Implicit knowledge in people and connectionist networks
Zoltan anq Josef Perner
There has been considerable debate about whether and in what way complex cognitive knowledge can be unconscious or implicit (e.g. Berry and Dienes 1993; Cheesman and Merikle 1984; Dulany et al. 1984; Holender 1986; Jacoby 1991; Reber 1967, 1989; Reingold and Merikle 1988). This chapter will focus on artificial grammar learning as a paradigm example of the acquisition of knowledge that appears to be implicit by some criteria but not others. In a typical artificial grammar learning task, subjects are first asked to memorize strings of letters generated by a finite state grammar (see Fig. 6.1). Subjects are then informed that the strings were actually generated by a complex set of rules, and the subjects are asked to classify new strings as obeying the rules or not. Typically, subjects can classify well above chance, but they find it difficult to say what the rules are ( Reber 1967). Initially, four methodological criteria by which subjects knowledge could be assessed as being implicit will be discussed. Then some ways in which representations can be implicit or explicit are described and related to the preceding criteria. Next a possible connectionist model of artificial grammar learning will be briefly presented. Finally, we will consider if connectionist models illuminate the way in which subjects knowledge can be regarded as implicit.
METHODOLOGICAL CRITERIA OF IMPLICIT KNOWLEDGE
The unconscious or implicit nature of subjects' knowledge can be assessed according to at least four methodological criteria, which are discussed below.