Pronouns Without Explicit Antecedents: How do We Know When a Pronoun is Referential?

Pronouns without explicit noun phrase antecedents pose a problem for any theory of reference resolution. We report here on an empirical study of such pronouns in the Santa Barbara Corpus of Spoken American English, a corpus of spontaneous, casual conversation. Analysis of 2,046 third person personal pronouns in fourteen transcripts indicates that 330 (or 16.1%) lack NP antecedents. These pronouns fall into a variety of subtypes. 88 refer to entities that are inferable from an activated frame or script, or are otherwise easily accommodated. In 110 cases, it could refer to a fact, proposition event, activity, situation, or reason which has been evoked by a previous non-NP. 92 cases of it were classified as pleonastic. In this paper we focus on some problematic subclasses of pronouns which could be analyzed as either referring to entities of various degrees of abstractness that were introduced by or implied in previous discourse, or as non-referential, including pleonastic. Such cases include possible truncated cleft pronouns, possible truncated extraposition pronouns, and certain non-specific uses of they.