Colin Phillips (University of Maryland)

Abstract: Our team focuses primarily on the moment by moment grammatical processes that underlie linguistic judgments of acceptability and interpretation. But in the course of doing this, we also run a lot of medium-scale acceptability rating experiments. The results are almost always pretty much what you’d expect, and they do not motivate the kind of hand-wringing that one finds in meta-linguistic debates. We also measure a lot of online and implicit correlates of linguistic judgments, using tools such as reading times, eye-tracking, electrophysiology, and speech rate analyses. Here also, the results generally align very well with untimed acceptability judgments. However, there are systematic cases where we find mismatches between rapid responses and untimed judgments (“grammatical illusions”), and we have investigated these in detail. In this talk I summarize our current understanding of the systematic properties of these mismatches, and what they tell us about how to understand standard acceptability intuitions, whether from experts or non-experts.

John Collins (University of East Anglia)

Abstract: Virtually all of the methodological debate on the evidential status of intuitions in linguistics has focused on so-called ‘syntactic intuitions’ ('acceptability judgments'). The paper offers a simple model of the evidential status of a class of semantic (not conceptual!) intuitions, and shows how such a model readily explains why intuitions in this department are evidential of linguistic facts. The paper then shows how the model transfers to cover (almost) all syntactic intuitions.

Anna Drożdżowicz (Aarhus University)

Abstract: Speakers’ intuitive judgements about meaning are taken to provide important data for many debates in philosophy of language and pragmatics. Is their use justified? While many embrace their use (e.g. Neale, 2004; Recanati, 2013), others have raised worries about their nature and reliability (Devitt, 2012; 2013). We are still lacking a systematic account that would explain the nature and evidential status of such intuitive judgements and address these worries. In this talk I propose such a positive account. I argue that speakers’ intuitive judgements about meanings of utterances provide valuable evidence for various debates in philosophy of language and pragmatics because they are importantly (causally) connected to the linguistic performance of speakers. Arguably, the connection is reliable due to a constant monitoring of the interpretations ascribed to linguistic utterances. The proposal draws on recent work in psychology of language. I call this the voice of performance view.

Jon Sprouse (University of Connecticut)

Abstract: Acceptability judgments are robust and replicable, at least when the question is whether there is a contrast between two (or more) sentence types. Because contrast-logic is the primary method in linguistics (and cognitive science more generally), this means that acceptability judgments are a robust and replicable data type for constructing contrast-based linguistic theories. But as many linguists have noted (including many of the participants in this workshop), there is information in acceptability judgments beyond simple contrasts. In this talk I'd like to point to some of this information, and discuss the ways that linguistic (and psycholinguistics) theories might evolve to begin to explain these other dimensions of the data. My hope is that together we can come up with an agenda for the next phase of research into the relationship between acceptability judgments and grammatical theories.

Georges Rey (University of Maryland)

Abstract: I defend a version of what Michael Devitt (2006) has called the “Voice of Competence” (“VoC”) view of linguistic intuitions, according to which at least the spontaneous intuitive reactions of native speakers of a language provide special evidence of their competence with the phonology and syntax of their I-language (I make no claims about semantics or pragmatics). Addressing objections and alternatives to the view raised by Devitt and, independently, by Peter Ludlow, I’ll defend a straightforward perceptual model of a VoC, according to which the intuitive reactions are caused relatively directly by non-conceptual structural descriptions of the relevant categories. I’ll supplement my earlier discussions of the topic by citing still further experimental evidence for it regarding, e.g, involuntary perception of native speech, its rhyme and register; “garden paths,” syntactic priming, “slips of the ear,” and syntactic “nonsense” (as in Lewis Carroll and Derridas). Devitt seems to claim that such phenomena can be explained merely by representations of semantic messages having syntactic properties, rather than representing them, but I argue he hasn’t yet shown this explanation to be a serious contender, particularly in cases of these phenomena that are perceptually vivid and trump plausible intended messages. A representation’s merely having certain, e.g., neural properties, after all, doesn’t explain why those properties should be perceptually real.

Steven Gross (Johns Hopkins University)

Abstract: Michael Devitt (2006a, 2006b, 2010, 2013) has issued a distinctive challenge to the value of intuitions as evidence in linguistics, one based on a hypothesis concerning their etiology—viz., that a linguistic intuition is a theory-laden “central systems” response, rather than the output of a modularized language faculty (the “Voice of Competence,” or VoC). This paper distinguishes Devitt’s challenge from others, explains its relation to his conception of linguistics more generally, and offers a novel reply. Whereas in previous work (Culbertson and Gross 2009, Gross and Culbertson 2011, Maynes and Gross 2015), the author has argued both that Devitt’s etiology hypothesis should be rejected and that his opponents need not commit themselves to VoC (i.e., Devitt’s option are not exhaustive), here it is argued that VoC may well be true for some linguistic intuitions. More specifically, Devitt’s challenge is used to frame and motivate the suggestion that error signals—of the kind explored in investigations of speech monitoring—may play a role in generating some linguistic intuitions. Along the way, it is emphasized that linguistic intuitions may not constitute a natural kind with a common etiology, and the view developed is compared with the broadly consonant views of Georges Rey (Rey 2013, forthcoming).

Karen Brøcker (Aarhus University)

Abstract: In the past decade, there has been a lively discussion on the question of why we are justified in using syntactic acceptability intuitions as evidence in linguistics (Devitt 2006, Culbertson and Gross 2009, Textor 2009, Rey 2013 etc.). More surprisingly, there has also been an extended debate over which view on the nature of these intuitions is the most common among generative linguists. This second question has received as much, if not more, attention than the justification question, and so far no agreement seems to have been reached. The voices of generative linguists have been largely missing in these debates. Of course, the answer to the justification question does not depend on how generative linguists in fact view syntactic intuitions. However, answering this question would bring focus back to the issue at the centre of the debate, the justification question itself. A better understanding of the common generative view on syntactic intuitions would also shed light on other interesting questions, for instance, to what degree is the so called expertise defence applicable to the common view?  And how can the use of syntactic intuitions as evidence be justified on that view? To answer the question about what generative linguists in fact think about the intuitions they use as evidence, I designed a questionnaire based closely on the views found in the debate and distributed it to linguists through LinguistList. In this talk, I present the first results of that study.

Frederick J. Newmeyer (University of Washington, University of British Columbia, and Simon Fraser University)

Abstract: Introspective judgments of acceptability have long been criticized for being both inconsistent and irrelevant. A number of publications, beginning perhaps with Cowart (1997), have addressed the former issue and have argued that such judgments, carefully collected, are generally consistent. It remains the case, however, that many linguists, typically those with a ‘usage-based’ orientation, question whether introspective data is or can be relevant to the construction of the correct theory of language. In their view, the ‘disembodied sentences that analysts have made up ad hoc, … rather than utterances produced by real people in real discourse situations’ (Tomasello) lead inevitably to the supposedly unrealistic complex abstract structures posited by generative grammarians. Usage-based grammarians assert that if one focuses on naturally occurring discourse, then grammar will reveal itself to be primarily a matter of memorized formulas and simple constructions. This paper challenges that view. Appealing to a 170MB corpus of conversational English, it argues that introspective data and conversational data do not lead to different conclusions about the nature of linguistic theory. 

Benjamin Bruening (University of Delaware)

Abstract: I present three cases studies where the field has been led astray by incorrect judgments of ungrammaticality.  In all three, it took corpus searches to reveal that the construction at issue was in fact grammatical. The first case involves adjectival passives.  Since Wasow (1977), adjectival passives were universally regarded as being ungrammatical with raising to object verbs.  Bruening (2014, NLLT) found numerous attested examples of raising to object in adjectival passives, contradicting all previous work.  Out-of-the blue constructed examples are consistently judged unacceptable, but minimally different examples in a context are acceptable. The second case study involves raising to subject and object inside nominalizations.  Chomsky (1970) claimed that nominalizations may not include raising.  In new work, I show that raising to subject and object are both attested inside nominalizations in electronic corpora.  A survey of acceptability using Amazon Mechanical Turk finds that naive speakers of English judge them to be acceptable. The third case study involves the VP anaphor *do so*, which has always been judged to be ungrammatical in a passive form (Bouton 1969, Hallman 2004, Houser 2010).  Once again, examples of passive *do so* can be found in electronic corpora.  Another survey using Amazon Mechanical Turk finds that passive *do so* is just as acceptable as active examples.  In all three cases, incorrect judgments of ungrammaticality have led the field in unwarranted directions.  In all three cases, the judgments were only found to be erroneous by consulting corpora.  Since context and word choice can greatly affect acceptability, it is important to look for naturally occurring examples of any construction that is claimed to be either grammatical or ungrammatical.

Ken Ramshøj Christensen (Aarhus University)

Abstract: There is an ongoing debate about the validity of expert intuition in generative syntax. I present data from experimental syntax showing that intuition is a valid tool but one which requires a quantitative approach. Furthermore, I present data showing that quantitative intuition data can be used address otherwise counter-intuitive interpretations of so-called linguistic illusions. While it is intuitively true that language usually makes sense, that it is meaningful, it is not always true. During parsing, we make intermediate semantically anomalous interpretations. Crucially, though, such counter-intuitive interpretations are only made if it does not violate the syntactic structure. This is a first indication that naïve intuition about the nature of language cannot be trusted. Furthermore, we can be systematically tricked by certain syntactic constructions, sometimes into believing that certain sentences that are meaningless are actually meaningful; at other times, we consistently misinterpret sentences that turn out to be very complicated. Because people disagree on the interpretations as well as on the acceptability of such examples, these counter-intuitive findings are only accessible with an experimental quantitative approach.

Jana Häussler (University of Wuppertal) and Tom S. Juzek (Nuance Communications)

Abstract: For decades, informal ways of obtaining acceptability judgments dominated syntactic theory, where linguists are their own informants. This has been criticised as substandard. However, Sprouse et al. (2013) presented quantitative results that suggest that informal and formal judgments concur to a large extent. They extracted marked constructions and their unmarked counterparts from the literature and then checked in a formal experiment whether marked items received lower ratings than their good counterparts. However, most authors consider more than just pairs, which should be reflected when comparing informal and formal methods. Consequently, we designed an experiment that tests items at large. We randomly sampled constructions from the literature, 2 × 100 sentences, one set from binary papers, distinguishing only two levels of acceptability (unmarked vs. *), and another from gradient papers, distinguishing more than two levels of acceptability (unmarked, *, and some additional level(s), e.g. ?, ?? and the like). Each set contained 50 *-items and 50 OK-items. We then tested the sentences in a formal experiment, using a 7-point scale. By and large, experimental ratings agree with the authors’ ratings, though not perfectly (rpb=.65). However, threshold tests show a significant number of violations: 14/100 (binary set) and 17/100 (gradient set). Furthermore, ratings cover the whole space rather than clustering at the two extremes of the scale. Bin tests identify 35/100 items (binary set) and 43/100 items (gradient set) in the mid-bin. The high number of violations suggests that there is a non-trivial mismatch between informal and formal results. Furthermore, instead of the S-curve as a noisy approximation of the step-function predicted by the assumption of a binary division (i.e. +/– grammatical), we found an almost linear increase. We conclude that gradience is not an epiphenomenon or artefact. This should be acknowledged in linguistic theory. Formal experiments will help to quantify differences in acceptability and thus should be part of the linguist’s toolbox.

Carlos Santana (University of Utah)

Abstract: I argue for a limited non-evidential justification of the use of native speaker metalinguistic judgments (intuitions) as data in linguistics. On my account, intuitions are not a source of scientific evidence, but can be a justified means of appealing to shared background theory. None of the leading defenses of construing acceptability judgments and truth-condition/entailment judgments as evidence, I argue, are supported by the preponderance of empirical and philosophical evidence. Appealing to a Quinean picture of scientific knowledge, I argue that having a shared background theory is necessary to the advancement of a research program. Intuitions in linguistics serve the function of delimiting what belongs to the shared background, and which questions are currently under the microscope. For intuitions to serve this purpose, however, they must be broadly shared. Data on acceptability judgments, particularly the experiments of Sprouse and Almeida, indicate that acceptability judgments usually qualify as broadly shared in this sense, thus justifying their use in linguistics. I present experimental data of my own on truth-condition judgments, however, which shows that semantic intuitions are not shared to the same extent, and so may not be justified in the same way.