Conceptual modelling for knowledge acquisition

The task of a knowledge engineer in encoding and formalizing an expert's knowledge for use in an expert system amounts to interactive conceptual modelling; thus the basic tenet of our work is that knowledge acquisition is a communication process with all the usual problems of natural language understanding (NLU), plus the additional problem that usually a domain expert finds it very difficult to articulate his or her expertise at all. Thus, knowledge acquisition is a human activity, but one that can make much use of the tools, techniques, and approaches that we are developing in NLU. We look upon language use as a knowledge-based process that depends on the mental models of the user as well as on the idiosyncratic and idiolectic use of language, both in generation and in understanding.

We have presented an algorithmic method for conceptual analysis of natural language text, and explicated the distinction between concepts, referents, and words in knowledge acquisition. As a result of our research, conceptual analysis was generalized to a more comprehensive technique that we have called sortal analysis. In this, we use the agent-centred meaning triangle to distinguish several ontological categories. We interpret NLU and knowledge acquisition from text as concept-cluster attachment performed by an agent. We model the agent with a three-level architecture consisting of verbal, conceptual, and sub-conceptual components. We are embodying this theory in software tools for knowledge acquisition. An important new re-interpretation of our work on these tools is to construe them as knowledge acquisition modules within a knowledge-base management system (KBMS). Their output is knowledge bases managed by the KBMS.

References:

Return to Research by Graeme Hirst and students