In a recent journalistic article in the German weekly "Die ZEIT" (Die Wörterjäger) the journalist says (original quote in German):

Ist eine Gruppe dagegen homogen, haben ihre Mitglieder weniger Probleme, die komplexen Regeln zu pflegen. Je kleiner eine Sprache, desto sperriger und komplexer.

Short summary: The smaller a speech community, the more complex the language.

Can this statement be confirmed or falsified by data from the world's languages?

  • Do you mean "in principle", or do you mean "did someone actually do the math?". There is a germ of truth to it. – user6726 Oct 9 '16 at 20:25
  • 1
    Related question: linguistics.stackexchange.com/q/19664/200 (see also comments) – melissa_boiko Oct 9 '16 at 21:17
  • 1
    The authors appear to believe that most communities have only one language. This is generally not the case, even -- perhaps especially -- with small speech communities, where almost everyone is usually conversant with several languages, and the people who speak them. – jlawler Oct 10 '16 at 2:14
  • @user6726: It is probably not possible to do it rigorously, but is there at least a serious survey (e.g. based on the WALS survey)? – jk - Reinstate Monica Oct 10 '16 at 9:39
  • dunno if that is ttue, but the converse is intuitively obvious: the more speakers the simpler the grammar. – mobileink Oct 10 '16 at 18:23

It is not clear that the idea can be rigorously tested, in part because "complexity" can refer to at least three different things, and in part because there's a big sampling problem in checking if there is such a correlation. The worst possible way to try to answer the question is via cherry-picked languages evaluated subjectively. Chinese is obviously really simply morphologicially and it has a huge number of speakers, and Lushootseed is really complex morphologically, and it has very few speakers (none, in fact).

You could start by generating a random sample of let's say 100 languages, and check the facts, but I've engaged in that game before and it turns out that the biggest problem is getting an answer for most of those languages – in my most recent 100-language sample, only 5 languages have over a million speakers (Xiang Chinese, Southern Azerbaijani, Western Farsi, Sylheti, Georgian), and probably less that 20% of them have sufficient descriptive material that you have a prospect of saying how complex the language's morphology is. Since there are huge numbers of tiny languages and many well-populated dialects of Chinese, it's not clear to me what an unbiased sample would be.

One kind of complexity is "number of affixes", so if a language has just one affix slot and 200 affixes that could go there, that would be kind of complex, though mostly trivial. A second kind of complexity involves numbers of slots, so that most Indo-European languages would not be very complex since there are relatively few slots for affixes. The final and IMO best metric is in terms of the system of rules, for example if you get suppletive portmanteaux of slot1a+slot2d, or if there are morphological conditions on affixation (such as "add X only on perfective verbs"). While I think that gets at real complexity, I don't see how we can count (and if we can't count, we can't numerically correlate, and thus we're back to subjective judgments like "Turkish is complex" or "Kalaallisut is complex".

Still, I think it is probably true, based on subjective experience.

| improve this answer | |
  • A modern idea how to measure complexity uses tools from Computational Linguistics like lemmatizers (are there also morphematizers available?). Train them on manually prepared gold standard data and measure how big the compressed model to faithfully lemmatize the training data becomes (because the model grows with the amount of training data, you either measure a function depending on the size of the training set or use a standardised size of it). – jk - Reinstate Monica Oct 10 '16 at 9:45
  • I think the problem would be getting GOLD standard data that doesn't filter out real complexity (via the "how the heck do you encode that" filter). I don't use these ontologies myself because I find them incomprehensible and onerous, and many of my colleagues in the field likewise say "Nice idea, but I have other fish to fry". – user6726 Oct 11 '16 at 16:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.