Continuous Acceptability, Categorical Grammaticality, and Experimental Syntax

Abstract

1. Introduction It almost goes without saying that acceptability judgments form a continuous spectrum. While many sentences are either clearly acceptable or clearly unaccept-able, a significant number of sentences fall somewhere in between in a gray area of partial acceptability. This fact has been explicitly admitted by linguists since at least Chomsky 1965. The working assumption adopted by most linguists over the past 40 years has been that these intermediate levels of acceptability are caused by properties other than grammatical knowledge. Linguists have assumed that grammatical knowledge is categorical — sentences are either grammatical or ungrammatical — and that the continuous spectrum of acceptability is caused by extra-grammatical factors (plausibility, working memory limitations, etc.). Of course, ideas such as strength of violation have been introduced into theories of grammatical knowledge at various times, for instance Huang's (1982) proposal that ECP violations are stronger than Subjacency violations or Chomsky's (1986) proposal that each barrier crossed leads to lower acceptability. However, with the notable exception of Optimality Theory (see especially Keller 2000, 2003), these analyses have been the exception rather than the rule. The past ten years or so have seen a major shift in attitudes toward intermediate levels of acceptability. With the increasing use of formal experimental methods for measuring acceptability — a methodology that has come to be known as experimental syntax — it has become almost trivial to detect subtle differences along a continuous spectrum of acceptability. This new power afforded by experimental syntax raises the question of whether the continuity of acceptability reflects a continuity in grammatical knowledge that should be captured by the theory of grammar, or in other words, whether the working assumption of the past 40 years should be abandoned (see especially Keller 2000, Fanselow et al. 2004). While the answer to this question is ultimately an empirical one that is far from being settled, this report presents two pieces of experimental evidence for a categorical distinction between grammatical and ungrammatical sentences. The first is a direct prediction of theories that assume categorical grammaticality. The psychological claim underlying theories of categorical grammaticality is that ungrammatical sentences have no licit representation, or in other words, cannot be constructed from the available mental computations. Grammatical sentences, Biolinguistics  Reports  119 on the other hand, have licit representations that can be constructed from the available mental computations. This predicts that extra-grammatical factors that affect the acceptability and are predicated on the existence of a representation, such as syntactic priming (Luka & Barsalou 2005), should not affect the accepta-bility of ungrammatical sentences. Section 2 presents results from Sprouse (2007) that confirm this prediction: Ungrammatical sentences, in particular island violations (Ross 1967), do not show a structural priming effect. The second piece of evidence comes from the experimental syntax technique magnitude estimation. Unlike traditional tasks such as the yes/no–task and the Likert scale task in which subjects must categorize their responses, magnitude estimation allows subjects to respond using the theoretically infinite continuum of values available on the positive number line (see Bard et al. 1996). By removing the categorization aspect of the task, one might expect that res-ponses would no longer show any categorical distinction between grammatical and ungrammatical sentences. Contrary to this prediction, section 3 presents evidence from Sprouse (2007) that subjects impose a form of categorization on magnitude estimation responses, and that this categorization appears to correspond with the grammatical/ungrammatical distinction. 2. Syntactic Priming and Categorical Grammaticality