A Companion to Chomsky. Группа авторов
later helping to found the field of cognitive science; and with her husband and long‐term collaborator, Henry Gleitman, for over 50 years fostered a continuing research community aimed at answering questions such as: When language input to the child is restricted, what is left to explain language acquisition? The studies reported here find that argument structure encoded in the syntax is key (“syntactic bootstrapping”) and that children learn word meaning in epiphanies (“propose, but verify”).
7.1 Background
7.1.1 Graduate School
From the beginning of my graduate school days I was exposed to the best of the best among linguists: Zellig Harris and his greatest student, Noam Chomsky, who had just moved to MIT when we met. They represent, at least to my understanding, two opposed trends in the study of language and its learner‐users. Harris was the pinnacle of structuralists, {believing} that what you saw (rather: heard) was what you got, and to my mind is the father of “big data” analysis as it appears in the current computational and psychological literature. For him, the sole legitimate approach was relative‐distributional analysis, and his theoretical aim, in these terms, was to find mechanical procedures that moved theory from phoneme to morpheme, thence to phrase, thence to sentence, and finally to connected discourse (Harris 1951, 1952).
One obstacle to identifying principles of coherent discourse was that sentences with the same information content could have very different syntactic structures. {Chomsky's} big step toward a solution was to render all sentences in a normal or “kernel” form, removing optional variations in form and style while preserving content. The relations by which one could get to the kernel forms of sentences were what was first known as “transformations.” One might think of this bottom‐up analysis as a proposal about how language acquisition works. That's how I understood it, but certainly and emphatically not how Harris {did}. He disdained “psychologizing” and meant to describe “language,” rather than the activities or acquirers of such a system.
In my first year of graduate school with Harris, I concretized my own idiosyncratic understanding of the Harrisian point of view in a game – perhaps, more correctly, an evangelical demonstration – which I called The Great Verb Game. It was designed to show that one could recover a specific verb, or small cluster, just from an enumeration of the surface structures in which it appeared. I would think up some “mystery verb” and write it down, covering it. The “players” were to guess it from hearing, essentially, subcategorization frames. So, after a little instruction, the player hears/sees the frame N1 V N2 from N3 and is to come up with an example (say, take: John took candy from the baby). Next the player was offered another such frame, understood to be a paraphrase or entailment of the first, maybe N1 V N3 from N2. Notice that take will not work in this second case ( John took the baby from the candy); it won't preserve meaning or might even be ungrammatical. So the player revises her guess. After a couple more clue frames, such as N2 V from N3, and N1 V N2 and N3 (and if the players are smart undergraduates or other verbal whizzes), they come up with the (or a) verb that survives this winnowing‐down procedure (in the present case, separate). Turning over my card, I now triumphantly show them “my” verb, hoping to convince them that the meanings are embedded in the structure in some way. In my first year or two as Harris's graduate student, I tried to formalize these ideas in terms of some derivational web (“chains of Harrisian transformations”), but this turned out to be a false step. In later years, however, these ill‐formulated but interesting ideas returned to me in the context of studies in word‐meaning acquisition, as I'll mention. But in my first years, this bit of verbal legerdemain so amused audiences that I think this is what caused Swarthmore College to give me my first job.
For the most part, though, my views were rapidly changing because of contact with, and reading of, Noam Chomsky (Chomsky 1957, 1959, 1965). His thinking and findings began to suggest to me, and to a large following of younger scholars who began to surround him, that these surface sentences at the core of Harris's thinking were mere artifacts of an underlying innate system that was only indirectly reflected in our everyday speech forms. Even more importantly, I became convinced of that approach, bolstered by the preliminary findings emerging from my (and collaborators') earliest studies of language in infants and young children. From an autobiographical point of view, I was dragged kicking and screaming toward Chomsky's view, which offended my innately empiricist leanings. Anyhow, I'll now try to say something about the work I did with many brilliant younger collaborators and where it has led me over the years.
7.1.2 What Do They Know and When Do They Know It?
Few, if any, humans behave like the ideal grammarians that these theories – even Harris's – purport to describe. First there is the question of what babies and toddlers “know” about language despite being unable to understand or produce much of the speech to which they are being exposed. The psychological literature, as it looked when I first examined it, described at most what children actually produced, as a function of age: so and so many words at age one, more at age two, and so forth; one‐word utterances at first (“Momma”) and two at a time later on (“Mommy sock”). In the early sixties, with the advent of Chomsky's first writings on the psychology of language, psychologists began to write little “grammars” of the child's primitive speech (see, e.g. Brown et al. 1968; Rodd and Braine 1971). But what is the status of these descriptions? Is children's speech, thus chronicled and organized, a reasonable proxy for what they know about their language – or about Language?
With Elizabeth Shipley, my brilliant colleague in mathematical psychology, I set out to probe the toddlers further, asking them, in effect, “Is the following sentence grammatical in your infant dialect?” To be clearer, we tried to find out if Mommy sock was an output fairly representing their linguistic knowledge at time of test. After all, when linguists try to build grammars, they never simply use actual speech as the data, but rather, or in addition, use direct or indirect judgments of grammaticality, aiming to get at what constitutes “your language,” discarding speech errors, and so forth.
Our finding, in brief (Shipley et al. 1969), was that every kid who spoke in two‐word sentences like Throw ball acted as if the same two‐word sentences from the mother to them were abnormal – they didn't obey those commands, but acted as if something bizarre had happened, whereas grammatical sentences like Throw me the ball, they simply obeyed.
This raises a second question. It may be [and Henry Gleitman and I showed, at some length (Gleitman and Gleitman 1979)] that you can't get these judgments in any coherent or full way from even the mature speakers who, if anybody, are the targets that the theory of grammar is, or should be, designed to organize and explain. Most people can't even tell you how black birdhouse, blackbird house, and house bird‐black differ in their meaning (that is, in the paraphrases they license). A question we raised concerned how a grammar built from judgments of one population – let us say, literary or otherwise book‐wise speakers of some standard dialect – can be used as the empirical basis for building a theory of all native users. Well, we did convince ourselves, in the end, that judgment‐giving itself, while useful if available, was not the only relevant source of data.
Yet a variety of uses of language, many of them important, are well correlated with the ability to give linguistic judgments – that is, to contemplate language more or less explicitly, even consciously. As we later showed with several other collaborators, including Paul Rozin and Kathy Hirsh‐Pasek (Gleitman and Rozin 1977; Hirsh‐Pasek et al. 1978), there is a cluster of skills including learning (or inventing!) an alphabetic writing system, making and understanding puns, and the like that are predicted by this “metalinguistic awareness” (Gleitman and Gleitman 1979). Something like these early generalizations is reminiscent of recent studies and approaches now called executive function, which also attempt to describe and explain certain users' access to aspects of their own cognitive functioning.
7.2 A First Look at the Input to the Child: Non‐Effects of Motherese
Now we move to the story of what I did in language