520 likes | 631 Views
An experimental investigation on the semantics/pragmatics of scalar terms. N. Katsos & R. Breheny University of Cambridge. Overview…. The semantics/pragmatics of Scalar terms Study 1: Off-line evidence on the psychological reality of defeasibility scale dependency
E N D
An experimental investigation on the semantics/pragmatics of scalar terms N. Katsos & R. Breheny University of Cambridge
Overview… • The semantics/pragmatics of Scalar terms Study 1: Off-line evidence on the psychological reality of defeasibility scale dependency • Competing linguistic accounts: Context Sensitive vs Default Generation Studies 2a-c: On-line evidence on the processing of scalar terms • Pragmatic accounts of numerals • Study 3: On-line processing differences disjunction vs the numerals • Architecture of the pragmatic system • From philosophical to psychological validity
Standard assumptions • Linguistically encoded meaning underdetermines Speaker meaning A: Was the food good? B: John or Bill got sick. >> The food wasn’t good. • Grice 1989: The constraints that guide inferencing from linguistic meaning to speaker meaning are pragmatic. We have to infer an implicature, in order to assume that the speaker is not violating conversational principles.
Implicatures of Scalar terms A: Was the food good? B: John or Bill got sick. >> The food wasn’t good. >> either John or Bill got sick, but not both A: Did they enjoy the bull-fights? B: John or Bill got sick. >>They did not have a good time. >> either John or Bill got sick, but not both Particularised Implicatures Scalar Implicature
On the properties of scalar sets • a. <or, and> • b. <some, many, most, all> • c. <zero, one, two, three> • d. <like, love, adore> • Asymmetric entailment relations. Terms on the right of the scale are more informative, i.e. true in a narrower set of circumstances
On the properties of scalar sets • Asserting a weaker term of the scale implies that the stronger ones are not applicable. A: Who got sick? B: John or Bill got sick >>It is not the case that John and Bill got sick
Pragmatics/semantics of scalars • Are such aspects of meaning really ‘pragmatic’? Unlike other types of inferences, Scalar Implicatures (SI) are: 1. Explicitly defeasible 2. Defeasible due to discourse goals 3. Scale dependent
Pragmatics/semantics of scalars • Explicitly defeasible • Can be explicitly cancelled without giving rise to contradictions: A: Who brought a cake? B: John or Bill brought a cake. In fact, both of them did.
Pragmatics/semantics of scalars • Defeasible due to discourse goals • Scalar Implicatures (SI) will not arise, if the discourse goal can be satisfied without them: A: Who could give me a ride to the patisserie? B: John or Bill both. (either John or Bill or both. Inclusive “or”)
Pragmatics/semantics of scalars • Scale dependency • SIs do not (usually) arise in a Downward Entailment (DE) context where the scale of informativeness is reversed: If John or Bill remember to bring the cake, our birth-day party will be saved! (if either John or Bill or both. Inclusive “or”)
Pragmatics/semantics of scalars • Scalar expressions have been standardly (Horn 1982) assumed to involve: • lower-bound semantics ‘X or Y’: either X or Y and possibly both (inclusive ‘or’) • upper-bound pragmatics through an SI ‘X or Y’: either X or Y but not both (exclusive ‘or’)
Psychological reality of the properties of Scalar terms • Are interlocutors implicitly sensitive to the fact that SIs are Defeasible and Scale Dependent? • Disjunction ‘X or Y G-ed’ • Create Upper Bound contexts where it is required to know precisely ‘Who G-ed’ • Create Lower Bound contexts where discourse goals will be satisfied ‘if at least one of the X, Y can G’ • Create Downward Entailment contexts where the inclusive interpretation of the disjunction is more informative.
Study 1: Items • UB: Jane asked Mary: Which colleague has used my photocopying card without my permission? Mary replied: George or Andrew from the next office ... • LB: Jane asked Mary: Who has a photocopying card to lend me? Mary replied: George or Andrew from the next office ... • DE: Jane told Mary: I urgently need a photocopying card. Mary replied: I think that if George or Andrew from the next office ... one, you should ask for it.
Study 1: Methodology • Participants were asked to fill in a missing verb either in singular or plural: is, are, has, have, does, do • Assumption: singular exclusive interpretation plural inclusive • 43 participants native speakers of British English • 24 items rotated in 3 lists
Fundamental issues • Off-line evidence that implicatures are indeed defeasible and scale dependent: Their computation is sensitive to context (discourse goals) and grammar (direction of entailment) • Given that SIs can be defeated, how are they computed? Competing linguistic approaches: • Computation by Default vs. Context Sensitive
Default SIs • SIs are attached to linguistic expressions • When a trigger is processed, the grammar (or a specialized pragmatic system) generates the strongest interpretation • At the stage where contextual assumptions are taken into consideration, the interpretation may have to be cancelled • SIs are default defeasible inferences
Context Sensitive Approaches • Emphasis on the importance of contextual assumptions: • SIs are only computed if there is some specific inferential benefit. • SIs are Context Sensitive inferences (will not be computed unless there is a relevant discourse goal) Interactive Approaches • Both sources provide activation (predicts equivocal behaviour when the sources conflict)
In an Upward Entailing Context Context Sensitive Check whether there is a relevant inferential benefit • If yes, then the SI is computed Architecture: single system sensitive to context Default The SI is always computed • If then it becomes evident that it is not warranted (discourse goals), the SI will be cancelled Architecture: two stage system 1st Stage: default operations (impenetrable to contextual assumptions) 2nd Stage: contextual assumptions are considered
Strong Interactive Strong Default Contextualist
Calls for experimental investigation • From pragmatics to cognition and from philosophical to psychological validity • Levinson 2000: ‘GCI Theory clearly ought to make predictions about processing... There is very little psycholinguistic work directly addressed to implicature and still less of this concerns on-line processing, but one may hope for rapid progress here’ (ibid:370) Also reasoning and acquisition. • Similarly Relevance theory, Chierchia (2005)
Study 2a: On-line Processing Assumption: all inferencing processes require a certain amount of time (potentially reflected in behavioural tasks like reading time and response time) • Default approach: Upper Bound ≤ Lower Bound S + I ≤ S + I + Cancellation In the L B the SI must be generated AND THEN cancelled • Context Sensitive: Upper Bound > Lower Bound S + I > S In the L B the SI is simply NOT generated in the first place
Study 2a: Items Disjunction: X or Y • Upper Bound: Jane asked Mary:/ Which colleague/ has used my photocopying card/ without my permission?/ Mary replied:/ George or Andrew/ from the next office/ has. • Lower Bound: Jane asked Mary:/ Who has/ a photocopying card/ to lend me?/ Mary replied:/ George or Andrew/ from the next office/ has.
Study 2a: Methodology • timed self-paced, segment-by-segment reading experiment, with comprehension motivated by questions after some of the sentences • 16 critical items rotated in 2 conditions • 42 fillers – 25 comprehension questions • items were presented in English • 40 participants, adult native speakers of British English
Study 2b: Items Disjunction: X or Y • Upper Bound context: John was taking/ a university course/ and working at the same time./ For the exams/ he had to study/ from short and comprehensive sources./ Depending/ on the course,/ he decided/ to read/ the class notes or the summary./ • Lower Bound context John heard that/ the textbook for Geophysics/ was very advanced./ No-one of his friends/ understood it properly./ He heard that/ to pass the course/ he had/ to read/ the class notes or the summary./
Study 2b: Methodology • timed self-paced, segment-by-segment reading experiment, with comprehension motivated by questions after some of the sentences • 12 critical items rotated in 2 conditions • 40 fillers – 25 comprehension questions • items were presented in Greek • 32 participants, adult native speakers of Greek
Study 2c: Non logical scale • Scales based on world knowledge A: Did you paint the house? B: I painted the roof. A: Did you meet her parents? B: I met her sister. Contextualists: Discourse introduces an object or a set. Asserting one part of it implies that “the rest” is not the case. It is difficult to see how the grammar or a default system could generate these inferences
Study 2c: Items UB: George went to pick up Mary from the station. He was covered in paint. Mary asked him: Were you painting the house? George replied:/ I was painting/ the roof/ with an insulating paint./ LB: George went to pick up Mary from the station. He was covered in paint. Mary asked him: What were you painting? George replied:/ I was painting/ the roof/ with an insulating paint./
Study 2c: Methodology • timed self-paced, segment-by-segment reading experiment, with comprehension motivated by questions after some of the sentences • 12 critical items rotated in 2 conditions • 40 fillers – 25 comprehension questions • items were presented in English • 20 participants, adult native speakers of British English
Study 2a-c: Conclusions • It is suggested: SIs are generated only where context warrants them. • The fact that it takes a significant amount of time to generate SI corroborates the findings of Noveck & Posada (2003), Bott & Noveck (in press) (reasoning). • There is a single pragmatic system sensitive to contextual assumptions
The pragmatics of Number • Numerals form a scalar set. <zero, one, two, three> • Asymmetric entailment relations Terms on the right of the scale are more informative, i.e. true in a narrower set of circumstances • Horn 1989, Levinson 2000, Numerals have an ‘at least N’ semantics and an ‘at most N’ implicature.
Similarities with SIs • The ‘at most N’ inference exhibits all three properties of implicature 1 defeasibility 2 sensitivity to discourse goals 3 scale dependency
Similarities with SIs • Defeasibility: Mary has three children. In fact, she has four. • Sensitivity to discourse goals: A: Families with three children can apply for tax exemption. B: Mary’s family can apply for it. She has three children. • Scale dependency: If a family has three children, they can apply for family benefits.
Differences with SIs Carston, 1998, Geurts 1998, Horn 1992 i.a. have pinpointed several differences between numerals and other scalar sets. • Different distribution with modifiers: Bill is hosting less than {?a few/ ?some / three} relatives Bill is hosting exactly {?a few / ?some / ?many/ three} relatives • Mathematical statements: (!) 3 + 3 = at least 6
Differences with SIs Approximation in numerals seems to affect the ‘at least’ interpretations: • Bill is carrying £100 in his wallet. vs. • Bill is carrying £101.5 in his wallet. • Everyday conversation: (!) I took six cigarettes with me, gave one to Fred and two to Ed, so I still have three (Sadock 1984)
Differences with SIs • Evidence from acquisition: numerals are processed differently from other scalar sets. Papafragou & Musolino (2003) found that children at the age of 5 accept ‘some of the Fs G-ed’ when the case is that ‘all the Fs G-ed’ but they do not accept that ‘4 of the Fs G-ed’ when the case is that ‘5 of the Fs G-ed’.
The pragmatics of Number Menu: • Neo-Gricean: • at least N semantics • exactly N through pragmatics • Independently motivated mechanisms • Carston 1998, Underspecification [X[two]] • Guerts 1998, context dependent Polysemy
Study 3: Disjunction vs Numerals • In the case of SIs, extra pragmatic processing, i.e. implicature generation, reflects on extra processing time (studies 2a-c). • Underspecification and polysemy are linguistically triggered processes that do not require extra processing time (Frisson & Pickering 1999).
The experimental approach Disjunction: X or Y • Upper Bound context: John was taking a university course and working at the same time. For the exams he had to study from short and comprehensive sources. Depending on the course, he decided to read the class notes or the summary at the end of the chapters. • Lower Bound context John heard that the textbook for Geophysics was very advanced. No-one of his friends understood it properly. He heard that to pass the course he had to read the class notes or the summary at the end of the chapters.
Study 3: Disjunction vs Numerals Numerals • Upper Bound context Peter and a friend were taking part in a geography contest. Peter could not see the score clearly and so he asked his friend how the opposing team was doing. His friend told him that they had managed/ to answer/ two questions/ in total./ • Lower Bound context Peter took part in a geography contest to win a ticket to the World Cup. He looked at the rules to see how well he had to go to qualify for the next round. He saw that he had/ to answer/ two questions/ in total./
Study 3: Methodology • timed self-paced, segment-by-segment reading experiment, with comprehension motivated by questions after some of the sentences • 24 critical items, 12 for each expression (disjunction - numeral) rotated in 2 conditions (Upper bound – Lower bound) • 42 fillers – 25 comprehension questions • items were presented in Greek • 40 participants, adult native speakers of Greek
Significant Interaction: Inference Type x Upper/Lower Bound F(1,39), 7.91, p< .001
Study 3: Results Numerals are processed different than standard scalar terms
Conclusions Study 1: linguistic assumptions on defeasibility of SIs enjoy psychological reality Studies 2a-c: SI processing is not compatible with the Strong Default account on SI generation Study 3: Processing data converge with data from acquisition that the interpretation of numerals does not involve an SI Studies 2 & 3: there is a single context sensitive pragmatic system Overall: Relevance and feasibility of experimental investigations for traditional linguistic debates.
Next steps… • Further studies: to investigate the DE conditions for SI and numerals to manipulate the interaction of context and DE the problem of neutral contexts
C + G +: Jane asked Mary: Which colleague has used my photocopying card? Mary replied: George or Andrew from the next office ... • C – G +: Jane asked Mary: Who has a photocopying card to lend me? Mary replied: George or Andrew from the next office ... • C + G -: Jane asked Mary: Which colleague has used my photocopying card? Mary replied: I don’t. I don’t know whether George or Andrew from the next office ... • C – G -: Jane asked Mary: Who has a photocopying card to lend me? Mary replied: I don’t. I don’t know whether George or Andrew from the next office ...