An Exemplar Dynamic Approach to Language Shift

1. Introduction

This squib presents an exploration of how Exemplar Dynamics (Hintzman 1986; Goldinger 1996; Pierrehumbert 2001) can be used to model the increased phonetic variability observed in language shift situations. It is based on a study of laryngealized resonants, as pronounced by three fluent speakers of St'át'imcets, a Northern Interior Salish language of British Columbia.

2. Background: Phonetic Variability

Phonetic variability refers to the variability with which we speak. This variability may be a function of socio-linguistic factors such as dialect, for example the intervocalic /t/ in city is often pronounced as flap [ɾ] in Canadian English but as glottal stop [ʔ] in British English. It may also be a function of linguistic factors. For example in Canadian English, the pronunciation of /t/ is a function of prosodic environment: it is aspirated [t^h] in top, flapped [ɾ] in city and glottalised [t^ʔ ∼ ʔ]¹ in nightlife. In Québecois French, /t/ varies as a function of segmental environment: it is affricated [t^s] preceding high front vowels /i/ and /y/, as in [t^sy] tu 'you sg.' but unaffricated elsewhere, as in [to] tôt 'early'. Finally, a certain amount of variability is intrinsic to speech, and results from the fact that our articulatory mechanisms are not characterized by robotic levels of precision.

Phonetic variability has recently become the focus of much research because of its implications for how sound structure is represented in a language's grammar. Generative models of phonology represent speech sounds lexically as single, abstract, invariant underlying forms. For example, [t^h] [ɾ], and [ʔ] in the English examples above would all be unified in their underlying form: /t/. Phonologically predictable surface variability is achieved in generative models through the application of rules or constraints within the Phonology module (1):

(1) Generative rules deriving surface representations of /t/

1. a.    (as in Canadian English city)

2. b. /t/ → [t^h] / #__v   (as in Canadian English top)

3. c. /t/ → [ʔ] / __ .C   (as in Canadian English nightlife)

Variability that is not predictable phonologically is generally left to the Phonetics module,² as is gradient variability (e.g., a function of frequency effects)³ and variability that is not language-internal (e.g., socio-linguistically motivated). As a result, generative models do not offer a unified account of different kinds of phonetic variability.

As an alternative to traditional generative models of phonology, usage-based models highlight phonetic variability. For example in Pierrehumbert's (1995, 2001, 2002) Exemplar Theory, sounds are represented lexically as sets (or clouds) of examplars, which correspond to variable tokens encountered by a speaker/listener in the everyday use of his or her language (see section 4 for further details). The variability among these tokens may be conditioned by linguistic, socio-linguistic, or other factors; all kinds of variability are treated as equal. As a result, and unlike generative models, usage-based models provide a unified approach to phonetic variability. Furthermore, because phonetic variability is directly encoded in the lexical representation of sounds, no rules or constraints are needed to achieve this variability.

Recently, generative and usage-based approaches have been merged in hybrid approaches to the lexical representation of sound, such as Goldinger's (2007) Complementary Learning Systems (CLS) approach. In such approaches, two kinds of information are encoded: abstract (corresponding to underlying forms within generative models) and episodic (corresponding to exemplars within usage-based models). Such hybrid approaches have great potential in accounting for natural language phenomena; time will tell how well these approaches hold up to empirical findings.

The purpose of this squib is to examine Exemplar Dynamics specifically with reference to language shift situations.⁴ These are situations in which a given language is no longer being used as the primary mode of communication, having been replaced by a more widely spoken language. It is proposed here that, because of the mechanism that underlies speech production in Exemplar Dynamics, it predicts precisely the kind of increased variability observed in language shift situations. Before embarking on a detailed discussion of Exemplar Dynamics and language shift (section 4), a set of data is presented (section 3) illustrating the nature and degree...