Adapting a State-of-the- art Anaphora Resolution System for Resource-poor Language

In this paper we present our work on adapting a state-of-the-art anaphora resolution system for a resource poor language, namely Bengali. Performance of any anaphoric resolver greatly depends on the quality of a high accurate mention detector. We develop a number of models for mention detection based on heuristics and machine learning. Our experiments show that, a language-dependent system can attain reasonably good performance when re-trained on a new language with a proper subset of features. The system yields the MUC recall, precision and F-measure values of 57.80%, 79.00% and 66.70%, respectively. Our experiments with other available scorers show the F-measure values of 59.47%, 49.83%, 31.81% and 70.82% for BCUB, CEAFM, CEAFE and BLANC, respectively