Monday, August 27, 2007

Darwinian Linguistic Evolution

There is a paper by Oudeyer and Kaplan, which discusses the evolution of languages in Darwinian terms. That is a refreshingly new (to me!) take on how languages may evolve. It applies the same Darwinian principles of heritability, variation and selection to individual phonetic words as well as associations between words and meanings.

The article makes use of computer simulations to inform their theory. Some of the take home from that article are:

  • For Linguistic coherence to evolve (that is one word referring to same meaning for different agents), the replication principle most suited is whereby the most frequently encountered word is repeated and thus gets fixated in the population. This scores over the use-the-last-heard phoneme rule and use-the-phoneme-as-per-frequency-in-usage rules.
  • The phonemes that can be easily confused (are liable to mutate more) with nearby phonemes get selected against and thus, selection leads to implicit evaluation whereby those phonemes that do not mutate (or mutate less) are preferred and get fixated.
  • In a population with agents coming and leaving, the population flux ensures that optimal words are used and sub-optimal done away with.
  • The linguistic phonemes (or words) that are used to represent concepts, break the acoustic space in such a way that their is least scope for confusion amongst the phonemes.
  • A trade-off happens between linguistic distinctively and robustness. Some words are long enough that they can mutate more, but are not easily confusable. Other frequently used words are short and do not mutate easily, but if they mutate than more confusion of meaning arises.

There are more such interesting information nuggets in the paper. So why don't you have a look at the original paper itself.

Hat tip: Babel's Dawn.

Sphere: Related Content

No comments: