Usage-based and rule-based approaches to phonological variation

Frans Hinskens, Ben Hermans, Marc van Oostendorp

Meertens Instituut, The Netherlands

Themed-panel

The difference between 'usage-based' and 'rule-based' approaches is currently a topic of debate in phonology (as well as elsewhere in linguistics). Roughly, usage-based models assume that language users store detailed phonetic information about the words of their language each time they hear them. Rule-based models (i.e. several generations of models of generative phonology, including Optimality Theory) on the other hand assume that language users base themselves on phonological rules which are to some extent abstract, always categorical, and generalize over many cases. Usage-based models have as their advantage that they seem better in capturing regularities with respect to, e.g. frequency and other 'gradient' phenomena including phonetic gradience, which is sometimes sociostylistically significant. Rule-based models, on the other hand, have as their primary goal to explain absolute and exceptionless regularities.

Usage-based models as well as the closely related exemplar-based approaches and cognitive grammar are inspired by connectionism, a school of thought in cognitive science which attempts to explain mental and/or behavioral phenomena as emergent processes of a network of mutually connected units, in this case of the brain. The notion of emergent process refers to the emergence and subsequent development of coherent patterns, structures and/or properties during the process of self-organisation of a complex system. In this approach regularities of several types as well as processes of language change are usually accounted for along quantitative lines on the basis of distributional and usage frequencies (or type and token frequencies).

The debate concerning usage-based and rule-based approaches raises many questions which are of interest to sociolinguists. For one thing, although a lot of older sociolinguistic work implicitly or explicitly embraced a rule-based view of phonology (witness e.g. the original meaning of the concept of the variable rule), recent times have witnessed interest in exemplar-based models for sociophonetics as well. Since both models thus have been implemented in the study of language variation, the time has arrived to discuss the relevant merits and disadvantages of each in order to deepen our insight into sociolinguistic phenomena. For another thing, one could perhaps suppose that the difference between 'rule-based' and 'usage-based' phenomena reflects to some extent the classical distinction between phonology and phonetics, or between lexical and postlexical phonology. Such distinctions play a role in a variety of sociolinguistic work; e.g. Labov's proposal that children acquire the phonology from their parents in the first years of their life, while later continuously fine-tuning the phonetics based on the speech of their peers. An important question, then, is whether distinctions of this type make sense from the perspective of usage-based grammar. Are they still relevant, and is it the case, for instance, that 'usage-based effects' are typical in certain domains of the grammar, but not in others? Or are they irrelevant, as some radicalist proponents of usage-based grammar should think, believing as they do that there is no modular separation whatsoever. If, however, distinctions of this type turn out to make sense after all, then we must ask how we can theoretically model them, and how we could test the resulting models empirically.

Yet another important aspect meriting discussion is that both rule-based models and exemplar-based models seem to face specific difficulties when encountering language variation. Rule-based models have been typically associated with ideas of linguistic universalism and idealisation and abstraction over individual data; thus, they do not always seem to fit well with the rough facts of life that sociolinguists are used to deal with. Furthermore, rule-based analyses typically deal with categorical distinctions between sounds, and seem less well-adapted to deal with gradient and fine-grained distinctions. On the other hand, as far as we are aware there is as yet no elaborate proposal of how to deal with the multiple dimensions of sociolinguistic reality within the space of 'exemplar clouds', i.e. words with similar phonological forms: given the large number of relevant variables, it is not clear whether these clouds can ever become dense enough to allow speakers to make any generalisation at all. Furthermore, exemplar based models seem to focus much more explicitly on language as a psycholinguistic reality, raising the question as to how this relates to the reality of language as a social phenomenon.

The panel contributions discuss theoretical, conceptual and methodological advances in the area, as well as relevant tests to break new ground regarding this topic, focussing on variation in the area of socially and/or geographically significant phonological and phonetic variation.

Relevance to the conference theme:

In the realm of variation of linguistic sounds, 'micro' and 'macro' may be seen as referring to small, gradient and fine-grained ('phonetic') vs. large, categorical ('phonological') differences or continuity versus discreteness. Both of these seem to play a role in sociolinguistic reality, and both of these can be argued to be needed. This panel is intended to instigate discussion as to what would be the best theoretical model incorporating all empirical results from both schools.