Discourse Prominence and Movement Constructions

One of the key requirements in understanding a text is deciding how successive utterances relate to one another. Anaphors help in this respect by picking out entities already existing in the discourse and indicating the referent to which the new information encoded in the utterance should be attached. Sometimes, the interpretation of an anaphor is determined entirely by the morphosyntactic features of the anaphor itself which fully constrain its interpretation (e.g., reflexives). However, at other times, the choice of a referent for an anaphor is free and listeners must rely on other factors to determine its interpretation. Current theories of pronominal resolution (McKoon et al., 1993; Grosz and Sidner, 1986; Grosz et al., 1995; Lappin and Leass, 1994; Kennedy and Boguraev, 1996) hold that in such cases, the pronoun is assigned (if possible) the best-matching most prominent referent in a list of discourse-prominent entities–that is, a list of entities invoked in the current discourse and ranked according to discourse prominence. Should later information prove that this assignment was incorrect, reanalysis may be necessary and certain consequences may result such as degraded discourse coherence. Hence, the process by which discourse entities are ranked is therefore a critical feature for pronominal resolution. In this paper, I examine two previously uninvestigated features—one syntactic, the other semantic—which may influence this ranking process. All else being equal, entities realized as syntactic subjects are preferred referents for pronominal resolution. This is illustrated by the difference in discourse coherence between (1) and (2).