The goal of the research described here is to develop a system, called KRITON, that can acquire domain-specific problem-solving knowledge by interviewing experts, analyzing protocols, and understanding documents.

KRITON interviews the expert using the repertory grid approach, that is, it presents triples of semantically similar concepts and asks the expert to distinguish two of the concepts from the third. If this proves difficult, the system tries laddering, that is, it asks the expert for important concepts supported by examples, supertypes, and subtypes.

KRITON analyzes transcribed protocols of problem-solving sessions in five steps: using the expert’s pauses to segment the session into chunks, extracting propositions in the form of operations and arguments from the text, evaluating the appropriateness of each proposition, resolving pronominal references, and ordering propositions according to their appearance in the protocol.

KRITON analyzes relevant documents by interacting with an expert. The expert, based on keyword statistics on articles, selects portions of texts for propositional analysis using the same tools used in protocol analysis. Since documents are normally not as problem oriented as protocols, the expert adds goal information to the results of the analysis.

This paper raises some major questions. First and foremost, how does KRITON understand protocols and documents when unrestricted text understanding is one of the great unsolved problems in AI? Second, when KRITON interviews experts using the repertory grid and laddering approaches, how does it avoid sounding like a lunatic hopping from concept to concept? Does KRITON also solve the hard problem of designing conversational human-machine interfaces? Although a prototype has been implemented in InterLisp, no examples of output are given, so it is impossible to see how the system behaves or how it answers the concerns just raised. Either this paper is more promise than practice, or KRITON is a true breakthrough in artificial intelligence applications.