指代消解中代词待消解项识别研究

ON IDENTIFICATION OF PRONOUNS TO BE RESOLVED IN COREFERENCE RESOLUTION

以基于机器学习的指代(Anaphora)消解平台为基础,研究代词待消解项识别问题。挖掘能区分代词是否为待消解项的特征集,总结归纳具有规律的代词待消解项的句法结构,使用机器学习的方法将二者结合生成代词待消解项过滤器并将其加入到代词指代消解平台。在ACE2003基准语料上测试过滤器自身性能及对代词指代消解的贡献。实验表明过滤器具有较高的准确率,能明显地提高代词指代消解系统的性能。

This paper studies the identification of pronouns to be resolved on the basis of a machine learning based coreference resolution platform.A filter of pronouns to be resolved is generated with machine learning method by combining these two： to mine sets of features which are able to discriminate whether the pronouns are the items to be resolved or not,and to summarise and educe syntactic structure of pronouns to be resolved with rules,and they are add onto the pronouns coreference resolution platform.The performance of the filter and the contribution to pronouns coreference resolution are tested with ACE2003 benchmark corpus.Experiment shows that the filter achieves higher precision rate and the performance of pronouns coreference resolution system can be improved outstandingly.