ABSTRACT
‘Mixed errors’ are defined as errors that are similar to the correct response on more than one dimension and whose probability of occurrence is greater than a simple stages model would predict. Two examples of them are given: visual— semantic errors in word reading and semantic-phonological errors in spontaneous speech. Alternative models, based on activation and comparator ideas, that maintain aspects of the stages approach and yet predict a high rate of mixed errors are compared in terms of how well they can account for segmentation errors induced by visual masking. It is shown that visual similarity helps to induce segmentation errors, and that semantic cueing can only induce them if the cue is consciously perceived. The results are interpreted within the framework of the activation approach.
