Discourse model representation of referential and attributive descriptions

Abstract

De nite descriptions (as in The murderer of Smith is insane) can have at least two interpretations: a referential one, in which insanity is predicated of a particular individual who killed Smith, and an attributive one, in which insanity is predicated of whoever it is that killed Smith. Experiment 1 manipulated shared knowledge and focus on speci c entities, the verb in the sentence, and whether the description was de nite or inde nite. Each factor in uenced interpretation of the description. Experiment 2 con rmed that changing the verbs alone affected reference choice. Experiments 3 (ratings) and 4 (reading times) indicated that both referentially and attributively introduced entities are conceptually singular (better as antecedents of singular than plural pronouns) while generically introduced entities are conceptually plural. Thus, the difference between the discourse representation underlying referential and attributive interpretations does not hinge on a difference in the number of tokens being instantiated.

When anaphor resolution fails: Partial encoding of anaphoric inferences

“…According to this ''No Encoding'' hypothesis, although definite noun phrases (e.g., the dessert) were used anaphorically by the characters in the passage (i.e., Wanda knew that the dessert was a tart), the noun phrases were not treated as anaphors by the readers. Instead, given the low level of activation of the antecedent when the anaphor was processed, readers treated the anaphors as new information, rather than as co-referential (e.g., Onishi & Murphy, 2002). This hypothesis seems plausible given that definite noun phrases are often used to refer to entities that are new (Fraurud, 1990;Poesio & Vieira, 1998).…”

How incorrect information delivers correct search results: A pragmatic analysis of queries

“…To provide an explanation for such success above, we apply a well‐known distinction in pragmatics, the attributive/referential distinction, initially developed by the philosopher Keith Donnellan (1966). Although studies have applied Donnellan's distinction to computational models of reference and discourse (Birner, 1991; Kronfeld, 1986; Onishi & Murphy, 2002), there has been, with the exception of Ng's adaptation of Habermas (2002), little application of pragmatics to actual information retrieval (IR) interactions.…”