Martin Haspelmath: David, thanks for engaging in various discussions on Facebook over the years. Most recently, I had a conversation with Elena Anagnostopoulou about innate principles, and she mentioned parasitic gaps as a convincing example of something that is explained by innate principles. Then I was asked how I would deal with the universal properties of parasitic gaps, and I replied that I wasn’t sure what exactly a parasitic gap is. To test the universality of the phenomenon, I‘d need a definition that applies equally to all languages, using the same criteria. If so, I‘d try to find out how likely it is that this phenomenon is universal. Then I‘d consider a range of possible explanations (because there must be one – if it‘s universal, it can’t be a random effect). Some explanations are inherently unlikely and thus not preferred (e.g. that we‘re born with parasitic gaps), but nothing can be excluded.
David Pesetsky: You could have a look at Culicover’s (2001) paper “Parasitic gaps: A history”
Martin: Thanks, David. This does not answer my question, though, because it does not give a definition of “parasitic gap”. It starts in a way that’s quite typical of (generative) linguistics: “The parasitic gap construction is exemplified by the following examples…” – so we’re supposed to form a general concept on the basis of an example from English. But this doesn’t tell me what a parasitic gap construction is in general. (You may say: For that you need a theory. And then I would say: So testing the claim that parasitic gaps are universal must wait until we have the right theory, i.e. not in our lifetimes.)
David: That’s how the summary article starts, but by the time it ends, we have in hand a collection of properties that characterize the construction called “parasitic gaps” cross-linguistically. That’s why communication on the topic is possible among linguists (which it is), including investigations of questions such as “does sentence Y in language X involve a parasitic gap” that can yield “yes” and “no” answers (as well as more complicated answers between yes and no, in troublesome cases that we are still somehow able to communicate with each other about). Don’t see the problem you see.
Martin: I admit that I haven’t read it to the end (yet) – life is short. But why doesn’t he say what a parasitic gap is at the beginning? My feeling is that his method doesn’t allow him straightforward definitions of terms. General definitions are seen as results of research, and research starts with particular languages. Parasitic gaps are a marginal phenomenon, so I haven’t studied them in great detail. But Culicover is a great linguist (his 2005 book Simpler Syntax, coauthored with Ray Jackendoff, explains some of the ways mainstream generative grammar has gone astray), so maybe I’ll find the time to read the paper soon. But since he mentions almost no non-European language, I have very little hope that he can get a deeper understanding. Depth of understanding has never come from looking at some corners of a range of European languages, as far as I can see.
David: For some work on parasitic gaps in non-European languages, see Baker (1991) on Mohawk, Branan & Sulemana (2019) on Buli, and Ershova (2019) on West Circassian. Baker (1991: 573) says: “The fundamental structural relationships found in Mohawk are essentially identical to those of English, including the fact that the subject of the verb asymmetrically c-commands the object. Thus, under the analysis offered here, Mohawk confirms Speas’ (1990) conjecture that the principles of phrase structure are universal. The principles that govern coreference and extraction are also the same in both languages – including Condition C, the Condition on Extraction Domains, the Weak Crossover principle, and the principles that govern parasitic gaps]. […] “The configurations that these principles apply to are not always the same, however. […]”
Martin: Many thanks for all these references! There’s a huge generative literature, I know – but again, almost of it has the problem that it makes claims that are not readily testable. You first need an “in-depth” analysis, based on the right theory (which we don’t have). This methodological problem does not go away, even if you cite 50 papers on individual languages.
David: You’ve hit the nail exactly on the head, but for some reason don’t agree that it’s a real nail and it needs to be hit on exactly that head. Yes indeed, the claims are not readily testable. And yes indeed, you first need an “in-depth” analysis. That’s the whole point of our work, and why what passes for “typology”, while sometimes useful in generating guesses about promising generalizations or correlations, often does not look like maximally useful research to people like me — since uncovering the real generalizations does require in-depth prior analysis. And (this is crucial) in-depth prior analysis works: you can learn whether a gap in some language patterns with parasitic gaps in English or, say, with null topics in Mandarin, but it requires a wider effort within the language under investigation. That’s what we do. That’s what feels like productive linguistics to folks like me. (And yes indeed, the question is “how can a speaker learn that”. We advance on that question each time we discover something new about the nuts and bolts of a particular language and languages more generally, but we need lots of help solving that problem. All the more shame that the field is so fragmented and we’re complaining about each other rather than pitching in to help each other with it.)
Martin: Good, so there is at least something there that I seem to have understood correctly 🙂 (And incidentally, this conversation did not start with me complaining, but with me praising Elena for giving so much attention to my 2004 paper – even though she doesn’t seem to have understood what I was doing in that paper.) The problem that I have is the circularity: We make assumptions, build frameworks on them, make framework-based descriptions of languages, claim universality of our frameworks on the basis of these descriptions. I don’t see any hypothesis-testing in this process.
David: “We make assumptions, build frameworks on them.” In general (a few Chomsky exceptions aside), no. We make discoveries, build frameworks on them. Islands, parasitic gaps, etc. These are discoveries independent of framework. (For example, note that researchers who claim they are developing distinct frameworks talk about some of the same phenomena. Just not self-styled functionalists and typologists for the most part.)
Martin: What I meant was, for example, the idea that most syntactic relations are expressed as binary trees with movement, that variable-binding relations must be due to c-command, or that vocabulary items are “inserted late”. As far as I understand, these are generally thought of as productive assumptions, or hypotheses about innate principles, not as discoveries – or would you disagree?
David: My point is that there’s no circularity. Yes, how we describe discoveries uses terms whose relevance is justified by other discoveries. But the paths that lead from the primary discoveries to these descriptions are transparent. So a statement of a claimed universal property of language that mentions structures or abstract relations can be replaced by a much more complicated statement in which the justification for positing the relevant structure or abstract relation is substituted for the structure’s name or the name of the relation. That’s what we teach in our intro courses, after all — e.g. you’ve learned the arguments for VP; now we can talk about VPs instead of repeating the arguments every time. Of course, being human, we often mess this up; there are often multiple options and we can’t explore them all; and unexpected discoveries require us to question our conclusions (and retrace our steps). But the effort is legitimate and productive: nothing circular, I think.
Martin: Many thanks, David, for this conversation. Till next time!
[Postscript: In the following blogpost, I explain in more detail why I think that our cross-linguistic claims must be readily testable, and why we cannot based them on in-depth analyses: https://dlc.hypotheses.org/2282]