Recent Focused Series »

Indo-European Origins
Northern California
The Caucasus
Imaginary Geography
Home » Cultural Geography, Latin America, Linguistic Geography

Does Pirahã Allow Recursion?

Submitted by on November 26, 2013 – 8:14 pm 24 Comments |  
piraha_mapThe previous GeoCurrents post examined the alleged lack of numbers in Pirahã. But the uniqueness of this language has been claimed to extend far beyond the lack of counting. Dan Everett, the preeminent expert in Pirahã, has made four claims about both the language and the culture of the Pirahã that would make them unique. First, Everett argues that the Pirahã language does not allow embedding (or recursion more generally) and lacks key grammatical structures, such as ‘relative tenses’. Second, Everett insists that the Pirahã have several peculiar cultural gaps, such as the lack of creation myths and fiction and the absence of any individual or collective memory of more than two generations past; he also claims that they have the simplest kinship system yet documented, and that they have remained completely monolingual despite more than 200 years of regular contact with Portuguese-speaking Brazilians. Third, Everett proposes that “Pirahã culture severely constrains Pirahã grammar” (Everett 2005: 622). In other words, he seeks to furnish a common explanation for both the grammatical and the cultural peculiarities of Pirahã. Finally, Everett views his cultural explanation for the supposed linguistic gaps as a challenge to foundational ideas in linguistics:

“These constraints lead to the startling conclusion that Hockett’s (1960) design features of human language, even more widely accepted among linguists that Chomsky’s proposed universal grammar, must be revised. With respect to Chomsky’s proposal, the conclusion is severe – some of the components of so-called core grammar are subject to cultural constraints, something that is predicted not to occur by the universal-grammar model.”

These four claims have stirred a great deal of controversy, particularly in the popular media (see the article in The Economist and this one in The Chronicle of Higher Education). But Everett’s key arguments have also been challenged by other researchers, most notably by linguists Andrew Nevins, David Pesetsky, and Cilene Rodrigues (2009). Based on Everett’s own data, Nevins, Pesetsky, and Rodrigues argued that the “inexplicable gaps” of the Pirahã language are illusory, nonexistent, or not supported by adequate evidence. In addition, based on the work of Brazilian anthropologist Marco Antônio Gonçalves, Nevins, Pesetsky, and Rodrigues argue that Everett’s characterization of Pirahã culture as exceptional fails as well. According to Gonçalves, Pirahã do have creation myths and are not strictly monolingual. Moreover, even assuming that Everett is correct in his characterization of the Pirahã language and culture, Nevins, Pesetsky, and Rodrigues show that any linguistic and cultural gaps are not linked. The peculiar constructions and grammatical oddities found in Pirahã are actually encountered elsewhere, in languages as varied as German, Chinese, Hebrew, and Adyghe, whose speakers do not share the unusual cultural properties ascribed by Everett to the Pirahã. Finally, Nevins, Pesetsky, and Rodrigues asserted that “even if such a connection [between language and culture] should exist, it poses no conceivable challenge to the proposition that some features of [Universal Grammar] are unique to language”, thus requiring no revision of Chomsky’s model of language, as entailed by Everett. Thus, Nevins, Pesetsky, and Rodrigues conclude that Pirahã “emerges from the literature as … a fascinating language – but at the same time, it is just a language among other languages of the world, a claim that casts no aspersions on Pirahã.”

Let’s consider more closely one of the most crucial issues of this controversy: whether Pirahã allows recursion. One of two things may be understood by the term “recursion”: either putting something inside of something which is then put inside of something else, or putting objects of the same type inside each other. In his delightfully humorous and culinarily informed post, Daniel Harbour appetizingly describes these two senses of “recursion” as “dumpling-borsht recursion” (filling inside dumpling inside borscht) and “turducken recursion” (chicken inside duck inside turkey). Recursion understood in the former sense is obviously present in all languages: sounds are combined into morphemes, morphemes are combined into words, and words are combined into sentences. Thus, the really crucial type of recursion is the latter type, in which objects of the same type are embedded inside one another, like Russian “matryoshka dolls”. Dan Everett also used this definition of recursion: “putting one [multiword] phrase inside another of the same type …, e.g., noun phrases in noun phrases, sentences in sentences, etc.” (Everett 2005: 622). Constructions that involve embedding in this sense include, among others, recursive possession (noun phrase inside a noun phrase) and clausal objects (a clause inside a clause). The following English examples illustrate the relevant structures, which are bracketed in the examples below:

Recursive possession:                          [Mary’s brother’s] canoe has a hole.

Clausal object:                                     John knows [how to make an arrow].

Both types of recursion are potentially infinite, at least as far as grammar is concerned: witness Mary’s brother’s wife’s cousin’s…or John knows that Bill believes that Mary thinks that Dan has said…(The famous “This is the house that Jack built” rhyme illustrates another type of a potentially infinite recursion structure involving relative clauses.)

As it turns out, the Pirahã language does not have the constructions corresponding precisely to those found in the English examples above. But Pirahã is not unique in lacking recursive possession or clausal objects. Moreover, Pirahã has variations of these constructions that are found in many other (unrelated) languages.

Consider recursive possession first. A prenominal possessor (‘Mary’s brother’) is possible in Pirahã, but a possessor noun phrase may not itself contain a possessor (the asterisk before the second Pirahã sentence means that the sentence is ungrammatical):

xipoógi hoáoíi hi xaagá.
Xipoogi shotgun his be
‘That is Xopoogi’s shotgun.’ (Everett 1986: 205)


*kó’oí hoagie kai gáihií ’íga.
Ko’oi son daughter that true
‘That is Ko’oi’s son’s daughter.’ (Everett 2005: 630)


This ban on recursive possession, contrary to Everett’s claims, is not peculiar to Pirahã or other similarly “exotic” languages. It is found in such a familiar language as German, where Hans-ens Auto ‘Hans’ car’ is grammatical, but an attempt at recursive possession such as *Hans-ens Auto-s Motor ‘Hans’ car’s motor’ is not. This prohibition does not mean that German speakers cannot express the latter meaning; they can say the equivalent of ‘the motor of Hans’ car’ instead. So what seems to be ruled out in German is recursion with noun phrases (i.e. noun phrase inside noun phrase inside noun phrase). This prohibition against possessor recursion in German has been explained not by the impossibility of embedding but by limitations on the genitive case. The same effect is found in other genitive environments in German, such as the direct objects of particular verbs and prepositions. Simply put, genitives do not stack in German. Crucially for our discussion of Pirahã, the fact that German appears to show the same restriction suggests that “whatever syntactic switch turns off prenominal possessor recursion in German is also at work in Pirahã”, as Nevins, Pesetsky, and Rodrigues (2009: 368) put it. They also make the obvious observation that “the culture shared by most German speakers is more similar to that of most English speakers than either English-speaking or German-speaking cultures are to the culture of the Pirahã”. Thus, the explanation for the lack of recursive possession cannot be formulated in cultural terms.

The argument based on clausal embedding is similar: the way in which clausal objects are structured in Pirahã is similar to the patterns found in other languages that do not share the cultural peculiarities of Pirahã. Clausal objects in Pirahã are characterized by two peculiarities. First, such objects must contain a special morpheme -sai which is said to make the object clause more nominal; for example, grammatical notions such as tense, aspect, and agreement typically found in clauses cannot be expressed by such a nominalized clause. In this respect, -sai is similar to the English gerundive –ing. The way a Pirahã speaker says ‘He really knows how to make arrows’ is more akin to the English ‘He really knows making arrows’:

hi ob-áaxáí [kahaí kai-sai]
his know-intensifier arrow make-nominalizer
‘He really knows how to make arrows.’ (Everett 1986: 263)


But again, using nominalized clausal objects is not unique either to Pirahã  or to other Amazonian languages. This pattern is found in many other languages, including Quechua, Inuktitut, and Adyghe, as well as in Turkic languages, such as Tatar, where the suffix -w corresponds to the Pirahã -sai and the English -ing:

min [ sineŋ alma aša-w-ıŋ-nı ] bel-ä-m.
I your apple eat-nominalizer-2sg-acc know-present-1sg
‘I know that you ate an apple.’ (literally, ‘I know your apple eating’)


The second peculiarity of Pirahã clausal objects, also identified by Everett, is that such clausal objects follow the verb rather than precede it, like noun objects do. Note that in the above Pirahã example, the noun object kahaí ‘arrow’ precedes the (nominalized) verb kai-sai ‘make’, but the clausal object kahaí kai-sai follows its verb obáaxáí ‘really know’. However, this combination of Object-Verb order with noun objects and Verb-Object order with clausal objects is also far from rare cross-linguistically. Among languages that have the same pattern are German (illustrated below), Hindi, and Wappo (an extinct Yukian language, once spoken in California). In the German examples below, the object is bracketed and the relevant verb is boldfaced.

German Object-Verb (noun object):

Hans hat [die Kinder] gesehen.
Hans has the children seen
‘Hans has seen the children.’

German Verb-Object (clausal object):

Hans sagte, [dass er die Kinder gesehen hat].
Hans said that he the children seen has
‘Hans said that he has seen the children.’


Upon examinaing various properties of Pirahã that Everett claims to be specially constrained by the speakers’ culture, Nevins, Pesetsky, and Rodrigues conclude that

“if speakers acquire the same types of languages whether their home is a German city, a village in the Caucasus, or the banks of the Maici River in Amazonas, Brazil, we have discovered just the kind of disassociation between language and culture that sheds light on the nature and structure of UG”

And while this conclusion runs contrary to Everett’s view of Pirahã as being unique among human languages, it is extraordinary in its own way. Given how closely language and culture seem to be connected at the first glance, it is tempting to see them as inseparable. It is also easy to think that certain cultural peculiarities of a given group—such as lack of numbers—may entail that their language is “primitive”. The discovery that this is not actually so is therefore most remarkable—and the next GeoCurrents post will consider this issue in more detail.

In the meantime, Happy Thanksgiving, Happy Hanukkah and a joyous weekend to all our readers!



Nevins, A., Pesetsky, D., Rodrigues, C. (2009). Pirahã Exceptionality: A Reassessment. Language 85(2), 355-404


Previous Post
Next Post

Subscribe For Updates

It would be a pleasure to have you back on GeoCurrents in the future. You can sign up for email updates or follow our RSS Feed, Facebook, or Twitter for notifications of each new post:

Commenting Guidelines: GeoCurrents is a forum for the respectful exchange of ideas, and loaded political commentary can detract from that. We ask that you as a reader keep this in mind when sharing your thoughts in the comments below.

  • Gerhard Jaeger

    More could be said about genitives and object clauses in German. A few thoughts:

    - Stacked genitives in German are not completely ruled out. At least in my idiolect, the following is okay:

    (1) Peters Vaters Freund

    Peter-GEN father-GEN friend

    ‘Peter’s father’s friend’

    I also found this phrase on the web – just google it in quotation marks.

    Perhaps more importantly, German another possessive constructions which is entirely productive:

    (2) Der Rand von dem Hut von dem Freund von dem Vater von Peter

    the rim of the hat of the friend of the father of Peter

    ‘Peter’s father’s friend’s hat’s rim’

    I don’t know about possessives in Piraha, but if Everett is true that their *only* possessive construction is not recursive, then this is quite unlike the situation in German, where there is a recursive possessive construction.

    Regarding the placement of object clauses, things are also a bit more subtle. Consider the two examples given above (repeated as (3) and (4)):

    (3) Hans hat [die Kinder] gesehen.
    Hans has the children seen
    ‘Hans has seen the children’

    (4) Hans sagte, [dass er die Kinder gesehen hat].
    Hans said that he the children seen has
    ‘Hans said that he has seen the children’

    The two sentences do not form a minimal pair since (3) uses a periphrastic tense (with an auxiliary plus participle) while (4) uses a synthetic tense (with the main verb in past tense). Using a synthetic tense in (3) gives you verb-object order with a nominal object:

    (3′) Hans sah [die Kinder]
    Hans saw the children

    If you put (4) into a periphrastic tense, both orders are – at least marginally – possible.

    (4′) a. ??Hans hat, [dass er die Kinder gesehen hat], gesagt.
    Hans has that he the children seen has said
    b. Hans hat gesagt, [dass er die Kinder gesehen hat].
    Hans has said that he the children seen has
    ‘Hans has said that he has seen the children:.’

    It is usually assumed that the OV-order (as in 4a) is basic and VO is derived via extraposition, a syntactic operation that puts heavy phrases at the end of the sentence.

    So to make a long story short, word order in German is rather complex because several syntactic processes interact. Still, fundamentally German has object-verb order both for nominal and for clausal objects.


    • Asya Pereltsvaig

      Thanks for your comments, Gerhard. Indeed, more could be said about any of the languages I’ve mentioned, so thanks for bringing up these details on German.

      Re: the possessive constructions. As far as I know (and I might be wrong on that), Piraha does not have the analog of “the motor of the car”. If it did, it would be interesting to see if were recursive. As it is, we just can’t know. But when it comes to prenominal possessives, they do seem to be more restrictive. In Piraha they don’t stack. In German, they seem to be more difficult to stack than the postnominal possessives. I don’t know about (1) and what makes it better, but I asked about similarly structured examples in one of my classes where I have a couple of German speakers, and they rejected them. There must be something about (1) that makes it okay and I am not sure what it is. Do you know? But anyway, German isn’t the only language where such a pattern obtains: postnominal possessives stack but prenominal ones don’t. Same thing in Russian: they can get awkward for processing reasons but not ungrammatical, e.g. oshejnik sobaki uchitel’nicy Peti (the leash of the dog of the teacher of Petya). Prenominal possessives don’t stack (there’s probably a morphological reason for that).

      Re: the verb and object order, you are right, I should have used consistent tenses to neutralize the V2 effects. But nonetheless, there is a contrast in (4), whatever the reasons behind it are.

      • Gerhard Jaeger

        @prenominal possessives in German: Acceptability of these constructions is pretty gradient. The general tendency is that prenominal genitives are the better the higher the possessor is on the definiteness and animacy hierarchy, and the lighter it is (choose your favorite definition of lightness). So pronouns are perfectly fine, and an inanimate mass term like ‘water’ is totally impossible:

        (5) *Wassers Gefrierpunkt

        water-GEN freezing point

        ‘water’s freezing point’

        I think my (1) is still fairly okay because both possessors, ‘Peter’ and ‘Peter’s father’ are fairly short and very high in animacy and definiteness. In general, stacked genitives are bad because of their lengths, I guess.

        From a functionalist perspective, this is not all that surprising. These are just the good old tendencies ‘short before long’, ‘definite before indefinite’ and ‘animate before inanimate’ which are at work here (with obvious language specific parameterization).

        Anyway, you write ‘This ban on recursive possession, contrary to Everett’s claims, is not peculiar to Pirahã or other similarly “exotic” languages. It is found in such a familiar language as German…’. As pointed out above, this is not quite correct because German does have recursive possession – just not in the prenominal construal.

        • Asya Pereltsvaig

          Yes, my phrasing wasn’t perfect. However, the postnominal possessives likely involve a very different structure and I am not sure they qualify as the “turkducken” recursion. As for the prenominal possessives in German, it seems that they are restricted even before stacking comes into play (not unlike in English). The stacking apparently only intensifies those effects. That’s interesting.

  • Pingback: Rekursioon | suhtlus

  • bringmeabanana

    This post brings absolutely no new information to bear on the Piraha case and it ignores most of the literature. For two new papers, see: and

    Blogs are interesting things. They have allowed those who have not even carefully read the primary sources to pontificate a mishmash of misunderstanding.

    • bringmeabanana
      • Asya Pereltsvaig

        And I must agree with Dan Everett why exactly?!

    • Asya Pereltsvaig

      If there are specific points that you think I misunderstand, as opposed to disagree with, please point them out. As for the papers you’ve provided the links for (for which I am sure some of our readers would be thankful), I am familiar with them but I didn’t see the point in including them as to my mind they do not provide solid counterarguments that were worth including.

  • Ygor Coelho Soares

    Excellent post. Please, write more posts about Brazil’s linguistic, demographic and cultural features, as I know Brazilians are amongst the most frequent readers of GeoCurrents (or so I’ve read it in an old post here). Besides, Brazil still seems to us, Brazilians, to be very little understood and even known – besides the usual vague stereotypes – by foreigners, so GeoCurrents would be making an internationally useful service. ;-)

    P.S: Let me seize the moment and ask you if you could some day write a post on a somewhat “useless” but nevertheless very intriguing thing that keeps coming to my mind: in terms of probability, can we know who are the “Indo-Europeans who didn’t migrate”, i.e those IE tribes that simply kept living and developed in their Urheimat, even if much later they had to migrate somewhere else? Thanks!

    • Asya Pereltsvaig

      Thank you, Ygor. It does seem that a lot of our Facebook fans are from Brazil. And indeed we haven’t done too many posts on this fascinating country—we’ll try to do more in the future.

      As for your IE question, it appears pretty certain that no descendants of PIE remained in place ever since. Some might have migrated here and there and eventually came back to where it all started (Ukrainians, probably), but they weren’t there for the whole time. Was it just a thought or did you have a reason for this question, I am curious?

      • Ygor Coelho Soares

        Thanks for your reply, Asya. I have no particular reason for that. I just find the IE topics fascinating and have tried to know as much about it as possible (that’s how I came to find GeoCurrent out, btw). I find it intriguing that all IE tribes should’ve abandoned their native lands throughout the centuries, but considering the historic times of the most accepted Urheimat hypothesis (the Ukrainian steppes) I think the extreme instability and mobility of those lands may make that hypothesis (that all IE migrated or were displaced at some time) quite plausible.

        My question was also due to some claims I read that either the Balto-Slavic or, according to others, the Indo-Iranian subfamilies were the “last” IE languages to diverge, so that it meant their speakers were probably the “IE tribes who stayed” in their Urheimat for the longest time before eventually migrating/expanding. Have you heard anything about those claims and about their credibility? Thanks again!

        • Asya Pereltsvaig

          Indeed, the most commonly accepted theory is that PIE originated in the Ukrainian steppes and then various groups separated from it and migrated in various directions, with either the Balto-Slavic or the Indo-Iranian being the last ones left. However, even if it was the Balto-Slavs, they migrated northwest and from their the East Slavs (including modern Ukrainians) migrated “back”. This is greatly simplifying the matters but still…

          • Ygor Coelho Soares

            Thanks for your “simplified”, but also very clear and informed reply. :-)

          • Asya Pereltsvaig

            Thank you. A more detailed response will be forthcoming in Martin Lewis’s and my book, hopefully out next year!

  • möngke

    Everett’s 2009 response (which I think is in the same issue of Language as the Nevins et al article) is probably pertinent here. I read it a few years back but as far as I remember his basic point is that it’s easy to cherry-pick examples of non-recursion from a host of language, but Pirahã is ‘exceptional’ in that all of these features occur TOGETHER.
    Disclaimer: I’m probably biased here since I’m extremely sympathetic to anyone who tries to mount a defense against UG. I profess no love for Everett’s 2005 paper though. I believe it’s framed in a very simplistic way that does a huge disservice to more sensitive critiques of UG – see especially Levinson’s response in the Current Anthropology paper, p. 637-8.

    • Asya Pereltsvaig

      If none of the things that Everett mentions present a counterargument to Chomsky’s idea of UG, them being all present in Piraha simultaneously doesn’t make a difference…

      And interesting choice of words: UG isn’t attacking anyone so it doesn’t need to be defended against. It’s people like Everett, Levinson et al. who are attacking UG. Most of their arguments are besides the point though… The Lingua volume in response to Levinson et al. is a good example, but too big a topic to be discussed in detail in a comment.

      • Alon

        I’m not sure your argument obtains, Asya.

        Let’s simplify a bit, and say that Chomsky’s version of UG requires recursion (I know that’s not exactly it, but bear with me). It does not require that recursion be implemented in a specific syntactic way, just that there be a way to express recursion.

        Now, showing that a language lacks a specific form of recursion does not invalidate the UG thesis, of course. Thus, your German examples are compatible with UG. But showing that a specific language lacks all known forms of recursion is a much stronger counterargument. Any of Everett’s arguments, individually taken, is compatible with UG. The conjunction of them isn’t.

        It is still theoretically possible that Pirahã shows a hitherto unknown form of recursion, and therefore does not challenge the UG thesis, but that’s by no means a given— and I think it should be up to UG supporters to find evidence for it.

        As for your claim that “UG isn’t attacking anyone”, that’s transparently true inasmuch as UG isn’t a human actor. But proponents of UG have been known to be vicious in their criticism of opponents. (In all fairness, they’ve been equally vicious when the squabbles turned internal, as Postal has shown quite conspicuously.)

        • Asya Pereltsvaig

          I am afraid I didn’t explain the argument clearly enough. “Recursion” refers not to a particular set of constructions (symptom) but to the mechanism that allows such constructions to appear. However, as the German, Hindi, Adyghe, Tatar and other examples show, there is more to these structures than recursion. Hence, the lack of certain constructions does not mean that (the mechanism of) recursion is absent. Quite possibly (even probably) the Piraha have the cognitive mechanism of recursion. That we don’t see the clearest examples of recursive constructions is simply a result of orthogonal factors, each of which is also present in other languages.

          As for your claim that “proponents of UG have been known to be vicious in their criticism of opponents”, this just seem to be judgmental slur, with no supporting evidence. Most typically “vicious arguments” come from the anti-UG camp, but as proponents of UG have shown time and again, such “arguments” typically have no linguistic value whatsoever. Simply put, anti-UG advocates have not shown any LINGUISTIC (i.e. language-based) argument to support their position. “Doing linguistics” in this ridiculous manner is what really annoys the UG folks. All that such anti-UG work does is (unjustifiably) raise public interest and distract real linguists from doing real work that promotes linguistics…

          • Alon

            Your point came across the first time. True, the abstract cognitive mechanism of recursion does not necessarily entail any specific linguistic implementation. But if there is no positive linguistic evidence of any form of recursion in Pirahã (and I don’t think anyone has shown positively that there is), then I don’t see why we simply assume the cognitive mechanism should be there because the theory predicts it. In other words: absence of evidence is not evidence of absence, but it is a strong enough argument that the onus of proof should lie with proponents of recursion.

            I’ll stay on topic and leave the UG dispute for some other time, but I’ll say it pains me to see you can’t keep a level head about it (what need was there to SHOUT?). Robin Lakoff, Pieter Seuren, Paul Postal and a number of others have already made the case much more clearly than I can, anyway.

          • Asya Pereltsvaig

            I wasn’t SHOUTING, just emphasized one crucial word. I am surprised you’ve misunderstood that.

            Re: Piraha recursion, I do think that the Piraha have recursion. I think that the analysis of its embedded clauses as not really clausal is very weak in argumentation. I’ve been studying the so-called nominals in many languages and the more I do so, the more I see that the boundary between “noun phrases” and “clauses” is an artificial one due to early linguists (grammarians) starting with English. If you can define a clause in such a way that it includes everything analyzed as a clause (including “small clauses”) but not the “-sai” “things” in Piraha, do let me know. I’d be curious for my own research. My own take on this is that clauses express propositional content—but the “sai” constituents (let’s call them that for now) do also!

          • Alon

            The notion that the distinction between NP and clause is an artifact of our naïve grammatical theories seems intuitively very plausible to me, but I couldn’t spot a Pirahã clause if it bit me on the nose, so I don’t aspire to do more than try and follow the experts’ analyses.

          • Asya Pereltsvaig

            Uli Sauerland’s
            work is highly recommended in this respect…

  • Avery Andrews

    A bit of companion reading that is to my mind very relevant but not often mentioned in connection with Piraha is Newman and Gayton ‘Yokuts Narrative Style’, orig 1940, reprinted in Hymes (1964) _Language in Culture and Society_ pp 372-381; their point is basically that Yokuts has the resources to build various kinds of complex structures, but mature Yokuts style just doesn’t use them, on the basis that they are childish and frivolous.

    So, people with a relaxed attitude towards stipulative constraints, ie a belief that the power of statistical learning to learn things from indirect negative evidence is probably strong, can conjecture that a disinclination to use complex structures often can lead to their disappearance, of which Piraha appears to at least be an extreme case, even if it turns out in the end not to be absolute.

    People with a stronger belief in the necessity of a limited number of parameters otoh have much more of a problem with Piraha, and possibly an even worse one with German: given that the restrictions on prenominal possessives are so complex, how can they be learned? Tom Roeper has a proposal somewhere, but Gerhard Jaeger’s observations demolish it completely, it seems to me.

    • Asya Pereltsvaig

      Thank you for your comments, Avery. Re: Piraha, it is indeed important to stress the difference between not having some linguistic elements/structures and not using them. I am not sure if the “statistical learning” people make that distinction. As for German, I don’t see a problem with learning the relevant parameter—why exactly is hard to learn?