The language instinct (Steven Pinker)

[cartoon portrait] [grammar digram]

What makes an English sentence gramatically correct? Is it more correct to say "everyone returns to his old habits" than "everyone returns to their old habits"? Why do languages vary so much...or do they? Where does language come from?

In "The language instinct", Steven Pinker entertains these questions and many others in an impressively engaging survey of the cognitive scientist's approach to linguistics. He storms through the material. Chapters 4 to 7: structure and phonetics; chapter 8: language families, all the way up to Nostratic and "proto-world"; chapters 9 and 10: language acquisition and neurophysiology, with amazing anecdotes of the brain-damaged; chapter 11: language genes, and why animals can't learn language; chapter 12: a devastating critique of the language police, after which everyone returns, reassured, to their old habits. I'll get to chapters 3 and 13 later on.

In between the <hilariously ambiguous newspaper headlines> and <amusingly misheard song lyrics>, however, it is made perfectly clear that the world of linguistics is divided into the right and the wrong, and that we owe our understanding of the right to a new synthesis, "cognitive science", for which Pinker is to be our guru. Although in his zeal Pinker sometimes comes across as a know-it-all, he does an energetic job of persuading us that cognitive science really is science. Only in isolated chapters does he succumb to the standard temptation for popularizers of science: trying to convince us that his research topic has deep meaning. He is helped in maintaining this discipline by the fact that his material is intrinsically rich and of universal interest. In the hands of a master communicator like Pinker the ideas practically jump off the page directly into your brain.

The constructive Instinct

The positive picture that Pinker gives of the workings of human cognition is believeable, even common-sensical. Everyday experience indicates that people have innate abilities ("modules") for many mental tasks like face recognition and verbal fluency, just as they do for physical tasks. Much of the book consists of fleshing out the wonderful details of the language module, with all its related skills such as the production and decoding of speech.

Pinker quite reasonably emphasizes commonalities rather than differences being innate. This is particularly important given that most of us only notice such innateness when we attempt to overcome God-given aptitudes (often our own) for certain tasks such as mathematics. The modules can be amusingly independent: there are mathematicians who are almost inarticulate, fluent writers who never pick up the pronunciation of foreign words, and so on. The hard-wiring of certain specific talents (and character traits) must be obvious to anyone who has had to teach or to raise a child. It also seems scientifically plausible that natural selection has played a large role in the formation of these skills. So who is he trying to debunk? See below.

The other foreground element in the picture is "mentalese", the inner language that cognitive scientists apparently imagine our thoughts are couched in. But I was disappointed that it was never made clear what is constructively gained by treating the non-linguistic aspects of our minds as secretly linguistic. Cognitive Science's predecessor Artificial Intelligence was famous for its unmet promises. By now there should be some empirical successes based on mentalese, but Pinker really only uses the theory in his critique of linguistic determinism. So, as with the innateness question, we are lead to the destructive side of "The language instinct".

The destructive Instinct

As I mentioned above, Pinker sorts the world into right and wrong. The wrong are:

  1. The Standard Social Science Model (SSSM) , the "nurture not nature" ideology of the malleability of the mind. Pinker focusses particular criticism on anthropologists, whose minds are apparently so open that their brains fall out, and on linguistic determinism (the Sapir-Whorf hypothesis), which holds that people's world-view is partially determined by their language.
  2. Those who claim language is too special and complicated to have arisen by natural selection. He criticizes Stephen Jay Gould and Richard Lewontin for helping this viewpoint to spread.
  3. Those who agree that language is evolved by natural selection, but then infer that non-human animals must have this ability too. Pinker is specially scathing about the supposed acquisition of American Sign Language by chimpanzees.
  4. The language mavens who seek to impose artificial rules like not splitting infinitives. Even William Safire gets a dose of his own medicine.

The villain always hiding in the shadows is the Standard Social Science Model (SSSM). In fact, it is such a cartoon character that one is left wondering whether Pinker has described it fairly. His quotation (page 406) of J. B. Watson's claim that he could raise any child to be any type of adult surely represents the antiquated extreme of such doctrine. It would be very interesting to hear how psychologists and anthropologists react to Pinker's characterization of their world-view, since to the uninitiated layman or physical scientist the SSSM seems hardly worth bothering with.

And so to linguistic determinism and mentalese. Linguistic determinism states that thought is shaped by language. Pinker has a good time ridiculing the wackier versions of this idea, such as Whorf's suggestion that the Hopi have no conception of time proceeding smoothly between past, present, and future. And Yes, obviously there is such a thing as thinking one thing and saying another. But there is something weird about his confident assertions that
(1) thought is independent of language (or other communicative behavior) but is actually couched in its own internal language, mentalese;
(2) knowing a language is knowing how to translate mentalese into strings of words and vice versa, so linguistic determinism can be restated as claiming that our mentalese is (or is similar to?) our language.

The first assertion is unsettling because mentalese lacks one of the defining properties of a language: it cannot be used for communication. It might be better to call it a "coding" or "representation". And it may very well be one component in a good model of human cognition. But, as I will explain, I had a hard time seeing its relevance to larger philosophical questions. Pinker does not mention it, but there is a vigorous debate over the status of mentalese (see the Stanford encyclopedia of philosophy or this article by Larry Kaye).

The second assertion, apart from any reservations one might have about the relevance of mentalese, relies on ambiguity about the claims of linguistic determinism. Pinker mentions three possibilities:
(a) identicality: language determines thought precisely, word-for-word;
(b) concept determinism: language determines what we can think (Orwell's Newspeak);
(c) linguistic relativity:the form of our language merely influences what we tend to believe.

Pinker then unleashes a somewhat indiscriminate assault on all three. He treats concept determinism and linguistic relativity as if they were just as absurd as identicality, and presents mentalese as the long-awaited refutation of all of them. It seemed to me, however, that mentalese was irrelevant. Identicality is a straw man, killed off by his polemical asides without the heavy artillery of mentalese. Concept determinism is wrong for more general reasons that I'll describe below. Linguistic relativity emerges largely unscathed, mainly because Pinker sidesteps it: he does not mention any study of the correlation between peoples' beliefs and their language, but puts forward a "clinching experiment" that shows that physiology rather than language is the dominant influence on the learning of new color words.

One gets the sense by the end of the chapter that in trying to create and then discredit a sharp formulation of linguistic determinism, Pinker has overlooked its deeper incoherence. Identicality is obviously wrong. Linguistic relativity is a catch-all that without more specifics is too vague to discuss. And as for concept determinism, it doesn't even seem to make sense. How do you study a person's concepts apart from through their speech or other communicative behavior? So even if you find some interesting correlation, how do you decide if language has determined thought or vice versa?

Language and thought: two sides of the same coin

Thoughts have to be communicable to count as thoughts. After all, what sort of thoughts could not be expressed in language or behavior? Are they thoughts at all? How are they distinguished from emotions that cannot be expressed, or demonic possessions that cannot be expressed? This is exactly the point that underlies the Turing test: the sophistication of someone's thoughts is defined by what they can, under the fairest conditions, communicate. We can imagine a thought that is not communicated, or that is not expressed in language, but we cannot imagine one that could not be communicated. When you are not sure how to express youself, that means you have a half-formed thought, or a vague thought, not a crisply-defined thought that somehow defies expression.

At the beginning of chapter 3, Pinker trots out Wittgenstein's remark that a dog cannot think "perhaps it will rain tomorrow" as an example of a philosophical claim that animals lack consciousness. He lumps this in with the Sapir-Whorf hypothesis, and later dismisses the whole package as "wrong, all wrong", styling cognitive science as the defender of common sense (that thought is independent of language). There is indeed something wrong here. Consciousness is a separate can of worms from the language-thought relationship. And common sense tells us that doggie thoughts are as complicated as doggie communication skills, no more and no less. Thought is not identical to language, but the two are closely related by their very definition. Ironically, Pinker shows us that cognitive scientists feel the same way: because they are unable to discuss raw thought, they immediately imagine it as being couched in a fictional language, mentalese.

With this in mind, it is interesting to look at the examples that Pinker uses to argue for the primacy of thought over language. For most thought (and certainly most interesting thought) the medium of communication is language. But more primitive behavior can communicate very simple thoughts. This is what is happening in the example of a baby looking surprised when it sees one object suddenly turn into two. Pinker seems to imply that the baby is not just thinking "Wow!", but something more sophisticated, such as "I assume that things don't split into two like that, but this sense data is inconsistent with my assumption. I must look more closely." These sentences may be related to what is in a good cognitive science model of the baby's behavior, but to say the baby is actually thinking this is silly. We could just as well claim that a falling stone is thinking about Newton's law of gravitation, or, in the case of a particularly clever stone, Einstein's.

The same is true for measurements of physical phenomena that are not related to communication at all, such as some change of blood flow in the brain, or a Chomskian "trace" in a sentence. The fact that humans build languages in certain ways does not show that they think things like "Hmm, this is a topic-prominent language, so I'd better put the question word at the end." It is even worse to say that these abilities constitute knowledge, since for someone to be said to have knowledge of something they must be able to not only communicate it but also justify it. Cognitive scientists like to call this "tacit knowledge", as if it were a type of knowledge. But this terminology comes from the same stable as "involuntary choice".

To sum up, it is perfectly reasonable to construct a model of the human mind, using whatever kind of mechanisms successfully reproduce human behavior. To make such a model work would be a great achievement, and I would have loved to hear more about progress in this direction. But why over-reach by claiming that those mechanisms (as currently hypothesized) have an exotic reality as tacit knowledge and secret thoughts in a private language that we never knew we knew? It is as bad as a neurophysiologist telling us that thoughts are "really just chemical reactions", or a physicist telling us that our eyes tacitly know the theorems of optics because they can form an image on the retina.

The fun stuff

Finally, some fun with English grammar. Pinker discusses several points about English grammar that are confusing or even controversial. I don't know whether his account is as definitive as its tone would suggest, but it is illuminating to see how professional linguists approach these questions.

One point is the importance of heads: a phrase or compound has a head word that determines its grammatical properties. In "the last man off the boat" the head word is "man", since the whole phrase refers to a kind of man. To pluralize, apply the pluralization rules to the head: "the last men off the boat". Things get more interesting when you have a phrase with no head. The compound word "sabretooth" is headless: it does not refer to a kind of sabre, nor a kind of tooth. This explains why its plural is "sabretooths", not "sabreteeth". Other examples are "lowlife" and "pickpocket".
This shows why one feels ambivalent about phrases such as "John and me are coming too." It slips out quite naturally, but the pedant argues that "John and me" is the subject of the verb, so it should be "John and I". This would be true if both "John" and "I" were heads, in which case to make the phrase a subject, both whould be written as subjects. But actually "John and me" is a headless phrase, so one should not individually inflect either of its constituents. Pinker's discussion is not completely satisfying, however. Why is it OK to say "He and I are coming too" but not "Are you coming with he and I?"? And to make the phrase possessive would you really say "John and me's preference would be..."?. Is there a more complicated rule here, or something vaguer?

The thing I liked best was Pinker's analysis of "everyone returns to their old habits" vs "everyone returns to his old habits". It feels right to say that "everyone returns to their old habits", but the pedant argues that it breaks down like this:
every person[singular noun]
returns to his[possessive pronoun, agreeing with the singular noun it refers to]
old habits.
Pinker claims that it actually has a different structure:
every[quantifier] X[quantifier variable]
returns to X's[possessive form of quantifier variable]
old habits.
Other languages apparently have special words for quantifier variables, but English uses ordinary pronouns. In English the variable X is usually "he" or "they" (consider: "everyone has a place to which they long to return") and the possessive form can perfectly well be "their". [For more, see Henry Churchyard's singular their page.] Again, though, I am left wondering. Pinker has refuted the argument that "his" is uniquely correct in this context, but he has not given a uniquely correct alternative. Is this like the case of headless phrases, where there is no clear rule? Both examples seem to raise the very interesting possibility that the grammar of a language may contain vague preferences as well as sharp right-or-wrong laws.


Copyright © Mark Alford (2000)

alford(at)physics.wustl.edu

Mark Alford's home page

Valid XHTML 1.0!