Language as a complex system: the case of phonetic variability

Abstract Modern linguistic theories try to give an exhaustive explanation of how language works. In this perspective, each linguistic domain, such as phonetics, phonology, syntax, pragmatics, etc., is described by means of a set of rules or properties (in other words, a grammar). However, it is a long time linguists have observed that it is not possible to give a precise description of a real life utterance within a unique domain. We illustrate this problem with the case of phonetic variability and show how different domains interact. We propose then a two-level architecture in which domain interaction is implemented by means of constraints. 1. Introduction Recent evolution of linguistics (see for example [1] or [2]) leads to take into consideration the fact that language constitutes a complex system. First, and this pointis of particular importance in cognitive linguistics, language does not constitute an autonomous cognitive process. Many recent neurolinguistics experiments seem to confirm this aspect, in particular by means of imaging techniques. In the same direction, nd even more interestingly, modern linguisticsa stipulates now that linguistic domains such as phonetic, phonology, syntax, semantics, etc. are not autonomous modules and cannot be described separately. More precisely, each domain can have its own structuration, but linguistics has to describe their interaction as well. Our position is situated in this perspective: linguistic information is disseminated into the different domains. Each one contains a more or less important part of information vehiculated in the message or the utterance. But at an upper level, domain interaction also produces information: emphasis for example can result from a conjunction of prosodic and syntactic information. In the same way, morphologic, semantic and pragmatic domains often interact in the interpretation of some phenomenon such as politeness. In this perspective, we think that information has to be considered as coming both from each linguistic domain but also from domain interaction which constitutes in itself a meta-level. This approach relies then on a collaborative conception of domains relations more than a dependency hierarchization between them. One consequence is that in this approach, constraints controlling the construction of a given domain do not depend directly from another domain, but from this meta-level domain interaction. We propose in this paper to describe such a conception of linguistic information flow. In the first section, we illustrate this organization with the example of phonetic variability. We consider such phenomenon as a typical example of such an interaction: we will show how the quantity of information produced by domain interaction has direct consequence on phoneme production. In the second section, we propose a general framework explaining this two-level architecture.