A Ranking Approach to Persian Pronoun Resolution
Nafiseh Sadat Moosavi, Gholamreza Ghassem-Sani · 2009
Abstract. Coreference resolution is an essential step toward understanding dis-courses, and it is needed by many NLP tasks such as machine translation, ques-tion answering, and summarization. Pronoun resolution is a major and challeng-ing subpart of coreference resolution, in which only the resolution of pronouns is considered. Classification approaches have been widely used for corefer-ence/pronoun resolution, but it has been shown that ranking approaches outper-form classification approaches in a variety of fields such as English pronoun resolution (Denis and Baldridge, 2007), question answering (Ravichandran, 2003), and tagging/parsing (Collins and Duffy, 2002; Charniak and Johnson, 2005). The strength of ranking is in its ability to consider all candidates at once and selecting the best one based on the model, while existing classification methods consider at most two candidate responses at a time. Persian and its va-rieties are spoken by more than 71 million people, and it has some characteristic that make parsing and other related processing of Persian more difficult than those of English. In this paper, we have evaluated maximum entropy ranker on Persian pronoun resolution and compared the results with that of four base clas-sifiers.