Discriminating anaphoric and non-anaphoric definite nouns: A unified memory-based model

When readers encounter a noun, they rapidly associate it with a mental referent. Definite noun phrases ("the dog") often refer to referents that were previously mentioned in the text (anaphoric use), however such phrases are also frequently used to introduce a new referent into the text (non-anaphoric use; e.g., Poesio & Vieira, 1998). A model of preliminary (first-pass) referent assignment must explain: (i) how readers assign referents to non-anaphoric and anaphoric definite nouns, and (ii) how readers implicitly determine the status of each definite noun (anaphoric or non-anaphoric).; The present research addresses this Anaphoric Classification Problem, and assesses the potential for anaphoric nouns to be misinterpreted as new referents on encounter (misclassify-as-new error). Two accounts of referent assignment are considered: special-purpose versus memory-based. Under the special-purpose view, readers process definite nouns with an anaphoric bias, and strategically attempt to assign a discourse referent before resorting to new-referent assignment. Under this view, misclassify-as-new errors may be rare, though preliminary assignment may be prolonged for difficult anaphors (e.g., metaphoric anaphors) and non-anaphors (which violate the anaphoric bias). The present research, however, supports an unbiased view of preliminary assignment, in which general-purpose memory mechanisms bring a referent to mind (Gerrig & O'Brien, 2005).; A unified model, MEMBRAL (Memory-Based Referent RetrievAL), was developed and computationally implemented to address the classification issue. All referents in memory, including specific referent tokens and generic referent types, compete for retrieval. The referent with the highest activation (discourse referent vs. referent type) determines whether or not the noun is interpreted as an anaphor. MEMBRAL clarifies the influence of anaphor word choice (including metaphoric anaphors), and predicts that word choice will primarily affect the accuracy rather than duration of preliminary referent assignment.; Four experiments were conducted to investigate the accuracy and duration of preliminary referent assignment (2 eye-tracked reading studies, 2 cloze studies). Results were highly congruent with those in the MEMBRAL simulation. Thus, this research supports the view that preliminary referent assignment to nouns is memory-based, and confirms that a memory-based account is sufficient, in practice, to account for a high rate of success at preliminary referent retrieval.