ABSTRACT
The model to be presented predicts recognition performance in a situation where subjects respond to Dutch words consisting of three letters presented at different eccentricities in the visual field.
As basic features of the word stimulus the constituent letters were chosen, being about midway between elementary visual properties and global word-shape. This choice is empirically supported by a strong relation between letter recognition in words and in meaningless strings.
The perception of letters in such a case is taken to be position-specific, as letters in words are subject to interaction by neighboring letters dependent on eccentricity. By means of a probabilistic rule, letter recognition, given a stimulus word, leads to a set of letter strings as viable alternatives. From these strings only the real words are retained by means of a matching procedure for which the final response probabilities are predicted with the Constant Ratio Rule. Thus word frequency effects are not separately incorporated in the model.
All parameters of the model are supplied by earlier results on letter recognition. Though the reliability of these data is not optimal, predictions of the model compare favorably with responses obtained in a word recognition experiment. Tests of the model are described for correct and incorrect responses to words presented at four different eccentric positions.
