In this dissertation, I investigate the problem of grammatical variation among members of the same lexical category. Adopting the basic theoretical assumptions of Autolexical Grammar (AG), I conceive of lexical categories as groupings based on typical correspondences of information from different domains of grammar. Each lexical category is characterized by a prototypical association of syntactic, semantic, pragmatic, and morphological mini-categories. When items deviate from these default correspondences, defective grammatical behavior and/or grammatical behavior characteristic of more than one lexical category often results. Thus, given a multi-modular conception of category structure, a good deal of variation within lexical categories can be explained in terms of ‘mismatch' among the mini-categories of different modules.
Based on two case studies of English, I propose two major sources of grammatical variation within lexical categories: lexical category mismatch and phrasal category mismatch. The case of quantificational nouns illustrates the former while the case of predicate nominals illustrates the latter. In addition, I argue that some phenomena, such as stative verbs, generic nouns, and quantificational adjectives do not warrant a cross-modular mismatch analysis but still deviate from the default correspondences in ways that can be specified within a theory of grammar. Such phenomena are accounted for by allowing for variation within the major modules of grammar. In keeping with the spirit of AG, this variation is accommodated by positing subcategories within each module and default correspondences among the subcategories. Thus, subcategory mismatches within a module may account for some kinds of non-prototypical grammatical behavior. Finally, I identify four ways in which grammars cope with mismatch and other incongruity: by mixing of categorial features, by neutralization of categorial features, by override, and by constructional compensation for missing information.