摘要： 在自然语言生成系统的内容规划阶段，主要的手段是基于Schema的规划方法和修辞结构理论RST方法，Schema方法只适用于模式固定、细节略有变化的文本的生成，而RST方法则比较灵活，但是它的数据结构、规则库比较难建立。把这两种方法集成起来，并对Schema方法进行改造，使之能与RST有机地结合起来。在规划文本时首先用Schema方法规划全局的组织，然后用RST方法控制局部话语的连贯。这种结构可弥补采用单一方法的不足，具有双方的优点。

关键词: 自然语言生成, Schema, 修辞谓词, RST, 扩充转移网络

Abstract: In the content planning phase of natural language generation,the main means are Schema and RST.Schema adapts to the generation of text whose pattern is fixed and detail is a little variable.RST is much flexible,but its data structure and rules base are complex . An integrated approach combined the Schema and RST is presented . In this generation architecture, the Schema is a variant of McKeown`s Schema, it determines the global organization of the text, while RST controls the local coherence of the discourse. This structure covers the advantages offered by Schema and RST.

中图分类号: 

• TP182