Wednesday April 14, 2010

This was the day the workshops met. The occasion for me to come on this trip was the workshop “Does the math add up?” which was about mathematical modeling and simulation. In my wanderings on Tuesday, I found the street and building where the workshops were to be held. It’s behind a church down the main road, about a five to ten minute walk from the hotel. The buildings all run together, and this particular one had a rococo-style meeting room with curlicues and pastel colored walls.

The workshop was a sequence of presentations about 20 to 30 minutes each. They ranged from fairly concrete projects like mine to more general philosophizing and thoughts about whether we should have a database or web site to start combining various approaches. The starting point for the workshop was a position paper by the organizers de Boer and Zuidema. They put forth a lot of interesting ideas, but what struck me was how many times they mentioned ‘how did such-and-such a bit of language related anatomy evolve….’ So my project was about how you might start addressing that kind of question. My idea, which I’ve been working with off and on for a very long time now, is to deal fairly directly with biochemistry. I feel that I should give some background before I say more.

When you work with game dynamics, you model competing variants of a species but all of the anatomy and conversation is abstracted away: You only keep track of how well each variant communicates and derive some measure of reproductive success from that. Basically, you represent an organism’s ability to communicate as a relatively short table of numbers. Such a highly abstract model does give interesting and useful results. My dissertation advisor Martin Nowak, for example, wrote a couple of articles with his colleagues (Plotkin & Krakauer) about how in a noisy environment there is an upper bound on how well you can communicate using a single signal for each meaning. Eventually, you have so many things to talk about that you run out of distinguishable signals, and you have to switch to a combinatorial signaling system. However, there’s no way to use a model like theirs to actually see how a species discovers how to combine signals.

At an even higher level of abstraction is what I call ‘philosophizing,’ which you find a lot of in Pinker’s and Jackendoff’s writing and in some of the conference presentations. These are verbal arguments and sometimes thought experiments, and although they make some good points, I usually find them very unsatisfying. For example, what I recall from Pinker’s book The Language Instinct, he spends a very long time writing that language must be ‘adaptive,’ or ‘a result of selection,’ that is, it grants the species some reproductive advantage, so it was ‘selected,’ and that’s how we got it. Jackendoff’s book Foundations of Language is written for a different audience and goes into a lot more detail, and almost becomes mathematical.

The trouble with verbal arguments like these is that they often make claims that are a little too strong, so you have to sort of hedge in your writing, and it becomes a mess. For instance, Lightfoot argued (sort of in opposition to Pinker) that certain features that appear to be common to all human languages make it nearly impossible to express certain perfectly reasonable questions, which suggests that there are innate (= genetic) constraints on language that pointlessly reduce its ability to convey information, and therefore seem to be specifically maladaptive. So Pinker can’t get away with claiming that every aspect of language is adaptive, and maybe his chapter doesn’t really try to, I don’t remember, it’s been a while since I read it. So when he says that language is generally adaptive, what does he really mean?

Later at this conference, Fitch explained a little about how he and Hauser came to write their famous article with Chomsky, and how they had to keep going back and forth with Chomsky because they would read one of his articles or books and ask “This article says X. Do you really believe X?” and he would reply “No,” tell them that they misunderstood, and that he did believe something else that was consistent with what he wrote but maybe not how a typical reader would interpret it without further guidance from the author….

So that’s why I’m a mathematician. Verbal argument leaves out a lot of important details, or glosses over them, or makes some sort of general acknowledgment that there’s more to the story. I read them and feel like yeah, they’re making a pretty good point, but I’m not sure we’re making much progress here….

So you take all of those thoughts, put them in the blender, and out comes my latest crazy idea. I want to know how innate behavior and computation are encoded in the brain, and how that structure is encoded in the genome, and how a non-linguistic species evolves language as a behavior and computation. So, I wrote an artificial life simulation, where the virtual organisms run a sort of highly simplified biochemistry that can theoretically do any calculation (thanks to Luke for that idea!). I gave it some simple tasks, like communicating four bits of information using only two transmission chemicals. That means it has to use time, so it has to combine different signals (= level of the transmission chemicals). I’ve just started working on this, but I got enough figured out to present something at this workshop. Basically, it can evolve serial encoding of information, but surprisingly, it accomplishes identical sub-tasks in equally effective but different ways, and sometimes the calculation it comes up with accomplishes the task correctly but in an absurdly complicated way. So, when thinking about language evolution, you have to be open to the possibility that certain features of language are genuinely inexplicable, they’re just what evolution first came up with to solve certain problems. They aren’t specifically adaptive. Once I get some better software written for tracing genes, I expect to see genes that appeared and just sort of hung around until some much later mutation made them part of the solution—So then which mutation gets to count as ‘adaptive?’ Other evolutionary leaps may be fundamentally mundane, like separating useful mechanisms so they don’t interfere with each other. There are all these neat means of discovering useful calculations that just don’t appear in the verbal reasoning, and I’m not sure that they can, at least not until more projects like mine are done. And it’s this sort of weird simulation, partly about nerves signaling sequentially to muscles to form gestures, and partly about creatures signaling sequentially to each other to transmit complex information.

So, that’s what I did, and what I presented at this workshop.

All through the conference, I think a dozen people at least mentioned to me that they liked my project, and that they wanted to know more about it. A guy named Luke was particularly interested, so I sent him the source code and talked to him many times about many things throughout the conference.

I met a couple of people from University College London who were spreading the word about their upcoming conference on spatial models and language change. I had lunch with them at a café and I’d like to go, but this conference is of course in London, and it takes place during the school year, and after all the rest of what happened on this trip, I have to think twice before making a second trip to Europe like this so soon….

The main historical landmark in Utrect is the Domtoren that I described before. The opening lecture and reception were at a sort of hall in the building next door to it. It’s a beautiful old building with a chapel with a pipe organ, but they didn’t have anyone play it.

The scheduled opening speaker was unable to come, so they asked one of the other invited speakers, Maggie Tallerman, to take his place. She spoke in response to one of these “there is no universal grammar” papers.

Here’s some explanation. The term ‘universal grammar’ or ‘UG’ normally means either (1) the innate assumptions that children bring to the task of learning language, or (2) just the innate assumptions that apply only to language and absolutely nothing else, or (3) the features that are part of every human language, or (4) one of Chomsky’s approaches to generative syntax. Those are completely different concepts, so of course there’s mass confusion about the term, and the preferred definition changes from sub-field to sub-field and researcher to researcher and from decade to decade. This means that papers get published announcing that they have determined conclusively that there is no universal grammar, which is of course crazy and ill-defined. There is not enough input in any set of sentences that any child hears to specify a language unless the child brings some assumptions—this is called the ‘poverty of stimulus’ problem, so UG in sense (1) has to exist. What the authors usually try to prove is something off to the side; for example, that the assumptions children bring to language are all general principles that aren’t specific to language, that is, UG (2) doesn’t exist. What the authors usually say is something different; for example, some large group of researchers have been writing as if feature X of language (something like verb tense or word order) is innate, language specific, or otherwise present in all human languages, or maybe Chomsky posited it at some point, but some new explanation of X has come to light, so it is possible that children deduce X from the sentences they hear other people say, so X must not be innate, or it must be a consequence of some very general innate principle of learning. So they end up supporting a much weaker claim, that UG (2) might not contain X. Some of these authors like to write these articles in such a way as to sound like they’ve proved absolutely conclusively that UG (1) or (2) doesn’t exist, which stirs up controversy, and much tedious discussion and academic flaming follows. One of these articles must have come up again recently.

Tallerman spoke about it, and without looking at my notes I don’t even remember exactly what they were arguing about. I was very tired by that point, and I’ve become more or less immune to these inflammatory articles anyway. I ignore the fireworks, try to fish out their core argument, and see if it makes sense. I really think that the existence or non-existence of language specific features of the brain is a false dichotomy. I mean, can you really make a list of all the features of the body that are specific to, say, running and to no other feature of human life? So. What I found the most interesting in this presentation was that she had lots of examples of neat structures in languages that I’d never heard of, mostly Native American languages from the west coast. She had some examples about languages with almost but not quite completely free word order. She talked about a famous and controversial language that seems to have no recursion, but there’s still no complete consensus.

I didn’t really have dinner on Wednesday because I ate enough at the opening reception.