Abstract: Acquiring the meaning of natural language is a bottleneck to make deeper use of natural language processing (NLP). There are two main measures on analyzing meaning of natural language at conception-relation level: one is the method of extracting characteristic vectors based on statistics, and the other one is method of computing semantic similarities according to semantic dictionary like WordNet or HowNet. Both of the two methods have weakness when putting them to applications. The previous is only applicable to analyze the meaning of those materials with big granularitics such as paragraphs, documents or multi-documents, but is not fit for the applications at the level of sentences or words. The latter can deal all sorts of relations between conceptions easily, but when coming to complicated modified relations between conceptions and events, conceptions and conceptions or events and events, the semantic dictionary and computing method shall be extended. This paper presented a new method to structure semantic knowledge base(SKB) according to the modifying relation of real context; algorithm of computing unknown relations on the knowledge base was presented; we pointed out the way to design the rules of constructing natural language sentences under modifying relation and present the algorithm; in the end we made experiment on the platform developed in the light of the theory mentioned above and the result shows the theory is feasible.

Key words: NLP, Syntax, Semantic, Modifying relations, Knowledge base