On the Nature of Syntactic Variation: Evidence from Complex Predicates and Complex Word-Formation

The existence of substantive parametric variation in syntax, as characterized in Chomsky 1981, has been questioned in the more recent generative literature, notably in Borer 1984, Fukui 1986, and Chomsky 1993. This article provides converging evidence from child language acquisition and comparative syntax for the existence of a syntactic parameter in the classical sense of Chomsky 1981, with simultaneous effects on syntactic argument structure (e.g., verb-particle constructions) and complex word-formation (root compounding). The implications are first that syntax is indeed subject to points of substantive parametric variation as envisioned in Chomsky 1981, and second that the time course of child language acquisition is a potentially rich source of evidence about the innate constraints on language variation.*

1. Theoretical background

A central question for syntactic theory is whether crosslinguistic variation is a ‘deep’ domain of inquiry—in other words, a domain in which general, explanatory principles are operative. The question is logically independent of the existence of substantive universals of human language. In principle, points of syntactic variation could be limited to superficial, listed idiosyncrasies within an otherwise invariant and richly structured language faculty.

1.1. The nature of syntactic variation

The principles-and-parameters framework introduced in Chomsky 1981 permitted the statement of highly abstract constraints on crosslinguistic variation (in the form of parameterized principles of Universal Grammar, or UG), as well as “absolute” universals of human language (in the form of unparameterized UG principles). Indeed, early research in the P&P framework led to the proposal of a number of parameterized principles, each permitting two or more distinct parameter settings with broad consequences for the surface characteristics of the resulting grammar. Travis 1984, for example, proposed parameterized principles of head government and theta government, whose interaction with independently motivated universals of government theory accounted for complex patterns of crosslinguistic variation in word order.

More recent research within the generative tradition, however, has called into question the existence of broad, parameterized principles of the kind envisioned in Chomsky 1981. Notably, Richard Kayne (1984) and Luigi Rizzi (1982) have emphasized the importance of microparameters in accounting for the syntactic variation observed across closely related Romance dialects. Proposed microparameters are typically expressed in terms of lexically listed, morphosyntactic requirements of functional heads. The recent emphasis on microparameters, in combination with various empirical and conceptual problems that were discovered in earlier macroparametric proposals, has led many generativists to doubt the existence of genuine macroparameters of syntax (see among others Rizzi 1989 on the subjacency parameter).

A competing view has been championed by Mark Baker in his 1996 work on polysynthetic languages. Baker argues that both macroparameters and microparameters are needed to account for observed patterns of crosslinguistic variation. Baker attributes the greater explanatory role of microparameters in the research on Romance dialects to a methodological artifact: the operation of macroparameters is evident only if one compares genetically and typologically diverse languages, because closely related languages tend to be extremely similar in their macroparametric choices.

Questions of methodology are critically important for the additional reason that macroparameters, by definition, are more abstract than the surface characteristics of grammar that they help to determine. Thus, macroparameters are relatively unlikely to be discovered by simple crosslinguistic comparisons of surface properties, in the absence of a larger theoretical framework derived from fine-grained analyses of individual languages.¹ For example, the operation of a macroparameter could easily be obscured by the existence of two distinct arrays of parameter settings, each of which gives rise to similar surface constructions: a given language could present a spurious counterexample to a valid macroparametric generalization, by allowing (the semblance of) a particular surface construction without the predicted grammatical concomitants.

Hence, the present paucity of convincing macroparametric analyses may well reflect the limited number and variety of languages for which there exist detailed, theoretically sophisticated grammatical analyses; or indeed may reflect more general deficiencies in the grammatical framework theories that are currently available. To circumvent the limitations of a purely comparative approach, I add a novel source of evidence: the time course of child language acquisition.²

1...