Information Seeking for Object Identification.

In two experiments subjects learned to assign each of 100 cards to one of six alternative classes on the basis of four binary cue dimensions. Then they estimated the proportion of cards in each class and the relative frequency of each cue value. The extreme accuracy of the estimates showed that the frequency structure of the deck had been learned. Then, individual cards were selected and the subjects attempted to identify the card's classes by asking as few questions as possible about its cues. Optimal information seeking was prescribed by the Shannon-Fano encoding theory which takes the frequentistic structure of the deck in account. Individual subject's information seeking strategy was represented by a decision tree. The optimality of each tree was evaluated in terms of the Shannon-Fano theory. In the first experiment, subjects were merely requested to ask as few questions as possible. In the second experiment, cost for information and penalty for wrong identification were imposed. In both experiments all subjects used highly efficient trees, most of which corresponded to the optimal decision rule. Cost and penalty had only a minor effect. [ABSTRACT FROM AUTHOR]