Discovering Pronoun Categories using Discourse Information

Interpretation of a pronoun is driven by properties of syntactic distribution. Consequently, acquiring the meaning and the distribution are intertwined. In order to learn that a pronoun is reflexive, learners need to know which entity the pronoun refers to in a sentence, but in order to infer its referent they need to know that the pronoun is reflexive. This study examines whether discourse information is the information source that the learner might use to acquire grammatical categories of pronouns. Experimental results demonstrate that adults can use discourse information to accurately guess the referents of pronouns. Simulations show that a Bayesian model using guesses from the experiment as an estimate of the discourse information successfully categorizes English pronouns into categories corresponding to reflexives and non-reflexives. Together, these results suggest that knowing which entities are likely to be referred to in the discourse can help learners acquire grammatical categories of pronouns.