Event Structure and Discourse Coherence Biases in Pronoun Interpretation

In a story completion study involving transfer-of-possession passages (John handed a book to Bob. He _____), Stevenson et al. (1994) identified a pronoun interpretation preference that is compatible with two possible explanations: a superficial thematic role bias for Goals over Sources, and a deeper event structure bias toward focusing on the end state of such events. To distinguish these hypotheses, we conducted an experiment manipulating the salience of the end state by comparing the perfective (handed) and imperfective (was handing) forms of the transfer verb. We found that sentences in the imperfective yielded significantly more Source resolutions than those in the perfective, supporting the eventstructure hypothesis. Furthermore, we found that a significant proportion of Goal interpretations arise from continuations which use narration-based coherence relations. As such, the interpretation preferences are better explained at the level of event structure and discourse coherence, rather than by appeal to superficial heuristics. Interpreting Ambiguous Pronouns The question of whether pronouns are interpreted based primarily on surface-level morphosyntactic cues (subjecthood, first mention, recency, parallelism) or as a byproduct of deeper discourse-level processes and representations (inference, event structure) remains unresolved in the literature. These two views come together in a story-completion study by Stevenson, Crawley, and Kleinman (1994; see also Arnold 2001) where they test the influence of thematic roles on pronoun interpretation. One of their most striking results comes from their experiment with story completions like (1) with a transfer-of-possession context sentence followed by an ambiguous pronoun. The context sentence in these passages contains two possible referents for the pronoun, one that appears in subject position and fills the Source thematic role, and one that appears as the object of a sentence-final prepositional phrase and fills the Goal thematic role. They found that Goal continuations (continuations which correspond to a Goal interpretation for the pronoun) rivaled Source continuations in frequency. This result is unexpected in light of a variety of existing models of pronoun interpretation. The subject preference (Crawley, Stevenson, & Kleinman 1990) predicts that an ambiguous pronoun ought to be coreferential with the subject of the previous sentence, in this case the Source. The first-mention privilege (Gernsbacher & Hargreaves 1988) also points to the salience of the Source subject. The grammatical parallelism preference (Smyth 1994; Chambers & Smyth 1998) predicts that an ambiguous subject pronoun resolves preferentially to a subject antecedent, again the Source for examples like (1). Stevenson et al. consider two explanations for the promotion of the non-subject Goal. The first is a thematicrole-level preference which amounts to a heuristic ranking Goals above Sources. The second explanation is an eventlevel bias for focusing on the end state of the previously described event. The Goal is considered to be more salient than the Source with respect to the end state in transfer-ofpossession events. In this paper, we seek to separate out the thematic-role preference from the event structure bias in Stevenson et al.’s results. We further investigate whether the Goal preference can receive deeper motivation from mechanisms that are used in establishing coherence in discourse. Thematic-Role Preference or Event-Level Bias We designed an experiment to distinguish Stevenson et al.’s two hypotheses. By manipulating the verbal aspect of the context sentence, we kept the thematic role relations constant but altered the structure of the event. Taking passages like (1), repeated here as (2), we formed minimal pairs with the imperfective as in (3). (2) COMPLETED EVENT (PERFECTIVE) JohnSOURCE handed a book to BobGOAL. He _______ (3) INCOMPLETE EVENT (IMPERFECTIVE) JohnSOURCE was handing a book to BobGOAL. He ____ The thematic roles remain the same in examples (2) and (3), but the perfective verb in (2) describes a completed event which is compatible with end-state focus, while the imperfective verb in (3) describes an event that is an ongoing process, making it incompatible with end-state focus (Moens & Steedman 1988). Thus the thematic role preference predicts a Goal bias for both (2) and (3), while the event structure hypothesis predicts fewer Goal interpretations for (3) since the imperfective verb is inconsistent with a salient end state. The distinction between perfective and imperfective therefore allows us to distinguish between Stevenson et al.’s two hypotheses by separating out the intrinsic thematic role assignments of the verb from the end-state focus at the event level. Experiment and Methodology Following Stevenson et al., we used a story completion task to elicit continuations which were then evaluated to determine the participants’ intended pronoun interpretations. Participants Forty-eight monolingual English-speaking undergraduates at UC San Diego participated in the study for extra credit in Linguistics courses. Stimuli The twenty-one experimental stimuli consisted of a transfer-of-possession context sentence followed by an ambiguous pronoun prompt, as in (2) and (3). To manipulate the event structure, we varied the verbal aspect, using perfective and imperfective forms of the verb. For each verb, participants saw either the perfective or imperfective form, but not both. The Source referent always appeared in subject position, and the Goal was the object of a to-phrase. All verbs described physical transfer events (ex. hand, throw). We excluded verbs that described abstract or conceptual transfer (ex. show, teach) because they lacked a clear consequent or end state, even though they have been used in prior work on transfer of possession. The twenty-one verbs in the stimuli were classified along two dimensions: co-location of event participants and guarantee of successful transfer. We were interested in how properties of the event, such as the absence of a co-located Goal or the lack of guaranteed transfer, could affect the salience of the available referents. Figure 1: Transfer-of-possession verb classes Class 1: hand, give, bring, pass, deliver,