I’ve just read Michael Tomasello’s Constructing a Language (2003), and no, it’s not about conlangs.  It’s about how children acquire language.  It’s one of the best books on language I’ve read; also one of the most difficult.

As adumbrated in Chapter 1, the Generative Grammar hypothesis focuses only on grammar and claims that the human species has evolved during its phylogeny a genetically based universal grammar. 

Jeez, don’t open with a joke or anything.

Contra Chomsky

I already reviewed an essay of Tomasello’s; the book is an extended form.  In brief, he destroys the idea of the poverty of the stimulus.  Chomsky was reacting against Skinner’s stimulus-response notion of language, rightfully pointing out that there has to be a good deal of mental machinery dealing with language— language is nothing like a conditioned response.  His mistake was to assume that this machinery was innate— that children don’t hear enough language to deduce its principles.  As Tomasello shows, reviewing study after study, there is no evidence for this.  Children learn most easily precisely those constructions they hear the most often, and the more difficult constructions take longest to learn.

In later years Chomsky, faced with the vast array of human languages, elaborated the idea of parameters: there are a bunch of switches (OS or SO?  pro-drop or not?  AN or NA?) which determine a language’s particular grammar; all the child has to do is learn what the settings are for the language she hears.  Again, the evidence is against this.  Simply put, children don’t suddenly acquire settings; their competence increases slowly and (this is a key point) item by item, construction by construction. There’s not a sudden point where (say) they realize that English isn’t pro-drop.  They mix constructions with pronouns and those without. 

A major argument for innateness is that children learn languages “naturally”, supposedly within a window of opportunity and much easier than adults.  I’ve addressed this before, and Tomasello repeats some of the same objections, but adds a good new one: children have the advantage of having no first-language interference. 


How do children learn language?  We know most about the years from 1 to 4, which have been best studied.  Tomasello doesn’t believe in a language organ at all; he maintains that human language depends simply on human cognitive abilities, and the key one, appearing at about 2 years of age, is the ability to maintain joint attentional frames… that is, the child interacts with an adult, about some situation.  The key word is attention: the child only now can understand that others have mental states, and seek to affect them.  Animal language is all about expressing states: the animal is horny or hungry or wants to go home, or sees a predator.  Other animals may react to these expressions, but they’re not intended as communication— in fact the animal is quite likely to make the same expressions when alone.  What distinguishes human language is the ability to model other minds (and thus to try to affect them). 

Joint attentional frames are Tomasello’s response to Quine’s dilemma about ostension: pointing to a rabbit, do we mean the rabbit, the rabbit’s foot, the act of running, the color of the fur, or a bag of rabbit parts? 

Tomasello points out, by the way, that ostension is of less use than we might think in language learning.  Verbs, for instance, are most often used not to point out an ongoing action, but to describe one that just occurred or that’s about to occur— neither of these are things that can be pointed to.  Even nouns often occur when not present (“Where’s Daddy?”  “What does a cow say?”).

What the frames provide is meaning and context.  Basically, toddlers learn language because it’s the commentary to a situation they already understand.  (To put it another way, if you leave the TV on, they won’t learn about elections or American Idol.  There’s no attentional frame to give them a handle on the words from the TV, so they don’t learn anything from it.)  A child won’t learn ‘rabbit’ from a random act of pointing.  They learn the word in a familiar, information-rich context: playing with a pet, visiting a zoo, reading a book, whatever.  They pretty much already understand what the adult is doing and what the utterance means, and they can use that to figure out what any unfamiliar words mean.

Bag those trees

Tomasello rejects generative grammar and formal linguistics entirely.  The language organ hypothesis posits that children have a full adult understanding of grammar and only need to learn how to activate it.  This just doesn’t match the years-long struggle children have to acquire language and the mistakes they make.

How do they acquire language?  Item by item— and the items may be words, phrases with open slots, or entire constructions (e.g. passive voice).  The evidence is that they don’t learn to link up these items right away.  E.g., learning the verb hit, they don’t really have a concept of the verb’s subject and object.  They learn the word’s particular slots: hitter and hittee.  It’s only much later that they abstract out general syntactic categories like subject; and particular items may indeed remain as anomalies in adult speech. 

As an example, generative grammar treats questions as a transformation of statements… “Where’s the rabbit?” is related to locatives like “The rabbit is in the cage.”  But Tomasello points out that for many children, the first multi-word constructions they produce are questions: where X, what’s X?  They can hardly be transforming statements when they’re not producing statements yet.  Rather, they learn the questions because they hear similar questions from adults. 

Some aspects of language are delayed because they require more cognitive sophistication.  The proper use of pronouns and definite articles, for instance, requires an understanding of what other people know.  Young children use these features based only on what they themselves know.  There are items and constructions that aren’t mastered until well unto school age.

Further research needed

The book starts with words and simple constructions, and progresses to more complicateed ones.  It gets weaker as it goes on, not because Tomasello’s argument declines, but because the research gets thinner.  There just aren’t enough studies of how children learn the more complex constructions of their language.

Still, his usage-based linguistics is perhaps the first overall theory of language that strikes me as being on the right track in general.  He rejects generative syntax and innate linguistic competence entirely, and that may be going too far.  But as a heuristic, it’s completely correct: we should explain as much as we can with general cognitive abilities before positing language-specific ones. 

Chomskyan linguistics in particular seems like arid speculation verging on pseudo-science.  The whole idea of parameters, for instance, is an invitation to fool oneself: any anomalous data can be swept under the rug by adding a new parameter.  The best alternatives so far have been people who are more sensible (e.g. Lakoff and McCawley) but who still are very far from the neurochemistry of the brain.  I’ve long felt that we won’t be getting near the truth till linguistics is a lot more like color theory: read Hardin’s Color for Philosophers and note how much vision and color perception derive directly from facts about neurons.

Tomasello isn’t at the neural level yet, but he deals refreshingly in facts, both facts about child language acquisition and facts about human cognitive development.  It doesn’t exactly liven up the book, but a few decades of this and I think we’ll get a whole lot closer to exactly how we do this language thing we do.