Integration of referential scope limitations into Japanese pronoun resolution

Abstract

We propose a practical approach to the anaphora resolution of Japanese pronouns incorporating knowledge about referential scope limitations extracted from an annotated corpus A machine learning approach (deci sion tree) is utilized for the classification of the coreference relation of a given anaphor and antecedent candidates. The resolution scope of each pronoun is limited according to the relative distance distribution of the training data, resulting in increases in the classification accuracy and analysis speed by causing only a minor decrease in the recall perfor mance.