The Language Instinct: How the Mind Creates Language
How many languages do you speak?
I have some acquaintance with French, Hebrew, and Spanish, but am far from fluent and never rely on this knowledge when I write about language. When my research involves a foreign language, I collaborate with someone who is both a native speaker and an expert in the linguistics literature. In my general writings I occupy a high link in the linguistics food chain, relying on surveys of the world’s languages by experts on linguistic diversity, who in turn rely on experts in particular languages.
Do you still do empirical research?
Yes. Since I wrote The Language Instinct, I have studied past tense and plural inflection in children (including identical and fraternal twins), in neurological patients, in speakers of English, Hebrew, and German, and with the use of functional MRI and other neuroimaging techniques. Recently I have shifted my focus to semantics and pragmatics, pursuing questions that grew out of The Stuff of Thought.
My spouse speaks a language other than English (or We’re spending a year in a foreign country). Is there anything I can do to encourage my children to retain their second language as they grow up?
Children care more about their peers than about their parents, so send them to summer camps, after-school programs, or vacations with their cousins, where they will have to use the language with kids their own age.
My child is three years old and hasn’t said a word. What should I do?
Take him (it’s more likely to be a “him”) to a speech and language pathologist accredited by the American Speech-Language-Hearing Association (www.asha.org), ideally one affiliated with a university, clinic, or teaching hospital. If the child seems to understand language, and is bright and socially responsive, there is a good chance that he is simply a “late talker” (probably from genetic causes) who will outgrow his silence and end up just fine. (See also the recommendations in “Read on”)
The Language Instinct Today
MANY FIELDS OF SCIENCE enjoy a rapid rate of new discoveries and a high degree of consensus about what they mean. The study of language, unfortunately for linguistics but perhaps fortunately for The Language Instinct, is not among them. A book on human genetics or nanotechnology would be hopelessly out of date a dozen years after its publication, but I like to think that The Language Instinct is still a useful introduction to the science of language today. Of course the field has hardly stood still (nor have my opinions), and here are some reflections on the contents of each chapter in the light of developments since 1994.
Chapter 1: An Instinct to Acquire an Art. The two heroes of this chapter are very much in the spotlight today. Darwin’s influence has expanded in psychology, the social sciences, philosophy, medicine, and genomics (despite the retrograde efforts of the “Intelligent Design” movement), and my 2002 book, The Blank Slate, explored some of the ramifications of his influence. Chomsky is still the most influential living linguist, and his political writings have inspired a new generation of leftists (I recently saw a sticker on a lamppost that said, “Read Chomsky.”)
Many readers concluded from this chapter’s acknowledgment of Chomsky’s influence on linguistics that I am a “Chomskyan.” In some regards—the view that language comes from a mental system tailored for the computation of symbolic representations—I am. But in the chapter I hinted at some disagreements, and I have spelled them out in the years since. In The Blank Slate I explained why I don’t share Chomsky’s romantic view of human nature or the radical leftist-anarchist politics connected to it, and he and I have locked horns in a recent debate on grammatical theory and the evolution of language (more on this later). My views on language and mind are very close to those of my comrade in that debate, the linguist Ray Jackendoff, himself a former student of Chomsky’s. Jackendoff lays out a coherent vision for the science of language in his recent book Foundations of Language, which I endorse wholeheartedly.
Chapter 2: Chatterboxes. Many of the phenomena discussed in this natural history of language have made the news in the past dozen years:
In 2005 the linguist Daniel Everett described an Amazonian people, the Pirahã, whose language, he suggested, did not allow them to describe abstract subjects lying outside their immediate experience. That claim, though, was belied by many of his own observations, such as that “spirits and the spirit world play a very large role in their lives.” Though Pirahã is in some ways simpler than familiar European languages (it has a “one-two-many” counting system and simpler systems of tenses and pronouns), in other ways it is quite complex, with sixteen verb suffix classes and more than fifty thousand attested word forms. Everett has emphasized that, contrary to casual impressions, Pirahã is by no means a “primitive language.”
Black English Vernacular made the news in 1996 under the wacky name “Ebonics” when members of the Oakland School Board floated a suggestion that it be recognized as a language in their bilingual education programs. The linguists John McWhorter and Geoffrey Pullum have written excellent analyses of the resulting hooha.
Anne Senghas was a research assistant in the project that first studied Nicaraguan Sign Language, and when she came to MIT as my graduate student I suggested that she study this fascinating phenomenon in her thesis and beyond. She has since published beautiful quantitative analyses demonstrating that the younger children have indeed developed a new language with a discrete combinatorial grammar.
Many readers were surprised to read of cultures where parents say little to their small children, who acquire language from older siblings and peers. But their surprise is a symptom of what Judith Rich Harris has exposed as “the nurture assumption”—the dogma that children are socialized by their parents. In her famous 1998 book by that tide, she argued that the most important influence parents have on their children is at the moment of conception. Children acquire their culture, and develop their personalities, in their interactions with their peer groups and society. Many features of language acquisition bear this out: the dispensability of parental speech in language acquisition, the phenomenon of creolization, the emergence of Nicaraguan Sign Language, and the fact that children of immigrants always grow up with the accents of their peers, not those of their parents. These phenomena (together with the findings of behavioral genetics) made me receptive to Harris’s theory, and I wrote a foreword to her book and discussed it at length in The Blank Slate.
Though Chomsky is famous for advancing the hypothesis that language is innate, he has never laid out a systematic scientific defense of the claim, and his main argument, the “Poverty of the Input,” is far from watertight. Geoff Pullum and the philosopher Barbara Scholz, using large bodies of online text (the most prominent new methodology in linguistics), have shown that many constructions children allegedly never hear can in fact be found in reasonably sized samples of English. They don’t deny that a poverty-of-the-input argument can be made (and I think that the “Simon” study, Nicaraguan Sign Language, and Peter Gordon’s mice-eater study reported in chapter 5 are good candidates), but they point out, reasonably in my view, that they are harder to nail down than Chomsky and most of his followers assume.
Human genetics and cognitive neuroscience are two areas that have exploded since the book came out. Williams syndrome is now known to be caused by the deletion of a small region of chromosome 7 containing some twenty genes, which accounts for the heterogeneous symptoms of the syndrome. At least one of them, the gene for LIM-kinase 1, has been associated with the problems in spatial cognition. Though the language of people with Williams syndrome is, as I emphasized, less impaired than other cognitive functions, there is a lot of variation among the affected people. The hyperdeveloped language abilities of Crystal, though showing that language can dissociate from other aspects of cognition, are not found in all people with the syndrome.
Progress in understanding the genetics of K family has been spectacular. First a marker for the gene was identified (SPCH1), then the gene itself (FOXP2) together with the mutation that leads to the deficit, then its evolutio
nary history. The gene has counterparts in other mammals, but the exact sequence of the normal human version is unique to us, and it has been a target of Darwinian natural selection during the past 200,000 years. Currently its function in mammalian brain development is being studied intensively. We now know that it is a transcription factor that turns on other genes, and that the versions in other mammals affect neural circuitry for motor control, particularly the circuits for making sounds.
Despite my cautious characterization of what was known about the family’s syndrome in chapter 10, I have seen myself cited as claiming both that the affected gene was specific to grammar and that it affects nothing but control of the muscles of the mouth and face. The family members have since undergone many tests, and the truth lies somewhere in between: the affected members have deficits in articulation and in oral and facial motor control, and have, on average, reduced intelligence. But they also have impairments in language itself that cannot be reduced to these other deficits.
Though no single gene specific to grammar has been identified (and perhaps none ever will), it is increasingly clear that sets of genes will be tied, with varying degrees of specificity and overlap with other functions, to aspects of language ability. The psychologist Heather van der Lely has documented a group of children with a syndrome she calls “Grammatical Specific Language Impairment.” Unlike that of the K family, their deficit appears to be specific to language itself, indeed, specific to grammar: they are normal in overall intelligence, in interpreting complex sounds, in understanding words, and in using language in a natural way in social settings. Their syndrome is probably inherited, but the families are not large enough, or the pattern of inheritance clean enough, to pin down the genes. Coming from the other direction, my former student Karin Stromswold has reviewed a large literature showing that many kinds of variation in language ability, including language impairment and language delay, are highly heritable.
Chapter 3: Mentalese. When I wrote this chapter, the Whorfian hypothesis was largely out of favor among linguists and psychologists, but the pendulum has swung back, and there is now a lively neo-Whorfian movement. In The Stuff of Thought, I review this new research and argue that the problem with the idea that language affects thought is not that it’s entirely wrong but that there are many ways in which language can affect thought, and people tend to blur them together. In particular, people tend to confuse banal observations, such as that one person’s words can affect another person’s thoughts (if that weren’t true, language as a whole would be useless) with radical claims, such as that we think in our native language, and that the language we speak makes it impossible to think certain thoughts. In the new book I argue that the major conclusion in “Mentalese” is right—we think not in our native language but in more abstract media of thought.
Chapter 4: How Language Works. The machinery of syntax described at the end of this chapter is barely recognizable in Chomsky’s current theory, which he calls the “Minimalist Program.” Chomsky is famous within linguistics for overturning his theory every decade or so; the current release, depending on how you count, is 5.2, whereas the release I presented in this chapter was a stripped-down version of 3.2, the “Revised Extended Standard Theory.” Nonetheless, the picture of grammar presented here will be recognizable to anyone reading on in linguistics, because I pretty much stuck to features that would stand the test of time and translate easily to other theories. In my own work I have always favored theories with less exotic machinery than Chomsky’s (shallower trees, fewer traces, less movement), and whose structures are more visible at the surface, like phrases, lexical items, and constructions. Joan Bresnan’s theory is an example, and a more recent version with this flavor can be found in Ray Jackendoff and Peter Culicover’s Simpler Syntax.
The most stunning development since I wrote the chapter took place in 2004, when the Boston Red Sox won the World Series.
Chapter 5: Words, Words, Words. In two subsequent books, I have explored the world of words in far greater depth. Words and Rules: The Ingredients of Language looks at the combinatorial richness of word formation and its implications for the machinery of cognition. The Stuff of Thought: Language as a Window into Human Nature looks at the meanings of words and how they originate and spread.
In his lovely book How Children Learn the Meanings of Words, Paul Bloom argues that children have no mental mechanisms dedicated to word learning and learn a word the way they learn any other kind of fact. Children zero in on a word’s meaning by exercising their “theory of mind” or intuitive psychology, deducing what a sensible speaker is probably referring to in the context. Ray Jackendoff and I think this is not the whole story, for reasons we explained in our paper debating Chomsky.
Chapter 6: The Sounds of Silence. Speech recognition technology has advanced tremendously and is now inescapable in telephone information systems. But as everyone who has been trapped in “voice-mail jail” knows, the systems are far from foolproof (“I’m sorry, but I did not understand what you said”). And here is how the novelist Richard Powers described his recent experience with a state-of-the-art speech recognition program: “This machine is a master of speakos and mondegreens. Just as we might hear the…Psalms avow that ‘Shirley, good Mrs. Murphy, shall follow me all the days of my life,’ my tablet has changed ‘book tour’ to ‘back to work’ and ‘I truly couldn’t see’ to ‘a cruelly good emcee.’” Recognizing a large number of words from a large number of speakers is still a formidable engineering task.
The sound pattern of English and the logic behind the vagaries of its spelling are explored more deeply in Words and Rules, including the extraordinary suggestion by Chomsky and Morris Halle that English spelling “comes remarkably close to being an optimal orthographic system.”
One raging public debate involving language went unmentioned in The Language Instinct: the “reading wars,” or dispute over whether children should be explicitly taught to read by decoding the sounds of words from their spelling (loosely known as “phonics”) or whether they can develop it instinctively by being immersed in a text-rich environment (often called “whole language”). I tipped my hand in the paragraph in this chapter which said that language is an instinct but reading is not. Like most psycholinguists (but apparently unlike many school boards), I think it’s essential for children to be taught to become aware of speech sounds and how they are coded in strings of letters. Diane McGuinness’s Why Our Children Can’t Read is my favorite book on the subject. The tide came from the publisher’s marketroids; she originally called it The Reading Revolution, since it was about both a scientific revolution in reading research and the revolution in human history that gave us alphabetic writing.
Masaaki Yamanashi’s comment on Bill Clinton proved to be prophetic.
Chapter 7: Talking Heads. Anyone who has tried to search the Web using one of the engines that claims to understand English can verify that comprehending natural language is still an unsolved engineering problem. Ditto for the programs that claim to translate from one language to another. The Loebner Prize competition (erroneously described as a “Turing test”) continues to be won by uncomprehending programs using canned responses.
The subfield of linguistics known as “pragmatics,” which deals with the use of language in a social context and WITH phenomena such as politeness, innuendo, and reading between the lines, was covered in a scant three pages in this chapter. A deeper discussion, which links these phenomena to social and evolutionary psychology, can be found in a chapter called “Games People Play” in The Stuff of Thought.
Chapter 8: The Tower of Babel Daniel Everett, the linguist who documented the Amazonian language Pirahã, claimed that the language violates Hockett’s universals by providing no means to discuss events remote from experience, and by lacking the mechanism of recursive embedding, in which a word or phrase can be inserted inside a word or phrase of the same type. But the first claim, as I mentioned, is contradicted by numerous observations of the Pirahã way
of life, and the second is questionable as well. Pirahã allows for a degree of semantic embedding using verb suffixes and conversions of nouns to verbs (so one can express the thought “I said that Kó’oí intends to leave,” with two levels of semantic embedding), and one can conjoin propositions within a sentence, as in “We ate a lot of the fish, but there was some fish we did not eat.” The linguists Andrew Nevins, David Pesetsky, and Cilene Rodrigues have taken a dose look at Pirahã, and have disputed his claim that recursive syntactic embedding is absent from the language.
The notion of Universal Grammar continues to be debated, though in a half-full/half-empty way; the proverbial Martian scientist would still consider human languages to be extraordinarily similar compared with the countless ways one could imagine a system for vocal communication. In The Atoms of Language, the linguist Mark Baker presents an explicit empirical case for a Universal Grammar with a smallish set of parameters differentiating all human languages.
My suggestion that controversies about language families might be resolved by “a good statistician with a free afternoon” has been taken up by a number of biostatisticians, though of course it is taking them more than an afternoon. In several cases, computer programs in biology that look at genes from a number of species and construct phyiogenetic family trees have been applied to words from a number of languages in order to construct linguistic family trees. The programs are first tested on uncontroversial families (like Indo-European) to verify that they can replicate well-established trees, then they are set to work on the murkier families, yielding both trees and the approximate dates at which protolanguages may have split off from their ancestors. Recent analyses of the Indo-European languages have suggested that the speakers of the proto-language lived 8,000 to 10,000 years ago, a date consistent with the upstart “Out of Anatolia” theory in which they were Europe’s first formers. Most linguists remain skeptical, because this dating contradicts the results of “linguistic paleontology” (for example, Proto-Indo-European had a word for the wheel, which was invented only 5,500 years ago). The debate over whether the Proto-Indo-Europeans were early farmers or later horsemen rages on, though both theories might be right, and may be true of different historical periods.