In the field of Artificial Intelligence Natural Language Processing is one of the main research problems. We choose Chinese old literature form Chunlian as research object. Chulian auto generation system includes not only the work in NLP, but also some work in Computer Arts. So we hope that through combining human natural language we could do some research on human thinking. This thesis is mainly focused on sentence analysis. Sentence analysis includes two aspects, firstly, analyzing and getting syntactic structure; at the same time, obtaining deeper semantic relationship in sentence. As to Chinese, its grammar system is not as complete as that of English, and the interface between the syntax and the semantics is comparatively indefinite, thus the words in Chinese language emphasize the semantics more than the syntactic function. So the thesis pays more attention on the analysis approaches combining syntactic analysis and semantic analysis. In this thesis, we propose the interactive boot based approach (IBB), which is similar to human beings' pattern of language behavior and has two main points. First, the approach processes the sentence word for word, using the syntactic and semantic information autonomously. Second, the semantic and syntactic modules interactively guide each other's process. In order to make this model similar to human beings, we choose ATN Grammar to design a syntax processor. The semantic processor is based on the concept combination. The processing ability of this approach is very good. It can resolve the ambiguity in some sentences and get deeper semantic relationships. 'Considering the advantages of Neural Networks such as their strong learning ability, nice robustness, and so on, we apply them in sentence analysis. By neural networks, syntactic information and semantic information are combined indefinitely and stored in the networks. BP and SRN neural networks are successfully applied in distinguishing between legal and illegal sentence based on syntax and semantics. In additional, we propose a word choosing approach based Simple Recurrent Network. In the thesis, we also discuss some problems in the integration of symbolic system and connectionist system and introduce the integration methods used in our system. The whole frame of our system is also presented.
修改评论