Monday, May 01, 2006

More on language evolution   posted by agnostic @ 5/01/2006 09:46:00 PM

There's a recent paper in Nature (here), in which the authors claim that starlings show evidence of understanding recursion, the putative "core mechanism" of human language on Chomsky, Hauser, and Fitch's account (by no means consensus). I don't have university access, but there are plenty of good reviews: lay-directed ones at LiveScience and John Hawks, and somewhat technical ones at Tensor and LanguageLog.

Recursion in syntax is when a category X can be composed of another representative of the same category X, plus perhaps other stuff. For example: the rule "S --> S and S" says that a sentence (S) may-consist-of some sentence, the word "and," and another sentence, in that order. This result can feed into another application of the rule, like so: "I study linguistics and Razib studies biochem," "I study linguistics and Razib studies biochem and Darth studies psychometrics," ad infinitum, making sentences out of sentences. This is what allows human language to be so flexible -- in Wilhelm von Humboldt's classic phrase, it allows the "infinite use of finite means." [1]

The links above give good overviews, so I won't waste anymore space rehearsing the details (thus, minor "outside" reading required). The take-home message is that while not a negative result for other species possessing a putatively human-only syntactic mechanism, they're not necessarily strong positive evidence either, as the Tensor & LanguageLog entries show. For all we know, the starlings used short-term or working memory, not part of a language module. Unfortunately, the three fields of cognitive sciences / linguistics, psychometrics, and evolutionary biology don't talk an awful lot to each other (though see commentaries on LanguageLog here and here for mention of potential role of general intelligence in the starling study). Hence, the possibility that the results could reflect the starlings bringing some avine version of g to bear on the task wasn't fully explored.

My personal take on these sorts of studies, where the animals only succeed after tens of thousands of training trials, say more about animal psychometrics -- using general intelligence to consciously figure out murky patterns -- than about animal instincts -- effortlessly acquiring the skills & knowledge necessary to survive & reproduce. The fact that 9 of the 11 starlings learned the grammars suggests that they may be a bird version of a pretty easy IQ test. If, after tens of thousands of trials, a group of humans were trained to jump & remain in the air for several seconds of hang-time, that would say more about general athletic ability than about a basic human instinct for quasi-flight.

[1] Math / Comp Sci nerds can see how this works in old posts at my personal blog here and here.

See also: Chomsky interviewed by Ali G.

Update: Carl Zimmer writes about the paper, with quotes from many, in the May 2 NYT. Also, Chris at MixingMemory has a post on the topic here.