American Scientist

Pathways to Language: From Fetus to Adolescent. Kyra Karmiloff and Annette Karmiloff-Smith. xii + 256 pp. Harvard University Press, 2001. $27.95.

Seldom is the talent of a would-be Mozart or Einstein overlooked. But the real prodigies are the millions of ordinary young children who learn to talk every day. Your three-year-old child may know nothing of politics or calculus, but she's a genius at learning language.

Pathways to Language: From Fetus to Adolescent, written by noted language researcher Annette Karmiloff-Smith and her daughter Kyra Karmiloff, is a comprehensive introduction to the field of language acquisition. Pathways would serve as an excellent text in an undergraduate course on language acquisition or cognitive development and should be of interest to parents and other lay readers who are curious about how children learn language.

Chapters devoted to topics such as word learning and grammar learning review virtually all of the major studies in the field, highlighting how impressive children are at language learning (for example, they manage to learn nine words a day—try doing that in your German class), and at the same time helping readers to better understand how children accomplish this remarkable feat. A chapter on how children master the basic sounds of their language shows that the process of learning to speak begins long before birth. Another chapter documents children's abilities to understand how stories are told. And another explains the methods scientists use to probe what goes on in the mind of a child. (Even readers well versed in contemporary scientific methods may know little about the familiarization preference paradigm or how to test whether a fetus recognizes his or her mother's voice.)

Karmiloff and Karmiloff-Smith do more than simply report the facts; they interpret studies and theories through wise eyes, knowing the field and its pitfalls. For example, they note that the famous stage of "telegraphic speech" is really a misnomer: Telegraphs typically leave out words but include inflectional endings (such as plural –s and past tense –ed), whereas speech in early childhood is marked by the omission of inflectional endings ("Where car?," "Mummy shoe," and so forth). Another section describes a famous theory of language acquisition known as the "operating principles" approach. Proposed by Dan Slobin, this theory consists largely of sets of data-driven rules of thumb such as "pay attention to the ends of words." After working through several examples, Karmiloff and Karmiloff-Smith make a number of strong points against Slobin's theory. Perhaps the strongest is a concern that the operating principles, while useful, may be too distant from the grammars that children ultimately acquire to really yield an adequate account.

It is a shame, though, that Karmiloff and Karmiloff-Smith do not apply the same critical eye to their own theories. A section on neural-network models does not even mention that they have been extensively criticized (see Steven Pinker's Words and Rules or my own The Algebraic Mind), and the concluding sections on innateness and language impairments give up much of the balance achieved earlier in the book.

Learning language is clearly something special—human adults are not very good at it, and even the best nonhuman primates never get very far. Yet toddlers manage to acquire language even when their parents don't talk to them directly; deaf children will go so far as to invent their own sign languages if they are not exposed to sign at home. Children of immigrants turn limited "pidgin" languages into full-fledged creoles. For all these reasons, Noam Chomsky and others have argued that the ability to learn a language depends in part on an innate "Universal Grammar" or "Language Acquisition Device" that includes at least some machinery that is built in and specially dedicated to the task of learning language.

Surprisingly, Karmiloff and Karmiloff-Smith disagree with this argument. Although early in the book they seem to endorse a pluralistic position in which the ability to learn a language would be a mix of domain-general mechanisms (learning strategies that are not special to language) and domain-specific mechanisms (learning machinery that is special to language), by the end of the book they give little credence to the domain-specific, instead suggesting that the ability to learn language is largely a consequence of a domain-general ability to do fast auditory processing.

But the notion that an aptitude for fast auditory processing would suffice for learning language doesn't seem terribly plausible. As Karmiloff and Karmiloff-Smith themselves note, chimpanzees and other nonhuman primates share many of our basic auditory capacities. Yet even when immersed in human environments, our primate cousins are unable to learn language, suggesting that there is more to language than a talent for auditory processing.

The same sort of point arises when it comes to language impairments. The issue is whether they stem from general cognitive impairments or from mechanisms specific to language. Some language impairments are surely due to auditory impairments: If you can't hear a spoken language, you can't learn it; if you can't hear it properly, you may not learn it properly.

But language impairments can also be found in people who have normal cognitive capacities and no detectable hearing difficulties. These people are a subgroup of those with Specific Language Impairment (SLI); they have grammatical SLI (G-SLI). We don't yet fully understand what causes the difficulties that they face, but it does seem plausible that irregularities in their genetic makeup may contribute to the partial miswiring of their language faculties.

Karmiloff and Karmiloff-Smith rightly note that this logic is not airtight—the auditory tests of the G-SLI group generally come later in life, and it is possible that if people with G-SLI were tested earlier, subtle auditory difficulties would be revealed. But this seems to be a slender, almost desperate thread on which to hang their theory—all the more so because recent research by Dorothy Bishop shows that plenty of people with pronounced hearing impairments do just fine on the very tasks that the G-SLI subjects find difficult.

The book's final argument against domain-specificity seems to rest on a serious misunderstanding of genetics. Karmiloff and Karmiloff-Smith note rightly that people with Williams Syndrome have a certain genetic irregularity in every cell (deletions on chromosome 7), but they wrongly infer from this that the disorder itself is domain general. The crucial question is not whether every cell contains a copy of the abnormal gene (roughly speaking, every cell contains every gene) but whether every cell expresses that gene (hardly likely—many genes are expressed only in highly specific sets of cells.) And the biology matters—given that chimps and humans raised in nearly identical environments wind up so differently when it comes to language, it seems that we will have to understand the role of genes as well as environment if we are ever to understand what makes our children so gifted.

The real target here seems to be innateness. Logically, the question of domain-specificity is actually independent from the question of innateness. For example, the ability to play chess is highly specialized but clearly not innate. The capacity for memory is clearly innate, but not special to any particular domain. But most scholars who argue against innateness are really arguing against domain-specificity, and Karmiloff and Karmiloff-Smith are no exception to the rule. Indeed, a concluding section on innateness actually turns out to be largely a discussion about domain-specificity.

Although the authors don't quite get the questions of innateness and domain-specificity right, the book does offer a valuable summary of recent research in language acquisition. Readers wanting to get a taste of that research and how it is conducted could do no better than to start here.