The Atoms of Language

The Atoms of Language
Mark C. Baker

I must admit to a lifelong predilection to the aesthetic of reductionism [maybe that's why I ended up in particle physics; the ultimate end-of-the-road for reductionism]. Right or wrong, I find a certain beauty in the reduction of a messy, complicated and fundamentally unexplainable system down into an elegant set of constituent parts and rules.

I guess I'm not alone. Mendeleev, Mendel, Gell-Mann, Linnaeus, Schoenberg, and Chomsky all made their considerable fame through discovery of reductionist principles in their fields. I can't profess much aesthetic interest in the product of Schoenberg's reductions of music, but I spend almost all of my free time these days studying those of Gell-Mann, Mendel, Linnaeus, and Chomsky. There is no doubt that the principles of understanding given to us by these men are powerful.

This fascination of mine is probably why I ended up getting hooked on The Atoms of Language after casually flipping through it a couple weeks ago. It was intended as a supplemental book for an advanced linguistics course my wife took as a student in college. It's very dry, quite dense, and particularly narrowly focused. But it is a thorough reduction of human language in the Chomskyan style and as such really appealed to me.

Gell-Mann's discovery that all visible [and much of the invisible] material in the universe is composed of a very small handful of fundamental particles is the reductionist heritage of Mendeleev's periodic table. This reduction is clean, precise, and demonstrably true. The reduction of all life into species, genera, families, etc. as initiated by Linnaeus is, on the other hand, completely messy, controversial, and even arbitrary. Species are not a fundamental building block of ecosystems in the way quarks and leptons make up the entirety of any atom, molecule, or star. There are compelling practical uses for species and genera, but they exist more on paper than in reality because of the inherent messiness of ecosystems.

Chomsky's reduction of language falls somewhere between the intractable reduction of ecosystems and the neat and tidy reduction of matter. Although messy and strewn with exceptions, language is surprisingly reducible, lending itself shockingly well to a hierarchy of constituents and rules. This concept strikes me as terrifically profound. Whereas the species concept is mostly a construct created for practical purposes, the "atoms of language" are quite real and tell us a great deal about human nature and, importantly for me, what's going on in the head of my babies when they hear me talk.

In this way, Mark Baker's incessant use of an analogy between the periodic table and the reduction of language is apt. When he speaks of the "atoms" of language, he is not referring to the words we speak, which are in a sense the fundamental constituents of the language that we can hear when others speak [or morphemes, which are even more fundamental]. This language--what Chomsky calls "E-language" or external language--is mostly culturally transmitted. The atoms of the language that are really interesting are the parameters that set the rules for syntax within our "I-language" or internal language. What appeared, until Chomsky, to be a potentially infinite set of possible languages with a continuum of possible properties turned out to be the result of a nearly ludicrously small set of "parameters" which come pre-loaded in our brain and are set by experience with our native language as babies.

This is a powerful concept, and the book fulfils its role as an explicator of the known parameters extremely well. The details get tiresome for somebody reading it from my angle, just wanting to learn about human nature, but the big picture is presented well too. Here's a recap of my greatest insights from this book [since you probably won't read it]:

1. Based on surveys of known world languages, it is clear that not every syntactic structure possible is used. In fact, there are only a few syntactic systems humans use out of potentially thousands or millions. This is a clear indication, in the very least, that humans share an innate knowledge pertaining to syntax.

2. Chomsky likes to talk about the "poverty of the stimulus," which points out that children do not have enough language presented to them to learn every rule for every possible sentence in their native tongue. In fact, there are exactly infinite possible sentences. We would then never learn to speak if parameters didn't exist. This is not new to me, but through reading this book I realized that the evolutionary purpose [so to speak] of parameters is language acquisition. Acquisition is a two-fold task: (a) acquiring a lexicon and (b) setting a few parameters. This is an infinitely simpler proposition than memorizing every legal sentence configuration in the native tongue.

3. One of the most interesting new revelations I had from this book is that parameters are hierarchical. Some parameters take presendence over others, and some manifest themselves in obvious ways better than others. Baker constructs the rudiments of a parameter tree that will presumably at some point have every language on earth on its branches.

4. The lack of geographic inhomogeneity in syntax speaks volumes for the hypothesis of parameters. The fact that Indonesian, for example, has syntax essentially identical to English, yet German is quite different is completely incompatible with a fluid, infinite, culturally-transmitted grammar. Suppose language A diverged from language B a couple hundred years ago, and now has a completely distinct grammar. Also suppose that there exists a language C which is thousands of miles away and historically completely socially isolated from language A, but which has identical syntax to language A. The only explaination for this occurrence--which, by the way, is very common--is that language A swapped a parameter from its original form and since there is a finite number of parameter configurations now matches another isolated language by lucky accident.

5. It actually turns out that not every syntactic property is parametrically controlled. For example, the fact that English has reflexive pronouns [like himself and themselves] allows us to use the construction Bob said that Joe should buy himself a car. It is clear that it is Joe that should be buying the car because the pronoun himself attaches itself to the secondary subject Joe. If English didn't have reflexive pronouns, the sentence Bob said that Joe should buy him a car would be ambiguous at best and thus not possible. The lexicon of a language can influence the possible syntax without the involvement of parameters, but it requires special situations.

We have to be cautious with this, though. The evolutionist Daniel Dennett aptly described a very useful concept called "greedy reductionism." It refers to the compulsion to take reductionism well past its practical limitations. Describing the behavior of a bird in terms of the quantum mechanical interactions of its constituent electrons and nucleons, for example, is greedy reductionism. However, Baker doesn't fall into this trap. He aspires to, I think, but the field is so new and fresh that every reductive step is sound and useful up to this point. That's not to say that there is no controversy on the issue of the existence of parameters though.

Philip Lieberman is the modern anti-Chomsky whose ideas shaped a large portion of Christine Kenneally's The First Word, which I reviewed a few weeks back. In his 1984 book Toward an Evolutionary Biology of Language, Lieberman argues that one simple result of evolutionary biology sinks every language innateness theory. The argument is this: single random genetic mutations are common from one generation to the next. If all human grammar was pre-programmed in our genes, then some people with random mutations should be expected to display an inability to form certain syntactical structures or learn language at all. Such people do not exist, and thus, Lieberman says, the sum of human syntax is not encoded in our genes.

I have what I consider to be a simple refutation of this idea, but it requires a small retreat from a strictly Chomskyan theory. Chomsky views language as being a single entity which is whole, complete, and perfect. In his view parameters come as a bundled package; each is as necessary to language acquisition as every line of code is to a computer program [often removing one line at random will render the entire program useless]. Not only does this idea make it impossible to manufacture language through selective evolution, but it creates the exact problem that Lieberman takes advantage of in his attacks against the entire theory.

I however see parameters strictly as a learning tool; their biological purpose is to assist babies in learning language. If human language is limited to a certain subset of possible grammars, and if each child comes already knowing the settings that contribute to each possible grammar, then learning language becomes tractable. This allows an evolutionary theory of language since each built-in parameter increases a child's ability to learn language and presumably increases their reproductive viability. Additionally, Lieberman's complaint becomes moot: genetic mutation is likely to only effect a single parameter. Remove one parameter and all it does is make learning language harder for that one individual. One of the "startling" revelations in the post-Chomsky days is that human brains are very powerful. Not only are we pre-programmed with some things, but we can also figure things out other things for ourselves very well [like where the breaks between words occur, which is terrifically difficult and requires tremendous computation power]. General cognition can cover the holes in parameter space left by random genetic mutation.

Of course, parameters are studied from the perspective of systematics. That means that the parameter hierarchy describes [most of; see point 5] the differences between the syntaxes of languages, not how parameters can help children learn the bulk of the syntax of their native language. A meaningful parametric theory should have as its final goal a map of the innate choices in grammar each child is born with which teaches them their native syntax based on setting parameters through experience. This map is complete only when it contains every parameter a child must have in order to learn any possible human language. Baker's ultimate goal is a systematic description of the systematic relationship between the grammars of all languages. His parameter hierarchy is complete when it describes all of the observed differences between languages. This is actually a huge distinction and a potentially schema-breaking one for me. Baker's hierarchy shows only two parameters necessary to explain every observed difference between Mohawk and any other language. But I guarantee that a child learning Mohawk will have to flip more than two mental switches before being able to create every valid sentence in that language.

I don't see any evolutionary force in Baker's method; the diversity of languages is just a byproduct of the existence of parameters, and thus any description is akin to the reduction of life into species and genera; it is a useful description tool but doesn't hint at much that is more more profound [in that species per se are not the goal of evolution but survivability is]. I see Baker's method as analogous to the species method of description, and this acquisition-oriented method as analogous to a genetic description. I think his approach will break down at some point, or get mired in greedy reductionism. Shifting focus to discovering which parameters a child much identify and set to properly learn his native grammar is surely the path to follow.

I don't suggest reading this book for everyone. It does requires a small amount of introductory linguistics knowledge [for me, one introductory class and a few books, including most Pinker]. But beyond that, it's really intended as the definitive source on the workings of each parameter. There is little indulgent philosophizing about the implications of parameters; for that, go to Pinker. But if you, like I, enjoy the beauty of reductionism then you will enjoy Baker's relentless systematic destruction of the apparent infinite variability of language in this book. His brutally clean logical structure and impeccable analysis leaves analyticians like me very satisfied, even if in the end he strays off in a tangential direction.