Martin Haspelmath: Yakov, you have been following theoretical and comparative research in morphosyntax over more than four decades (we first met in Moscow in 1986), and you have been discussing these things with me on and off (for completeness, let me add that you are at the RAS Institute of Linguistics and at Russian State University for the Humanities). Most recently, you gave an initial opinion on my question of what “formal linguistics“ is – see this earlier blogpost. You seemed to agree that “formal linguistics“ is widely understood in the narrower sense of ”an approach that accepts (explicitly or tacitly) Noam Chomsky’s philosophy of language”.
Yakov Testelets: Yes, but Chomsky of the minimalist period particularly emphasizes that, unlike in his earlier Principles and Parameters framework, there is no more need in the “rich set of specific innate mechanisms”: the mechanisms he postulates now are the simplest of all imaginable.
Martin: Yes, this is widely known (and I wrote a blogpost about this as well), but I do not understand how this position is compatible with the practice of mainstream generative grammar. If a category is postulated for one language (e.g. DP, or vP, or Appl, or head movement), it is treated as being available also to all other languages. That presupposes the innateness of these categories, doesn’t it?
Yakov: It does not if there is an obvious way of reducing this rich array of categories to more fundamental and elementary ones. For example, the theory of phrase structure is abstract and ignores the differences among the “big four” of the major lexical classes: noun, adjective, verb and adverb. It treats them alike as lexical heads and so circumvents the problem of the universality of the “big four” and of the respective phrasal categories NP, AP, VP and AdvP. But if a category term is successfully employed for many different languages but cannot be easily reduced to more abstract and simple elements, the problem you have mentioned remains unresolved.
Martin: In my blogpost, I suggested that “naturalistic linguistics“ might be a better term for the Chomskyan approach, because it presupposes that many linguistic categories are part of human nature (with which we are born), not only part of human cultures (which we must learn).
Yakov: When we speak of “naturalistic linguistics”, we should remember that the natural sciences employ formal, i.e. mathematical models, therefore they are all “formal”. In this sense, “naturalistic” presupposes “formal”. Of course Chomsky’s ideal has always been the (post-)Newtonian physics, and not the physics of Aristotle.
Martin: OK, fine, though this does not interest me personally so much. I would rather compare the linguistic nature of humans with their psychological nature. Psychologists widely agree (it seems) that many aspects of psychology are due to innate natural kinds, e.g. the six basic emotions (surprise, happiness, sadness, anger, fear, disgust). I am not sure if this research is based on mathematical models. But for physics, the mathematical models are of course crucial.
Yakov: In Chomsky’s view, the “naturalistic” approach and physics share the reductionist ideal: At first sight, natural phenomena seem infinitely various, but the apparent huge differences are illusory, because they can all be explained by several fundamental laws that must be formulated in the mathematical language. Likewise, human languages seem to be tremendously different, and any separate language displays an enormous complexity that apparently cannot be fully captured by linguistic research. Nevertheless, the method of reduction must uncover simple underlying principles, because it has worked so well in physics.
Martin: Yes, that has been the research programme.
Yakov: Now some of the functionalists like Givón or Aleksandr E. Kibrik have similar beliefs: We can reduce all complexity to simple basic functionalist principles. Fillmore’s Construction Grammar, by contrast, claims that the complexity of human language cannot be dispensed with.
Martin: Isn’t this more a matter of emphasis? In my perception, Givón has emphasized that IF we want to understand anything in language structures, we must look seriously at their functions, but he has not really said that the functional principles explain everything. And Fillmore and the construction grammarians have emphasized that IF we want to explain all of language acquisition, we have to explain how idiosyncratic patterns are acquired as well. And this requires a mechanism that stores a huge amount of detail, in addition to being able to make generalizations. So I see these two approaches as compatible with each other (and also as compatible with Chomsky’s recent view that only very minimal mechanisms are specific to language; though not compatible with the practice of Chomskyan syntacticians).
Yakov: Of course, the two approaches are compatible, but it seems that every scientific school is characterized by an orientation at its own “model studies” that show to its followers how linguistic work must be successfully done. Model works in generative linguistics are utterly reductionist, the classical examples being the fourth chapter of Haj Ross’ 1967 dissertation in which he reduced the context conditions on different transformations to the few island constraints, Chomsky’s (1977) “On wh-movement”, and the like. On the other hand, a model study in Construction Grammar like Goldberg & Jackendoff’s (2004) paper on the resultative constructions in English is not reductionist: rather, the authors’ aim is to find an elegant and convincing format to represent the language complexity.
Martin: Yes, Goldberg & Jackendoff work in the Fillmorean tradition, and again, there is no incompatibility with Givón-style and Kibrik-style functionalist reductionism, I think. But I have another question: Do you have a comment on my observations (in the earlier blogpost) about Moscow’s Chomskyan linguistics? The term “formal Russian linguistics“ has recently come to be used, apparently in the sense of „Chomskyan linguistics of Russian“ (I referred specifically to Ljutikova & Tatevosov (2017)).
Yakov: Ljutikova & Tatevosov’s only inaccuracy is that they claim that Soviet formal linguistics started with Igor Mel’čuk. But in fact, it started in the late 1950s with the “grammatical modeling” approach (грамматическое моделирование) which strongly influenced Mel’čuk but was still different.
Martin: That’s interesting – I had thought that Mel’čuk and his generation were mainly influenced by generative grammar.
Yakov: That is only partly true. There was a group of linguists and mathematicians (Kolmogorov, Vladimir Uspenskij, Andrey Zaliznyak, Gladkij and others) who, unlike Chomsky, had mathematics as the ideal for a linguistic theory, not physics. They believed, partly under Hjelmslev’s influence, that they can transform linguistics into a rigorous science by constructing definitions, i.e. formal analogues, or models, of the traditional linguistic notions. Models, being applied to data, must bring the same results that the traditional informal notions do, although the latter are employed intuitively. They believed that when the array of the linguistic terms becomes as formal as the mathematical terms are, linguistics becomes “scientific”. Therefore they were constructing huge formal definitions of case, gender, agreement, subject, word, morph(eme), and the like. Later the approach proved useful for computational linguistics, but the theoretical program, in my view, failed.
Martin: That would not be the only failure in theoretical linguistics over the last century… Which kind of linguistics has been the most successful, in your view (I mean apart from historical-comparative linguistics, whose successes are completely undisputed)?
Yakov: The greatest success of theoretical linguistics, as I see it, has been its heuristic strength: it has helped the linguists to discover mountains of previously unknown facts and phenomena of tremendous importance – somewhat like Columbus’ idea of the western sea route to India which failed as theory but succeeded as a research program. Another success is that we have arrived at much narrower spaces of typological variation and, starting a linguistic work with a new “exotic” language, we know much better what to expect from it as regards e.g. its tense-aspect-mood categories, anaphora, case, word order, phrase structure, etc. than our predecessors in the field could know.
Martin: Yes, I completely agree with all this (as I wrote in my recent post about the innovations of generative linguistics). So we shouldn’t be depressed about our lack of progress, and instead focus on the good things that we have learned.
OpenEdition suggests that you cite this post as follows:
Martin Haspelmath (March 15, 2019). An interview with Yakov Testelets (Moscow). Diversity Linguistics Comment. Retrieved November 13, 2024 from https://doi.org/10.58079/nsus
I think, to make linguistics computable is ultimately a practice to statisfy the demand of electrical circuits, where a specific number is needed to turn on and off the electrical currents. Yet this could be far away from how human instinct works and further away from a humanly-acceptable linguistic theory.