Linguistic theory and language acquisition
This commentary concerns the relation between linguistic theory and psycholinguistic studies of child language acquisition. Crain’s elegant experiments ably demonstrate how linguistic theory provides concepts, categories, and data that can be fruitfully explored in language acquisition studies. The question that remains to be answered is what these studies tell us about linguistic theory—that is, about the nature of the language faculty. I will try to formulate a tentative answer on the basis of a more detailed discussion of “the parade case” for postulating innate linguistic constraints: the structure-dependence of grammatical rules.
The discussion of the structure-dependence of rules contains a speaker’s knowledge of the status of the following paradigm (from the original presentation in Chomsky 1971b, pp. 25–28):
(In the question examples the italicized copula is interpreted in terms of the underlined empty position in the sentence.) Assuming that the question sentences are derived by a transformational movement of the copula from the empty underlined positions to the sentence-initial position, the argument for the structure dependence of rules is given as follows. A child exposed to data in (1) has at least two options for postulating a transformation to derive (1b):
Option (3b) is a structure-dependent formulation because it requires a hierarchical analysis of the sentence to identify the subject NP adjacent to the copula. In contrast, (3a) is structure-independent because it relies solely on the analysis of the sentence as a linear string, that is, it makes no reference to hierarchical structure.
As (2) illustrates, more complex constructions (in this instance a subject that contains a finite relative clause) show that the structure-independent rule makes the wrong predictions. Taking (2a) as the underlying structure for (2b,c), the rule (3a) would generate the deviant (2c) and not the well-formed (2b). A child who formulates (3a) should produce sentences like (2c), presumably until explicitly corrected, or presented with sentences like (2b), which falsify the hypothesized rule. Because children do not make these errors, an innate constraint on structure-dependence must be guiding their acquisition of the question formation rule.
This account is based on an empirical assumption that children encounter (or pay attention to) simple sentences prior to those with subjects containing finite relative clauses. Whether this is actually the case has not been established. If children hear and process sentences like (2b) along with the simple cases before they formulate the question formation rule, then they have been presented with positive evidence against the structure-independent rule (3a).
Notice that there is yet another rule of question formation a child could hypothesize, namely (4):
This rule differs from the structure-dependent rule to the extent that it does not directly involve an analysis of the sentence in terms of its hierarchical structure as (3b) does with its explicit mention of subject NP. It is also different from the structure-independent rule (3a) because it does not involve a counting predicate like “first” or “left-most.” This counting property of the structure-independent rule explicitly ignores hierarchical structure. Sharpening the distinction concerning structure-dependence, we can take counting to be the salient property of structure-independent rules. Given this, (4) would be classified as a structure-dependent rule, because it is compatible with a hierarchical analysis of the sentence even though it does not directly refer to this analysis.
Rule (4), in contrast to rule (3b), will generate the deviant (2c), so the reason children do not produce such sentences cannot be explained by a constraint on the structure dependence of rules, as Crain’s discussion suggests. The explanation might hold if there were some principle of grammar that prohibited rules of the form (4). This is unlikely, since (4) is simply a more general version of (3b), without the contextual restriction (adjacent to a subject NP).
A transformational movement rule consists of an elementary operation (substitution or adjunction) that can be further specified in terms of a structural description and a structural change. The structural description indicates the special properties of a phrase marker to which the elementary operation applies, and the structural change indicates how the elementary operation modifies the phrase-marker. The elementary operation is part of universal grammar (UG). Any language whose grammar involves movement operations involves one of these elementary operations. The structural description of a transformation is a language-specific condition on the application of the elementary. Current syntactic theory has dropped the statement of these language-specific structural descriptions from the formulation of transformational rules because it has been shown that the behavior of transformations can be accounted for in terms of more general conditions on the output and application of elementary operations—conditions that are also part of UG. From this perspective, rule (4) is just an instance of the general rule (5):
where α stands for any syntactic category (e.g., tensed copula) and “move” is an abbreviation for the elementary operations of substitution and adjunction. More precisely, substitute α for β and adjoin α to β, where α and β are syntactic categories.
In a theory in which rule (5) accounts for question formation in English, a structure dependence constraint on the form of transformations does not explain the deviance of example (2c). Rather, there is a general principle of UG that accounts for the deviance of such constructions. Example (2c) violates the principle known as the Head Movement Constraint (Travis 1984), which restricts the movement of a lexical head to a head position that governs the phrasal projection of the head. In the structure underlying (2b, c), the head position to which the tensed copula moves governs the main clause copula but not the relative clause copula. The Head Movement Constraint also accounts for the deviance of questions in which an auxiliary verb has been moved into sentence-initial position across another auxiliary, as in (6a), in contrast to the well-formed question.
Since this analysis is posited at the level of UG, it is preferable to one that involves a language-specific rule like (3b), all other considerations being equal.
Note that the issue of structure-dependence is still relevant, since it is important to distinguish between rule (5) and structure-independent rules like (3a) that do not occur in the grammars of human languages as far as we know. Given our characterization of the structure-independent rules above, we can eliminate the possibility of such rules by a constraint on the form of transformations that prohibits the counting property. This is achieved by restricting the structural analyses that define transformations to Boolean conditions on analyzability a constraint that was proposed in some of the earliest work on transformational grammar. This restriction excludes quantificational statements that are needed to instantiate the counting property. Thus, we eliminate structure-independent rules by prohibiting quantificational statements in the formulation of transformational rules. In this way the structure-dependence of rules follows from a more specific constraint on the form of rules.
The analysis of structure-dependence given above should be instructive concerning the relation between linguistic theory and psycholinguistic studies of language acquisition. It demonstrates why it is crucial for psycholinguistic research to be informed by current developments in linguistic theory. It also shows that the kinds of studies reported in Crain’s target article do not tell us much about syntactic theory. For example, they do not distinguish between the analysis of question formation as given in rule (3b) as opposed to rule (5) plus the Head Movement Constraint. The relationship between linguistic theory and the kinds of language acquisition studies discussed by Crain therefore seems to be one-sided, in answer to the question raised at the outset. What these studies do show, however, is that quite young children demonstrate the kind of linguistic behavior we would expect if our theory of the language faculty (i.e., an innate UG with very specific properties) is on the right track. In this way they contribute to our understanding of human language and for this reason they merit our careful attention.
