Building Natural Language Interfaces for Rule-based Expert Systems

In this paper we discuss a semantics for translating natural language statements into facts of an underlying expert system, replacing the more conventional menu interface for gathering data from the user. We describe two issues that must be considered when building such an interface for an expert system. These issues are semantic processing of the user statements and the design of an interpreter for the expert system that efficiently utilizes the facts entered by the user. The semantic approach is based on verb categorization and hierarchical structuring of each category. The parsing algorithm based on selectional restriction is directly encoded into each verb class hierarchy. Next, we describe Director, an interpreter for rule-based expert systems that efficiently utilizes these facts for inferencing. Director uses a combination of forward and backward chaining that gives ful l consideration to each fact entered by the user and enables the system to process input in an efficient and focused manner.