Abstract
This study investigated the role of word class and gender during lexical access in language production. It was predicted that word class would constrain lexical access because it acts as the interface between the syntax and the lexicon. Gender, in contrast, should not constrain lexical access because it is a linguistic category that does not correlate with any semantic or syntactic information. These predictions were tested against contextual and noncontextual word substitution errors in a corpus of German slips of the tongue, as well as against verbal paraphasias produced by a German-speaking aphasic patient. The results indicated that in all three subsets, both word class and gender influenced the search through the mental lexicon to a reliable degree, with word class making a greater impact than gender. The model that best captured the empirical effects centered around the distinction between prelexical and postlexical features, assigning word class to the former and gender to the latter group. This distinction could be most naturally implemented in a parallel-interactive processing network. The creation of nodes and connections in this type of model was shown not only to respect functional principles, but also to occur on purely structural grounds. On the assumption that the information would be transmitted more or less reliably from one node to another, the aphasiological and speech error data could be readily accommodated within the same psycholinguistic model.