Anaphora Resolution of Demonstrative Noun Phrases in Medline Abstracts

This paper reports our investigation of machine learning methods applied to anaphora resolution for Biology texts. Our primary concern is the investigation of features and their combinations for effective anaphora resolution. In this paper, we focus on the resolution of demonstrative anaphoric noun phrases. We propose several novel features that we call highlighting features and consider their utility particularly for processing of abstracts. The system using the highlighting features achieved 78% accuracy on our test corpus for demonstrative anaphora. Use of the highlighting features reduces errors by 10% compared to the system without using those features.