In a recent blogpost, David Adger replied to my earlier post about “abandoning innateness”, trying to explain to me how one can be a mainstream generative grammarian (MGGer) and still say that most of the technical devices of one’s analyses are not innate. (I’m saying “MGGer” here, because practitioners of HPSG have long been explicit that they do not assume that the devices of their framework are innate; cf. Borsley & Müller (2021), in the forthcoming HPSG handbook).
I’m glad that this exchange is taking place, and Adger’s post is clear and helpful. However, it does not really address the main issue that I have with innateness-rejecting MGG, namely that many of its argumentations for particular analyses make good sense only if the elements that are posited can be assumed to be the same across languages AND to exist in advance. But unless an element exists in advance (as an innate component of the language faculty), it cannot be identified in different ways while still existing in different languages (cf. Haspelmath 2018: §7). For example, one cannot claim that (i) all languages have a contrast between adjectives and verbs (cf. Baker 2003), AND (ii) different languages draw the distinction in quite diverse ways, AND (iii) the distinction is not innate (sensibly, Baker does indeed claim that the distinction is innate). This is a simple point of logic: If we are not dealing with natural kinds, then the same elements in different languages must be identified in the same way. When dealing with adjectives, we often do this implicitly: adjectives are defined semantically as ‘property concept words’. But if we reject this semantic criterion (as linguists often do when talking about word classes), then we get different categories in different languages – unless we claim that adjectives are natural kinds (i.e. that they are innate).
So what I have claimed (in the blogpost, and more explicitly in my 2021 paper “General linguistics must be based on universals”) is not that there is a problem with innateness of grammatical knowledge (let alone a problem with abstractness, as Adger seems to think). Clearly, we have special cognitive abilities that allow us to use language and that are not shared with other species, and these features of our innate cognition may well include domain-specific aspects.
The problem is that many MGGers use highly specific formal devices that are identified differently in different languages while claiming to be agnostic about their innateness. This seems to be incoherent. Moreover, a much-touted selling point of the Principles & Parameters (P&P) approach was its restrictiveness: not everything can be described by its mechanisms. If the innate toolbox can do certain things but not others, then the gaps in attested languages can be explained (I called this the restrictivist approach in 2014).
But Adger’s blogpost does not really address these two key points. He makes four larger points, of which the fourth (on placeholders) seems to be somewhat novel – but let me first address points (I–III).
(I) First, he says that the Principles & Parameters approach “enriched our understanding of the syntax of lots of different languages”, but this is too vague – any other approach would have “enriched” it in some way. My question has always been: Have we made discoveries that have turned out to be robust and were confirmed by subsequent research from different perspectives? Did it indeed “uncover some deep theoretical generalizations about human language in general?” (Surely, there are many mid-level generalizations, as listed informally by Peter Svenonius and published formally by Roberta D’Alessandro (in 2019), but these are explicitly not meant to be “deep theoretical generalizations”.) I do think that it inspired interesting work which made good sense on the older view that the principles and parameters are innate (Roberts 1996; Baker 2001), but most of the ambitious claims of the 1980s and 1990s have not held up to further scrutiny (I wrote about this in 2008; even Baker 2008 recognized that his hopes of discovering more and more deeper parameters did not pan out).
(II) Second, Adger justifies Minimalism by saying that “the organization of things in the world is simpler than it looks because that’s been a successful strategy for deepening understanding in the past”, which is not really relevant in the present context (and quite doubtful – many insights have been gained by considering the possibility that the world is a lot more complex, e.g. 100 chemical elements rather than the four classical ones). That simplicity should be valued is completely uncontroversial. Of course, Darwin’s Problem (“How did the Narrow Faculty of Language get so complex, given the apparently brief evolutionary time over which it appeared?”) is relevant, as it should indeed lead us to expect few domain-specific aspects of innate cognition. But this does not favour MGG-style minimalism – any approach that minimizes innate mechanisms will do (including almost all non-Chomskyan approaches). And it is odd to say that it is good when we find that there are no or few observed limits on diversity (as in Legate’s 2021 paper on different kinds of passive-like constructions). In view of Plato’s Problem (Chomsky 1986), we should be happy to find limits on diversity, rather than finding that there are fewer limits than we thought. Such diversity would be expected on a view of language learning that makes the task appear more tractable, not on a classical Chomskyan view. (I’m beginning to wonder: Has Chomsky not only added Darwin’s Problem to his thinking, but has he also stopped viewing acquisition as a central issue?)
(III) Third, Adger says that the Narrow Faculty of Language (NFL, = the domain-specific innate cognitive prerequisites for language) can be reduced in a different way: by “making the primitives of the theory fewer and more abstract, but with wider empirical reach”. He cites Harbour’s work on person forms, and to the extent that this is successful, it is indeed a good example of NFL-internal simplification. This would be analogous to the simplification that chemists achieved with Mendeleyev’s periodic table by reducing the hundreds of types of stuff that were known to (al)chemists to four dozen elements. In other words, it is not different from the Roberts-Baker P&P approach. Baker (2001) elaborates in rich (an entertaining) detail on the suggested similarities between chemical elements and the natural kinds of syntax.
Adger may be right that some of the authors that I cited have this “increased abstractness” view in mind when they deemphasize innateness, which is indeed probably the case for Kayne 2013 (and maybe I should not have included this quotation). I find some of the other quotations (by the younger colleagues) harder to interpret as subscribing to a (more abstract) natural-kinds view (G. Ramchand: “(Perhaps) everything that all languages have in common is highly abstract, cognitive and not specific to language”; D. Ott:“What the truly universal categories are, and at what level of abstraction they can be found, is just an empirical question”).
Moreover, Chomsky’s motivation for Minimalism does not seem to have been the finding that it is actually possible to reduce the mechanisms (as turned out to be the case in chemistry during the period between Lavoisier and Mendeleyev), but primarily the desirability of reduction. Chomsky never claimed, as far as I know, that reducing phrase structure and movement to a single mechanism (Merge) was empirically motivated. And for Harbour’s person-number features, there was never a particularly strong case to begin with that the Harley–Ritter features (or any other feature set) is innate, because person systems make pretty good functional sense and could have evolved independently in different languages, regardless of biocognitive constraints. So I do not see any strong motivation either for the rich innateness of the 1960s-1980s, or for the more recent abstractness-increasing reductions. But more importantly, in the present context: None of this removes the incoherence noted in the fourth paragraph above. When an element is not considered innate, it cannot be identified by different diagnostics in different languages. (Compare the analogy in medicine: When a disease is not due to an established natural-kind cause, such as a pathogen, we cannot diagnose the same disease in patients that have different symptoms; see Haspelmath 2015.) But generative linguists do this all the time, as illustrated by the following quotation from a recent paper:
“While the division of the lexicon into nouns and verbs is likely universal, the diagnostics for lexical classes are language-specific and can be highly obscure or subtle.” (Polinsky & Magyar 2020: 3)
The more controversial mechanisms of MGG analyses all have the same problem. This can be seen nicely in the recent debate about NP vs. DP: While Bruening (2009; 2020) argues in favour of a universal NP (with determiners as dependents rather than heads), Salzmann (2018) and Preminger (2020) argue for a universal DP – but each of these authors base their highly general claims on rather different phenomena in different languages. If NP and DP are thought to be innate categories, then this is a meaningful debate. But if they are merely preliminary notational devices, then the whole discussion is pointless.
Borsley & Müller (2021: §2.3) make a very similar point when they discuss an analysis of case English PPs (e.g. with me) from a Minimalist textbook, illustrated in their Figure 28.1:
The problem here is not the “unnecessary” complexity of derivation, but the assumption of uniformity (despite the surface word order) for which there is no language-particular evidence. That English PPs are Hungarian-like is possible (just as it is possible that nominals are DPs in all languages), but only if all these mechanisms are natural kinds (= are innate). Borsley & Müller say that such analyses would be reasonable if rich innateness were an established fact, but I am much more modest: I merely demand that one recognizes that one needs to hypothesize/posit innateness.
Adger does seem to recognize the need for innate elements, but he downplays the significance of current MGG analyses in his fourth point, where he introduces “placeholders”:
(IV) The abstract analyses and the elements posited in them (such as vP, nP, DP, AgrP) are “placeholders for a better understanding”. Current theories are “uncertain” anyway, and subject to future change. Current analyses are “not the final word”, but “stepping stones which have taken us further than we were, but there’s a long way to go still”. It would not be realistic to assume that “our theoretical posits are the right ones”. MGGers “adopt particular hypotheses as long as they do explanatory work”, but it’s OK if they turn out to have other explanations (completely outside of general syntax, or maybe of the “Third Factor” type). There is no “innateness hypothesis”, and instead we should be open-minded.
All this sounds eminently reasonable to me, but I still do not understand how it justifies MGG-type analyses. What kind of “explanatory work” is done by vP or DP (or X-bar theory, or Case assignment under abstract Agr, or a “binding theory” that covers reflexive pronouns and “traces”) unless it is assumed that it is the final word? It is well known that a lot of the argumentation in MGG is “theory-internal” – this is what some people have described as the “moving parts of the theory”. A concrete example of this is Preminger’s (2020) article in which he argues against Bruening’s (2020) recent defense of the “NP Hypothesis” against the “DP Hypothesis”. Preminger’s data are limited to a particular construction in Hebrew, where a noun (such as ‘shortage’ or “victory’) takes two arguments, as seen in his (3a):
The ordering of these elements is fixed, and this is what needs to be explained. Preminger claims that in order to derive this sentence, one needs to assume some higher functional category (probably DP), and that Bruening’s “NP Hypothesis” is not compatible with these data.
But of course, this is so only if one makes certain additional assumptions, and Preminger considers one alternative assumption in a footnote (n. 7):
“Assuming that the External Argument is not generated in a specifier position would … throw our understanding of the structural correlates of theta-assignment into chaos. We must therefore consider whether the payoff of such a move would justify the price. As best I can see, the answer is a resounding no.”
This kind of argumentation seems quite typical of MGG to me – one relies on proposals of the earlier literature and builds on these, positing as few additional mechanisms as possible. But this approach would make sense only if the earlier “theoretical posits” are thought to be “the right ones” – in other words, if these elements are (innate) natural kinds.
(Note that there is an even simpler way of describing the Hebrew facts, using ternary branching as in HPSG and Simpler Syntax (Culicover & Jackendoff 2005: Ch. 4); but this is not even considered as an option by Preminger, even though it would seem to be the most straightfoward description of the Hebrew facts. Preminger seems to have great confidence in the innateness of binary Merge.)
If there is no rich grammar blueprint (as had been thought in the 1960s), and if our current theories are uncertain and subject to future change, would it then make sense to write a paper about a simple Hebrew construction and use it to argue for a bigger point? Apparently Preminger still holds on to the earlier 20th century view (like Baker 2001; cf. Levin & Preminger 2015, which builds on Baker’s dependent case theory), but in any event, this kind of argumentation has long been extremely widespread in MGG, and inexplicably, this did not change after Chomsky began to suggest that much less is innate than had been thought earlier.
So when it comes to the larger picture, I have no disagreements with Adger – but when we get to the actual practice of analyzing grammatical patterns, MGG work keeps relying on the assumption of a rich grammar blueprint with many elements that must be given in advance (the kinds of things that are taught in textbooks such as Adger 2003). This is all well as long as one adopts the traditional restrictivist UG view (of which Roberts 1996 and Baker 2001 presented a fascinating vision). But if we question the presence of rich innate grammatical knowledge (which is now widely rejected because of Darwin’s Problem), and if we adopt Adger’s stance concerning the great uncertainty that surrounds all our work in theoretical syntax, it seems wiser to treat our grammatical metalanguages as notations, not as substantive theories (or as innate frameworks) – and to dissociate our general theories from the notations.
Adger, David. 2003. Core syntax: A minimalist approach. Oxford: Oxford University Press.
Adger, David. 2021. On doing theoretical linguistics: A reply to Haspelmath. Theoretical Linguistics (to appear). (https://ling.auf.net/lingbuzz/005691)
Baker, Mark C. 2001. The atoms of language. New York: Basic Books.
Baker, Mark C. 2003. Lexical categories: Verbs, nouns, and adjectives. Cambridge: Cambridge University Press.
Baker, Mark C. 2008. The macroparameter in a microparametric world. In Biberauer, Theresa (ed.), The limits of syntactic variation. Amsterdam: Benjamins.
Bruening, Benjamin. 2009. Selectional asymmetries between CP and DP suggest that the DP Hypothesis is wrong. Proceedings of the 32nd Annual Penn Linguistics Colloquium. Philadelphia: University of Pennsylvania. (http://repository.upenn.edu/pwpl/vol15/iss1/)
Bruening, Benjamin. 2020. The head of the nominal is N, not D: N-to-D Movement, Hybrid Agreement, and conventionalized expressions. Glossa: a journal of general linguistics. Ubiquity Press 5(1). 15. (doi:10.5334/gjgl.1031)
Chomsky, Noam A. 1986. Knowledge of language: Its nature, origin, and use. New York: Praeger.
Culicover, Peter W & Jackendoff, Ray. 2005. Simpler syntax. Oxford; New York: Oxford University Press.
D’Alessandro, Roberta. 2019. The achievements of Generative Syntax: a time chart and some reflections. Catalan Journal of Linguistics 0(0). 7–26. (doi:10.5565/rev/catjl.232)
Haspelmath, Martin. 2008. Parametric versus functional explanations of syntactic universals. In Biberauer, Theresa (ed.), The limits of syntactic variation. Amsterdam: Benjamins. (Accessed May 27, 2016.)
Haspelmath, Martin. 2014. Comparative syntax. In Carnie, Andrew & Sato, Yosuke & Siddiqi, Dan (eds.), The Routledge handbook of syntax, 490–508. London: Routledge. (https://zenodo.org/record/344909)
Haspelmath, Martin. 2015. Defining vs. diagnosing linguistic categories: A case study of clitic phenomena. In Błaszczak, Joanna & Klimek-Jankowska, Dorota & Migdalski, Krzysztof (eds.), How categorical are categories? New approaches to the old questions of noun, verb, and adjective, 273–304. Berlin: De Gruyter Mouton.
Haspelmath, Martin. 2018. How comparative concepts and descriptive linguistic categories are different. In Van Olmen, Daniël & Mortelmans, Tanja & Brisard, Frank (eds.), Aspects of linguistic variation: Studies in honor of Johan van der Auwera, 83–113. Berlin: De Gruyter Mouton. (https://zenodo.org/record/3519206)
Kayne, Richard S. 2013. Comparative syntax. Lingua 130. 132–151. (doi:10.1016/j.lingua.2012.10.008)
Legate, Julie Anne. 2021. Noncanonical passives: A typology of voices in an impoverished Universal Grammar. Annual Review of Linguistics 7(1). (doi:10.1146/annurev-linguistics-031920-114459)
Levin, Theodore & Preminger, Omer. 2015. Case in Sakha: Are two modalities really necessary? Natural Language & Linguistic Theory 33(1). 231–250. (doi:10.1007/s11049-014-9250-z)
Polinsky, Maria & Magyar, Lilla. 2020. Headedness and the lexicon: The case of verb-to-noun ratios. Languages. Multidisciplinary Digital Publishing Institute 5(1). 9. (doi:10.3390/languages5010009)
Preminger, Omer. 2020. Functional structure in the noun phrase: revisiting Hebrew nominals. Glossa: a journal of general linguistics. Ubiquity Press 5(1). 68. (doi:10.5334/gjgl.1244)
Roberts, Ian. 1996. Comparative syntax. London: Edward Arnold.
Salzmann, Martin. 2018. Revisiting the NP vs. DP debate. Leipzig University. (https://ling.auf.net/lingbuzz/003823)
As an outsider, one of the things I see is an asymmetry between MGG and (lato sensu) functional papers: the former basically make sense to people who are governed by the MGG tradition and bound by its rules, whereas functional papers make sense to anyone with a minimal grasp of linguistics (while noting that even this much is a rare accomplishment).
In this sense, functional theories are more primitive. in the same way that planetary astronomy is more primitive than nuclear physics, or the conflict of laws is more primitive than contracts. But nuclear physics has an empirical and experimental underpinning that allows it to progress beyond primitivity, whereas MGG (as far as I can see) turns its back (and always has) on those underpinnings.
Yes, that sounds right. Chomsky has often compared his approach to physics (hasn’t he even adopted “standard theory” from physics?), but so far, most of the highly general claims of generative grammar remain at a speculative level. Intriguing and inspiring, but not solid results.
Again, there was some discussion on Twitter: https://twitter.com/haspelmath/status/1381570953201729536. David Adger said: “Actual practice is investigation of the extent to which this is possible. So you hold things constant and investigate.” To which I replied: “Maybe I fail to grasp the rationality of „holding things constant“ when these things are extremely preliminary anyway. And especially when there are much more straightforward options (such as ternary branching) available. It seems reckless, but I do see that it can be fun!” Adger’s answer: “Have always liked to live life on the edge! More seriously, I believe the world is too complex for us to understand, except thru building understandable theories that thereby bring us closer to it. Keeping things to a ‘straightforward’ level doesn’t always deepen understanding.” – This is a good and honest answer, and it confirms my general impression that generative linguists are happy to speculate without worrying too much about empirical considerations (“life on the edge”). This can be fun, but the risk of going off in wrong directions is of course very high.
The periodic table is a different simplification: not the discovery of the chemical elements, which mostly came earlier, but the discovery that their physical and chemical properties become predictable if they’re just placed in a simple two-dimensional spreadsheet of the right size – to the point that there were unfilled cells in it which predicted the existence and the properties of further elements that were indeed later discovered (or made in colliders).
True, Mendeleyev’s systematization was just the culimination of a longer process. But still, throughout the 18th and 19th centuries, chemists found that previously known stuffs were made up of smaller elements, and the evidence for the elements kept converging. Alas, we have almost nothing of this sort in linguistics, it seems.
Since I am mentioned in this post, allow me to offer my two cents.
On the more general issue, I tend to side with Adger, in that I see no incoherence between positing, say, vP, and viewing it not as a piece of innate machinery unto itself but as a placeholder for whatever piece of innate machinery it is derived from. I fail to see how this is incoherent; the methodology is as follows. Suppose we have good reason to believe vP is universal (note: this is not ‘universal’ in the Greenbergian sense, as in “there will be evidence for this in every single language we check”, nor will the evidence we do find necessarily always be of the same kind). We can then proceed *as though* vP were universal, treating it as a de-facto innate category while simultaneously working to derive this de-facto state of affairs from some deeper mechanism, eventually dissolving the need to posit vP itself as part of the innate machinery.
Now, you can reject the premise (that we have good reason to believe vP is universal). That’s a separate argument. But I don’t see the incoherence.
On the larger issue, I think the theoretical perspective you attribute to “younger colleagues” – citing Gillian Ramchand and Dennis Ott – is probably indeed pervasive among Ramchand’s milieu (PhD: 1993), which includes Adger (PhD: 1994). But Ott (PhD: 2011) is something of an outlier, in my opinion. I think most (or at least many) people in this research community who got their PhDs after 2005 or so (I got mine in 2011) think that this “slim UG” stuff that Chomsky has been on about for the last ~25 years is a bunch of wishful thinking, and on the big-picture issues, we are much closer to the views you attribute to Roberts 1996 and Baker 2001 than any of the views found in Chomsky’s writings post-1990. See here – https://omer.lingsite.org/blogpost-post-minimalism – for my (admittedly probably parochial) opinion.
As for my Hebrew DP paper, it’s true that I don’t consider HPSG (nor LFG, nor CCG, nor …). The paper is an 8 page response to a very particular claim embedded in a very particular framework, and so it’s responding to that. (You’re right that, as a biographical matter, I do subscribe to the binary-branching assumption in my work; but one does not go back to first principles in every paper one writes, much less a response paper. I also don’t motivate the existence of ‘noun’ in this paper…) I happen to believe that there are good reasons to reject HPSG in favor of, um, well what to call it is a complicated question (see link above). But again, not every paper in HPSG begins with an argument against the competing frameworks, either.
Finally, I would be remiss if I didn’t point out that the observations in the paper are old. As the paper explicitly and repeatedly notes, the relevant observations (that is, the consequences of these particular empirical facts in Hebrew for theories of the universal structure of noun phrases) are due to Elizabeth Ritter and are 30 years old. The paper was simply reminding the relevant sliver of the literature of the existence of these facts and arguments.
Thanks for this comment – very interesting!
I have the impression that generative linguists do not really consider the question whether one should posit labels such as NP or DP as issues that have to do with innateness, which is already evident from the fact that the concept of “DP” is not applied to all languages. What I find difficult to grasp, however, is how notions such as DP function within the theory, what their empirical status is and how one can use such notions in the explanation of empirical phenomena. See for example the discussion about DP/NP in my paper on explanation in syntax (https://ling.auf.net/lingbuzz/005664) where Bailyn (2012) refers to Bošković (2004, 2005) who argues that the lack of a DP shell around NP in Russian is exactly what allows for particular word order such as split-scrambling: “the equivalent movement is impossible in English because of the blocking effect of the DP shell itself.” (Bailyn 2012). In this case a DP phrase, literally seems to acts as a “shell”, or an absence of a DP seems to act as the absence of a “shell”. This means that something like a DP in a phrase structure is seen as something with an actual ontological status: it occupies a position in the phrase structure, which can act as a “shell” which “blocks” “movement”. But how can a theoretical concept block something? Is this a metaphorical way to point at the presumed correlation between the presence of determiners and rigid word order (and absence of determiners and free word order)? And if not, how should we understand this? What I need is an analysis of what is meant by good explanation.
Your article was very interesting. The fact that Generative and Functional approaches to linguistics are incommensurable in Kuhn’s sense explains why there is so little collaboration and understanding across the groups.
In the conclusion of your paper, you mention that discussing interesting data patterns that everyone agrees on, and discussing it in as a theory neutral way as possible might help with collaboration, and this definitely makes sense to me.
That left me wondering: what sort of data do you think would work for this purpose? Are there any existing large source of syntax data that is relatively theory neutral? To me it seems that the World Atlas of Language Structures could provide such data, and Martin Haspelmath’s idea of “comparative concepts” is useful for collecting theory neutral data in syntax, but I’m curious whether you agree that these are sufficiently theory neutral for these purposes.
Thanks for your question! I definitely think that the data you mention are sufficiently theory neutral, even though one would also have to ask a generative linguists whether s/he agrees. In fact, I think that generative linguists often provide data that are theory neutral (as do functional linguists) even if often people do not read each other’s work. As I see it, the problem starts as soon as one mixes up describing things an explaining and analyzing things within a specific theory or formalization. The DP versus NP case I mentioned is a good example. I personally do not find these notions very clear, nor do I understand how one can use them in the explanation of single or even all languages. It would be good to first focus on the question whether there is a correlation in languages of the world between the presence of determiners and a more rigid word order (in fact, one would need a definition or measure for that, which is probably not that easy). And only after having established that one can starting thinking about explanations and formalizations/ representations. Probably, functional linguists will then come up with a very different explanation (if there is something to explain of course) than generative people, but at least the data we all agree on will be clear. And if one listens to one another we can also learn from the type of data that are interesting to take into account.
The final step would be to have a discussion amongst the different groups about what a good theory is, a good explanation is, and what good theoretical concepts are, for specific languages, and for language in general. That is what Martin Haspelmath is doing with people from the generative framework. My impression is that even if they do not accept what he is saying, it will make their theories and explanations better, since it forces them to think with an open mind about fundamental things such as: What is a good theory? What is a good explanation?, etc. What strikes me in the argumentation of generative people is that they often refer to things like ‘good explanation’, ‘explanatory work’, ‘explanatory hypotheses’, ‘explanatory adequacy’, ‘good model’, ‘understanding’, etc. but they do not make explicit what they mean by that. It would be nice to have more discussions about that to see whether different schools of linguistics can converge to some extent with respect to these questions. (And of course, I am not a generative linguist, so, perhaps, generative linguists will have the same type of questions about non-generative theories, which is only good, as long as there is discussion about important topics).