Band 50, Heft 1-2 - Reflections on Theoretical Linguistics; Issue Editors: Hans-Martin Gärtner, Manfred Krifka

In a pioneer paper, Featherston (Featherston, Sam. 2007. Data in generative grammar: The stick and the carrot. Theoretical Linguistics 33. 269–318) advocated the use of better controlled data in theoretical linguistics. Despite diverging on many aspects, most syntactic theories are now testing their hypotheses with more data than a few linguists’ intuitions. I will examine the consequences of this empirical turn on two syntactic phenomena: long-distance dependencies (LDD) and ellipsis. In a series of recent experiments (Liu, Yingtong, Elodie Winckel, Anne Abeillé, Barbara Hemforth & Edward Gibson. 2022. Structural, functional and processing perspectives on linguistic islands effects. Annual Review of Linguistics 8. 495–525), most of the syntactic constraints (‘island constraints’) on LDD have shown less crosslinguistic variation and more cross-construction variation than previously thought. Corpus and experimental data have also shown elliptical clauses to be more flexible than expected under deletion-under-identity theories (Poppels, Till. 2022. Explaining ellipsis without identity. The Linguistic Review 39. 341–400). These are challenges for most syntactic theories, which call for taking discourse factors more seriously into account.

Linguistic typology is an all-embracing discipline central for inductively-based cross-linguistic generalizations, supported by language facts. Firsthand investigation of previously undescribed languages from regions known for their linguistic diversity helps expand our knowledge about the nature of language and the parameters of cross-linguistic variation. We explore the options of marking commands in a non-main clause and the issue of associative non-singular number in Yalaku and in Manambu (from Papua New Guinea), before turning to nominal aspect and non-propositional evidentiality (exemplified with Tariana and Jarawara, from Brazilian Amazonia). Previously undescribed languages help typologists expand and test our analytic frameworks

Large language models are better than theoretical linguists at theoretical linguistics, at least in the domain of verb argument structure; explaining why (for example), we can say both The ball rolled and Someone rolled the ball , but not both The man laughed and * Someone laughed the man . Verbal accounts of this phenomenon either do not make precise quantitative predictions at all, or do so only with the help of ancillary assumptions and by-hand data processing. Large language models, on the other hand (taking text-davinci-002 as an example), predict human acceptability ratings for these types of sentences with correlations of around r = 0.9, and themselves constitute theories of language acquisition and representation; theories that instantiate exemplar-, input- and construction-based approaches, though only very loosely. Indeed, large language models succeed where these verbal (i.e., non-computational) linguistic theories fail, precisely because the latter insist – in the service of intuitive interpretability – on simple yet empirically inadequate (over)generalizations.

We review some of the main goals of theoretical linguistics in the tradition of Generative Grammar: description, evolvability and learnability. We evaluate recent efforts to address these goals, culminating with the Minimalist Program. We suggest that the most prominent versions of the Minimalist Program represent just one possible approach to addressing these goals, and not a particularly illuminating one in many respects. Some desirable features of an alternative minimalist theory are the dissociation between syntax and linear order, the emphasis on representational economy (i.e. Simpler Syntax) and an extra-grammatical account of non-local constraints (e.g. islands). We conclude with the outline of an alternative minimalist perspective that we believe points to more satisfactory accounts of the observed phenomena.

In recent decades, semanticists moving beyond truth conditional meaning and sociolinguists moving beyond correlations with social structural categories have developed a shared interest in what is commonly called social meaning. This raises the possibility of a comprehensive theory of meaning, moving away from the exclusively propositional model of grammar, and treating language not just as structure but as practice. The paper briefly traces the emergence of social meaning in sociolinguistics, and discusses some issues involved in a broader theory, such as what counts as social meaning, the notions of context and intention, and the need to adopt the Peircian sign.

Some recent publications have made the suggestion that Large Language Models are not just successful engineering tools but also good theories of human linguistic cognition. This note reviews methodological and empirical reasons to reject this suggestion out of hand.

Given the centrality of partial predictability to linguistic experience, it plays a strikingly minor role in theoretical linguistics. For many, partial predictability is to be set aside: the job of linguistic theory is to explain the infinite generative capacity of language and the semantic compositionality that accompanies it. For others, partial predictability is evidence that such an approach is missing the point. But surprisingly little attention is devoted to understanding how partial predictability actually works. We argue that linguistic theory should recognize partial predictability as a central design feature of human language, and propose a strategy for doing so.

In this article, I briefly explore how theoretical linguistics and philosophy are interconnected. I focus on three possibilities, and argue that the fields are most harmonious when utilised in critical reflection of a particular target, a format officially adopted in Theoretical Linguistics since 2002.

After more than sixty years of research, it is now widely accepted that sign languages are real languages, sharing key properties with spoken languages. This means that spoken and signed languages together comprise one natural language system in some sense. But that is not the whole story. Here I probe more deeply into the two systems, and focus on the differences between them -- differences that are pervasive, systematic, and predictable. Taking the existence of two identical articulators in sign languages, the two hands, as a case in point, I show how the physical channel of transmission profoundly influences linguistic structure. Further support for the characterization of language proposed here, different systems in the same faculty, comes from the newly emerging sign language of the Al-Sayyid Bedouins. The Whole Human Language can only be fully understood by admitting and elaborating two types of language in one language faculty, and by acknowledging the fundamental role of the body in determining language form.

This paper highlights a small selection of cases where cross-linguistic insights have been important to big questions in the theory of semantics and the syntax/semantics interface. The selection includes (i) the role and representation of Speaker and Addressee in the grammar; (ii) mismatches between form and interpretation motivating high-placed silent operators for functional elements; and (iii) the explanation of semantic universals, including universals pertaining to inventories, in terms of learnability and the trade-off between informativeness and simplicity.

This paper first introduces the standard recipe for deriving quantity implicatures in the neo-Gricean framework. Then, it compares this pragmatic stance with the grammatical view that argues that scalar implicatures should be generated via an operator in syntax. After showing how the grammatical view can derive canonical scalar implicatures, motivations for this view are discussed which include embedded implicatures, obligatory scalar implicatures concerning the Hurford Constraint, and Free Choice inferences. This paper finally examines basic tenets of the grammatical view and points out three potential problems for this approach.