Modeling language change: a population and learning perspective

Research in generative grammar generally rests on the idealization of language learning as instantaneous and perfectly accurate. For many purposes, this idealization is useful but when we try to understand the causes and course of grammatical change, it can become misleading. If language learning is perfect, then language change and language acquisition cannot be intimately connected. Indeed, generative theorists have often proposed that language change is driven by drifts in adult usage, which have an effect on grammar only at the end of a change, when the frequency of primary data for an original grammatical analysis becomes too low to support learning, so that child learners postulate a new grammar. Evidence from our historical research, however, indicates that grammar changes arise, not at the end of a period of frequency change but at the beginning; that is, changes begin with a change in grammar, after which usage frequencies shift gradually as the new grammar spreads through the language community. This result raises the question of the cause of the initial grammar change, which we propose is due to errors in learning, not only by adults acquiring a second language (as might be expected in cases of grammatical change due to language contact), but also by children learning their first language in cases of endogenous language change.