A Tri-Level Knowledge Representation Model for NLP

In this paper we propose a tri-level knowledge representation model based on the view that the process of language expression is the process of changing the form of knowledge from its natural form existed in human´s brain to its language form. In such a process, there exists a middle-form of knowledge that represents the key information needed as the base of the language form. Such key information can be recognized or retrieved from the natural form or form the language in NLG or NLU respectively. They are organized in a form known as the scene structure. In computer implementation, the natural form of knowledge may be simulated by various domain-specified KRs that offer special computing ability for domain-related knowledge. The scene structure of knowledge acts as a bridge or an interface between domain KRs and the natural language with regard that it can be dual-directively mapped with both of them