A model of the lexical representation of polymorphemic words is presented. In this model, such words are stored in stem form for the purposes of matching with an incoming letter string or utterance. Once a match is made, information about the complete word becomes available. Experimental support for this position is described in a review of past research into morphological decomposition. In addition, though, two new studies are presented which produce results that are difficult for the model to handle.

This paper presents a summary of experimental studies that I have carried out which suggest that an affixed word is represented in the lexicon in terms of its stem morpheme (see Taft 1985 for a more detailed description of this research). It will conclude, however, with a couple of recent studies that possibly herald a weakening of the very strong model that I have posited in relation to morphological representation in the lexicon. What I mean by'representation in the lexicon'must be clarified. The lexicon can be subdivided into (at least) two components: an input system and a central system (see Forster 1976; Morton 1980). The input system comprises the representations of words that are used to find a match with the incoming stimulus. There must in fact be two input systems, one for when the stimulus is visually presented and one for when it is spoken, since the matching is based on the physical properties of the stimulus. Having achieved a satisfactory match between a presented word and a representation in the input system, information about that word is now made accessible from the modality-free central system. This is information about the pronunciation and spelling of the word, as well as the meaning and syntactic characteristics of the word. According to the above description of the lexicon, therefore, there are several different representations that may be called'lexical representations'. First, there is the representation in the input system used for