Grammar competition, speaker models and rates of change

A critical reappraisal of the Constant Rate Hypothesis

Abstract

The notion that rates of replacement of an old grammatical op- tion by a new one are identical across linguistic contexts during a period of change (the Constant Rate Hypothesis, CRH) has attracted considerable attention in the historical syntax literature. Here, I argue that any inferences made about change processes using models of the constancy or variability of rates of change must be conducted in a way that balances three considerations: (i) empirical fit, (ii) model complexity and (iii) ontological interpretability of model parameters. Five models involving constant or variable rates of change are examined with respect to three datasets with the help of the Akaike Information Criterion in an effort to explore how model selection can be carried out rigorously. Although this technique balances empirical fit and model complexity in a principled manner, thus offering an improvement over the statistical methods traditionally used in the examination of constancy of rates of change, it cannot weed out models which fail the criterion of ontological interpretability. I argue that such models should be excluded from consideration on a priori grounds. What remains can be termed ‘speaker models’: mechanistic models whose components have interpretations in terms of the representation of knowledge of language and language use. Whether the CRH is such a model or can be derived from one remains an open question.