Monday, February 11, 2013

AI for the Evolution of Languages

There's an interesting article at New Scientist about a machine-learning algorithm that can be used to recreate the sounds of extinct ancestor languages.  That, of course, is very cool, if intrinsically hard to verify.

I would be interested, though, in whether the algorithm could work the other way.  Given the languages of today, including some reasonable guesses about how they will mix in the future (based on trade, population, etc.), can reasonable extrapolations be made for how a language will evolve into the future?  This could be tested, for example by taking the languages of 1750 or A.D. 1000 and trying to recreate modern English.

As far as I know, all attempts to extrapolate the language of the future have been either outrageously vague (as with the Eloi of H.G. Wells), more a statement of philosophy than an extrapolation of philology (1984), the haphazard introduction of a few words, or (more commonly, e.g. Star Trek) ignored altogether.  The language of 2313 should be mostly comprehensible, but nevertheless weird-sounding to us, as our speech would be to an English speaker from 1713.

Of course, if the resulting science fiction is wildly popular, it might very well alter the evolution of the language, either in a self-fulfilling or (more likely) self-frustrating way.


  1. I hate to say it can't be done. But there are just way too many variables and too many unknowns.

  2. I suspect you're right. This program probably only works when there are several daughter languages, each giving some information about the single parent language. The mere fact that one parent language can give rise to multiple daughter languages (for example, Latin -> French, Spanish, and Romanian) shows that going the other direction there is no unique solution. Add to that the incorporation of words from other languages -- something English is particularly prone to -- and new words that have to be created to deal with new technologies, and it's a total mess. Still, it would be an interesting thing to try. It would still fit a common sci-fi theme: one possible future, but not necessarily an inevitable one.

  3. Yeah, I agree it can't be done - and would take a long time to verify, but the idea is cool.