Retour vers liste

Détail de la contribution

Auteur: Ezra KESHET

Coherence and Donkey Anaphora

Abstract/Résumé: Kehler et al. (2008) show the effect of discourse coherence on pronoun resolution with stimuli like the following: (1) Samuel threatened Justin with a knife, and a. ... Erin blindfolded him with a scarf. [Parallel] b. ... Erin stopped him with pepper spray. [Result] c. ... he blindfolded Erin with a scarf. [Parallel] d. ... he alerted security with a shout. [Result] The continuations in (a-d) vary by pronoun placement (subject/object) and coherence relation (Parallel/Result). Result variants were biased towards non-parallel pronoun-antecedent relationships to balance expected antecedents. Overwhelmingly, participants chose antecedents consistent with establishing coherent discourses rather than, e.g., parallel syntactic structure (i.e. "Justin" in (d) and "Samuel" in (b)). Their study used only referential pronouns, though, not bound or donkey pronouns. Donkey anaphora (Geach 1962) theories in the E-type tradition (Evans 1980, Heim 1990, Elbourne 2005) posit quite different analyses for donkey versus referential pronouns, while DRT-type theories (Kamp 1981, Heim 1982) maintain a closer relationship between the two via machinery like discourse referents. If donkey resolution behaves like referential resolution, it might indicate a DRT-like theory is correct, although much work would remain. The current study addresses this question with stimuli such as the following, using donkey anaphora in place of referential: (2) Every time a thief stabbed a cop in the kitchen, a. ... an accomplice shot him in the hallway. [Parallel] b. ... a detective arrested him in the yard. [Result] c. ... he shot a detective in the yard. [Result] d. ... he radioed for back-up in the hallway. [Result] Eight such sets were designed. In the initial study, eight participants on Amazon Mechanical Turk each saw one variant from each set, therefore seeing each type twice, presented thus: (3) Every time a thief stabbed a cop in the kitchen, a detective arrested him in the yard. Who did the detective arrest? (a) the thief (b) the cop The subjects strongly preferred to resolve donkey anaphora consistent with coherence, choosing subject antecedents 7 of 16 times for type (a) sentences, 15 of 16 times for (b), 13 of 16 for (c), and 0 of 16 for (d). (0, 16, 16, 0 would have been perfectly consistent with coherence.) Two full factorial ANOVAs, with mean choice of subject antecedent as dependent variable, pronoun position and coherence as factors, and subject (F1) and stimulus set (F2) as random variables, show a significant interaction between pronoun placement and coherence [F1(1,8)=39.076, P<.0001; F2(1,8)=24.200, P=.002], as expected if subjects chose by coherence. Future studies will increase the number of stimulus sets and subjects, compare definite description to pronoun resolution, and test reaction time.