Allowable variability: a preliminary investigation of word recognition in Navajo

Because speakers do not produce uninflected or 'base' forms, and listeners do not hear them, the shape of the word lexicon in languages with highly productive word formation processes directly addresses the conflict between morphological theories which assume the primacy of word formation processes (Anderson 1992, Bybee and Moder 1983,) and theories of word recognition such as the Cohort theory (Caramaza, Laudana and Romani 1988, Marslen-Wilson 1978) which assume words are stored. How does a relationship between inflected forms, or between inflected forms and their more abstract base, get established? One (common) assumption is that less fluent speakers have less complete grammars and their mistakes reflect their less complete or 'imperfect' knowledge of structure. Since the productive morphology indicates a more complex word processing device and presumably a more complex word lexicon, these errors may reasonably reflect the principles that underlie the organization of the lexical system. In this study, designed to test the feasibility of this strategy, we produced a list of 100 Navajo forms, half of which were 'correct' Navajo words and half were 'incorrect', containing mistakes that less fluent Navajo speakers actually made. The Navajo verbs were categorized into 5 groups, reflecting five types of commonly occurring errors. We found that all speakers accommodated ‘errors’, with differences in the kinds of errors more and less fluent speakers tolerated. The results bear on the issue of the role context, fluency and morphological structure in the recognition of morphologically complex words.