Chomsky now rejects universal grammar (and comments on alien languages)

That our colleague Noam A. Chomsky no longer argues for a rich innate universal grammar (UG), containing many dozens (or even hundreds) of substantive features or categories, is old news. In Hauser, Chomsky & Fitch (2002), the authors say that the domain-specific faculty of language (=FLN) comprises only the property of recursion, nothing more. (This may still be called “UG”, and the headline of this post may thus be a little imprecise – but what I focus on is that Chomsky no longer argues for a rich UG of the sort that would be relevant for the ordinary grammarian and, e.g., for syntax textbooks.)

Why is this worth repeating? Because of a lot of current research on grammar is still based on the assumption that there is a rich set of innate features and categories, not only in phonology. Generative linguists routinely ask questions such as “Is this form in my language a verb or an adjective?”  – “Is this construction noun incorporation or something else?” – “Is this nominal in the spec position of VP or in some other position?” – “Is this element a clitic or an affix?” (cf. Haspelmath 2015). Such questions only make sense if it is assumed that all these category types are given in advance (a priori), and that every form in a language must belong to one of these pre-established categories.

Of course, it could be that Chomsky is wrong, but if so, one would like to see someone saying this clearly. Very strangely, what I observe is anti-Chomskyans rejecting universal grammar (e.g. Evans & Levinson 2009; Ibbotson & Tomasello 2016), and Chomskyans defending universal grammar in some unclear abstract sense – whereas Chomsky himself seems to largely agree with the anti-Chomskyan view. (I really pity newcomers to the field of linguistics – they must be terribly confused by what is going on.)

In a blogpost earlier this year, Norbert Hornstein, who is generally a staunch defender of the Minimalist programme, said that (surprisingly for me, and for him as well) he actually agreed with me that categories and features are probably not innate natural kinds:

I found the whole discussion to be disconcertingly convincing and believe me when I tell you that I did not expect this. MH and I do not share a common vision of what linguistics is all about. I am a big fan of the idea that FL is richly structured and contains at least some linguistically proprietary information. MH leans towards the idea that there is no FL and that whatever generalizations there might be across grammars are of the Greenberg variety.

Hornstein still thinks that there is a “rich UG”, and he says that “the arguments that I generally present for something like a domain-specific UG involve structural conditions on well-formedness like those found in the theories of Subjacency, the ECP, Binding theory, etc. … And all of these forms of argument lose traction when the issue involves features, categories and their innate status.” – apparently not realizing that his favourite arguments for UG also rely on categories and features. In any event, even if he is not as radical about a lean UG as Chomsky’s 21st century writings (where nothing apart from recursion is UG), Hornstein’s view is equally incompatible with current practice in generative grammar.

Thus, generative grammarians currently seem to lack an ideological superstructure. Noam Chomsky is treated as a nice guy who still gives philosophical interviews to students, but does not seem to be relevant to linguistics anymore.

Curiously, the extent to which Chomsky is now out of touch with mainstream generative practice is highlighted by different ways of thinking about alien languages (yes, languages of hypothetical extra-terrestrial beings). On the occasion of the release of the movie “Arrival” in 2016, linguist Jessica Coon (whose university office was used for shooting a scene with the linguist/actor Amy Adams in the movie) was interviewed about alien langages, and her answer reflected the old 20th century view that there is a rich innate (human-)universal grammar:

Question: So if universal language theory only applies to humans, there’s a real danger that if an alien race started communicating we’d have no hope of deciphering it?

Jessica Coon: Yeah, definitely. When people talk about universal grammar it’s just the genetic endowment that allows humans to acquire language. There are grammatical properties we could imagine that we just don’t ever find in any human language, so we know what’s specific to humans and our endowment for language. There’s no reason to expect aliens would have the same system. In fact, it would be very surprising if they did. But while having a better understanding of human language wouldn’t necessarily help, hopefully it’d give us tools to know how we might at least approach the problem.

(The interview was published at and accessed in January 2017, but is no longer available.)

If UG consists of highly specific but accidental properties (with no particular relation to the communicative functions of languages), then there is indeed no reason to think that alien languages would conform to UG.

By contrast, Noam Chomsky’s recent comments bring human languages and alien languages closer together. At a conference organized earlier this year in Los Angeles, by an organization interested in extraterrestrial intelligence, Chomsky said:

“the Martian language might not be so different from human language after all”

This makes good sense if there is no (or only a minimal) UG, because then the alien languages would be constrained by the same factors as human languages (including the laws of physics, which are presumably in force everywhere in the same way).

When I was invited to give a talk at ConSOLE 25 in Leipzig in 2017, shortly after watching the movie “Arrival”, I couldn’t resist commenting on Coon’s views on alien languages – and what I said was not that different from the views expressed by Chomsky (again confirming that I am also a minimalist, of sorts).

I said the following (see my handout here):

We wouldn’t expect aliens to have the same representational (=UG) constraints as humans, because presumably they have different brains and minds. But their languages would be expected to be subject to very similar functional-adaptive constraints as human languages, if the languages are used for communication in much the same way as humans use their language.

What’s different from Chomsky’s thinking here is that I appealed to adaptation to the communicative function of languages, but Chomsky has come to embrace some kind of functionalist (or at least domain-general) thinking as well:

“With the conceptual barriers imposed by the format framework overcome, we need no longer assume that the means of generating structured expressions are highly articulated and specific to language. We can seriously entertain the possibility that they might be reducible to language-independent principles, whether or not there are homologous elements in other domains and organisms. We can, in short, try to sharpen the question of what constitutes a principled explanation for properties of language, and turn to one of the most fundamental questions of the biology of language: to what extent does language approximate an optimal solution to conditions that it must satisfy to be usable at all, given extralinguistic structural architecture?” (Chomsky 2005: 9-10)

I don’t really understand what Chomsky means here, and why he doesn’t mention the communicative function of languages (which has a strong and fairly obvious effect on the shapes of all languages), but in any event, it is clear that Chomsky’s thinking since at least 2002 is not really compatible with the practice of mainstream generative grammar.


Chomsky, Noam. 2005. Three factors in language design. Linguistic Inquiry 36(1). 1–22.
Evans, Nicholas & Stephen C Levinson. 2009. The myth of language universals: Language diversity and its importance for cognitive science. Behavioral and Brain Sciences 32(05). 429–448.
Haspelmath, Martin. 2015. Defining vs. diagnosing linguistic categories: A case study of clitic phenomena. In Dorota Klimek-Jankowska & Krzysztof Migdalski Joanna Błaszczak (ed.), How categorical are categories? New approaches to the old questions of noun, verb and adjective, 273–303. Berlin: De Gruyter Mouton. []
Hauser, Marc D., Noam Chomsky & W. Tecumseh Fitch. 2002. The faculty of language: What is it, who has it, and how did It evolve? Science 298. 1569–1579. doi:10.1126/science.298.5598.1569.
Ibbotson, Paul & Michael Tomasello. 2016. What’s universal grammar? Evidence rebuts Chomsky’s theory of language learning. Scientific American.


11 thoughts on “Chomsky now rejects universal grammar (and comments on alien languages)

  1. On Twitter, David Adger made the following set of comments (
    (1) Why do people persist in the ‘only recursion’ thing? It’s recursion and the mapping’s to the interfaces. Read the bloody paper.
    (2) Syntax builds hierarchical structures, and there are rules about how those structures are interpreted semantically and phonologically. Those rules are the mappings to the interfaces (binding theory, principles of linearisation, etc). They depend on syntactic hierarchy, hence are specific to language, and specific to the species, though they map into domains outside language itself, like reference, sequential order, etc. These ideas have been around since early 90s. They’re really not difficult.

    When I asked about the “substantive features and categories such as “anaphor”, EPP, or vP – in other words, a rich UG”, which I thought was not assumed to play a role anymore, he replied:
    (3) So that’s the “substantive” issue of what’s in UG, vs the architectural one addressed in the Hauser et al paper (cf substantive/formal universals). These are hypotheses & everyone is working to reduce them from what is apparently descriptively needed, to what is explicable, e.g. reduction of count/mass & telic/atelic to same feature. Notion of change of state in evidentials & aspect. Anchoring to discourse via tense/person/deixis. Reduction of number systems to 3 features. The task is to reduce to a minimal set that can both explain and describe.
    (4) But in any event the point is: please don’t simplify to ‘recursion only’ it’s. It not what that paper says and not what the theory says.

  2. Sir Norbert the Surnameless says he mostly agrees with this post, but my understanding of the details gives out about two paragraphs in.

  3. Dear Martin:
    As someone has already said above, it seems that you are confusing FLN (the potentially specifically human FLB component) with UG, which is but part of the human genetic endowment that makes us able to learn and use any language.
    See here for a review of that matter:
    But there is another central confusion in your post. You write that “Generative linguists routinely ask questions such as “Is this form in my language a verb or an adjective?” – “Is this construction noun incorporation or something else?” – “Is this nominal in the spec position of VP or in some other position?” – “Is this element a clitic or an affix?” (…). Such questions only make sense if it is assumed that all these category types are given in advance (a priori), and that every form in a language must belong to one of these pre-established categories.”
    But it is simply not true that questions like these only make sense if it is assumed that all these categories are given in advance, and that every form in a language must belong to one of these pre-established categories. Or don’t functionalist grammarians wonder if such a word in such a language is a noun or a verb? You seem to suggest that the “pre-established” categories of generativist grammarians are less acceptable than those of functionalist grammarians because supposedly the former would have a biological basis, and now it turns out that Chomsky says they do not. But that is not serious. I guess you do not want to imply that Chomsky or someone else in the past claimed that the difference between a clitic and an affix has to do with the human genome or that there are genes that when activated produce languages with affixes … Affixes are undoubtedly not determined by UG, but they would be part of the theoretical universe of the generative grammarian even if it were confirmed that human beings are made of musical notes and not of cells, simply because they appear in languages.
    The task of ordinary generative grammarians is not to determine which parts of the assumptions they use in their work are biologically determined (it is necessary to wear a white coat for that), but the task of constructing theoretical models of the underlying structure of languages, and empirically falsifying those models. Nobody gives the ordinary grammarian a list of pre-established categories (or no more than the functionalist grammarian), but the linguist constructs a theoretical model based on the data and properties of languages and formulates hypotheses about the principles that explain them (ie, ordinary science).
    Even though, as Hauser, Chomsky and Fitch suggest, FLN was an empty set, we would still need theories about what an adjective is, about where the direct object is in German, or about what is the difference between a clitic and an affix. Generative grammar is a branch of the science of language, not a creed based on the revelations of a guru.

    • I do indeed want to say that it has been claimed (implicitly) that the difference between an affix and a clitic has to do with the human genome (though of course the differences between languages have nothing to do with gene activation, but with cultural differences). This is what I refer to as “a rich UG”: the (widely held) assumption that children are born with a (genetically determined) knowledge of a rich set of categories, features and architectures. Generative grammarians make the assumption that there is such a rich set of innate tools (natural kinds, like innate emotions, see This assumption is not made by non-generativists, e.g. descriptive grammars like those that can be downloaded here:, and hence the two approaches differ enormously, to the extent that practitioners on both sides cannot understand each other and attend different conferences. I have written about this in 2010 ( and again in 2018 (, and in many blogposts and other papers.

  4. Actually, Chomsky does not reject a rich UG at all, because UG = the genetic endowment for language.

    This is from a recent 2017 paper:
    “The child’s biological endowment is known as universal grammar (UG).”

    The minimalist perspective has been to minimise what is considered specific to language on the computational level, but that does not equal a minimal UG. UG doesn’t say anything about specificity for language of a particular phenotypic trait because that would establish a direct genotype to phenotype link which is simply biologically nonsensical as one would ignore development. So, when domain-general concepts enter into computations of the language faculty in the human mind/brain they can still be domain-specific despite their domain-general evolutionary and maybe even developmental origin. Whether sth. is completely specific to language, that’s a different question to ask and concerns FLN in the sense that you mentioned in your post. But, as of now, no one knows what this “specificity” would entail at the genetic level, so it is a bit of a nonsensical question to ask as the answer remains a truism: only humans have language, so sth. about our genome is “special” insofar as it enables us to develop a language faculty whereas other animals do not.

    In other words, humans have UG, a genetic propensity for language that is quite rich, but it need not be and, given our understanding of the evolution of the language faculty as of now as well a our general understanding of how genetics works, quite likely cannot be 100 % specific to language alone.

    Chomsky’s remarks on alien languages concern the nature of the computational system and, given that he thinks that language may actually be an optimal solution to the interfaces with the sensorimotor and conceptual-intentional system, it is natural for him to assume that such a computational system could have arisen elsewhere and operate on similar principles as the way the system works is actually determined by physics, not biology. But that only refers to the computational system (i.e. the language faculty), aliens may still have a completely different conceptual-intentional system, so that the atoms entering into computations of their language faculty maybe could not be recognised by us. So, Chomsky’s argument is not about the kind of functionalism you have in mind at all, he simply arrives at a similar conclusion because for him the core of language rests upon physical principles which should be invariant and thus may also have entered into the evolution of an alien language faculty.

    Long story short, you conclusion that Chomsky’s thinking is not compatible with mainstream generative grammar is incorrect. His current thought is very much in line with the main ideas driving the generative enterprise and with mainstream generative grammar.

  5. On the last quote, and your comment on it. I think one coud indeed see minimalism as a kind of functionalism, but the ‘function’ of language is a different one than what most ‘functionalists’ have in mind. It seems to me that Chomsky would contest (also now) that the primary function of language is communication. He has given many arguments for that in the past (trying to show that some system aspect of language is bad for communication).

    I think for Chomsky the ‘function’ of syntax (and sometimes he confusingly sometimes calls the thing other people call ‘syntax’ ‘language’) is to pair sound and meaning, phonology and semantics. This is what needs to be done in the most efficient way, with some computational measure of efficiency.

    Many people would say that this is a strange ‘function’. In terms of evolution, Chomsky’s argument is that the function of language must have been primarily a cognitive one: individuals who had Merge could think in a way that other’s could not and this gave them some evolutionary advantage. Only when Merge had spread through a community could it be used for communication, as some kind of secondary function.

    The next question then is how pairing sound and meaning would have a cognitive advantage. Strangely, this question is rarely discussed. I at least no of no place in Chomsky’s work where he directly answers that question. The only solution to the problem I know is by the British philosopher of language Noel Burton-Roberts who claims that giving phonology to thoughts helps one to consider one’s own thoughts. This would then be the main difference between man and animal.

    I think that is at least an interesting thought. To my mind, there is no escape from the conclusion that human language is at least partly shaped by its use in communication. But my objection to a lot of functionalists work is that it takes a very simplified view of language, in that it assumes that communication (sometimes even only the restricted form of communication that is transmission of thought) is the only function. I think we would all be better of if we understood that language has several functions and is somehow shaped by each of them.

    By the way, your phrasing “the assumption that there is a rich set of innate features and categories, not only in phonology” suggests that it is somehow uncontroversial in some circles that phonological UG is rich, but I actually don’t think I know anybody who would endorse such a simple view. To the contrary, the discussion between those who are interested in cognitive views and those interested in functionalist (which in phonology often means: phonetic) views is much older in phonology, has always been much less polemical, and also more fruitful, leading most people to take a fairly nuanced view I think. The only extremists are those who would simply deny that there is anything like phonology at all, since we can explain everything by phonetics and the workings of diachrony; but as I said, the idea of a very rich UG has not been popular for a long time. One might interpret Prince and Smolensky (1993) in such a way (because it poses a universal and innate set of constraints Con), but the doubling of the content of that set with certain phonetic findings was very soon recognized. I am not saying that the problem has been solved in a way that can boast consensus, but I think almost everybody is aware of it.

    • Thanks, Marc – I’m actually aware that not everyone in phonology accepts a rich UG, but the difference between phonology and syntax is that there are some fairly specific proposals for segmental phonology, going back to the 1950s. You can look up a list of (possibly innate, possibly universal) phonological features on Wikipedia. There is nothing of that sort in morphosyntax, so it seems that the question should be equally controversial among syntacticians. Maybe it is, but the two types of syntacticians don’t seem to talk to each other much (and I don’t know what type Chomsky is – maybe I’m wrong in interpreting him as saying different things from what he was saying in the 1960s-1980s).

      And I agree with you of course that communication need not be the only function of language. I don’t know who says this. Language(s) may have many functions. The question is not primarily which functions languages have, but what explains their properties, and it is clear that one cannot understand their properties without considering communication, I think.

      • Hi Martin,

        Just a word about universal features in phonology. My understanding of phonological typology, starting from Ian Maddieson’s groundbreaking (1984) Patterns of Sounds and continuing till today, is that there is very little that is universal about sound systems in terms of segment inventory. (This is backed up by databases much larger than SPA and UPSID, such as PHOIBLE).

        There do seem to be good typological generalizations, but nothing that needs to be seen as specifically linguistic AND innate.

        Also, Jeff Mielke’s (2008) Emergence of Distinctive Features shows nicely that all proposed sets of distinctive features are actually very bad at accounting for the phonologies of the world’s languages because of the prevalence of ‘unnatural classes’ in languages’ phonologies. So that even calls into question the idea that distinctive features are useful for describing phonologies in a comprehensive way.

        And even fundamental Hockettian design features of language, like ‘duality of patterning’ (Martinet’s ‘double articulation’) have been argued by Blevins (2012) and others to not be an all-or-nothing inherent property of language, but rather an emergent one.

        So all in all, it doesn’t look like there is a better case for a specific set of innate phonological features than for morphosyntax, beyond what we get from the fairly wide range of causal factors that phoneticians, phonologists, and “correlationists” have looked at or may look at in the future.

  6. Really interesting. I’d like to read more about this issue as I teach Linguistics at university level.

  7. Just a couple of comments:
    (1) I don’t think the question of FLN being small and containing only recursion is a 21st century Chomsky view, at all.

    I realise interviews are tricky and respondents are not typically perfectly careful. But, it is difficult to misunderstand Chomsky, in the discussion pasted below, as saying anything except that he thinks the evolutionary leap forward was not necessarily a big one. [Note: It is possible that in other places he was explicit in saying that what evolved recently was quite big. If so, citing that would be useful to me, and perhaps other readers.]

    Here is Chomsky in 1979-1980 (From “The Generative Enterprise: Revisited”. p. 49). The book I am citing has two parts, an older part from 1979 and a newer part from around 2002-2004, which is more recent. The section I am citing from, if I understand the book’s format correctly, is from the older part.

    Q: Somehow the evolutionary leap which we are talking about seems to be quite a big one. Yet, the period of differential evolution has been comparatively short.

    Chomsky’s Answer: It could turn very heavily on this one fundamental change, namely the development of the capacity to deal with discrete infinities. It just could be that much of it turns on that. We do not know, but that might reflect a small change in the brain. It could, for example, be a consequence of the increase in brain size or complexity. And if that system develops and links to a conceptual system, it really offers a total new world. For all we know that has happened once in the history of the universe. There is not particular reason to believe that it has happened elsewhere, so those possibilities and in fact even likelihoods make this a very important central issue….

    (2) The question of what evolved recently and without analog in other animals ( potentially FLN as per Hauser, Chomsky and Fitch) is logically different from the concept of UG. So, talking about a light FLN doesn’t quite say if UG itself is rich or not. So, I found that part of your post a bit confusing. In fact, I often find such discussions confusing, as people seem (at least to me) to be mixing up different concepts. As far as I can see, people aren’t referring to the same things in many cases, and maybe that is the source of the discrepancy.

    Honestly, I think I am perhaps being a bit sloppy here too. It is of course possible to separate the concept of specific to our linguistic capacity from a domain-general capacity. As far as I can see, there is nothing antagonistic between something being innate but domain-general, and still being part of UG. Again, the confusions (or assumed issues) might be arising from different usages of the terms (particularly, the term UG).

    It also matters what “leap” we are talking about. There could be one “leap” in that there was a change without an analog in the animal kingdom. Yet, many other things, which are not different in fundamental character from capacities that other animals have (FLB, maybe), might have still been “fine tuned” for the linguistic capacity. So, FLN could be light, but UG defined broadly could still be rich.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.