The Atoms of Language

parameters are hard to believe

Posted by Martin Ames Harrison on Sat, Sep 2, 2017

I read Baker’s The Atoms of Language, which makes a case for the existence of parameters as knobs on a Universal Grammar machine (a copy of which, Chomsky says, sits in each human brain), and here are my thoughts.


First, let’s praise this work. I want to understand language in the way that Baker, Chomsky, Pinker and the like want to understand language. Their goal is a noble goal. I want an axiomatic framework for describing natural language. I want to wrangle the enormity of it. Language acquisition, for instance, is mysteriously quick in children, and I want to know how it happens. Once, in kindergarten, I was led to a table on which many baskets of markers - each holding several markers of a single color - had been placed, and was told “Take one of each.” I could not parse that phrase, “one of each”. It scrambled my brain, and my response was to pick a single basket and to take all of the markers from that one basket. Once I was corrected, I knew not only the teacher’s intent (which a non-autiste would have inferred from context), but also the general rule for using “each”. The power of that phrase lies in the universal quantifier baked into it, which might be the reason I found the phrase so difficult. I was learning how to let a variable run through all the values in a given set. That is advanced mathematics.

And grammatical constructions often seem logical until one begins to probe. Then things begin to feel arbitrary. Usage which at first seems as plain and as literal as could be is seen to be little more than convention or idiom. So what is grammar and where does it come from? Are some grammars better than others? Is there a universal one? These are really interesting and worthwhile questions. They’re also deep questions in that answering them will demand that we answer pretty much all the hardest questions about minds and intelligence. In fact, one of the reasons I am now so fixated on NLP is that I believe it to be at the heart of the whole AI project.

Baker introduces a metaphor at the opening of Atoms. He compares parameter theory to the part of chemistry so elegantly depicted in the periodic table. He develops this metaphor throughout the entire book, in a bit too much detail and with too much commitment, going so far as to map Chomsky onto both Democritus and Mendeleev, and earlier linguists onto the alchemists. It would be more fitting, I think, to map Chomsky onto someone like Kepler-on-his-worst-day, who believed that the relative positions of celestial bodies must be determined by the divine geometry of nested platonic solids. And this brings us to the first of my two problems with Atoms.

Universal Grammar & Parameters

Why should there be a universal grammar the variants of which comprise the grammars of all human languages? There are different ways to think about this “why”. Cynically, we can ask why one would wish for this to be the case. Ironically, Baker answers this version of the question by accident but very much to my satisfaction - more on that in the forthcoming paragraphs.

More innocently, we can ask: “What have we observed which leads us to think there could be a universal grammar?” This question has some good answers the strongest of which is that children readily learn new languages. Even multiple languages at once, without getting confused. And the details of this truth are more impressive still: there are grammatical milestones early in acquisition which appear in the same order and in the same time frames across language families. Children appear to learn first whether a language is polysynthetic, later head directionality and so on. This is fascinating, but does it suggest that a built-in grammar is being tuned by the child to align with the peculiarities of his language? It suggests to me that rules are being established in the child’s mind, but not that ‘settings’ (Baker’s term) are being selected, however unconsciously, by the child.

But, regardless of how the mind works or how it came to work that way, there is a much more mathematical obstacle in the way of my accepting parameter theory.

A Misnomer

The space of parameters does not exist. One cannot say what constitutes a parameter, but it’s so much worse than that. Even accepted parameters are logically entangled not just with each other but with lexical parameters - the existence/nonexistance of certain word types in a given language. They cannot be the atoms which are sought. If there are atoms of language, they do not exist at the level of grammar analysis that had been carried out by the time Baker wrote Atoms.

Stretching the atomic metaphor, Baker points to what is essentially a decision tree and calls it the beginning of parameter theory’s periodic table. But if the values of parameters must be assessed via decision tree, and if that decision tree can never be fully fleshed out, even in the abstract, then we are not talking about a model that has parameters. We are just exploring various grammars and finding little rules which help to distinguish, partially, between them. Only ever partially, that has to be stressed.

Knob vs. Readout: Fine-Tuned Universe Confusion

So if parameters, even in the most rudimentary stages of parameter theory, are not knobs, then what are they? They are more like the ‘laws’ of physics, and let me hasten to explain why I use scare quotes here. I am not saying that the laws of physics are uncertain or tentative (they are, but that’s not my point and is not a weakness of physics), but that they should not be regarded as rules dictating action in the way that control flow statements in a computer program dictate procedures. Yet this is the interpretation you’ll usually hear tacitly encoded in laymen’s statements about the way our universe works. We speak of laws as if they are agents driving action. We hit upon the existence of dimensionless values, deriving them painstakingly in a quasi-mathematical process which integrates principles, observations and intuitions. But then, we utterly flip it around, remarking quite foolishly that if any of these characteristic values were to be set differently, we would be left with a smoking wreck of a universe uninhabitable except by the most fundamental particles (and very often, not even those). The laws are more like readouts on a visual display. Likewise, the so-called parameters of these linguists have values, and even have different values in different languages, but they are not settings - they emerge from our efforts to analyze a natural process. We should not assume that this relationship may be inverted. You would not attempt to refuel your car by reaching into the dash and rotating the needle of the gauge.

Let’s look at some informal objections to the arguments offered in Atoms.

Equalism & Motivated Reasoning

It is tainted with informal fallacy. In some places, there is no effort to conceal this. Once, Baker hints at an allegation of racism. This is irritating because he does not stick his neck out and take the risk of actually saying it. Instead, he crouches behind “borders on” to fling the turd of “racism”. This is in the context of efforts to explain why polysynthetic languages are spoken only in less developed societies. Later, he is more forthright, announcing a commitment to avoiding the implications of variable expressiveness in human langauges.

It is true that some languages are more expressive than others, and he nearly says so. In a mixed up and equivocating conclusion, he argues that the richness of linguistic diversity, as found in poetry, for instance, is a solid basis for valuing linguistic pluralism. But the hypothesis of a universal grammar reminds us that the differences between languages are not so daunting. He does not say that differences in expressiveness are all a wash, though I think he means to suggest this. The idea of UG is attractive because it really presupposes some degree of equalism. But that only makes it attractive, not any more true than the alternatives.


I recommend The Atoms of Language, but I think its main idea is incorrect. It is fascinating and hard. It is not a dumb book, and a dummy will get lost in the more detailed grammar analyses. And it is sincere, for the most part.

I am now reading Deacon’s The Symbolic Species. That book is longer and its style is downright masturbatory. But I love the ideas. I will write on that one, too, probably in installments because it is so long and I don’t want to lose my insights. (This review is missing a lot that I noticed while reading Atoms but could not recall a week later.)

comments powered by Disqus