Automatic identification of pronominal Anaphora in Turkish texts

Abstract

Anaphora identification is an important problem especially for its impact on anaphora and coreference resolution systems. In this paper, a system that automatically identifies anaphoric pronouns in Turkish is presented. The proposed system takes a decision tree learning approach, that of Quinlan's C 4.5, where a corpus examination is carried out to determine linguistic features specific to Turkish which are to be used by the decision tree learner. The proposed system is significant especially for its ease of incorporation into any anaphora resolution system for Turkish. The system is evaluated on two different Turkish text samples and its performance on these samples is close to that of human identification.