Tags

, , , , ,

About a week ago, I read and posted on a summary piece on cultural evolution research in PLoS Biology. The reviewer introduced me to Simon Kirby‘s work, which I found remarkable. Kirby and colleagues setup an experiment, one that observed the evolution of an artificial language from a set of random terms to an ordered, naturally adapting system in ways that assured its reproduction.

I didn’t know when Kirby was to publish his work, but lo and behold in this week’s issue of PNAS, I saw “Cumulative cultural evolution in the laboratory: An experimental approach to the origins of structure in human language,” by Simon Kirby, Hannah Cornish, and Kenny Smith. The experiment involved showing subjects illustrations that were associated with nonsense words.

The subjects were asked to play a game of Memory, by trying to recall the terms with the illustrations. Regardless of the accuracy of their recollections, the associated terms were used as a foundation of the group’s subsequent language training. This was done over and over, and low and behold, detectable patterns began emerging. Terms began to be used to describe whether an illustration pictured horizontal movement or a bouncing object. The following graphs document the transmission error and measure of structure over each generation:

Transmission Error & Measure of Structure versus the Number of Generations

Transmission Error & Measure of Structure versus the Number of Generations

Clearly there’s some pattern forming. But, Kirby and team understood that these emerging languages were simplistic and limited. So the team switched it up a bit, and discarded duplicate words. This represented a sort of selection, which gave structure and allowed the language to be remembered. Throughout 10 generations, the grammar of laboratory language went from meaningless, ad-hoc bunch of words into an expressive mode of communication. The speakers didn’t change, it was the change in the meanings behind the terms. The following graphs document the transmission error and measure of structure over each generation with selection:

Transmission Error & Measure of Structure versus Numer of Generations with Selection

Transmission Error & Measure of Structure versus Number of Generations with Selection

So how did the subjects screen out their own linguistic predispositions? Most humans are exposed to at least one language, which would clearly bias them and affect their abilities to give structure to a set of gibberish. In other words, the ‘selection’ applied could have been favoring structures that matched existing languages.

Kirby said that’s not really a concern, because that languages that emerged in his experiments do not have much in common with the extant languages. And since the emerging languages resembled those from computer models, which did not have preexisting languages to muddle up the waters, then we’re not to worry. Kirby concludes that the,

“The best explanation for our results is the cultural system ‘discovering’ adaptations for all aspects of the transmission bottleneck rather than merely mirroring the native language of our participants.”

    Kirby, S., Cornish, H., Smith, K. (2008). Cumulative cultural evolution in the laboratory: An experimental approach to the origins of structure in human language. Proceedings of the National Academy of Sciences DOI: 10.1073/pnas.0707835105
About these ads