December 21, 2003

Harmonic rhythm in Turkic

In his post on Dating Indo-European, Bill Poser observed that "lexical replacement is a small part of language change". This naturally raises the question of whether it is possible to use other aspects of language change as "clocks" for assigning dates to (unobserved but reconstructed) stages of language history.

The short answer is "no, at least not yet; and maybe never..." However, there is certainly some interesting recent work on the dynamics of other kinds of language change besides lexical replacement. These include changes in the overall sound system (phonology), in the patterns of word formation (morphology) and in the ways of putting words together into phrases (syntax). None of this work (as far as I know) has the explicit goal of defining a clock that could be applied to dating in linguistic reconstruction. Instead, the goals are simply to understand large-scale long-term historical change better -- a topic that has interested researchers for hundreds of years -- and also to see how well different theories about language structure and use stand up in a historical context.

"Language change has at various times been seen as linear -- that is, languages are progressing or decaying monotonically -- or cyclical -- that is, languages pass through a life cycle of birth, maturity, death and rebirth. However, modeling language change in a formal way has led to a recognition that it is a complex dynamical system (Lass 1997): the interaction of individual speakers leads to emergent, global population characteristics of a language that are neither linear nor cyclical."

That is a quote from a recent paper (Emergent Behavior in Phonological Pattern Change. Artificial Life VIII. MIT Press, 2003) by Mark Dras, David Harrison and Berk Kapicioglu, who have surveyed the history of (loss of) vowel harmony in various Turkic languages over the course of the past millennium or so, and have also explored ways to model the development and loss of vowel harmony as emergent properties of interactions in an artificial speech community. An earlier paper by the same authors gives more details: Agent-based modeling of vowel harmony, Proceedings of NELS 32 (2002).

Reading those papers, and other recent works on both empirical and theoretical aspects of the dynamics of long-term language change, I come to two provisionally negative conclusions about the prospects for finding reliable "clocks" in such changes. First, there are not nearly enough data points for us to be able to say much about the distribution of rates of change for various kinds of linguistic phenomena. Second, language change may turn out to be like climate change (and many other non-linear dynamic systems), in that trends can operate on a remarkably wide range of time scales.

The first problem will be solved as more research is done. The same new research will help determine whether the second problem is a show-stopper or not.

Let me stress that Harrison, Dras and Kapicioglu are not looking for a clock, and I am not criticizing them for not finding one! Rather, I'm asking whether work of this kind might lead to temporal estimates for processes such as harmony loss, in cases where the existence of the process can be inferred (e.g. from a set of related contemporary languages with partial harmony, and a confidently reconstructed ancestor with fuller harmony).

Posted by Mark Liberman at December 21, 2003 12:09 PM