Five Ways to Investigate Speech Origins

The basic problem of studying the origins of language is, to understate matters, language leaves few fossils. There are five techniques for getting around this problem Tao Gong told audience members attending the “Ways to Protolanguage” conference in Torun. The presentation was co-authored by Bernard Comrie and William S.-Y. Wang. The presentation itself was too focused on method to find much space on this blog, but its list of techniques was very valuable.

Study Language Acquisition in Children

Children go through a variety of stages as they move from making animal cries to human sentences. It is plausible to suppose that these steps are similar to the stages of language evolution, but that plausibility comes from the old saw, “Ontology recapitulates phylogeny” and that bit of wisdom must be applied cautiously. (The issue was discussed more fully in an earlier presentation at the conference. See: Does the Recapitulation Principle Apply?)

Adopt a Comparative Approach

Examine the communication systems of other animals, especially primates. (Unmentioned were birds, but birdsong learning appears to be relevant to learning how to vocalize. Nothing in primate communications seems to compare.) The problem is that primate communications bear almost no relation to human speech. (Unmentioned was the negative comparison. It has been very helpful on this blog to notice things that apes do not do.)

Perform Artificial Language Learning Experiments

Test to see what communications systems emerge under artificial conditions. However, human subjects already know how to talk and it is difficult to find ways to keep that existing skill from influencing the results. (Not discussed was the use of non-human subjects. Work with a variety of apes has established they have some intellectual capacity to use language. Therefore the absence has to be attributed to something else. It has also established that no new mental abilities had to evolve to at least get our ancestors using single words.)

Study Linguistically Impaired Individuals

If we could understand how language goes wrong, we might have some sense of language’s neurological and genetic dependencies. But there are not many such cases. (Left out of the discussion was autism, which is a fairly widespread language disorder. But we have no good idea of what has gone wrong in autism.)

Computer Simulation Studies

The presentation was titled, “The Necessity of Computational Simulation Studies,” and the bulk of the presentation focused on the importance of this technique. Simulations use computers to test out ideas. They can illustrate mechanisms and how they interact. They can also identify unexpected problems, and predict possible evolutionary trajectories. (For an example of a simulation study see Simulation Suggests Steps to Speaking of the Unseen)

Derek Bickerton boasts of his armchair approach and often complains that computer simulations only get out what they put in. Even if that is true, however, it does not abolish the point that simulations show what you are putting in. More serious is the tendency to oversimplify what are complex matters. But there seems no way to deny that simulation has long proved to be a valuable arrow in the researcher’s quiver,

Comments

[uzza]

Deaf adults who grow up with no exposure to language constitute a near ideal group for comparison, differing from normal subjects only in the one parameter of not having language, and they are anything but rare. Why are they not sought out and studied more?

[Slawomir Wacewicz]

I think the issue is more complex with the word 'language' being the tricky part here (again). There may be deaf adults with minimal exposure to *language*, but with lifelong history of functioning (also communicateively) within a modern symbolic culture, which is a powerful 'corrupting' factor.

As to simulations, in my (nonexpert) opinion the biggest problem is that pointed out by Bickerton, i.e. sensitivity to initial conditions. In most cases it is impossible to know if your model is good or is missing one subtle confounding factor that when added would totally mix up the results. The way I see it is that simulations are extremely valuable but only in those rare cases when you know for sure there is a limited number of parametres. Still, what it shows best is not a specific result but rather *general conditions* for something taking place, like some behaviour being an ESS or some language form propagating faster than others.