AUTOMATIC CIDNESE TEXT GENERATION BASED ON INFERENCE TREES

This paper describes a method for the generation of a coherent and continuous Chinese text from an inference tree. We argue that it is important to include information of rhetorical relations as part of the knowledge representation scheme in a rule-based expert system shell, in order to facilitate text generation of the inferred relationships. Applying the Rhetorical Structure Theory(RST) defined by Mann and Thompson[5,6], a set of rhetorical relations for Chinese rule-based inferencing is proposed. We observe that the rhetorical structure for an inference tree will be transformed after the inference tree is reasoned(or proved) by an expert system. Rules governing such transformation are derived. We also give an algorithm that can systematically generate multiple sentences of coherent Chinese text on the basis of the transformed rhetorical structure involving conjunctively and disjunctively conjoined constituents in Chinese.