Physics (and other science) envy in linguistics

09 Feb 2017 | last updated on 10 Jul 2017 | all notes

One thing that struck me as I was compiling the literature review chapter for my thesis was that, in comparison to other social sciences (as far as I can tell, it’s hard to get an objective grasp on such things when one is so deeply embedded in a scientific field), a vast amount of terminology in linguistics is literally borrowed from the natural sciences (particularly physics).

This is not to say that linguistics doesn’t have any terminology of its own – it absolutely does, but interestingly this ‘native’ linguistic vocabulary is largely limited to the domain of the descriptive (e.g. concepts such as case, ergativity, aspect, declension) which, as it happens, are also situated in synchrony.

As soon as one moves over to the more historical, diachronic aspects of linguistics, metaphors (or rather analogies) prevail: in contemporary writing one sees talk of pressures and forces that cause language change and shape languages, which might or might not encompass the emergence of all sorts of things (presumably in complex ways). Why do I frame this as physics envy, rather than as normal and productive adaptation of concepts between fields? Because, although linguists have been doing this for over a century, so far no terminology seems to have stuck (or at least none seems to have led to a solid and productive integration of those concepts in linguistics).

Sciencing (historical) linguistics

The development of the ‘sciencing’ of linguistics is interesting in itself, particularly because it wasn’t quite there yet at the dawn of modern Western linguistics and philology, which was way more historical in nature. In Grimm’s Deutsche Grammatik (1819) he speaks of ‘Regel’ (rule) and ‘Gesetz’ (law), but by association the meaning of the former (clearly in the sense of grammatical rules) is intended to have a descriptive or prescriptive (cf. human made ‘laws’ such as customs and legal laws) rather than explanatory (cf. laws of nature) character.

The first signs of a ‘sciencing’ of 19th century historical linguistics can be found in August Schleicher’s analogy of languages as ‘natural organisms’ (1850) which, by extension, were thought to obey ‘natural’ laws (Ross and Durie 1996, 14). Not everybody bought into this metaphorical rhetoric though, with Freeman (1909) claiming that the study of language “is strictly an historical study” (p.237-238), and Sapir (1924) ceding that “Linguistics has neither the sweep nor the instrumental power of mathematics”.

Analogies with physics really kicked off in the Atomic Age. One of the terminologically most stunning examples is Martinet (1952) taking the concept of chain reactions from physics and applying it to vowel shifts, introducing the distinction between ‘push-chains’ and ‘drag-chains’ with it. In their French original (1955), the two are actually referred to as ‘chaîne de traction’ and ‘chaîne de propulsion’, leaving no doubt about the framework of physical determinism this novel account of sound changes was aspiring to. (Doing a Google search of the French terms yields results about the mechanics and electrics of motorisation, whereas their more widespread English counterparts of ‘push chain’ and ‘pull chain’ reveal very mundane sorts of chains soon followed by articles on sound change, showing that the strongly analogical origin of the terms is certainly more obscured in English.) The links of a chain and a set of vowels do not have anything in common of course, but the analogy is fully consistent with Roman Jakobson’s “quest” (1965) to “physicalize” linguistic (particularly phonological) systems that Joseph (2000) (p.182) argues is actually mere rhetoric rather than substance.

Very much in the spirit of the time, the final blow that kicked linguistics into its subtle yet well-established idolatry of physics might have been the concept of explanatory adequacy introduced in Chomsky’s Aspects. Rather than just taking a metaphor from physics to describe some linguistic concept or process, he based his yardstick for ‘good’ linguistics work on criteria of scientificness derived from natural sciences, specifically those pertaining to deterministic prediction. (In another interesting physical twist, he also framed predictions on the macro-level as being based on underlying particles or components – UG – that are difficult if not impossible to observe directly.)

Chomsky’s narrow conception of language did not just completely remove it from the speech community, it also erased from much modern linguistic thinking the fact that languages are historical entities that should primarily be understood as such. Distancing linguistics from the cultural relativism associated with the arts and humanities, much research in the second half of the 20th century strived instead towards some cognitive universalism that aligned it with the scientific rigor of cognitive science, which was taking shape as a field in the same period. Even the staunchest opponents of Chomsky’s work are prone to adopting overly deterministic evaluation criteria, so that this way of thinking linguistics is not limited to generativist work at all: Greenberg’s (1963) quest for language universals too can be seen as an attempt to establish some degree of certainty within linguistic theorising, only that his are grounded in directly observable linguistic reality.

This borrowing of scientific terms and concepts is not limited to physics, with molecular and evolutionary biology easily securing 2nd place. While on one hand the discovery and breaking of the ‘genetic code’ has continuously given rise to analogies between genetics and linguistics (see for example the references in Shanon 1978), the quantitative advances in population genetics have seen many linguists adopt concepts such as selective ‘fitness’ (and with them the highly idealised notion of a largely static fitness landscape) to ‘explain’ language changes. But even on a purely descriptive level linguists are inclined to adopt mathematical concepts and apply them to superficially similar-looking linguistic data without much scrutiny: Altmann’s A law of change in language (1983) managed to firmly establish the logistic curve as the mother of language change trajectories by showing its good fit to all of two data sets (the second of which only covers growth up to the 26% mark). With the advent of complex systems science, physics and biology now even join forces, for example in Lieberman et al. (2007) who do not just find “exponential decay rates” in the history of English past tense regularisation, but even “evolutionary decay”. While the term might sound compelling, the fact that it is no more specific or enlightening than the characterisation of unimodal distributions as the frown of the Cheshire cat on the following page reveal it to be nothing more than a fanciful metaphor.

Ditching the (terminological) cargo cult

What’s my point? Linguists have not just been coining metaphors for language change processes based on terminology from the natural sciences, but have been superficially borrowing with it the supposed explanatory power of these terms. There is absolutely nothing wrong with transferring concepts between fields, as long as the productivity or usefulness of those concepts in the target field is validated on empirical grounds. This does not appear to be the case in linguistics where much of the imported terminology does not perform a clear predictive or falsifiable function, yet the terms are still perceived as valid and insightful due to their relevance in another (prestigious) field.

This is not just a vanity problem. I think the fact that borrowing or even one-off metaphorical extension of concepts from the ‘hard sciences’ is both easy and highly regarded comes at the expense of linguistics-specific theory building that would actually be of lasting relevance to the field which, across its many subdomains, currently shares no common framework (let alone terminology) to think about language change. (Note how this is not actually seen as a problem by many linguists who are fond of the idea of a highly modular organisation of language, particularly among those focussed on language synchronically.) The red herring of determinism that physics-derived concepts (as well as simple readings of ‘evolutionary’ accounts) are imbued with obscure the important fact that languages are historical entities (Blute 1997) for which not just the concepts borrowed from physics, but even the scientific criteria used to evaluate ‘natural laws’ (as understood naively by linguists), are wholly inadequate.

These are just some of my own thoughts on recurrent encounters with terminology from physics in linguistics; much more extensive things have been written on this, especially as it pertains to the period of 1930-1990, in Koerner (1978), Koerner and Asher (1995), Koerner (2002) (chapter 5 onwards) and in particular Koerner (1995) (chapter 3). For insightful remarks about the historic status of language change today see Durie & Ross (1996, ch. 1-2) and particularly Holm (2007).

References

Comments