Building a Natural Language Interface to Expert Systems

The primary goal of this research is to build a general semantics that will allow the mapping of user statements and questions into facts and goals of the underlying expert system. The semantic interpreter we propose to build will impose a structure on the underlying expert system, while being linguistically based and to some extent transportable. This last criterion requires that it separate domain dependent and independent information. Some of the main features of our semantic approach include linguistically based, hierarchically structured verb categories, a parsing algorithm that is encoded directly into the hierarchies and a mechanism to deal with semantically incomplete input. All of these features are discussed in detail later in the paper. A secondary contribution of this work is to build an inference engine, Director, for expert systems that meets the requirements imposed by a natural language interface. Figure 1-1 shows an overview of the total system as we envision it