Marrying Boas and Chomsky: Davis, Gillon and Matthewson on “formal” diversity research

I was happy to see the recent methodological article in the (online-only) “Perspectives” section of Language by Henry Davis, Carrie Gillon and Lisa Matthewson: “How to investigate linguistic diversity: Lessons from the Pacific Northwest”. The three authors (henceforth, DG&M) defend the approach of their very interesting work on Salishan, Wakashan and Tsimshianic languages, e.g. on the semantics of determiners and quantifiers. The main point of their paper is that elicitation-based negative evidence is often crucial for discovering the full depth of linguistic diversity, and that therefore generative-style research is superior to non-generative language description and survey-based typological work.

There are many aspects of their rich and provocative paper that are worth discussing (and I have discussed some in my published reply, Haspelmath 2014; see also Dryer 2014 and De Reuse 2014 for further replies), so here I’d like to limit myself to one aspect, the significance of labels for communities of linguists (“C-linguists” vs. “D-linguists”, “formal linguists”) and iconic figures like Boas, Greenberg and Chomsky that different linguists identify with or react negatively to.

I have myself long struggled to reconcile the particularistic Boasian approach (with its emphasis on doing justice to each individual language, rather than pressing them into a preconceived traditional mould) with the universalist Greenbergian approach (with its amazing discoveries of cross-linguistic regularities), but I now believe that the two are not incompatible, though not as relevant to each other as has often been thought: In Haspelmath (2010), I argue that the two approaches require two different sets of notions, Boasian descriptive categories (language-specific) and Greenbergian comparative concepts (cross-linguistically applicable).

But how can one bring Boas together with Chomsky? DG&M say in §8 of their paper that they “draw equal inspiration from Boas and Chomsky” (p. e217), which may appear puzzling at first. The answer turns out to be that they adopt only a few selected aspects of the generative programme, which other linguists would not regard as particularly crucial for pigeonholing their work into the “generative” or even “formal” category.

Surprisingly, DG&M adopt Levinson & Evans’s (2010) distinction between “C-linguists” (C for Chomsky) and “D-linguists” (D for diversity), but they interpret the “C” in a very loose sense. In their footnote 1 (p. e182), they distinguish a narrow and a broad interpretation of “C-linguistics” and say:

“The narrow version is explicitly equated in Levinson & Evans (2010) with the minimalist program by Chomsky (1995 and subsequent work), and therefore excludes most generative fieldworkers. The broad version, by contrast, encompasses a range of theoretical frameworks, which differ (sometimes substantially) in their architecture, but share a commitment to certain methodological principles and practices. These include a theory-driven approach to fieldwork, in which predictions are generated by an explicit formal model and then tested against data from a target language; a commitment to the use of grammatical intuitions, particularly in order to elicit negative data; and a view of the grammar as a set of representations in the minds of individual speakers. It is the latter interpretation of C-linguistics that we defend here.” (emphasis mine)

But the three characteristics of their broad view of “C-linguistics” are not specific to linguistics in the Chomskyan tradition (i.e. generative linguistics, in more common parlance). All good fieldworkers who are interested in grammar use a “theory-driven approach”, which crucially involves the creation of explicit models and the testing of hypotheses (this is emphasized by De Reuse (2014) in his published reply). All good fieldworkers rely to some extent on elicitation, negative evidence and thus on intuitions about well-formedness, and most contemporary linguists have no problem with saying that grammars are (also) mental entities (though this plays little role in their day-to-day work). Thus, interpreted in a broad sense, almost all contemporary grammarians would end up as “C-linguists”. (It is true, though, that many grammar-writers these days emphasize the role of corpora of natural speech, not only because corpora are more readily available, but also because it is often felt that elicitation needs to be complemented by evidence from corpora – a point that DG&M do not deny).

So why is it so important for DG&M to defend “C-linguistics” and “the generative program”? It seems that this has a sociological explanation: They see themselves as members of the social group of “formal linguists”, and in this group, the Chomskyan approach has an enormous prestige, even if it is not really practiced. So when reading Evans & Levinson’s polemical paper, they felt that they were included in the group of “C-linguists”, even though in actual practice they are much more Boasian – thus, their paper is about diversity, not about universals, and they make no attempt to solve the problems that their discoveries create for generative universalists. (This is particularly striking in the case of the Principle C violations, where an account in terms of UG has long seemed plausible to many.)

In theoretical semantics as well as in theoretical syntax, the work of MIT linguists (such as Irene Heim, cf. Heim & Kratzer 1998) has become very influential, and this approach is often called “formal semantics” (somewhat oddly, because there is no relation to linguistic form, as far as I can see). DG&M clearly see their work as belonging to this tradition, and in fact they use the term “formal” three times as often as the term “generative”. They talk about “formal linguists”, “formal theories”, “formal investigation”, and even “formal elicitation” and “formal fieldwork”. I find this way of talking very opaque, and I can only take it to mean that they associate with the formal-semantics and thus generative group. But this is a social fact, with no clear intellectual relevance.

In addition to being critical about Evans & Levinson’s attack on UG approaches, DG&M are also critical of large-scale cross-linguistic work such as WALS. But this is not a specifically Chomskyan (or “formal”) sort of criticism: Large-scale comparativists are often suspected of treating individual languages superficially and of making errors. Oddly, DG&M seem to overlook that grammar-mining comparativists necessarily rely on descriptive grammars written by hypothesis-testing fieldworkers like themselves. But nobody has suggested that large-scale surveys somehow employ a different or better method that can make do without hypothesis-testing and elicitation.

A different question that DG&M do not ask is why linguists who identify with the “formal” or “generative” group rarely write descriptive grammars that could be mined by large-scale comparativists. It seems to me that this is because of the (categorial universalist) assumption that all grammars are ultimately made of the same categories. Evidently we have not found those categories yet, so writing a full grammar would be premature, and we have to be content with writing papers that make proposals about specific phenomena. Boasians are not constrained by this assumption: They can make their own language-specific categories and thus can write full grammars. Thus, in practice it is much more difficult to combine the Boasian and the Chomskyan approach than the Boasian and the Greenbergian approach.


Davis, Henry, Carrie Gillon & Lisa Matthewson. 2014. How to investigate linguistic diversity: Lessons from the Pacific Northwest. Language 90(4). e180-e226. [preprint]

De Reuse, Willem. 2014. How field linguists have been investigating linguistic diversity: Commentary on Davis, Gillon, and Matthewson. Language 90(4). e227–e231.

Dryer, Matthew S. 2014. Competing methods for uncovering linguistic diversity: The case of definite and indefinite articles (Commentary on Davis, Gillon, and Matthewson). Language 90(4). e232–e249.

Evans, Nicholas & Stephen Levinson. 2009. The Myth of Language Universals: Language diversity and its Importance for Cognitive Science. The Behavioral and Brain Sciences 32(5). 429–448.

Haspelmath, Martin. 2010. Comparative concepts and descriptive categories in crosslinguistic studies. Language 86(3). 663-687.

Haspelmath, Martin. 2014. Descriptive hypothesis testing is distinct from comparative hypothesis testing: Commentary on Davis, Gillon, and Matthewson. Language 90(4). e250–e257.

Heim, Irene & Angelika Kratzer. 1998. Semantics in generative grammar. Malden, MA: Blackwell.

Levinson, Stephen C. & Nicholas Evans. 2010. Time for a Sea-Change in Linguistics: Response to Comments on `The Myth of Language Universals’. Lingua 120(12). 2733–2758.

6 thoughts on “Marrying Boas and Chomsky: Davis, Gillon and Matthewson on “formal” diversity research

  1. As one of those who has not written their descriptive grammar, I’m pretty sure that the causes of my deliquency do not include issues with the category system, for two reasons: a) introspection reveals nothing of the kind b) attendance at descriptive seminars at ANU based on field work in various places never reveals any horrible problems with category systems .. ‘is this thing an adjective or a stative verb’ is about as bad as it gets, and that would not cause any generativist to lose sleep and stop writing their grammar.

    I suspect that the reason is more in the direction of wanting to examine limited and apparently connected issues in considerable depth, as opposed to meeting some kind of (often rather low) standard of adequacy about a very wide range of things. Both kinds of work have their place, I suggest.

    • Yes, people often say that the distinction is between “in depth” and “wide range” study of language structures, but that does not explain why most generative papers adopt a very specific formal framework for their in-depth studies. There’s no reason why one should not study a particular phenomenon in a language in depth without assuming that the language has the same sort of categories and mechanisms (vP, Agree, c-command, etc.) that other languages have. So I think the explanatory direction is the other way round: If you are a member of the community that assumes universal categories and mechanisms, then these narrow-range, “in depth” studies are prestigious, because your proposals about modifying the framework might get adopted by others. If you are a member of the community that does not make these assumptions, then writing a grammar is prestigious.

      • I think you’re right about the appeal of the mechanisms, but not about the categories, and making a somewhat wrong cut wrt sociology vs esthetics/psychology; why would anybody with enough brains to get a PhD at all care about the esteem of either kind of community if membership and prestige was what they were interested in? If prestige was the motivation, they wouldn’t be doing linguistics of any kind at all. Rather, I think, some people are inherently very interested in language, find certain ideas and approaches more congenial than others, and gravitate towards communities that practice them.

        It is perfectly possible to be fascinated by the idea of mechanisms behind grammar without really caring that much about the universality claims, but once you understand Chomsky’s ‘Logical Problem of Language Acquisition’ it seems perverse (to me, at any rate) to conjecture that the child decides whether something like the classic ‘Subject to Subject/Object Raising’ phenomena are best accomodated by classic transformations, LFG-style lexically-driven structure sharing, or something completely different. A universal mechanism that does a reasonable job for all the phenomena seems a better bet (and it seems to not be classical transformations, but something more in the nature of structure-sharing).

        Another feature of the psychology is that the choice between alternative generative mechanisms is often decided by complex and unusual examples; for some, that is a bug, for others, a feature. But these oddities to occur, and people do manage them, and it’s not obviously crazy to be interested in that.

        Why most people prefer to play with ideas associated with Chomsky rather than the currently available alternatives (LFG, HPSG/Berkeley Construction Grammar, numerous flavors of Categorial Grammar and apologies to what I’ve left out) is another question. A possible answer is that failed attempts to over-universalize can often be instructive, but doing nothing rarely is, and of course by now, the pile of interesting Chomskian analysis is seriously large, so merits careful looking at whether you started out approving it or not.

        And wrt the categories: I certainly don’t want to belittle the theoretical aspects of justifying them (cherry-picking, etc), which I don’t think have been adequately treated, but, in practice, I don’t see it making that much difference. Everybody just seems to assume the traditional categories (perhaps modified a bit by areal wisdom), plus whatever tweaks seem to be called for by whatever they’re looking at — minimalists doing something on Greek or Spanish no different in this way from descriptive PhD students doing a language of PNG. I think it might also be relevant that the computational systems that I’m aware of that actually manage to learn something sort of like a grammar from something sort of like the data that language might learned from seem to have universalistic category assumptions (e.g. Kwiatkowsky et. al. CCG Bayesian grammar learning experiments). Maybe these assumptions can be dispensed with, but this hasn’t happened yet.

  2. Reading the Davis/Gillon/Matthewson paper I feel that they use formal essentially as a general term for a ‘theory-driven approach to fieldwork’. It is exquisitely ironic (assuming the best intentions of the authors) that their paper does not refer at any point to the rich resources for theory-driven fieldwork developed in the department of (of all people!) Levinson — see L&C Field Manuals and Stimulus Materials. They paint a picture of freewheeling, butterfly-catching D-linguists where in actual fact a lot of the most widely used materials for the stimulus-based and theory-driven field elicitation of linguistic data have been developed by D-linguists.

  3. “formal” has nothing to do with “form” but with a mathematical formalism. A formal theory is a theory expressed in a formal language, that makes use of formal logic. It is also not synonymous with “generativist” since a generative theory can have no clear formalization (and they often don’t).

    • Hi MGN! Yes, “formal” in “formal semantics” seems to come from “formalism”. But Davis, Gillon and Matthewson (2014) do not talk about formal languages or formal logic, and even “formal semantics” occurs only once in their paper. It seems that “formal linguistics” refers to a community of linguists, not so much to a special methodology or a set of assumptions.

Comments are closed.