Research On Chinese Zero Pronoun Resolution Based On Word Embedding And LSTM

Chinese zero pronoun resolution is an important task in Natural Language Processing,and it is helpful for computer to understand natural language. The purpose of Chinese zero pronoun resolution is to find the antecedent of the zero pronoun in a sentence. Previous approaches to Chinese zero pronoun resolution mainly use lexical and syntactic information, but the semantic information is ignored.With the development of deep learning technology, word embedding is widely researched, at the same time some deep neural networks such as recurrent neural network(RNN), long short-term memory(LSTM) are widely applied in many NLP tasks and achieve good performance.In this paper, we focus on using word embedding and LSTM model to resolve Chinese zero pronoun. We propose a novel framework based on word embedding and a LSTM based model for Chinese zero reference resolution respectively. Word embedding is an important semantic carrier, so we focus on modeling the task in semantic level by using it. The experiments show the effectiveness of using word embedding and LSTM model for Chinese zero pronoun resolution. This offers a new thinking of resolving Chinese zero pronoun.We construct a framework of linear classification based on word embedding, including keywords extraction strategy, definition of the sample format, training word embedding and building linear binary classification model. In addition, to make full use of lexicon and context information, we propose a model based on bidirectional LSTM, which includes constructing a special neural network for the special task, the special processing of word embedding and various optimization methods. Experiments show that our approaches can effectively use semantic information and have a better performance than the traditional methods.