How Readers Retrieve Referents for Nouns in Real Time: A Memory-based Model of Context Effects on Referent Accessibility

When a reader encounters a noun, she tends to rapidly associate the noun with a mental referent (representation of entity in question). Our computational model confirms that a memory-based account is sufficient to account for a high rate of success at preliminary referent retrieval. Definite noun phrases ("The dog") can be used anaphorically to refer to referents already mentioned in the text, but they also frequently introduce a new referent into the mix (Poesio & Vieira,1998). An adequate model must account for how a reader makes an explicit or implicit decision about each noun's anaphoric status. We suggest that LTM contains both generic referent types and specific referent tokens, which simultaneously compete for retrieval via resonance. Our ACT-R simulation operationalizes the memory-based view to model the pre and post-noun activations of referents in memory. It predicts which referent will be retrieved (i.e. the most active), and consequently whether an anaphor will be initially treated as a new referent. The influence of anaphor word choice is explained, and encompasses metaphoric anaphors. Simulations results are congruent with human performance in our eye-tracked reading study, in which regressions to reanalyze an anaphor are indicative of the incidence of preliminary error.