Returning to the obvious: the ubiquity of categorical rules

William Labov

University of Pennsylvania, USA

TP187: Usage-based and rule-based approaches to phonological variation

Recent work in the framework of exemplar theory has demonstrated that speakers can retrieve a great deal of social information from their stored memories of lexical forms. It is argued that categories are indexed by probabilistic reference to regions of exemplar space, populated with remembered words. In this model, the influence of exemplars on production and perception is necessarily dependent on their token frequency, so that word frequency strongly determines phonetic output, and change is linked to lexical diffusion.

Much of this evidence on frequency effects has been drawn from lenition processes like auxiliary contraction, where the amount of reduction is proportional to lexical frequency. When we examine changes in progress that are not linked to lenition, the situation looks quite different. The widespread fronting of /uw/ and /ow/ across North America shows a categorical constraint on fronting before liquids independent of frequency or lexical membership. If we trace vowel systems from the most conservative to the most advanced, we find a sudden shift to systems where all vowels before /l/ and /r/ remain back, and all others are fronted, with a gap of several hundred Hz separating the two distributions. A similar categorical situation is found in the development of the nasal system in tensing of short-a, where all vowels with nasal codas are raised to high front position and all others remain low front.

Proponents of exemplar theory argue that social information is inextricably mixed with phonological and grammatical information in the remembered words. Nevertheless, multivariate analyses regularly report independence of internal and social factors. Such independence seems to hold for some of the most prominent sociolinguistic variables like the T-V pronouns of power and solidarity. For these variables, each token has a clear social impact, unlike the stochastic effects of (ING) and (r). It appears that the information on which form has been used is stored independently of the part of speech used. Support for this hypothesis is found in the recent work of Wallace on thou and you in the 16th to 18th centuries. Across time, the frequency of thou was strongly correlated with social factors, but no evidence of the influence of part of speech or other grammatical categories was found. It would follow that important social information is stored in a form independent of whether speakers used the words thou, thee or thy.

Although we do not want to lose the profit that has been gained from the study of probabilistic distributions, a dispassionate overview of change and variation shows that it rests upon a solid foundation of categorical behavior.

Session: Themed Panel (part 2)
Usage-based and rule based approaches to phonological variation
Friday, April 4, 2008, 15:45-17:15
room: 03