Formalizing the Social Expert's Knowledge

Certain types of data, for example, texts and free-form interviews, tend to lend themselves to being analyzed in a qualitative fashion because they are hard to interpret and hence code. The ability to analyze and code such data often requires one to have an in-depth knowledge of the sociocultural environment related to this protocol, to be, in effect, a social expert. Social expertise involves both knowing more than the novice about the sociocultural environment, and knowing the level of consensus among members of the sociocultural environment to each piece of information. Knowing the level of consensus allows the social expert to make explicit the information that is implicit in the verbal protocol. It is often prohibitively expensive, and sometimes not even possible, to provide novice coders with the extensive training necessary to interpret and code such verbal data as would an expert on that sociocultural environment. By integrating theories of the nature of knowledge, social expertise, and coding it is possible to develop a cognitive foundation on which an expert system for coding verbal protocols can be built. In this article a two-stage computer-assisted procedure for coding verbal data is presented. This procedure minimizes the amount of training needed by novices by utilizing the expert's knowledge of social knowledge to improve the novices's coding of the verbal data. In stage 1 the novice uses CODEF, a computer-assisted procedure for coding verbal data on a particular topic as a knowledge base. In stage 2 the knowledge base is processed by an expert system, SKI. This expert system has as its knowledge base social knowledge and as its inference engine a cognitive model of the way the social expert uses this knowledge. SKI uses the expert's knowledge of social knowledge to explicate implicit information in the protocol and thereby diagnose and correct errors of explication in the novices coding of the protocol. By employing SKI it is possible to increase the level of reliability and replicability in the coded data without having the expert code each verbal protocol and without spending extensive time training the novice. This research is indicative of the way expert systems and the findings in artificial intelligence can be used in the social sciences.