The semantics-pragmatics interface. How it works, why we need it, and where it is

FACULTÉ DES LETTRES Département de linguistique

The seman)cs-‐pragma)cs interface

How it works, why we need it, and where it is

Jacques Moeschler Department of linguis)cs University of Geneva

hBps://sites.google.com/site/moeschlerjacques/

1

https://sites.google.com/site/moeschlerjacques/


A roadmap

1. the seman)cs-‐pragma)cs interface 2. possible S-‐P interfaces 3. the func)on of S-‐P interface 4. the loca)on of S-‐P interface

2


1. the seman)cs-‐pragma)cs interface

3


The seman)cs-‐pragma)cs interface

The necessity of a Seman)cs-‐Pragma)cs Interface (SPI) is due to the following empirical facts: A. some pragma)c inferences (Conversa)onal Implicatures

CIs) are triggered by linguis)c items; B. pragma)c meaning seems to be more than non-‐truth-‐

condi)onal, as explicatures; C. pragma)c meaning can be determined by truth-‐

condi3onal meaning (causal connec)ves, nega)on).

4


A. Linguis)c and pragma)c meanings: implicatures

Generalized conversa3onal implicatures (GCIs) raise the issue of the encoding of pragma3c meaning: (a) Are conversa)onal implicature

part of the seman)c meaning? (b) Are they contextually

triggered? The first op)on leads to the ‘pragma3c meaning by default’ solu)on: a CI is triggered by a default inference.

The second op)on leads to the contextual solu)on: a CI must be contextually licensed or contextually blocked. 1. Some elephants have trunks.

?? +> not all elephants have trunks +> all elephants have trunks

2. Some of my students passed the pragma8c exam. +> not all of my students passed ?? +> all my students passed

5


Predic)ons

The default approach predicts that CIs should not be costly, since they are default inference.

But it has been shown that they are (Noveck 2001). The contextual approach predicts that CIs are favored in some contexts and blocks in others.

Experimental approaches on scalar implicatures shows that the contextual approach makes be?er predic3ons than the default one (Noveck 2001, Reboul 2004, Noveck & Sperber 2007, Noveck & Reboul 2010).

6


ConsequencesThe avantage of the default approach is that CIs are aBached to lexical meaning.

Its disadvantage lies in its erroneous predic3ons. The disadvantage of the contextual approach is that pragma3c meaning is not calculable without accessing contextual assump)ons.

Its advantage is that it makes correct predic3ons. SPI is clearly defined in the default approach, whereas it is unclear in the contextual one:

What is represented in lexical meaning? One op)on is the Relevance-‐theore)cal difference between linguis3cally encoded and communicated inferred concepts (ad hoc concepts). But what is linguis3cally encoded?

7


B. Explicatures vs. CIsExplicatures are pragma3c meaning which are not conveyed implicitly.

An explicature is an assump)on which is a development of the logical form encoded by the u?erance (Sperber & Wilson 1986, Moeschler & Reboul 1994).

1. Mary is happy: she finally met a bachelor. CIs are tradi)onally defined as non-‐truth-‐condi3onal meanings.

They do not contribute to the truth-‐value of the uBerance: non-‐truth condi)onal meaning is cancelable.

2. John fell and Mary pushed him, but not in this order. Explicatures are pragma3c truth-‐condi3onal meanings: they determine the truth-‐value of the uBerance: 3. It’s always the same at par8es: either I get drunk and no-‐one will talk to me

or no-‐one will talk to me and I get drunk.

8


ConsequencesThe no)on of explicature reduces the area of CIs, and breaks the clear-‐cut border between seman)cs and pragma)cs.

There are pragma3c meanings which are developments of logical forms and are truth-‐condi3onal.

Explicatures should not be cancelable, as CIs are. In fact, explicatures are cancelable: 1. Abi and Fée climbed the Roche de Solutré [together] 2. Abi and Fée climbed the Roche de Solutré, but not together. 3. Abi and Fée did not climb the Roche de Solutré together, but

separately.

9


C. Pragma)c meaning determined by truth-‐condi)onal meaning

Conversely, there are pragma3c meanings which are dependent of truth-‐condi3onal meanings. This is the case with temporal and causal meaning of connec3ves like and and because:

P and Q: in order for P and Q to mean P and then/because of this Q, both conjuncts must be true. P because Q: in order for P because Q to infer that Q CAUSE P, both P and Q must be true.

10


Empirical evidencesAnd temporal meaning can be defeated: in this case, what is evaluated is not the truth vs. falsehood of proposi)ons, but the temporal rela)on between them (Wilson & Sperber 2012): 1. What happened was not than Peter leT and Mary got angry but that Mary got

angry and Peter leT. Because causal meaning cannot be defeated: what can be false is either the effect, or the causal rela3on: 2. John did not fall because Mary pushed him

a. John did not fall, and the reason is that Mary pushed him (before he could fall)

b. It is not because Mary pushed John that he fell, but because he slipped down. Hence, and and because ‘presuppose’ the truth of the proposi)on they connect (Blochowiak 2014).

11


Provisory conclusion

SPI is more complex than the tradi)onal Gricean pragma)cs predicts: the Gricean criteria defining the border between Seman)cs and Pragma)cs are ruled out:

the truth-‐condi)onal vs. non truth-‐condi)onal aspect of meaning the cancela)on criterion the implicit vs. explicit aspect of meaning.

12


2. Possible SPIs

13


Possible SPIsWhat are the possible S-‐P interfaces? There are two classical answers: a. the pragma3cs as output of the linguis3c system b. the pervasive pragma3c intrusion into seman3cs Both proposals are unsa3sfactory: a. the first solu)on implies a step by step processing (from

syntax to pragma)cs), and cannot account for pragma)c intrusion, neither for parallel processing;

b. the second solu)on cannot account for the rela3on between explicatures and implicatures, and leads to the Gricean circle.

14


a. The linear modelThe first solu)on is a kind of linear model:

seman)cs in the output of syntax pragma)cs is the output of seman)cs

Two big issues: seman3cs is an interface of grammar, not an output; pragma3cs does not belong to the linguis)c system: it is not an input system (Fodor 1983), but belong to the central system of the mind (Sperber & Wilson 1986).

NB: for RT (revised version), there is a pragma3c module, including a comprehension and an argumenta3ve module

15

structural representa)ons

syntax

seman)cs

logical forms

pragma)cs

complete interpreta)on


Seman)cs as an interface

In a formalist framework (for instance the Minimalist Program), logical forms are the interface of the computa3onal system, as phonological forms are.

In Hauser, Chomsky & Fitch (2002), they are defined as the sensory-‐motor and the conceptual-‐inten3onal interfaces. The assump)on is that phonological formes (PF) and logical forms (LF) are interfaces of the grammar, and belong to the faculty of language in the broad sens (FLB), whereas FLN (faculty of language in the narrow sense) is restricted to recursion.

This picture contrasts with the defini)on of language as form-‐meaning pairs, where no hierarchy between formal or seman)c structures dominates and Jackendoff’s approach -‐ the flat phonological, syntac3c and conceptual structures (2002), implying 5 types of interfaces.

16


b. Linguis)cs and pragma)cs

One strong assump)on of pragma3c theory is that pragma3cs is not a component of linguis3cs:

Pragma)cs is part of the central system of the thought (Sperber & Wilson 1986) -‐ pragma)c is not a module and not devoted to specific tasks.

It means that pragma3cs deals with different inputs (supposed be translated is a same format) processed by the inferen)al central system. Linguis)c informa)on is one among other types of informa)on processed by the central system of the thought. What then is the rela)on between linguis3cs and pragma3cs?

17


Pragma)c intrusionOne main issue with SPI is pragma3c intrusion. The concept of pragma8c intrusion implies that pragma3c interpreta3on affects seman3c interpreta3on. Classical data arguing for pragma)c intrusion (Levinson 2000): condi3onal perfec3on (Geis & Zwicky 1971)

1. If you mow the lawn, I’ll give you five dollars. conjunc3on bu?ressing (Atlas & Levinson 1981)

2. John turned the key and the engined started. bridging (Clark & Haviland 1977)

3. John unpacked the picnic. The beer was warm. inference on a stereotype (Atlas & Levinson 1981)

4. John said ‘Hello’ to the secretary and then he smiled. nega3ve strengthening (Horn 1989)

5. I don’t like Alice mirror maxim (Harnish 1976)

6. Harry and Sue bought a piano

18


What’s wrong with pragma)c intrusion?

Pragma3c intrusion, in a neo-‐Gricean perspec)ve, implies that pragma3c inferences contribute to truth-‐condi3ons.

For instance, in bridging, reference resolu)on (as a pragma)c process) determines the full proposi)on to be truth-‐condi)onally evaluated.

This yields the Gricean circle: Implicatures (what is implicated) are computed on the base of the proposi)on expressed (what is said). Implicatures determine the proposi)on expressed (truth-‐condi3onal meaning).

19

what is said

what is implicated


Composi)onal seman)cs Indexical pragma)cs

Gricean Pragma)cs I Disambigua)on, reference assigna)on, specifica)on, etc.

Seman)c interpreta)on

Output: sentence meaning, expressed proposi3on

Gricean Pragma)cs 2 Indirec)on, irony and tropes, etc.

20

Output: speaker meaning, proposi3on meant by the speaker

Output: seman3c representa3on

Levinson (2000)


Implica)ons

Pragma3c intrusion shows that the border between seman)cs and pragma)cs is porous. Some aspects of pragma)c meaning are truth-‐condi)onal (explicatures), other are not (implicatures). Other rela)ons need to be taken into account, as entailment and presupposi3on, in order to fix SPI (Moeschler 2013).

21


3. The func)on of the S-‐P interface

22


Why do we need SPI?

SPI has as a main func)on to allow quick and efficient informa3on transfer from non-‐linguis)c sources to linguis)c ones, and vice-‐versa. Contextual informa3on is generally required for proposi)onal enrichment to trigger implicit and explicit inferred meaning. Linguis3cally encoded meaning is the star)ng point to enrichment process to access reference, inferred conceptual representa)ons, and implicatures.

23


One example: scalar implicatures

Scalar implicatures (SIs) are a classical case of SPI, allowing predic)ons about its func)on. SIs triggered by quan)fiers are closely connected with their logical meaning, as represented by the logical square (Horn 2004):

I +> not-‐A = O O +> not-‐E = I

Subcontraries in the logical square implicates each other.

24


the (augmented) logical square

25

A E

I O

contraries

subcontraries

contradictories

∀x

∃x

¬∃x

∃x¬

all none

some some not

nega)onaffirma)on

universals

par)culars

cannot be true together

cannot be false together

one true, the other false

not nonenot all


What is the issue?

In what follows, I will assume the strong connec3on between par3culars, but propose another (par)al) analysis. I will insist on what is linguis3cally encoded (seman8cs) and what is inferred (pragma8cs). The results will be that the pragma3c meanings of some and some…not will be interpreted as explicatures.

26


The seman)c and pragma)c meanings of par)culars

What can be the seman3cs and the pragma3cs of some and some…not? Three assump)ons:

a. there is a strong connec3on between both par3culars: this rela)on is expressed by a complement opera)on;

b. their seman3cs is defined as what is truth-‐condi3onally incompa3ble with each par)cular: i. some is seman3cally (logically) incompa3ble with no: they are

contradictories ii. some…not is seman3cally (logically) incompa3ble with all: they are

contradictories c. their pragma3cs is given by their incompa3bility with their upper-‐bound

correlates.

27


A Boolean seman)cs and pragma)cs for some

Seman3cs of some Xs Y the intersec3on between X and Y (the sets denoted by X and Y) is not empty; some Xs Y is seman3cally incompa3ble with no X Y

X ∩ Y ≠ ∅

Pragma3cs of some Xs Y X is not included in Y, because there must be a sub-‐set of X which is not in Y some Xs Y is pragma3cally incompa3ble with all Xs Y

X ⊄ Y

As a result, some Xs Y has as pragma)c meaning its explicature only some Xs Y

28


Seman)cs and pragma)cs of some…not

Seman3cs of some Xs not-‐Y the complement of the intersec)on between X and Y (the sets denoted by X and Y) is not empty some Xs not-‐Y is seman3cally incompa3ble with all Xs Y

∁(X ∩ Y) ≠ ∅

Pragma3cs of some Xs not-‐Y the intersec3on between X and Y is not empty some Xs Y is pragma3cally incompa3ble with no X Y

X ∩ Y ≠ ∅

As a result, some Xs not-‐Y has as pragma)c meaning its explicature only some Xs not-‐Y

29


An informal analysis

30

some Xs are not Y all Xs are Y

some Xs are Yno X isY

X Y

O and I trueO false

I and 0 true I false

pragma)cally incompa)ble

X Y

X Y

X Y

seman)cally incompa)ble

seman)cally incompa)ble


Summary

The seman3cs of some Xs Y is the pragma3cs of some Xs not-‐Y, and vice-‐versa. What is the difference with the implicature analysis? The main difference lays in the truth-‐condi)onal vs. non-‐truth-‐condi)onal pragma)c meaning.

31

seman)cs pragma)cs

some Xs Y X ∩ Y ≠ ø X ⊈ Y

some Xs not-‐Y ∁(X ∩ Y) ≠ ø X ∩ Y ≠ ø


The role of explicature in uBerance comprehension

The interpreta3on of par3culars is directly dependent of their truth-‐condi3onal meanings. Their meanings are crucially context-‐dependent: the not-‐all and not-‐none interpreta)ons can or cannot be triggered, depending on what the context is. The rela)on with their seman3cs is based on an exclusion condi3on.

32


The exclusion condi)on

The processing of the seman)cs and pragma)cs for some and some…not is the following: 1. exclude the incompa3ble seman3c meaning 2. exclude the incompa3ble pragma3c meaning 3. enrich the pragma)c meaning by explicature. This procedure yields a specifica3on reading by narrowing the seman3cs of the par3culars.

33


An informal heuris)c

34

some

no some or all

all only some✕

some… not

all some…not or none

none only some…not✕

seman)cs

pragma)cs

✕✕


A general explana)onReadings by specifica3on of par3culars (only some, only some…not) receive a cogni)ve and communica)ve explana)on. a. Communica3ve explana3on: it would be a viola)on of the first maxim

of quan)ty in saying some while meaning all. From a Relevance-‐theore)cal point of view, saying some while meaning all would allow the addressee to unjus3fied inferences, giving rise to false conclusions.

b. Cogni3ve explana3on: the par33on reading for some and some not allows an efficient and rapid processing, avoiding useless cogni)ve processes.

The predic)on is that nega3ve par3culars are not more costly cogni3vely than posi3ve one.

35


4. The loca)on of

SPI

36


Where is the S-‐P interface?

SPI is mainly a linguis3c issue: seman3c meaning is the locus of pragma3c processes. How can conceptual (and procedural) meaning be located in seman)c meaning? How is this representa)on compa)ble with a contextual approach? The argument will be based on temporal and causal connec3ves meanings.

37


Three connec)ves with different seman)c and pragma)c meanings

How to explain the differences in seman)c and pragma)c meanings between parce que, donc, et (because, therefore, and)? 1. Jean est tombé parce que Marie l’a poussé.

‘John fell because Mary pushed him.’ 2. Marie a poussé Jean, donc il est tombé.

‘Mary pushed John, therefore he fell.’ 3. Marie a poussé Jean, et il est tombé.

‘Mary pushed John, and he fell. The hypothesis is that the difference is not in the meanings encoded by connec3ves, but in the layer of meaning they encode. All connec)ves encode at some level a CAUSE rela)on and allow inferring the fac3ve vs. non-‐fac)ve status of the proposi)ons.

38


Connec)ves and their inferen)al meaning

In all cases, temporal and causal inferences are obtained, but with different seman3c and pragma3c paths. Some contents are the result of entailments, others are the result of explicatures or implicatures.

When a proposi)on is entailed, it must be true. When a proposi)on is developed as an explicature, it allows assigning a truth-‐value to the uBerance. When a proposi)on is an implicature, it can be cancelled.

1. Jean est tombé parce que Marie l’a poussé a. John fell & Mary pushed him b. Mary pushed John CAUSE John fell

2. Marie a poussé Jean, donc il est tombé. a. John fell b. POSSIBLE (Mary pushed John

CAUSE John fell) 3. Marie a poussé Jean, et il est tombé.

a. John fell & Mary pushed him b. POSSIBLE (Mary pushed John

CAUSE John fell)

39


Some examplesDonc and et are compa)ble with situa)ons where the CAUSE rela)on is explicitly given as possible but not certain.

1. Marie a poussé Jean, donc il est peut-‐être tombé.‘Mary pushed John, therefore he may have fallen.’

2. Marie a poussé Jean, et il est peut-‐être tombé.‘Mary pushed John, and he may have fallen.’

3. ?? Jean tombé parce que Marie l’a peut-‐être poussé.‘John fell because Mary may have pushed him.’

With parce que, the causal rela)on can be denied: 3. Jean n’est pas tombé parce que Marie l’a poussé, mais parce qu’il a

manqué une marche.‘John did not failed because Mary pushed him, but because he missed a step’.

40


A summary

Entailment Implicature Explicature

parce que P Q Q CAUSE P

donc P ♢(P CAUSE Q)

et P Q ♢(P CAUSE Q)

These connec)ves trigger different speaker’s commitment about the truth of the proposi3ons expressed: a. P is entailed by all connec3ves – parce que, donc, et. b. Q is entailed by parce que and et. c. The CAUSE rela3on is an explicature with parce que, and an implicature with iconic

order under the scope of a modal operator (et, donc) – it can be cancelled. Hence, the same informa3ve content is seman3cally and pragma3cally distributed in different ways.

41


Accessibility and strength What is the impact of the type of inference on u?erance interpreta3on? Two criteria can be used in order to rank inferences: their seman3c vs. pragma3c nature, and the type of inferred content: implicit vs. explicit. Accessibility and strength are two criteria allowing to differen)ate between entailment, explicature and implicature:

accessibility explicature > implicature > entailment strength entailment > explicature > implicature

42

entailment explicature implicature

seman)c + – –

pragma)c – + +

explicit – + –


Accessibility and strength Seman3c rela3ons (entailment) are less accessible, but stronger. Pragma3c rela3ons (explicature and implicature) are more accessible, but lighter (they can be more easily denied) Explicatures are stronger than implicatures because implicatures rule out a logical constraint seman)c rela)ons cannot have (see the entailment rela)on):

the impossibility to accept the rela)on to be true when P is true and Q is false.

43


entailment and implicature

44

P Q P implicates Q

1 1 1

1 0 1

0 1 0

0 0 1

P Q P entails Q

1 1 1

1 0 0

0 1 1

0 0 1


ConsequencesLinguis3c lexical items are the (main) locus of SPI. SPI can be made visible by the con3nuum of seman3c and pragma3c rela3ons as entailment, (presupposi)on,) explicatures and implicatures.

Causal connec3ves show how atomic and complex proposi)onal meanings are distributed in the seman)cs-‐pragma)cs con)nuum. Quan3fiers as some and some…not show how basic seman)c rela)ons (inclusion, intersec)on) are distributed at the seman)c and pragma)c level.

45


A general picture

46

entailment explicature implicature

–inferen)al +inferen)al

+truth-‐condi)onal –truth condi)onal

+speaker commitment –speaker commitment

–contextual +contexual

SEMANTICS PRAGMATICS


many thanks to…This work is part of my new Swiss Na)onal Science Founda)on research projet LogPrag (Seman8cs and pragma8cs of logical words). It explores the rela)on between logical and pragma)c meanings of connec)ves (et, ou, si) and nega)on. My collaborators are

Joanna Blochowiak (post-‐doc researcher) Karoliina Lohiniva (PhD student)

47


Thanks for your aBen)on

48

FACULTÉ DES LETTRES

Département de linguistique

ReferencesAtlas J. & S.C. Levinson (1981), « It-‐clezs, informa)veness, and logical form: radical pragma)cs (revised standard version) », In P. Cole (ed.), Radical Pragma8cs, New York, Academic Press, 1-‐61. Blochowiak J. (2014), A theore8cal approach to the quest for understanding. Seman8cs and pragma8cs of whys and becauses, PhD thesis, UNiversity of Geneva. Clark H.H. & S.E. Haviland. (1977), « Comprehension and the new-‐given contract, in R.O. Freedle (ed.), Discourse Produc8on and Comprehension, Hillsdale, Erlbaum, 1-‐40. Fodor J.A. (1983), The Modularity of the Mind, Cambridge, MIT Press. Geis M. & A. Zwicky (1971), « On invited inferences », Linguis8c Inquiry 2, 561-‐566. Hauser M, N. Chomsky & W.T. Fitch (2002), « The faculty of language. What is it, who has it,. and how did it evolve? », Science 298, 1569-‐1579. Harnish R. (1976), « Logical Form and Implicature », in T. Bever, J. Katz & T. Langendoen (Eds.)(1976), An Integrated Theory of Linguis8c Abili8es, New York: Crowell, 464-‐479. Horn L.R. (2004), « Implicature », in L.R. Horn & G. Ward (eds.), The Handbook of Pragma8cs, Oxford, Blackwell, 3-‐28. Horn L.R. (1989), A Natural History of Nega8on, Chicago, The Chicago UNiversity Press.

Levinson S.C. (2000), Presump8ve Meanings, Cambridge, MIT Press. Moeschler J. (2013), « Is a speaker-‐based pragma)cs possible? Or how can a hearer infer a speaker’s commitment? », Journal of Pragma8cs 43, 84-‐97. Moeschler J. & A. Reboul (1994), Dic8onnaire encyclopédique de pragma8que, Paris, Seuil. Noveck I. & D. Sperber(2007), The why and how of experimental pragma)cs, in N. Burton-‐Roberts (ed.), Pragma8cs, London, Palgrave, 141-‐171. Noveck I. & A. Reboul(2010), Experimental pragma)cs : A Gricean turn in the study of language, Trends in Cogni8ve Sciences 12/11, 425-‐431. Reboul A. (2004), « Conversa)onal implicatures: nonce or generalized? », in Noveck I. & D. Sperber (eds), Towards Experimental Pragma8cs, Basingstoke/NY, Palgrave Press, 322-‐333. Reboul A. (2007), Langage et cogni8on humaine, Grenoble, PUG. Sperber D. & D. Wilson (1986), Relevance. Communica8on and Cogni8on, Oxford, Blackwell. Wilson D. & D. Sperber(2012), Meaning and Relevance. Cambridge, CUP.

49

The semantics-pragmatics interface. How it works, why we need it, and where it is

Documents

Transcript of The semantics-pragmatics interface. How it works, why we need it, and where it is