appropriate (conservative use)appr(c, s)A = true

Speaker non-appropriate(creative use) appr(c, s) A = false

Hearerappropriateappr(c, s)B = true

Successful react(c, s)H = true)

Default Case reinforce [c]A, [c]B

Successful react(c, s)H = false

Unnoticed creativity revise [c]A, reinforce [c]B

Unsuccessful?

Unsuccessful?

HearerNon-appropr.appr(c, s)B =

false

SuccessfulAccommodated

Conservative Use reinforce [c]A, revise [c]B

SuccessfulAccommodated Creative Use revise [c]A and [c]B

UnsuccessfulUncovered discrepancy ? (negotiation)

UnsuccessfulFailed Creative Use reinforce [c]A and [c]B

Meaning accommodation

• For each construct used in an U, the addressee in a dialogue is (usually) expected to react if he thinks a construct in U was incomprehensible or inappropriately used– Clarification Ellipsis: negative understanding

• If a breakdown occurs during interpretation of U by B, it may be due to a mismatch between – the situation in which c was being used by A, and – B’s usage disposition for c (or B’s reaction disposition,

if it is different from the usage disposition)

• The addressee B may now – either reject this use of c explicitly: negative

feedback on understanding or acceptance level

– or quietly alter B’s usage disposition for c so that c can be counted as appropriate after all.

• The latter process we may call usage accommodation, or meaning accommodation

Two variants of accommodated use

• Accommodated creative use– Speaker non-appropriate, Hearer non-appropriate, successful– New use not “appropriate” according to speaker but speaker tries it

anyway– revise [c]A and [c]B

• Accommodated conservative use– Speaker appropriate, Hearer non-appropriate, successful– Not really creative since the speaker followed her appropriateness

judgement, but hearer had not heard that use before– reinforce [c]A, revise [c]B

• Example (of either of the above)– (in 1991 or so)– A: What are you doing?– B: I’m surfing the web– A: ... Ah, OK.

Conceptual pacts

• When a DP uses a construct c creatively, and this use is accommodated by the other DP, the result is an ad-hoc dyadic coordination w.r.t. the use of c

• C.f. Brennan (2003): a conceptual pact– temporary, flexible agreements to view an object in a

particular way (... specific to a particular partner).– may result in the emergence of lexical entrainment

• The repeated use of expressions by people in the same conversation

Emergence of meaning in interaction

• The meaning of a construct c in a linguistic community L emerges from the coordinated use of c by the members of L– Meaning is inherently social, and arises out of coordinated

behaviour in a linguistic community– In interaction, members of a community “mould” each others’

usage dispositions by giving feedback and accommodating usage

– This keeps language use sufficiently coordinated for meaning to arise

– If we take care of dispositions, accommodation and feedback, meaning will take care of itself

• By modelling plasticity of usage dispositions, we have indirectly modelled semantic plasticity

Alignment of meaning in interaction

• Not only situation models etc. are aligned

• Linguistic resources are plastic and can be dynamically updated in dialogue

• Feedback and accommodation are basic mechanisms that can account for this

Meaning and representation

• Meaning emerges from the coordinated use of linguistic constructs

• The intra-organism mechanisms underlying linguistic behaviour are irrelevant except insofar as they facilitate inter-organism coordination

• Intra-organism “representations” have meaning only indirectly, by being causal related to the interpretation and production of meaningful verbal behaviour

• In this sense, (private) mental “representations” derive their representational status only from being related causally to (social) linguistic representations (e.g. spoken utterances)

• Meaningful linguistic behaviour does not (a priori) require intra-organism “representations”, although they may do so in practice

Alignment

Alignment

• ”In dialogue, the linguistic representations employed by the interlocutors become aligned at many levels, as a result of a largely automatic process” (Pickering & Garrod 2004)

• Two DPs are aligned on a certain level when they “share the same represention” on that level

• Levels:– Situation model– Semantic representation– Syntactic representations– Lexical representations– Phonological representations– Phonetic representations

Situation models

• “A multi-dimensional representation of the situation under discussion”– including space, time, causality, intentionality, and

reference to main individuals under discussion• Other levels appear to reflect the Saussurean

linguistic sign– Semantic representation– ----------------------------------– Syntactic representations– Lexical representations– Phonological representations– Phonetic representations

Situation models vs. DGBs and Information States

• The situation models are “assumed to capture what people are ‘thinking about’ when they understand a text”

• Presumably, when people are understanding a text they are thinking about, among other things, (something like) questions and propositions

• This sound a lot like a Dialogue Gameboard– i.e. the part of the information state assumed to be

shared– DGB includes questions under discussion, jointly

comitted-to propositions• So alignment of situation model appears to imply

coordination of the DGB– taking the same questions as being under discussion,

the same propositions as being comitted to, etc.

Representation as “symbolic store”

• One meaning of “representation”

• A symbolic, possibly formal, representation that plays some causal role in behaviour

• Very similar to the traditional notion of representation in formal semantics and AI

Representations as “talking-as-if”

• “Alignment occurs at some level when interlocutors share the same representation at that level.”

• What does it mean to share the same representation on the level of the situation model?

• One other possibility, indicated by several examples: representation as “talking-as-if” (or “taking-the-perspective-of”)– A and B share the same representation on the situation model

level of X if they talk in the same way about X• Example: (P&G 2005)

– talking about a maze as if it were an arrangement of figures, or as if it were a network of paths...

– ... indicates differing representations of the spatial aspect of the situation model (P&G 2005)

• “maze as an arrangement of figures” vs. “maze as a network of paths” seem to refer to two different conceptualisations – semantic systems

• This indicates that sharing representations involves sharing semantic systems

• Being aligned w.r.t the situation model implies coordination of semantic systems (in addition to coordination of all other levels)

• The situation model involves a certain conceptualisation of the situation

• Insofar as feedback and accommodation are ways of achieving coordination of CG, – they also contribute to coordination of situation model

• If our rough outline of semantic plasticity is correct, – feedback and (meaning) accommodation appear to be ways of

achieving coordination on the level of the semantic system, and thus of the situation model

• Nevertheless, it seems to be useful to make a distinction between these two aspects of the situation model – common ground vs semantic system (“registers”)

• The possibility of coordination of CG depends (at least analytically) on (sufficient) coordination of semantic systems

Dealing with divergergence

• Divergence = lack of coordination• When an utterance is not understood or accepted, this may be due

either to – diverging beliefs (disagreement about facts), – diverging registers – or both

• It is sometimes necessary to coordinate registers before one can even figure out whether there is a disagreement– If the registers are not sufficiently coordinated, there is no way of telling

whether one disagrees about the facts• In practice, the distinction is not clear-cut

– Even though disagreement about beliefs analytically presupposes coordination of registers, convergence w r t registers and beliefs can proceed in parallel (but not independently)

• Accommodation and feedback can concern either DGB, register, or a little of both

Summary

• Feedback and accommodation as mechanisms for coordination of...– Dialogue Gameboard

• Question accommodation and feedback in Issue-based dialogue management

– Semantic system (register)• Meaning accommodation and feedback in a semi-formal account of

semantic plasticity

• Suggested an interpretation of P&G (2005)– Situation model as including DGB and register– Corresponing to two senses of alignment in the sense of sharing

representations• Sharing DGB• Sharing semantic system

• Problem with this interpretation:– Hard to tease apart semantic representation

from situation model (in the sense of conceptualisation)

– Perhaps this aspect of the situation model can be regarded as a system of semantic representations

Is alignment automatic?

• What does automatic mean?– Outside volontary control; action X is

automatic if one cannot help doing X– Nonconscious: action X is automatic if one is

not aware of doing X

• Each of these definitions seem to rely on notoriously troblesome notions (consciousness, free will)

Alignment?

• Alignment = grounding + priming?

Feedback

• Used in both processes

Feedback levels and alignment levels: rough correspondence

• Situation model– pragmatic understanding level

• Semantic representation– semantic understanding

• Syntactic, lexical, phonological, phonetic representations– perception

• (What about acceptance/rejection and attention levels?)

Learning and routinization

• Conscious learning

• Imitation learning

• Beginner vs. expert

The physical symbol hypothesis (Newell &

Simon 1976)• ”A physical symbol system has the

necessary and sufficient means for intelligent action.”

• All intelligent behaviour can be captured by a system that reasons logically from a set of facts and rules that describe the domain

Knowledge representation

hypothesis (Smith 1982)• Any mechanically embodied intelligent

process will be comprised of structural ingredients that – (a) we as external observers naturally take to

represent a propositional account of the the knowledge that the overall process exhibits, and

– (b) independent of such external semantic attribution, play a formal but causal and essential role in engendering the behaviour that manifests that knowledge.

Symbolic and nonsynbolic/subsymbolic

• According to Dreyfus, Brooks and others, cognition is not based on symbolic representations– Dreyfus claims that human language understanding

relies on a background which cannot be represented as a set of facts

• Still, human language is the prototypical symbol system so human cognition must involve symbolic representations

• How does symbolic and nonsymbolic cognition interact in language understanding and dialogue in general?

ISU accounts and KR hypothesis

• Most (if not all) formal theories of dialogue, and certainly those put forward in the ISU framework, rely on formal symbol manipulation

• If they are to be regarded as models of human cognitions, they rely on the validity of the KR hypothesis

• Note: ISU accounts of human linguistic congition are not ruled out by the fact that the brain operates on the subsymbolic level of neural network processing

• It is possible that the job of neurons in the brain is to implement a Turing Machine which can then handle cognition purely by symbol manipulation (as in a digital computer)

ISU approach

• Information State Update approach• Information States are abstract datastructures• Insofar as ISU is to be regarded as a proposal

for a theory of human linguistic cognition...• ... it appears to claim that major aspects of

human linguistic cognition are based on symbol manipulation mechanisms

• Alternatively, ISU-style accounts can be regarded as simplified symbol-based descriptions of processes which in reality are based on subsymbolic processing

Semantic plasticity

• Does semantic plasticity operate discretely or not?– I.e., is semantic change a question of either/or, or is it a gradual

process

• Discrete account– Semantic features as defaults that can be overridden– Implementable in ISU framework– (But where do semantic features come from?)

• Nondiscrete account– Based on some account of generalisation and learning– A computational account could be based on some machine

learning mechanis

Mechanistic vs. intentional

• P & G distinguish mechanistic from non-mechanistic, or intentional accounts which “present analyses of psycholinguistic processes purely in terms of their goals”

• But Issue-based dialogue management, and several related accounts, are– mechanistic, in the sense of computational

(what other senses are there?)– intentional, in the sense of including goals and

intentions in the analysis

How are these distinctions related?

• Intentional / Mechanistic

• Conscious / nonconscious

• Symbolic / subsymbolic

• Computational / non-computational

• Discrete / continuous

Recap

• Question accommodation– May contribute to an account of how DPs deal

with misalignment– Formulated in an ISU framework, as a part of

Issue-based Dialogue management

• Meaning accommodation

Important questions

• How is the mechanistic / intentional distinction related to similar distinctions?

• How does symbolic and subsymbolic congition interact in human linguistic cognition?

• How can these mechanisms:– accommodation – feedback

• ...explain these phenomena:– alignment of situation models– plasticity and the emergence of shared meaning

• We used feedback and accommodation to sketch accounts of alignment and

How are new words learned?

• Typically, the learning process begins when one first notices the word being used in some situation. – In this case, the set Sc is created with a single member.

• The redundancy of language ensures that each such situation will provide more insight into the meaning of the new word– But we might not be able to understand the meaning directly

• Eventually, the learner tries using the new construct and monitors the reaction to decide whether this was a successful (appropriate) use

• In a sense, we never stop learning language; if we stopped, our language would soon become outdated and a social obstacle

What are linguistic intuitions?

• We can simulate someone saying s and use our usage dispositions associated with the constructs of s to guide a decision whether s is correct or incorrect

• Again, our intuition does not necessarily determine our response

Related areas

• Alignment (Pickering & Garrod 2004)– ”In dialogue, the linguistic representations employed by the

interlocutors become aligned at many levels, as a result of a largely automatic process”

• Poststructuralist discourse analysis– Langue (language as system) is continually affecting and

affected by parole (language use)– Studies (among other things) how in political discourse certain

words are used in certain ways to alter their meaning (e.g. ”feminist”, ”communist”); there can be political struggles over meaning

• ”Meaning as use” (Wittgenstein 1952, Kripke 1982)– Replace truth conditions with assertability conditions (usage

conditions)– Focus on activity-specific language games

• Meaning potentials (Halliday 1978)– Activity-specific set of semantic options– Roughly similar to activity-specific sets of dialogue moves

• Constructive semantics and pragmatics (Lorenz, Kambartel 1976)– ”… language structures are throughout reconstructed not as

objects but as part of a rational practice.”– Opposes itself to traditional formal semantics

What are linguistic rules?

• Formalised generalisations over observed behaviour and linguistic intuitions

• Can be used normatively if one so wishes• They do not necessarily play a causal role in

generating behaviour (although they might; this is an empirical issue)

• We can follow a rule without knowing the rule, and without there being any representation of the rule in our minds/brains

In what sense is language a social system of signs?

• Signs are social conventions• The linguistic behaviour of the members of a linguistic

community must be sufficiently similar to allow them to coordinate their behaviour sufficiently to achieve their individual and collective goals

• We may refer to these regularities as social conventions• We need not be aware of these conventions• An individual might not conform completely to all

conventions, as long as he conforms sufficiently to a sufficient number of them