Research of Coreference Resolution with Semantic Information

With the rapid development of Internet and computer technology, various kinds of information have been increasing explosively and the demand for precisely located information gives a strong impetus to the NLP research. As a critical and hot research topic in NLP, coreference resolution plays an important role in many NLP applications, such as text summarization, machine translation and multi-language information processing while it also much depends on the advance of various NLP techniques, such as part-of-speech tagging, named entity recognition and syntactic parsing.While previous research focuses on application of lexical and syntactic information in coreference resolution, this paper systematically explores various kinds of semantic information and their application in coreference resolution.First, a noun phrase coreference resolution system is built using Support Vector Machine (SVM), which improves over Soon (2001). The experimental results in MUC-6 and ACE2003 show that our baseline system outperforms comparable systems.Then, various kinds of semantic information, such as semantic class information, semantic role information and semantic similarity, are systematically studied and incorporated into the baseline system. Evaluation on the ACE2003_NWIRE Corpus shows that semantic information can largely improve the performance by about 4 in F-measure and our system achieves 58.8 in F-measures after adding various kinds of semantic information. It also shows that the three kinds of semantic information are quite complementary.Finally, this paper also evaluates above semantic information on other corpora to justify their robustness.