Linguistic intuitions and the puzzle of gradience

This chapter addresses the question of whether gradience in acceptability should be considered evidence for gradience in grammar. Most current syntactic theories are based on a categorical division of grammatical versus ungrammatical sentences. In contrast, acceptability intuitions, that is, the data used to build those theories, have long been recognized to be gradient. The chapter presents two experiments collecting acceptability ratings for 100 sentences extracted from papers published in Linguistic Inquiry. The results show a gradient pattern. It is argued that this gradience in acceptability is highly unlikely to be due to methodological and other known extra-grammatical factors. Unless another factor can be identified, it seems reasonable to assume that the observed gradience comes (also) from the grammar. Furthermore, the chapter presents a proposal concerning diacritics, according to which the traditional asterisk is reserved for ungrammaticality only, and a new diacritic (“^”) indicates reduced acceptability.

Acceptability ratings cannot be taken at face value

This chapter addresses how linguists’ empirical (syntactic) claims should be tested with non-linguists. Recent experimental work attempts to measure rates of convergence between data presented in journal articles and the results of large surveys. Three follow-up experiments to one such study are presented. It is argued that the original method may underestimate the true rate of convergence because it leaves considerable room for naïve subjects to give ratings that do not reflect their true acceptability judgments of the relevant structures. To understand what can go wrong, the experiments were conducted in two parts. The first part had visually presented sentences rated on a computer, replicating previous work. The second part was an interview where the experimenter asked the participants about the ratings they gave to particular items, in order to determine what interpretation or parse they had assigned, whether they had missed any critical words, and so on.

Can we build a grammar on the basis of judgments?

This chapter is a contribution to the ongoing debate about the necessary quality of the database for theory building in research on syntax. In particular, the focus is upon introspective judgments as a data type or group of data types. In the first part, the chapter lays out some of the evidence for the view that the judgments of a single person or of a small group of people are much less valid than the judgments of a group. In the second part, the chapter criticizes what the author takes to be overstatements and overgeneralizations of findings by Sprouse, Almeida, and Schütze that are sometimes viewed as vindicating an “armchair method” in linguistics. The final part of the chapter attempts to sketch out a productive route forward that empirically grounded syntax could take.

Linguistic intuitions again

Linguistics takes speakers’ intuitions about the syntactic and semantic properties of their language as good evidence for a theory of that language. Why are these intuitions good evidence? The received Chomskyan answer is that they are the product of an underlying linguistic competence. In Devitt’s Ignorance of Language, this Voice of Competence answer (VoC) was criticized and an alternative view, according to which intuitions are empirical theory-laden central-processor responses to phenomena, was defended. After summarizing this position, the chapter responds to Steven Gross and Georges Rey, who defend VoC. It argues that they have not provided the sort of empirically based details that make VoC worth pursuing. In doing so, it emphasizes two distinctions: (1) between the intuitive behavior of language processing and the intuitive judgments that are the subject of VoC; and (2) between the possible roles of structural descriptions in language processing and in providing intuitions.

A defense of the Voice of Competence

Michael Devitt has argued against the view that the intuitive verdicts on which linguists routinely rely result from a special “Voice of Competence” (VoC) whereby the spontaneous intuitions of native speakers provide special evidence of their internal representation of the phonology and syntax of their I-language. This chapter defends VoC by analogizing it to the spontaneous reactions of subjects in vision experiments that provide special evidence of the representations in their visual systems, an analogy for which there is substantial experimental evidence but that requires a number of distinctions that Devitt overlooks, e.g. between a grammar and a parser and between conceptual and non-conceptual content. Moreover, it is argued that his alternative model of speakers’ reactions in terms of sentences merely “having” rather than representing properties fails to explain how those properties might be integrated into a speaker’s psychology so that that speaker ineluctably “hears” her language as language, with all the constraints that language imposes.

Semantic and syntactic intuitions

The chapter argues that so-called semantic and syntactic intuitions, as appealed to as evidence in linguistic theory, are two sides of the same coin. Speaker-hearers do not have direct intuitive access to linguistic facts, but instead have intuitions about what can be said with a sentence; it is the theorist’s task to sieve such coarse-grained reports to discover the contribution of semantic and syntactic principles to the content associated with a sentence. The chapter contends that this conception better accommodates the complexity of linguistic phenomena vis-à-vis intuition. It also shows how the position is defensible against two troubling linguistic phenomena: one where we find interpretation without grammaticality, and another where we find grammaticality without interpretation. Both cases are illusory, at least as so described. In both, grammaticality (syntactic soundness) is aligned with semantic interpretability.

Do generative linguists believe in a Voice of Competence?

For more than a decade, there has been lively debate on the question of why we are justified in using acceptability judgments as evidence in linguistics. More surprisingly, there has also been extensive debate over what the received view on this issue is among generative linguists. This second question has received as much attention as the justification question, if not more; yet no agreement has been reached so far. The voices of generative linguists themselves have been largely missing in these debates. Do generative linguists in fact believe in a Voice of Competence, as Devitt suggests? This chapter presents the results of a study that surveyed generative linguists to answer just that question. These results, overall, offer evidence against the hypothesis that VoC is the received view in current generative linguistics.

Introduction

The introductory chapter presents the current status of the debate concerning linguistic intuitions, starting with their early use in the Chomskyan tradition. It introduces the two main questions discussed in the volume: the justification question, which asks for a theoretical rationale for using linguistic intuitions as evidence in the study of language; and the methodology question, which asks whether formal methods of gathering intuitions are epistemically and methodologically superior to informal ones. The introduction also provides summaries of the remaining chapters and explains how they contribute to the debates raised by these two questions.

A user’s view of the validity of acceptability judgments as evidence for syntactic theories

The primary goal of this chapter is to discuss the validity of acceptability judgments as a data type. The author’s view is that acceptability judgments have most, if not all, of the hallmarks of a valid data type: syntacticians have a plausible theory of the source of acceptability judgments, a theory of how to leverage judgments for the construction of syntactic theories using experimental logic, and a set of evaluation criteria that are similar to those used for other data types in the broader field of psychology. At an empirical level, acceptability judgments have been shown to be relatively reliable across tasks and participants, to be relatively sensitive, and to be relatively free of theoretical bias. Therefore the author’s view is that acceptability judgments are at least as valid as other data types that are used in the broader field of language science.

The relevance of introspective data

Introspective judgments of acceptability have long been criticized for being both inconsistent and irrelevant. A number of publications have addressed the former issue and have argued that such judgments, carefully collected, are generally consistent. It remains the case, however, that many linguists question whether introspective data are or can be relevant to the construction of the correct theory of language. In the view of many usage-based grammarians, the sentences made up by analysts rather than real-life utterances lead inevitably to the supposedly unrealistic and complex abstract structures posited by generative grammarians. The chapter challenges that view. Appealing to a 170 MB corpus of conversational English, it argues that introspective data and conversational data do not lead to different conclusions about the nature of linguistic theory.