TOWARDS A MODULAR APPROACH TO ANAPHORIC PROCESSING semantic

TOWARDS A MODULAR APPROACH TO ANAPHORIC PROCESSING: semantic operations precede discourse operations Arnout Koornneef

HYPOTHESES • Variable binding is cheaper than coreference • In ambiguous situation variable binding has precedence over coreference

SEMANTICS vs. DISCOURSE previousresearch • Piñango, Burkhardt, Brun and Avrutin (2001) • Dual task experiment: participants listened to critical sentences and performed lexical decision task • Rationale: tasks tap into same resources

STIMULI variable binding Everyonei hopes that the tenants will pay himi the rent before… coreference The landlordi hopes that the tenants will pay himi the rent before…

RESULTS • No difference in control condition • Longer RT’s in coreference condition in experimental condition • Variable binding is cheaper than coreference

DIFFERENT INTERPRETATION coreference condition The landlordi hopes that the tenants will pay himi the rent before… • the landlord c-command him: Variable binding and coreference are both possible • Longer RT’s because of interference between two available processes?

OBJECTIVES • Support for hypothesis 2: in an ambiguous situation variable binding has precedence over coreference • Additional support for hypothesis 1: variable binding is cheaper than coreference

HYPOTHESIS 2 • In an ambiguous situation variable binding has precedence over coreference

EXPERIMENTAL ITEMS • variable binding context A working day in the factory is always very tough. Especially today a lot of working men, among them the very old man Paul, could barely cope. Every working man who just like Paul was running out of energy, thought it was very nice that he could go home early this afternoon. After a hot shower things would probably look better. • coreference context A working day in the factory is always very tough. Especially today the very old man Paul could barely cope. Every working man who knew that Paul was running out of energy, thought it was very nice that he could go home early this afternoon. After a hot shower things would probably look better.

PREDICTIONS • variable binding context he is bound by every working man context supports dependency no re-analysis • coreference context he is bound by every working man context does not support dependency re-analysis he refers to Paul • Longer reading times (regression path or second-pass durations) in coreference context

EYE-TRACKING MEASURES • First-pass duration: time spent in a region before moving on or looking back • Regression path duration: time from first entering a region until moving the eyes beyond that region, includes regression time • Second-pass duration: duration of re-fixations • Total fixation duration: the sum of all fixations in a region

EXPLANATION OF EYE-TRACKING MEASURES DIFFERENT MEASURES The clown thinks that he is… 1 2 5 3 4 6 7 Reading Times for word 3 (thinks) First-pass duration = 3 + 4 Regression Path duration = 3 + 4 + 5 Second-pass duration = 6 Total duration = 3 + 4 + 6

FIRST-PASS DURATION binding: Every… thought it… that he could go home early this… co-ref: thought it… that he could go home early this… Every…

SECOND-PASS DURATION binding: Every… thought it… that he could go home early this… co-ref: thought it… that he could go home early this… Every…

TOTAL FIXATION DURATION binding: Every… thought it… that he could go home early this… co-ref: thought it… that he could go home early this… Every…

RESULTS HYPOTHESIS 2 • Significant differences in first-pass, second -pass and total fixation duration: longer reading times in coreference context condition

CONCLUSION • Results support hypothesis 2: in an ambiguous situation variable binding has precedence over coreference • Consistent with the more general claim: semantics before discourse

PROBLEMS • Are the ambiguous pronouns interpreted as we intended? • Is the context before critical pronoun strong enough to bias the interpretation?

TWO WEB-BASED CONTROL EXPERIMENTS 1) Ask participants how they interpret the pronoun 2) Ask participants to finish the story

CONTROL EXPERIMENT 1 • In variable binding context quantified referent is chosen 88. 2% of the time • In coreference context proper name is chosen 89. 5% of the time • The pronouns were interpreted as we intended

CONTROL EXPERIMENT 2 • Participants finish a story and indicate how the pronoun should be interpreted • A working day in the factory is always very tough. Especially today a lot of working men, among them the very old man Paul, could barely cope. Every working man who just like Paul was running out of energy, thought it was very nice that he. .

CONTROL EXPERIMENT 2 • In variable binding context story continues about quantified referent 89. 1% of the time • In coreference context story continues about proper name 78. 5% of the time • Pronoun in coreference condition initially more ambiguous

SOLUTION • Use subset of items with no difference between two conditions • 24 of original 36 items • Variable binding 85. 4% and coreference 89. 5%

SECOND-PASS DURATION binding: Every… thought it… that he could go home early this… co-ref: thought it… that he could go home early this… Every…

CONCLUSION • Results still support hypothesis 2: in an ambiguous situation variable binding has precedence over coreference • Consistent with the more general claim: semantics before discourse • Consistent with a modular view towards anaphoric processing

HYPOTHESIS 1: HYPOTHESIS 1 EXPERIMENTAL CONDITIONS experimental conditions • variable binding A working day in the factory is always very tough. Especially today a lot of working men could barely cope. Every working man who was running out of energy, thought it was very nice that he could go home early this afternoon. After a hot shower things would probably look better. • coreference A working day in the factory is always very tough. Especially today the very old man Paul could barely cope. Paul was running out of energy. It was very nice that he could go home early this afternoon. After a hot shower things would probably look better.

HYPOTHESIS 1: HYPOTHESIS 1 CONTROL CONDITIONS control conditions • similar to variable binding condition A working day in the factory is always very tough. Especially today a lot of working men could barely cope. All the working men who were running out of energy, thought it was very nice that they could go home early this afternoon. After a hot shower things would probably look better. • similar to coreference condition A working day in the factory is always very tough. Especially today the very old man Paul could barely cope. Paul who was running out of energy thought it was very nice that he could go home early this afternoon. After a hot shower things would probably look better.

PREDICTIONS • Longer reading times in coreference condition than variable binding condition • If there is interference between two processes: longer reading times in two control conditions compared to experimental conditions

EYE-TRACKING MEASURES • First-pass duration: time spent in a region before moving on or looking back • Regression path duration: time from first entering a region until moving the eyes beyond that region, includes regression time • Second-pass duration: duration of re-fixations • Total fixation duration: the sum of all fixations in a region

EXPLANATION OF EYE-TRACKING MEASURES DIFFERENT MEASURES The clown thinks that he is… 1 2 5 3 4 6 7 Reading Times for word 3 (thinks) First-pass duration = 3 + 4 Regression Path duration = 3 + 4 + 5 Second-pass duration = 6 Total duration = 3 + 4 + 6

REGIONS OF ANALYSIS 1 initial region variable binding: Every working man who was running out of energy, 1 initial region coreference: Paul was running out of energy. 2 pre-critical region variable binding: thought it was very nice 2 pre-critical region coreference: It was very nice 3 critical region: that he 4 spill-over region: could go 5 pre-final region: home early 6 final region: this afternoon.

SECOND-PASS DURATION binding: Every… co-ref: Paul… thought it… that he could go home early this… It was… that he could go home early this…

SECOND-PASS DURATION binding: Every… co-ref: Paul… thought it… that he could go home early this… It was… that he could go home early this…

RESULTS HYPOTHESIS 1: 1 • No significant differences in first-pass, regression path and total fixation duration • Significant difference in second-pass duration in first spill over region: longer reading times in variable binding condition • Only variable binding condition differed from control conditions

CONCLUSION • No support for hypothesis 1: variable binding is cheaper than coreference • No interference between variable binding and coreference • Why the difference between Piñango et al. and these results?

GENERAL DISCUSSION

GENERAL DISCUSSION hypothesis 11 HYPOTHESIS • No direct support for hypothesis 1: variable binding is cheaper than coreference

EXPLANATION? FOLLOWING BURKHARDT • Everyonei thinks that hei is not funny. • Every clowni thinks that hei is not funny. • The clowni thinks that hei is not funny.

EXPLANATION? FOLLOWING BURKHARDT • Everyonei thinks that hei is not funny. quantified non-referential • Every clowni thinks that hei is not funny. quantified referential • The clowni thinks that hei is not funny. referential

EXPLANATION? PROCESSING COSTS 1) quantified non-referential (everyone) is purely variable binding 2) referential (the clown) is coreference 3) quantified referential (every clown) is coreference associated with additional processing costs because the information about the set, provided by the restrictor clown, has to be transferred to the pronoun (e. g. , Burkhardt, 2004)

IMPLICATION • We used quantified referential dependencies in our variable binding condition. • This can explain why variable binding is more difficult than coreference in our stimuli • Similar results by Carminati, Frazier & Rayner (2002)

GENERAL DISCUSSION hypothesis 22 HYPOTHESIS • Support for hypothesis 2: in an ambiguous situation variable binding has precedence over coreference • Consistent with the more general claim: semantics before discourse • Consistent with a modular view towards anaphoric processing • The results for hypothesis 1 require further research

That’s all for today