Evidence for Gradient Salience: What Happens with Competing Non-salient Referents during Pronoun Resolution?

Proceedings of the 2006 Australasian Language Technology Workshop (ALTW2006), pages 91–98. Evidence for Gradient Salience: What Happens with Competing Non-salient Referents during Pronoun Resolution? Ralph L. Rose Faculty of International Communication Gunma Prefectural Women’s University Gunma, Japan rose@gpwu.ac.jp Abstract The necessity of a gradient approach to salience ranking of referents introduced in a discourse is evaluated by looking at (unbound) pronoun resolution preferences when there are competing non-salient referents. The study uses a sentence- completion technique in which partici- pants had to resolve pronouns (“John sprayed the paint on the wall and then it ...”). Results suggest that a gradient salience model is necessary. Syntactic and semantic prominence effects on pronoun resolution were also compared with results showing that semantic prominence (i.e., agent > patient) determined the salience ranking of competing referents. 1 Introduction A pervasive theme in theories of discourse coher- ence is the concept of salience. It has proven to be a useful means of explaining how particular enti- ties seem to receive some preferential treatment in both the production and perception of a discourse. For instance, it has long been observed that entities realized in certain structural positions (e.g., gram- matical subject or first-mention) are preferred en- tities for topic continuation (Giv´on, 1983)—that is, they are preferentially referred to in the sub- sequent utterance. Similarly, pronominal refer- ence to entities realized in certain structural po- sitions (again, subject position, for example) is preferred to reference by repeated name (Gordon et al., 1993). In order to account for these ob- servations, it has often been theorized that in the speaker’s and hearer’s mental representation of the discourse, these entities are salient (similar terms include focused or given). (1).To illustrate this line of thinking, consider The pronoun in the second clause is pref- erentially interpreted as referring to LUKE rather than MAX. This has been observed in numer- ous psycholinguistic investigations (cf., Hudson- D’Zmura and Tanenhaus (1997); Mathews and Chodorow (1988); inter alia). In a simple salience- based account, it is hypothesized that LUKE is a salient entity after the first clause and that refer- ence to salient entities should be pronominal. (1) Lukei hit Maxj and then hei/#j ran home. While many studies have investigated differ- ences between pronominal reference to salient and non-salient entities, I have found no studies that have looked explicitly at what happens when a salient entity is not compatible with the pronoun, but more than one non-salient entity is. This is one of the main themes of the present study. Putting it as a question, what happens when there is compe- tition among non-salient entities for pronoun inter- pretation? The answer to this question has some wider implications for how salience is to be un- derstood. In particular, the answer to this ques- tion leads to conclusions about whether theoretical models require a gradient model of salience rank- ing or whether a categorical model is sufficient. In the following background section I will discuss this primary question further and introduce two related questions …

Content truncated to 500 words