An expert system acquires and generalises experiences

For classical expert systems, knowledge has to be formulated explicitly, while our system, TXPS (Trainable eXPert System) acquires knowledge by experience, structuring sequences of facts into hierarchical schemas with constant or variable terminal and non-terminal facts. The inferencing mechanism compares input facts with those in the stored schemas and continues the most similar schema by producing its subsequent facts as output. TXPS generalises schemas by searching for analogies between the values of variable facts in different schemas. After words and clauses have been trained as facts of schemas in the context of some natural language sentences, TXPS can apply this syntactical knowledge to a variety of sentences which have not been trained. As an intelligent agent, TXPS detects errors made by commercial speech recognition systems and translates their outputs into a foreign language, into data for communication with computers and databases, or into signals for the control of technical systems.