Anaphora Ambiguity Detection Method Based on Cross-domain Pronoun Substitution

Pronoun anaphora ambiguity is very common in natural language descriptions, especially in specilized fileds such as computing, medicine and aerospace. When multiple antecedents appear before a pronoun word, readers with different background knowledge often have completely different understandings on a same word. In order to reduce such misunderstandings caused by ambiguity in the process of document propogation, we usually use manual methods to check the ambiguity of reference, which however cannot meet the increasing needs of detection with the development of various disciplines. In this paper, we propose a method to intelligently detect sentences with anaphora ambiguity. First of all, we identify criteria for ambiguous sentences and then use word embeddings to further detect ambiguity. Specifically, we propose a pronoun substitution strategy based on coreference resolution, and combine this strategy with word embedding techniques to generate a cross-domain anaphora ambiguity detection method. Finally, we carry out experiments on aerospace documents, which verify the effectiveness of our proposed method in anaphora ambiguity detection.