Dependency Parsing Of Spoken Chinese Based On Graph-based Model

In spoken dialogues,it is important to understand speaker's intent in order to better interact with a user.Researches have shown that the dependency parsing of spoken Chinese is helpful for Spoken Language Understanding.However,existing researches have mainly focused on western spoken languages,Japanese and so on.Although there are a lot many dependency parsing researches of Chinese,these researches pay more attention to the parsing of written Chinese.Little research has been done for spoken Chinese in terms of dependency parsing.Thus,this paper pays more attention to the dependency parsing of spoken Chinese.First,there are few public spoken Chinese corpora with syntactic annotation.Thus,we build a new spoken Chinese corpus named D-ESCSC.D-ESCSC is built by adding new dependency relations special to spoken Chinese based on a written Chinese annotation scheme.Second,experimental results show that dependency parser on written text is not suitable for spoken Chinese.It is necessary to built a new dependency parser for spoken Chinese.Graph-based model is a model that often used in dependency parsing.The key is to design new features for spoken Chinese dependency parser.First,a thorough analysis of spoken Chinese has been done.Six typical characteristics of spoken Chinese are found,e.g.translocation,repetition,duplication and omission.Then,a new atom feature related to punctuation and three feature templates are proposed to improve the graph-based dependency parser for spoken Chinese.Experimental results on spoken Chinese corpus show that the atom feature and three templates really work and the new parser outperforms the baseline parser.To our best knowledge,it is the first work to report dependency parsing results of spoken Chinese.