A Corpus-Based Analysis of Coreferential Recency Effect in Japanese Discourse for Tracking Dynamic Topic

This paper describes an analysis of the recency effect of terms appearing in a corpus of meeting transcripts in Japanese. The aim of the study is to improve a method for tracking dynamic topics that have been developed for visualizing a long meeting transcript. The recency effect observed from the appearance of terms in discourse context, the coreferential recency effect, is an important factor to deal with dynamic transition of topic. Statistical analysis using a corpus of meeting transcripts quantitatively clarifies the relationship between linguistic features and the coreferential recency effect.