#they're described as looking morphologically similar in the book | Explore Tumblr Posts and Blogs

ukfrislandembassy · 7 months

Text

I think linguistics has a huge methodological problem, in that while there exist some formalised methodologies for conducting investigations, it's clear that you don't have to pursue any of them to actually get ahead academically because linguists aren't actually as scientifically rigorous as they would like to think.

This has knock-on effects for more general experimental design, because it is extremely common to find studies which purport to prove/disprove some hypothesis, but when you actually look at the methodology they use you find that it doesn't actually provide evidence for the point they're attempting to make.

Of course the most common place to find this kind of stuff is in the Generativist literature, which is awash with claims the phenomena they describe is only explainable through the assumption of Universal Grammar/a Language Organ/whatever, when in fact non-innatist models work just as well. But this isn't the only part of the field that is susceptible to this.

So for instance there was a paper put out in Science recently that claims to have found evidence to counteract the idea that social structure can have an impact on the structure of the language (most notably explicated by Peter Trudgill in his 2011 book Sociolinguistic Typology, give it a read if you can find it). They take demographic data from Ethnologue and compares this with a couple of metrics which supposedly quantify linguistic complexity.

Leaving aside the (sometimes glaring) issues with the way the metrics are defined (they perhaps could have done with reading Dahl 2004 The Growth and Maintenance of Linguistic Complexity), we can already spot a major problem with using these correlations to make claims regarding the Sociolinguistic Typology hypothesis: the phenomena in question are explicitly diachronic in nature, but the data being used to assess them is solely synchronic.

What do I mean by this? It's been suggested that English morphology is less complex relative to e.g. German because of contact with the large numbers of Old Norse speakers that settled in the Danelaw. The trouble is, of course, that all those Norse speakers later shifted to English, so this large community of L2 learners won't show up in the data, despite the effects still being apparent after over 1000 years. Similar points can be made about most of the languages where these kinds of effects have been proposed: it was precisely the process of these languages spreading rapidly and acquiring a large speakerbase in a disorganised fashion that cause these changes to occur. Indeed, the authors specifically note that education and official language status are factors that would mitigate against the failures of L2 transmission that are claimed to cause these simplification effects, but don't then seem to acknowledge that this is a major issue for extrapolating the correlations anywhere further back than the 19th Century.

I think there are several things going on here which have led to this kind of thing being possible. Firstly, linguistics has always had an uneasy continuing relationship with the rest of the humanities and parts of the sciences (particularly neurobiology and acoustics), because it touches on so many of them at once. Secondly, and relatedly, we've had a fairly significant period where the core theoretical concerns of linguistics were essentially driven by a philosopher (Chomsky), which created a model of research that explicitly rejected empirical work and argued for much of its base assumptions (such as 'Poverty of the Stimulus') almost entirely by thought experiment. Empirically minded subfields continued to exist (Trudgill for instance made his career as a sociolinguist), but empiricism has been slow to make a return to theoretical linguistics. As a result, it seems to me that, at least in typology, people are still getting to grips with the idea of formulating hypotheses and actually working out tests that actually assess those hypotheses. Thirdly, because again of Chomsky, linguistics has had a strongly mathematical character in terms of its conception of language, and this again still lingers in the design of these kinds of studies. Everything has to be quantified in such a way that a simple computer model can work with it, but the trouble is there's frequently little room for the nuances of individual languages (e.g. Irish and Welsh are both counted on WALS as having the same value for inflection on adpositions, but actually a detailed study of the two languages reveals their systems work really quite differently, as I'll perhaps post on another time) which can therefore have knock-on effects for the analysis of any correlations thus derived.

#linguistics #rant

104 notes · View notes