Martin Haspelmath: Elena, it was only last week that I became aware of your 2017 paper on the Person Case Constraint in the Companion to Syntax. (The “PCC” refers to the unacceptability of clitic combinations as in French *il me lui présente ‘he introduces me to her’.) I’d like to thank you for discussing my 2004 paper (“The Ditransitive Person Role Constraint”) in such great detail in that overview article – that’s really wonderful, and exceptional for a generative paper.
Elena Anagnostopoulou: Thank you very much for contacting me. I am glad the PCC survey gives us the opportunity to communicate because I always read your work with great interest and find your generalizations thought-provoking and exciting. It was therefore natural to include your approach in my survey; I have given it a lot of thought over the years and I felt that my paper would have been incomplete without it.
Martin: I have one question to you: You contrast my proposal (which I framed as “usage-based” or “grammaticalization-based” at the time) with “grammar-based” accounts (which you argue for in the end). By grammar-based, do you mean UG-based?
Elena: I was assuming a distinction between explanations that try to use generalizations over a symbolic system that manipulates abstract representations and is (to a large degree; for exceptions see Charles Yang‘s work, among others) isolated from frequency effects on the one hand (hence “grammar-based”) and theories that focus mainly on distributional and statistical generalizations over the data that can be seen as arising from putting the grammatical system to use, on the other hand (“usage based”). To my mind, frequency effects are a matter of preferences of the speakers and in that sense not “grammar based”. Even if I did not believe in UG, I would still consider your explanation non-grammar based in this sense explained.
Martin: But I do agree that grammatical systems are very largely isolated from frequency effects at the synchronic level. The very fact that my paper talks about “grammaticalization” implies that I do not think that the PCC effects are somehow extragrammatical – it’s just that I look for ways of explaining them without appealing to a rich innate set of categories or architectures for grammar. As for the grammar vs. usage distinction, I agree with Frederick Newmeyer, who said: “grammar is grammar, and usage is usage” (Newmeyer 2003). So in that sense, I’m all for grammar-based explanations of speaker behaviour (cf. also my 1993 grammar of Lezgian, which makes no reference to usage).
Elena: Yes, if Newmeyer is taken to mean by “grammar” a symbolic, discrete, recursive computational system, this is the distinction I had in mind.
Martin: But how do we explain the grammars themselves? In particular, why do we think that phenomena like the PCC in different languages should be treated together? It seems to me that most linguists think that the similarities are not accidental, but due to some general factor. The way I understand it, in generative grammar, one assumes that this general factor is innate categories (features, operations, architectures), while this is not the case in Greenbergian approaches.
Elena: My understanding from reading the typological literature is that they do try to explain universals, but by appealing to functional principles, processing load etc. Generativists in the Chomskian tradition try to account for variation in terms of innate principles.
Martin: Yes, so it does seem that “grammar-based” is not the right characterization of what separates your approach from mine. What I was trying to do in the (2004) paper is explain grammars with reference to some general factor outside the grammars (namely universal usage tendencies), but one cannot really describe the standard generative approach as merely “grammar-internal”, because it is just as interested in similarities between languages as the Greenbergian approach. So the generative explanation must be grammar-external in this sense, too – in other words, it must appeal to innate categories. We seem to need something like a contrast between “innate-principle-based” vs. “other-based”.
Elena: As I said, we try to account for variation in terms of innate principles. But, depending on who you talk to, parameters are a different issue. Some people assume the so-called Borer-Chomsky Conjecture (named after Hagit Borer), according to which variation is placed in the lexicon. And the lexicon, or parts of it, is not necessarily innate. So, for example, in the case of PCC one could assume that what is innate are the general mechanisms of Merge and Move and what is parametrized are the feature structures present on e.g. functional heads, that are not necessarily the same across languages. In fact, people have suggested this in different forms. They have talked about “parametrization in the feature structure of Probes”. In my 2017 article, I tried to report the work from the perspective of the position it takes with respect to the various versions of the PCC trying to focus on the empirical coverage different theories have and the correlations they can express, and I did not get into what parts of the different theories directly appeal to innateness and what parts have to do with the parametrization of feature structures and the lexicon. I happen to believe that it is an empirical matter to decide whether e.g. feature geometries capturing markedness relations are universal or not. For Person one could probably say it. For Gender it is way more difficult.
Martin: I know that some people say that only Merge & Move (=or “only (in-/ex-ternal) Merge”) plus Agree are innate, but I haven’t understood how that translates into a general account of general PCC tendencies – if all the features are language-particular and their interaction is parameterized, you couldn’t predict general effects. The literature is really confusing, but it seems that quite a few additional innateness assumptions are made, e.g. about person feature geometry (as you note). (For example, Jelinek (1993) seems to have assumed that local persons are innately specified as “definite”, so the latter also needs to be a substantive innate feature.)
Elena: For many of the phenomena that you have analyzed in terms of universal usage tendencies, my hypothesis is that they arise from the way features are encoded in the grammar and how the mechanisms assigning case, agreement and triggering movement interact with these features. Variation is not unpredictable, there are limited options and the hope is to be able to understand the limitations. I agree that the literature is sometimes confusing, but this also depends on specific people and how they use the theory. Some people are extremely clear. I do agree with you though that usage frequencies are probably very important for historical change because grammars change on the basis of the input and the more frequent an input is the more it can influence the grammar of speakers and especially of children acquiring grammar. So, I was thinking that the correlation you are suggesting between frequency of use and grammaticalization is something one needs to consider in the connection with the PCC. It is still an empirical matter, though, which approach is correct for each phenomenon. Frequencies cannot explain all syntactic regularities or phenomena. There are for instance rare phenomena hardly if ever attested in corpora (e.g. parasitic gaps) for which people still have very clear intuitions. Regarding the PCC, I talked about Wackernagel pronouns and binding as two infrequent phenomena that in addition appear to be sensitive to the same type of restrictions.
Martin: Clearly, frequency differences do not explain everything, but maybe they explain everything that is universal. You mention parasitic gaps, which are not frequent, but is there anything that is universal about them?
Elena: I do not know of any language with overt wh-movement that doesn’t have parasitic gaps, and the same goes for strong islands, Principle C effects, variable-binding and many other phenomena.
Martin: Yes, some of these do seem to be universal, but generativists also often make highly general claims about non-universal phenomena. It seems to me that part of the “talking past each other” phenomenon has to do with unclarity about one’s goals. My goal is exclusively the explanation of universals, because most language-particular phenomena are historical accidents. (So I don’t think that language-particular analyses need to be highly general.)
Elena: Linguistics would be a much better place if we could constructively talk to each other. I personally learn a lot from work on large-scale generalizations and I admire typologists for their generalizations and their effort of giving the linguistics community tools like data-bases and grammars, to teach us how to work with quantitative data and to highlight historical and areal factors in language variation and change, among many other things. But I also find it very exciting to discover that properties that look language-specific are not mere historical accidents, especially when they correlate with phenomena that look very different at the surface but can be argued to boil down to the same principles at the appropriate level of abstraction. And I am interested in phenomena that are rare in the input but turn out to be extremely robust crosslinguistically and probably universal. In any event, I enjoyed this communication and hopefully there will be more opportunities to talk to each other in the future.
Martin: Thank you very much for this conversation. Till next time!