Sunday, October 2, 2016

More on the irrelevance of Everett's research to UG, and a long critical review of Wolfe's lousy book

Here are two pieces you might find interesting.

First, Chomsky has recently been interviewed by the NYT about the Everett/Wolfe junk we have spent far too much time on (Thx to Chomsky for the quote). Here he is asked about Everett's work.

You have mentioned one paragraph that Wolfe got right in his book...what was in that paragraph? Was it an explanation of your work? Why do you think we're seeing this resurgence of analysis? You must get fairly tired of defending your work?!
It was a paragraph in which he quoted my explanation to him of why his crucial example, the Amazonian language Piraha, is completely irrelevant to his conclusions and claims.  The reason is quite simple.  Whether the alleged facts about the language are correct or not (apparently not), they are about the language, not the faculty of language, while the general principles he thinks are being challenged have to do with the faculty of language, explicitly and unambiguously, including the work he cites, and speakers of this language share the common human faculty of language, of course, as illustrated by their fluency in Portuguese.  So his entire article and book are irrelevant to his claims.  To take an analogy, if some tribe were found in which everyone wears a black patch over one eye, it would have no bearing on the study of binocular vision in the human visual system.  The problems in this work extend far beyond the total irrelevance of his examples to his claims, but I won’t elaborate here.I’ve been defending the legitimacy of this work, extensively and in print, for 60 years.  In earlier years the discussion were with serious philosophers, linguists, cognitive scientists.  I’m sorry to see that the resurgence you mention does not begin to approximate that level, one reason why unlike earlier years, I don’t bother to respond unless asked.
I have italicized the most important point: Note: it does not matter if Everett is right because his claims are irrelevant even if correct. This is the critical point and one that has, sadly, been obscured in most discussions. Not that the point has not been made. It has been. Rather, the point is quickly made and then the falsity of Everett's claims are discussed at length. This leaves the appearance that the empirical issues matter to the big point at hand. How? Because the space dedicated to the arguing for their falsity swamps that dedicated to their irrelevance. People conclude that if it really didn't matter then why spend all that time worrying about whether the claims are true. Chomsky's reply is economical and to the point. My suggestion: if you really want to debate the ins and outs of Piraha do so in a separate very technical paper that makes clear that it has nothing to do with UG.

Here is a second review online of Wolfe's book brought to my attention by Peter Ludlow. It is very amusing. I especially like the last paragraph for it identifies the real scandal in all of this.
I’m not worried about Chomsky, however, no more than I’m worried about Darwin’s position in future histories of science. Chomsky’s position too will be just fine. I do worry about how we will look in those histories, however. Because from where I sit the rampant anti-intellectual responses and the failures to distinguish nonsense from solid science in the attacks on Chomsky’s work look more like harbingers of a new dark age, one that rejects thoughtful scientific probes into human nature and levels charges of a new kind of apostasy– the apostasy of using one’s mind instead of gut instinct. And I suspect that, from the perspective of future intellectual historians, Chomsky’s ability to produce this last great piece of work against the backdrop of our new dark age will make his achievements seem all the more impressive.
The shoddiness of our high brow press is the real scandal, and it is one that deserves much more serious attention.


  1. Do you have a link to the NYT interview? Haven't been able to find it online.

    Chomsky's statement that Everett's claims are irrelevant even if correct is factually true, those claims being about the weak generative capacity of a particular (type of) grammar. However, we should ask ourselves why, at the same time, we so readily accept data about weak generation (called "acceptability judgments") as evidence for our theory in many other cases, and even try to give them a scientific look and feel (in what's called "experimental syntax"). I think one can make the argument that Chomsky's criticism really applies to a lot of work in our field, not just that of a certain braggadocious fieldworker.

  2. I'm not sure that Everett's claims are about weak generative capacity (the set of strings in Piraha). They could, at least, also be about the set of structures and the mechanisms for generating those structures (e.g. a normal grammar with limited selectional relations for verbs, no relativization, etc.). Assuming they are correct. I think the point is that, whether we care about strings or structures, what matters is the mechanism in the human faculty of language, as opposed to the particular I-languages of adult consultants.

    The issue of how we get information as scientists about these I-languages and, then, via theorizing, about the nature of the faculty, requires us, i think, to examine all sorts of evidence, including acceptability judgments which can take us a long way, or even experimental work. That evidence is obviously never dispositive, but if it's tied to theoretical positions, it's pretty helpful! Or have I misunderstood your point?

    1. Here's how I see it. Everett's facts could be facts about use, in which case they're irrelevant. In the best case (from his perspective), they are facts about Piraha grammar, but even in this case they're necessarily facts about weak generation. Nothing follows for the strong generative capacity of human I-language (unless, of course, it could be shown that their I-language were really different, such that they couldn't acquire other grammars -- as far as I know, this is known to be false).

      As for acceptability judgments, I think it's too simple to just say that anything is fair game and potentially good evidence. There's a lot of stuff we exclude for good reasons as irrelevant, because it's not the kind of stuff we want our theory to be about (coughing in the middle of sentences, whatever). Same thing with acceptability: do we want our theories to be theories of acceptability of sentences (weak generation)? Chomsky has argued for a long time that the answer is no, and for good reasons I think (the prime one being that we're interested in strong generative capacity, not in the "language" [set of legal strings] weakly determined by I-language if there even is such a thing). The field has almost entirely ignored this shift in his perspective, without any arguments as far as I'm aware. But at the same time everybody happily (and rightly, I think) dismisses Everett's claims as irrelevant. That's the inconsistency that I perceive and meant to draw attention to.

    2. ("other grammars" should've been "infinite grammars")

    3. I must be missing something. Everett's claim is that a single language isn't doing something (in fact, his observations about the special properties of Piraha pretty consistently involve it failing to do something that other languages usually do). That's fundamentally different from claiming that a language *does* do something, such as probing upward, or controlling backward, or dissociating abstract and morphological case, or whatever. Every persuasive instance of a language doing something bears on UG and FL. If it can be persuasively argued that *no* language does something, where you might have expected languages to sometimes do that, that could bear on UG and FL as well. But for one language to not do something doesn't clearly bear on either.

    4. Is there much work on strong generative capacity by the math ling folks with clear results, etc.?

    5. Not sure there is. I don't know about more recent developments, but at least traditional Formal-Language Theory (as reviewed, e.g., by Fitch & Friederici) is only concerned with weak generation, which is why it's mostly irrelevant to linguistics.

    6. @Peter: not sure I understand the connection between what you say and what I said -- assuming this was in reaction to my post, not David's.

    7. Yes, I was reacting to your "why ... we ... accept ... acceptability judgments as evidence for our theory". I don't see any contradiction (implied by your "however") between saying (i) that Piraha's purported lack of any XP-over-XP recursion does not bear on the HCF conjecture that Merge is a saltational property and (ii) that acceptability judgments can provide evidence for a theory of grammar, or for a theory of a grammar. So I must be misunderstanding you.

    8. @Peter: both Everett's claims about Piraha and claims about acceptability as standardly used in the field are at best about weak generation: whether or not certain kinds of sentences are "sentences of the language". (That's different from the kinds of questions you mention, e.g. whether or not some language has upward probing is not something that's determined by acceptability judgments.) Acceptability is some vaguely defined, informal property of strings. But we don't assume that the grammar actually generates strings; something like a string language is at best (very) weakly determined by an I-language. And yet, everybody seems to assume that acceptability is something we want our theory to model (as far as I know, Chomsky is the only one who has explicitly argued against this). At the same time, everybody (again, rightly) rejects Everett's claims because they have no bearing on strong generation -- but neither do acceptability judgments.

    9. Α problem with strong generation is that it has not been connected to anything empirical since we gave up on the idea of direct intuitions about grammatical structure decades ago (still hunting around for a reference from the 60s or early 70s when this idea was taken seriously, but can't find it); what we do have instead is intuitions about semantic properties and relations such as entailment. So Salles has demonstrated clearly that she had an informant who could understand recursive possessives and had a firm intuition about how they should be expressed (same word order as Portuguese, for what that may be worth), but this doesn't really establish grammaticality. For example, syntacticians afaik accept that 'a proud of her children mother would not send them to such a school' is ungrammatical in English, although I'm pretty sure that most English speakers understand them. So it's delicate.

      Another point is that if you are assuming a strong theory UG, as I think most people did back in 2005 when this issue broke out, the question of what to do about gaps in the range of structures that X-bar theory or your favorite replacement for that would predict was a lot more troublesome than it is now, since, most people thought, UG would have to be craftily engineered so that a subset principle would make them learnable.

      However since then, a number of ideas have emerged as to how to exploit 'indirect negative evidence', so that, in effect stipulative restrictions can be learned if they are 'useful' enough.

      My final observation is that afaik nobody who has comparable experience and knowledge of Piraha comparable to Everett's has popped up and said that he's just wrong about the data ... not even his ex. So I don't think the empirical story is over yet.

    10. @Dennis: I'm sorry, I'm still confused. It sounds as if you are comparing Everett's results, i.e. interpretation of data, with methods, i.e., the collection of data -- apples and oranges. As far as I know, Everett's methods for data collection are the same as those of anybody else working on languages not natively spoken by linguists or college students. At the first level of analysis, as well, his methods are orthodox; his analysis of the data as showing that Piraha lacks XP-over-XP recursion is not different in kind from other linguists' analyses that some other data show that Tsez has agreement across clause boundaries, for example, or that Malay has pronouns that can be locally bound, or that Icelandic has case-marked PRO. I thought that it was at this level of analysis that various people are suggesting that Everett's work does not bear on Chomsky's (and other people's) conjectures about Merge and so on. It might bear on a theory of subcategorization (see David Adger's comment), and it might bear on a theory of acquisition (see Avery Andrew's comment), but it doesn't bear on the conjecture that Merge is at the core of syntax, and of the emergence of language in the species.

    11. @Dennis "But we don't assume that the grammar actually generates strings; something like a string language is at best (very) weakly determined by an I-language. "

      I don't follow this -- could you expand? I understand that the view is that the grammar generates structures, but if there is a well defined set of well-formed structures and a well defined function from structures to strings, then won't the grammar/I-language determine the set of strings?

    12. @Peter: I wasn't comparing Everett's results with methods. I was trying to say that generally accepted methods (collection of acceptability judgments) yield the same kind of data that we (rightly) dismiss as irrelevant when they're coming from Everett. So I was comparing types of data to types of data. Both at best tell us something about weak generation, but in one case (acceptability judgments) we standardly make a leap of faith and assume that they somehow also tell us something about strong generation, whereas we are (rightly) reluctant to make that leap in the other case. Sorry if I'm doing a bad job at articulating this.

      @Alex: well, I think the general view (certainly Chomsky's) is that the grammar generates something like sound-meaning pairs, or perhaps sound-meaning-structure triples. In either case, the connection with "strings" is extremely tenuous: the strings we use as stimuli are still very different from the objects at the PF-interface (even more so if Chomsky is right in his speculation that the entire mapping to externalization is entirely separate from the rest). So there's no real sense in which the grammar determines a set of legal strings (an E-language in Chomsky speak). Despite the customary leap of faith, the notion of "well-formedness" in your phrase "well-formed structures" has nothing to do with intuitive "well-formedness" of strings (the structures licensed by the grammar could be all structures generable by Merge, for instance), which consequently has no real status in this theory.

      Chomsky has made these points repeatedly and explicitly at least since the 1993 survey paper with Lasnik:

      "[A] 'formal language' in the technical sense [is] a set of well-formed formulas; in a familiar variety of formal arithmetic, (2 + 2) = 5 but not 2 + = 2) 5(. Call such a set an E-language [...]. In the theory of formal languages, the E-language is defined by stipulation, hence is unproblematic. [...] [I]t is a question of empirical fact whether natural language has any counterpart to this notion, that is, whether [an] I-language generates not only a set of [structures] but also a distinguished E-language [...]. [T]he concept of E-language [...] has no known status in the study of language [...] and [is] perhaps not empirically meaningful at all." (Chomsky & Lasink 1993: 508)

      Or, more recently:

      "In Chomsky 1986, I suggested that the term I-language be used instead of grammar, in one of the senses of this ambiguous term. I added that any other concept of language might be called a variety of E-language ("E" for external). The latter term has come to be used to refer to a (finite) corpus of material, or to some infinite object generated by the I-language, usually a set of well-formed formulas WFF. This usage is unfortunate, and should be abandoned, I think. [...] A set of WFFs is a formal language, determined by some finite generative procedure. Apart from its derivative character, it is not even clear that such a set can be coherently identified for human language [...]." (footnote 5 of his paper in the 2014 Recursion volume edited by Roeper & Speas)

      By the way, none of the above is meant to be some kind of "Chomsky said it, so it must be right" statement. But I think these are important points and they deserve discussion.

    13. Thanks Dennis; I am familiar with these comments by Chomsky, but I have never really followed the argument (there isn't one really in those passages). I accept that the link may be "tenuous" and "derivative" etc etc but it still seems that there must be some close link. I don't really understand the issues here: I am not talking about intuitive well-formedness of strings but a theoretical notion allied to that of grammaticality which we would define as being the image of a well-formed stucture under the appropriate mapping... but I still feel I am missing the point (again!). Could you give me an example of a generative grammar that does not define a set of well-formed strings?

      (I assume we aren't talking about the distinction between sets and probability distributions or fuzzy sets of some sort..)

    14. I don't know what you mean by "a generative grammar" here (not playing dumb here, I really don't understand). I'm not aware of any model of GG that determines a set of well-formed strings, even though of course it's often informally presented in that way. If you take a look at Collins & Stabler's formalization, for instance, strings play not role. Syntactic objects (≠ strings) are defined as derivable over a lexicon (and "derivable" has nothing to do with "acceptable"); period. My (perhaps false) impression is that this is the general consensus, again despite misleading informal presentation (cf. the incoherent term "grammaticality judgment", which is all over the place).

      Why must there be some close link between a grammar and some set of formulas if the grammar doesn't generate formulas?

    15. @Alex: Chomsky at one point held a position according to which he wanted the grammar to assign a structure to every sequence of words. I believe the motivation for this was because he wanted there to be one route to semantic interpretation, as opposed to a canonical one plus error correction subroutines, say. At any rate, if this is the assumption, then, while the notion of weak generative capacity is of course meaningful, the WGC of any reasonable linguistic grammar is simply \Sigma^*.

      You or I might immediately wonder about closure properties of the grammar, such as intersection with regular sets, which would then allow the notion of WGC to become very important and useful even under this assumption; it would allow us to investigate the kinds of strings that could be assigned a coherent structural property.

      It is unfortunate that the people that disparage the use of formal language theory in linguistics miss such obvious and basic points.

    16. I just meant something like a TAG or HPSG or Stablerian MGs .. these all define structures and associated sets of strings.

      I haven't read the Collins and Stabler paper carefully, but they define PHON* as being sequences of phonological features (i.e. strings) and a function Transfer_PF that maps the syntactic objects to strings. So I may well be wrong, but that formalism seems to define a set of WFFs as well.

    17. @Alex: yes, but even if your grammar includes some kind of algorithm that derives linearly organized objects on the PF side, this set of WFFs (if you will) has nothing to do with the informal notion of well-formedness presupposed in experimental investigation of "acceptability". There's simply no logical link between the latter and issues of strong generation (e.g. some linearization algorithm). In the case of Evertt's Piraha claims, everybody's quick to point this out, but it's conveniently ignored in other contexts -- that's all I meant to draw attention to above.

      @Greg: can you elaborate a little on those obvious, basic points that would make clear to the formally naive (like me) why FLT is relevant to what I'm interested in as a syntactician? Or perhaps point to relevant literature? Thanks!

    18. I wouldn't say it has nothing to do with acceptability. What I am talking about is a fairly classic idea of grammaticality, which is one of the factors that affects acceptability (along with some processing factors like length and ambiguities and center-embedding etc. ).

      If you don't have that then how do we use the grammar to account for the linguistic *data* -- corpus data and acceptability judgments?

      I think part of our misunderstanding/ miscommunication may be just about what counts as being in the grammar ... I assume that the mapping to the interfaces, to use the minimalist term, is part of the grammar.

    19. @Alex: "If you don't have that then how do we use the grammar to account for the linguistic *data* -- corpus data and acceptability judgments?"

      But see, this is precisely the question: is this the kind of data we want the theory to account for, rather than, say, form-meaning association/constrained ambiguity (what can[not] mean what)? You're here simply presupposing that acceptability is something to be accounted for, but if the grammar doesn't determine a language in any relevant sense (set of legal strings), as argued by Chomsky, then what is the basis for this assumption (other than that it is a convenient and traditional one)?

    20. @Dennis: I am also a bit confused about what you are trying to say here, but I am wondering if this is just because there are a few similar but distinct points being run together (each of which is perhaps familiar to most readers here, taken in isolation). Namely, the following points seem to be floating around in the discussion:
      (1) The object of interest is not a set of things, but a finite mental object that generates those things.
      (2) The things that are generated are not simply strings, but rather are hierarchically structured expressions.
      (3) The grammatical status of such an expression is not simply a binary yes/no or in/out issue, but rather has various degrees and kinds of deviance.
      (4) The grammatical status of an expression is just one of many factors that go into an acceptability judgement.
      These, I suppose, all contribute to complicating the link between findings about strings and the conclusions that we all care about. But is there something beyond these that you are trying to point out? If so, could you try to situate it relative to these more familiar points?

    21. Dennis, so I think there are two points.

      The first is about the distinction between acceptability of a string versus acceptability of a string under an interpretation; i.e. whether a grammar defines just a set of strings or a set of string meaning pairs. (Maybe we need to loosen our definition of string a little?)
      So if we have a grammar that defines an association between strings and meanings, and says that some strings are ambiguous and some are not, then presumably some strings will receive zero interpretations. The grammar will therefore divide the strings into those that have zero interpretations and those that have one or more interpretations. We call the latter grammatical and the former ungrammatical.

      This uses two questionable assumptions: one, that not all strings receive an interpretation, and two that the grammar defines a set of form meaning pairs (rather than something graded/fuzzy/probabilistic).

      The other point is about whether linguists needs to account for acceptability judgments. So some people, e.g. Geoffrey Sampson, have argued that these reflect a meta-linguistic faculty rather than directly reflecting linguistic competence, and that linguists should focus on spontaneous or elicited language use -- basically looking at corpora. But that is not I assume (on sociological grounds) the argument here.

    22. Thanks both for your responses, I'm glad to see that people are interested in having this discussion.

      @Tim: wrt your (3/4), my point is not that deviance is graded/fuzzy/multifactorial -- that's old hat, and it misses the more important point (made by Chomsky): acceptability/deviance is a hopelessly informal notion, and it has no logical connection to what we take the grammar to strongly generate. As far as I can see, the only way to coherently claim that we want our grammar to be a model of acceptability is to accept the premise that the grammar is a formal language that determines a set of legal strings, and the speaker is essentially like an automaton that can recognize the legal strings (once we control for performance factors etc.). This was, quite clearly, the view adopted in Syntactic Structures, but it was explicitly abandoned later on, at least by Chomsky -- see the above quotes, and others. And again, perhaps he's wrong, but then I want to see arguments. If we say instead that the grammar is a model of constrained ambiguity rather than acceptability, then we should stop pretending that observations about string acceptability are data for our theory (as I have done in my own work, so this is all quite hypocritical :-)). Which means we should stop pretending, for instance, that we have good generalizations about things like islands; we don't.

      Does this make sense? What do you think?

      @Alex: again, I just don't think "acceptability of a string" is what we understand our theory to be a theory of, once we drop the FLT-inspired Syntactic Structures era view of grammar and speaker. What you call "acceptability under an interpretation" is just not acceptability, it's form-meaning pairing. And here, too, we can of course ask the question to what extent the grammar's form-meaning pairing is reflected in our intuitions about meanings associated with strings, but at least then we're talking about something that seems to be closer in character to what we understand our grammar to be a model of. But note that many of the putative "generalizations" in the field don't work this way, and this was precisely my point of departure in this thread: why do we on the one hand agree that claims of the form "language X doesn't have recursion" are irrelevant to what we're interested in (strong generative capacity), while writing paper after paper about claims of the form "language X doesn't have object shift/scrambling/..."?

      So yes, I would deny that linguists need to account for acceptability judgments, on a strictly literal reading of that statement, but not for Sampson's reasons (and I certainly don't share his conclusions) but rather those given by Chomsky. If he's wrong (or my interpretation of his statements), I'd like to know why.

    23. @Dennis: Now is a good time for me to answer your question to me: why is FLT useful for linguistics. The pithy answer is that FLT is not and has never been about string sets. It is about the kinds of regularities that various mechanisms are able to describe. Even if one were to adopt the Chomskyian position that the grammar should assign a structure to every string, one might still wonder whether the mechanisms of the grammar are able to treat (say) the passives as a natural class. If not, then you will have a hard time making sure that you are correctly assigning them the right kinds of meanings, and not, say, the odd Swahili sentence as well.

      The set of patterns generable by natural mechanisms often have a sort of internal coherence. [Closure properties; Abstract families of languages] Often, if you can describe two patterns, you can describe their union. [Closure under union] Importantly for the above example, often, if you can describe a pattern, then you can directly describe any subpattern thereof that can be given in a certain finite state way. [Closure under regular intersection] (This is like the comprehension axiom of ZF, but where the predicate is formulable in a restricted way.)

      One reason people tend to focus on strings is that strings reveal/force surprisingly many properties of structure. (Another is that strings are directly observable.) I think this is the answer as well to your question:

      why do we on the one hand agree that claims of the form "language X doesn't have recursion" are irrelevant to what we're interested in (strong generative capacity), while writing paper after paper about claims of the form "language X doesn't have object shift/scrambling/..."

    24. Dennis: "it misses the more important point (made by Chomsky): acceptability/deviance is a hopelessly informal notion, and it has no logical connection to what we take the grammar to strongly generate." If this is the case, why are books and papers on Minimalist syntax liberally decorated with stars? (Just like books and articles in all the other generative frameworks.) Such intuitions are clearly in fact, together with intuitions of semantic properties and relations, the major source of evidence in syntax, and this requires the product of 'strong generative capacity', whatever that may be taken to be, to be a major determinant of acceptability and/or perceived grammaticality. This can or course be mediated through something called 'phonology', but that has to be substantially constrained in order to endow the whole system with empirical content.

      The idea of the grammar determining a set of structures, each of which 'yields' a string, and determines a system of semantic properties and relations is surely an oversimplification, but it seems like a good place to start, and lots of decent work going on today encounters no motivation to shift to anything more sophisticated (and complex).

    25. Interesting discussion. I've been incommunicado, but let me jump in and say one thing re Dennis O's claim re Everett:

      "everybody (again, rightly) rejects Everett's claims because they have no bearing on strong generation -- but neither do acceptability judgments."

      I don't think that E's claims have much to do with either strong or weak generation. These are properties that individual Gs have; they weakly generate strings and strongly generate structures (with possible mooing to strings). Everett's claims are claims about Piraha Gs and his claim is that they fail to contain certain kinds of rules (ones that embed Ss within Ss). Hence they don't generate structures nor strings of a certain sort. However, as Chomsky makes clear his claims regarding Merge are not about Gs at all but about FL/UG. Our FLs can create Gs with recursion. This requires an operation like Merge to be able to do this. Piranha speakers like all other humans can acquire Gs which are clearly recursive in the required sense. Hence, E's claims about Chomsky's claims about Merge are irrelevant. This is the main fact about E's research; not wrong, irrelevant as they are discussing a different object.

      I believe that it is very important to make this point clear for as soon as we focus the dispute on whether or not E got the Piraha facts right, we seem to be conceding that it matters. But as Chomsky notes, it doesn't. It is refractory to the main issue of interest.

      Of course, this does not imply that facts about Piraha, whatever they turn out to be, are uninteresting. But they are NOT relevant to the big issue that E puts them in service of.

    26. @Greg: Thanks for elaborating. My knowledge of FLT is very limited and mostly informed by surveys such as that by Fitch & Friederici. From those I got (apparently falsely) the impression that FLT is indeed exclusively concerned with string languages, and as such of limited to no relevance for linguistic theory. Chomsky has expressed this view in various places as well, and so has a recent paper co-authored by Norbert (so perhaps he wants to chime in):

      "[W]ork on artificial language learning that is based on formal language theory, with its restriction to weak generation (strings), makes it largely irrelevant to the strong generation (structures) of all natural languages. [...] [W]e do not see too many applications of formal language theory to have been all that profitable in clarifying concepts, formulating and testing hypotheses, or opening new frontiers of research."


      I take it you disagree with this and take FLT to be more generally about mechnisms giving rise to regularities. Fair enough, I have no reason to take issue with this. But I don't see how it really answers my question why facts about string admissibility should necessarily be viewed as evidence for a theory that isn't a theory of string admissibility (IF we follow Chomsky in this regard). So yes, we think we can glean something about structure from strings, but that doesn't mean our theory should be a model of string admissibility/acceptability. Yet, most of the field seems to assume that it is.

      @Avery: Fair question (your original one). But notice that the stars don't always mean the same thing. Sometimes they're supposed to indicate that speakers turn up their nose at this kind of sentence; sometimes they're supposed to indicate that some externalized form doesn't have a particular interpretation. But this isn't always equivalent, and the two meanings are typically conflated, and this is precisely where I feel that the field needs a discussion about what it is that the theory is supposed to be a model of -- string acceptability? sound-meaning associations, including in "deviant" cases (as argued by Chomsky)? both? neither and something else instead? Just saying that speakers have some vague sense of acceptability of strings (which we furthermore KNOW is not a direct reflection of grammaticality in the technical sense) doesn't make it data for the theory.

      @Norbert: I agree with your characterization of the situation, but to me that does have to do with weak vs. strong generation. Suppose Everett's empirical claims were accurate. Then they are claims about inadmissible strings in Piraha (those deriving from structures with clausal embedding, whatever). This is the strongest argument Everett could make (the facts would be compatible with a weaker scenario, where Piraha Gs do license the relevant string but speakers don't use them). We don't care, as you say, even in this case because it doesn't inform our theories of strong generation, i.e. how humanly possible G construct expressions. Yet, at the same time, we standardly accept claims like "English doesn't have scrambling" to inform our theories of strong generation. Why? What's the difference between the two claims? I think that on Chomsky's view of what a grammar is and does (which we can disagree with, of course!), neither type of fact is strictly speaking data for our theory of grammar.

    27. Two different points:
      (i) "Then they are claims about inadmissible strings in Piraha (those deriving from structures with clausal embedding, whatever). This is the strongest argument Everett could make (the facts would be compatible with a weaker scenario, where Piraha Gs do license the relevant string but speakers don't use them)."

      Correct, that could be true and we would try to ferret out the differences. They don't use (produce) them, but they do understand them. Or they parse them like we parse such sentences using fancier techniques. At any rate, if Piraha judgments are that these are uniformly unacceptable and group with other kinds of unacceptable sentences that we believe have a grammatical source then we would have PRIMA FACIE evidence that these are ungrammatical in Piraha. In fact, so far as I know this is the kind of argument we standardly make and I find it to be a good one. Indeed, if Chomsky's practice is any indication, so dos he (and you if I recall a talk I heard you give). This reasoning is defeasible, but it is well grounded so far as it goes.

      (ii) "at the same time, we standardly accept claims like "English doesn't have scrambling" to inform our theories of strong generation. Why? What's the difference between the two claims?"

      I think that the difference is that we don't conclude much about FL/UG from the absence or presence of scrambling in G of English, whereas this is exactly what Everett did (of course, I am abstracting from the fact that "scrambling" is a construction etc and that these are not FL/UG primitives). So, the question is what one infers from the absence of something in a given G. The moral is that one cannot conclude much just by noting the absence. If that is your point, you are completely right.

      Next question: can one conclude that a certain operation/derivation IS absent in a given G. I don't see why not. Prima facie evidence comes from the standard acceptability judgments etc. But this can be overtaken by noting that certain other data require this operation (say it allows for an apparent evasion of minimality effects or inverse binding) and you conclude that simple judgments of acceptability on the primary data (strings generated via scrambling) are bad. So you infer it exists despite the surface evidence being against this conclusion (think successive cyclic movmt in English vs Irish). So, again, there is a difference.

      Last point: I too am having trouble with the what you are driving at: what data do you think does/could/should inform theorizing in syntax? If it is not acceptability, what then? If it is acceptability with caveats, what do you take these to be? I think we all agree that these data are NOT dispositive. We all agree that a G is not a theory of acceptability but a theory of the generative mechanisms. But you seem to find these responses wanting. Ok, how?

      Btw, I have found this discussion interesting, so thx. N

    28. This comment has been removed by the author.

    29. @Dennis: So is your main point the mismatch between the methodology that takes as its data the acceptability of *strings*, and the fact that the object of interest generates (structures that mediate) *string-meaning pairings*?

      If so, then I think the reason I don't see this as a huge problem is that in practice I don't think the methodology actually is based on facts about strings detached from interpretations. A reported acceptability judgement is almost always actually acceptability-relative-to-an-interpretation, it's just that we rarely bother specifying what the interpretation we're concerned with is. A common reason for not bothering to specify it is that there are actually zero interpretations that the string is acceptable with; this is the case that can sometimes seem like "unacceptability of a string alone" (and perhaps it's even handy to define string-acceptability this way, i.e. derived from string-meaning acceptability). Sometimes authors leave it not entirely implicit, and instead just say something like "sentence (5) is unacceptable (under the relevant interpretation)". Perhaps the messiest is when we use something like traces or labeled brackets to indicate the intended interpretation of a string, and then put a star or not on the front of it to indicate the acceptability of the string with that interpretation, for example:
      (1a) * How_i do you wonder [whether Mary fixed the car t_i]?
      (1b) How_i do you wonder [whether Mary fixed the car] t_i?
      I always read this as:
      (1a) * ("How do you wonder whether Mary fixed the car", which x is such that you wonder whether Mary, in manner x, fixed the car)
      (1b) ("How do you wonder whether Mary fixed the car", which x is such that you, in manner x, wonder whether Mary fixed the car)
      Indices in binding examples serve the same purpose, and this is something that we all get exposed to in a first semester syntax course.

      So while I do think this is something that we could bear to be more explicit about, I can't think of any specific cases where it's caused serious confusion (it certainly causes occasional brief confusions in casual discussions) or been detrimental to progress, and I can't see how it's connected to the Everett issues. Could you try to connect the dots for me?

    30. In response to both Norbert and Tim:

      If we think what a grammar does is determine a set of legal strings (an E-language in Chomsky's usage), then we're forced to hope that there are observable indicators of well-formedness in this sense that will constitute the data for our theory. On the other hand, if we think what a grammar does is (say) associating sounds with meanings, including for expressions that we might informally characterize as "deviant" or "degraded", then we have to hope that there are observable indicators of sound-meaning correspondences. So in either case there is, of course, a leap of faith, since we're trying to explain observables with theories of unobservables; that's unavoidable.

      As Tim points out, in many cases the difference between the two views may be academic, in that we often just use acceptability as a shorthand to say that something does or doesn't mean something. But I don't think it's just terminology. Substantive notions like "overgeneration", for instance, become meaningless unless you subscribe to the former view and assume that there is some relevant notion of well-formedness determined by the grammar (as Chomsky has pointed out somewhere, I forget where). And certain kinds of putative empirical generalizations cease to be generalizations strictu sensu. Take islands. Currently what we know is "speakers perceive sentences of the form X as somehow degraded". If assigning expressions a status like "somehow degraded" is something you want your grammar to do, then this may be a true generalization for you. But if you don't think that this is what a grammar should do, e.g. because you think of it as a theory of constrained sound-meaning correspondence, then you won't pretend that we have a good idea of what islands are (as people like to say), unless you can turn such raw behavioral observations into generalizations about sound-meaning associations of the kind computed by the grammar. This may well be possible in many cases, but a) it requires additional work, and b) it's not what the experimental syntax literature does, for instance, where acceptability is treated as an explanandum (unless I'm completely misreading it; but it's telling that the incoherent term "grammaticality judgment" even appears in the subtitle of Schütze's recently re-released book). So this would be a case where carelessness about these issues has caused confusion, in my view.


    31. And yes, I do think that this view of acceptability as an explanandum has been "detrimental to progress", too. Now things necessarily become even more subjective, of course, but to give you a concrete example: I believe it is precisely the Syntactic Structures-style view of grammar that gave rise to the whole "last resort" model of syntax, where features trigger operations which otherwise can't apply. Why? Because in this way we can fairly easily exclude "unwanted" derivations, which are by hypothesis those yielding illegal strings. The apparent attractiveness of the underlying premise, that admissibility of strings (acceptability) is a more or less direct reflection of what is or what isn't generated (grammaticality), is, I believe, the reason for why this feature-driven view of syntax is still so prevalent, despite the fact that it is nothing but a massive, spectacular failure: virtually all movement-driving features, perhaps with the exception of those triggering wh-movement, have absolutely no explanatory force because of their hopelessly stipulative character. In my book, this example tells us that we should ask ourselves if the obsession with acceptability has led to the kinds of theories we want. (I don't expect the above evaluation to be overly popular here, but I think if we're honest, a massive failure is what we have to call it if we are at all interested in explanation.)

      As I said below in response to David, my point with all this is not to be contrarian or poo-poo our field (@Norbert: you're right about the hypocrisy, guilty as charged). But at the very least we have to admit that we tend to be pretty sloppy with these things.

      @Tim: my connection to Everett was simply the observation that patterns of acceptability are necessarily E-language generalizations, and could thus be rejected on similar grounds as Everett's claims as irrelevant to theories of I-language. That was perhaps a little bit polemical. But hey, it gave rise to this discussion. :-)

    32. @Dennis: To be honest, I cannot make sense of what you are saying. I think what follows is partially relevant, but, again, the meaning of your claims elude me.

      Any grammar will define a set of structures. To be useful at all, there must be some way of associating with these structures something observable, like the way they are pronounced. It is a brute fact that no one says, understands, or reads in a fluent way strings of words like "tree big a there's backyard my in." If your theory of grammar can't distinguish between this string and "in my backyard there's a big tree", then you're in big trouble.

      Currently, there is no sophisticated linking theory which takes us from structures defined by the grammar to acceptability judgment behaviour. But we have loose intuitive linking theories which allow us to make gross predictions about such behaviour in many situations.
      We have better theories about linking structures to idealized truth conditions, which provide idealized descriptions of certain semantic judgments. If we didn't have any way or intent of making contact with data, I maintain, we would not be doing anything worthwhile or meaningful.

    33. First, to return to one of my points from before, I worry that there's some confusion coming from running together relatively familiar points:

      The apparent attractiveness of the underlying premise, that admissibility of strings (acceptability) is a more or less direct reflection of what is or what isn't generated (grammaticality), is, I believe, the reason for why this feature-driven view of syntax is still so prevalent ...

      This seems to be objecting to the idea that grammaticality is the only factor that goes into determining acceptability.

      Substantive notions like "overgeneration", for instance, become meaningless unless you subscribe to the former view and assume that there is some relevant notion of well-formedness determined by the grammar (as Chomsky has pointed out somewhere, I forget where).

      This seems to be objecting to the idea that grammaticality is about a binary distinction rather than degrees and kinds of deviance.

      But moving to the specifics of your recent comments:

      Currently what we know is "speakers perceive sentences of the form X as somehow degraded".

      No, I think what we know is that speakers perceive string-meaning associations of form X as somehow degraded. If all we had was the fact that the string "How do you wonder whether Mary fixed the car" is perceived as degraded, then we wouldn't have developed a theory that involves the ECP as we know it, for example.

      If assigning expressions a status like "somehow degraded" is something you want your grammar to do ... But if you don't think that this is what a grammar should do, e.g. because you think of it as a theory of constrained sound-meaning correspondence, then ...

      I don't see the distinction here. The standard model, as far as I'm aware, is that the grammar generates structured objects, and each of those objects (i) has certain degrees and kinds of deviance (perhaps none), (ii) can be mapped to a string (i.e. "interpreted at the PF interface"), and (iii) can be mapped to a meaning (whatever that may be) (i.e. "interpreted at the LF interface"). This is a theory that assigns certain expressions a status like "somehow degraded", and also induces a constrained sound-meaning correspondence.

    34. Both (et al.): sorry that I haven't been able to make myself clear, as I see looking at Tim's paraphrases above. My bad. I don't think I'll be able to clarify here in any way that takes us forward. Just to bring this to a hopefully reconciliatory conclusion (on my part), I don't object at all to the idea that our grammar should distinguish "tree big a there's backyard my in" from "in my backyard there's a big tree" (to take Greg's example). But I do think that there is a meaningful question to be asked even if we agree on this, namely whether we want our theory to say that "tree big a there's backyard my in" is not a sentence of the language, or rather something about why "tree big a there's backyard my in" doesn't mean "there's a big tree in my backyard". And perhaps my insistence on these distinctions turns out to be irrelevant, as the general consensus here seems to be.

      Thanks for your input!

    35. @Dennis: I think that any reasonable grammar formalism that could distinguish sentence pairs like'tree big a there's backyard my in' from 'in my backyard there's a big tree' will assign to the latter a simple characteristic structural property (let's call it 'goodness') in virtue of which one can speak of the set of 'good' structures. We will be able to rewrite our grammars so that they generate exactly the set of good structures. As I suspect this to be the case, I have a hard time seeing what meaning the question you pose should have, namely (reformulated):

      should we use the first grammar or the second grammar

    36. But I do think that there is a meaningful question to be asked even if we agree on this, namely whether we want our theory to say that "tree big a there's backyard my in" is not a sentence of the language, or rather something about why "tree big a there's backyard my in" doesn't mean "there's a big tree in my backyard".

      Yes, I agree that this is a meaningful question. I would say that having as our target a theory that says that "tree big a there's backyard my in" is not a sentence of the language is not a good idea. Instead we should have as our target a theory that says that this string has zero meanings associated with it (i.e. there are zero structured objects (that have any less than severe deviance) for which interpretation at the PF interface produces this string). I think current practice is pretty much in line with this; although I would admit that the way we casually speak about it sometimes obscures this and that this, on the whole, is not a good thing. (And also, for what it's worth, a large amount of work in what might be termed "modern formal language theory", such as Greg's work, is exactly in line with this view, too.)

    37. @Tim: I'm not sure I agree with you when you say:
      having as our target a theory that says that "tree big a there's backyard my in" is not a sentence of the language is not a good idea

      From a purely linguistic perspective, we have no reason to assign any particular structural configuration to the above reversed sentence. No argument about constituency will be persuasive, no argument about interpretation, nothing.

      I say this because most grammar formalisms are such that any normal analysis written in them will not assign any structural description to that sentence. So you would have to add it in. But seems to be no reasonable way of so doing.

      I agree very much with the spirit in which Mark Steedman says that syntactic structure is the trace of the algorithm delivering the meaning. Syntactic structure is the representation of our predictions about something. But the reversed sentence has no meaning, nor does its parts, we read it in a disjointed unintoned manner. In short, this sentence has none of the properties about which we might want our structures to help predict. Why then insist upon assigning it one?

    38. Yes, I agree with you, I think. I'm certainly OK with a theory that assigns no structural description to that string (i.e. generates no structural description that has that string as its PF interpretation). So in a way the result that this is "not a sentence of the language" comes out as a consequence of the fact that we decide not to assign it any structural description, for the reasons you mention. The point I wanted to make was just that those reasons you mention are the driving force, rather than any goals about having something be "in the language" or "not in the language".

    39. How can you decide something like that a priori? Suppose Merge applies freely. Then we can build a structure that linearizes as "tree big a there's backyard my in", but the interpretation system couldn't assign it a coherent interpretation. But from the point of view of the grammar there's nothing good or bad about it. I guess I just don't understand where you want to draw the line. It's clear that we *want* the grammar to assign interpretations to all kinds of "deviant" expressions.

    40. Maybe a way out of Dennis' puzzle is to say that 'grammaticality' is not the thing we are studying, but an apparently almost inevitable byproduct of the theories we know how to construct of what what we're really interested in, which is what strings get what interpretations under what circumstances, plus various other things we might want to say about them, such as stylistic observations.

      So it just not very interesting exactly how the theory rejects "tree big a there's backyard my in", as long as it doesn't say that this is a normal, or even intelligible, way to see "there's a big tree in my backyard".

      I suggest that the idea that the grammar designates certain sentences as unacceptable by assigning them no structure is not so much informal and intuitive as rather a first approximation, which can be improved in many ways without much effort.

      So we can explain the intelligibility of much grammatically defective but intelligible learner English (and the unintelligibilty of other samples of L2 English ) if we have a theory of constraint relaxation, whereby, if the parser fails when all of the constraints are active, it sees what it can do when some of them are deactivated. This might be a bit easier in a 'declarative' theory like LFG than in Minimalism (I put a facility like this into my toy WLFG parser in the 1990s), but it certainly can't be truly hard for anybody.

      Center-embedding limitations otoh can be dealt with by adding the statement that even if a center embedded structure is generated by the grammar, it's bad anyway if the depth of center embedding is too great. If there really is an effect whereby center embedding of greater than depth 1 is avoided (ie rarer than you would expect simply on the basis that the statistics of phrase structure is context free), it is again not such a big deal whether we call this 'performance' or 'competence'. Ditto for the universal superadditive effects that seem to be associated with some of the island constraints.

      Perhaps someday there will be a theory that will do everything that we can do and more, but doesn't use any concept even remotely like grammaticality, but there is no such theory today, and all attempts to put together something that doesn't induce a grammaticality concept seem to me (and most other people here, I'm sure) to be gravely deficient.

  3. @dennis. "everybody seems to assume that acceptability is something we want our theory to model". I don't think that's at all true. No one thinks (well noone sensible) that our theory *models* acceptability judgments. Acceptability judgments are evidence, to be evaluated as to their usefulness in theory building. So take center embedding, not acceptable with 5, say, but we discount that acceptability judgment because it would make our theory rubbish. Or am I still missing the import of your comment?

    1. @David: Yes, you're right; I should've said "acceptability modulo what we take to be interfering factors, such as memory limitations". But those aside, I think the prevailing attitude is that there is some notion of well-formedness that we want our theory to capture. Do you agree?

    2. Well the theory is a theory of a mental object (the language faculty), and that mental object engages in various kinds of behaviours, like judging sentences as acceptable or rhyming or whatever, but the theory isn't a theory `of' well-formedness, or indeed of rhyming, right? We use the various behaviours that the mental object engages in as evidence for our theory of that object. Does anyone think we have a theory of the behaviours by doing that? We might have an *explanation* of the behaviours but the theory isn't a theory of them. Maybe that's where the slippage you're identifying is? People loosely talk about explaining patterns of acceptability, and hence, even more loosely, having a theory `of' those patterns. But I guess what they really mean is that the theory is of the language faculty, and that theory enters into explanations of acceptability patterns (even modulo memory etc).

    3. @David: I don't think that's really the point, if I understand correctly what you write.

      I'll try again:

      If our theory of grammar is supposed to model acceptability of strings (modulo the usual), then we're committed, it seems to me, to the view that the language determines a set of legal strings (or well-formed formulas). Most of the field seems to subscribe to this view, which was indeed the view articulated in Syntactic Structures. Since then, Chomsky has taken the position that no relevant notion of well-formedness exists for natural language.

      Once we accept the post-SS proposition that a) we're interested in strong generation and b) what the grammar strongly determines is (not a set of legal strings but) an infinite set of structure-sound-meaning associations of sorts, then acceptability at best (!) becomes a very weak and indirect source of evidence for our theory, since it now no longer is what the theory is strictly speaking a theory of. As Chomsky says somewhere in the 1993 MP paper, on this view the only empirical criterion for the theory becomes whether or not it correctly associates sound and meaning, including in all sorts of "deviant"/"unacceptable" cases.

      Both of these views can inform theories the mental object we call FL. The question is what we take that mental object to do, which in turn has implications for what is and what isn't valid data for our theory.

      Sorry, I'm beginning to feel that I'm not good at conveying this point, which seems to me a pretty simple yet fundamental one. It was also not my intention to be overly contrarian here. I just think these issues should be cleared up!

    4. @Dennis: Are there particular grammatical principles you have in mind that do not predict any difference in the acceptability of strings?

    5. @Tal: No, because that would presuppose an answer to the very question we're discussing here, i.e. whether a grammar says anything about acceptability of strings. You can say yes or no to that question, either for reasons of methodological convenience or ontological commitment, but it's going to affect what kind of theory you construct and what is and what isn't data for the theory.

    6. If I understand correctly, you're making a distinction between what the theory is of -- say, neurotransmitters -- and what predictions you might be able derive from it (let's say, when controlling for other factors), in this case behavior or mood. But it doesn't seem right to say that grammar doesn't say anything about the acceptability of strings, given a theory of how acceptability judgments are made.

    7. @Dennis. Is the issue just the meaning of `acceptability'? I'm assuming this means `acceptability under an interpretation', since that's 95% of what is used in theoretical argumentation these days, and I guess that's what most of the posters on this thread are assuming. So the judgment is whether string S is acceptable with interpretation M. Are you just making the point that without the `under an interpretation' bit, acceptability judgments are fairly useless? If so, I think most people will agree with you.

    8. @David: I disagree with that assessment. Most work in syntax simply conflates grammaticality and acceptability into some notion of well-formedness that defines the class of expressions to be characterized by the grammar. At least this is what I read into statements like the following, from your textbook:

      "[A] sequence of words is called a string. Putting a star at the start of a string is a claim that it isn’t a grammatical sentence of the language in question." (Adger, Core Syntax, p. 4)

      Here are some further examples:

      "We say that an utterance is grammatical if native speakers judge it to be a possible sentence of their language." (O'Grady & Archibald 2016, p. 139)

      "The psychological experiment used to get to [the speaker's knowledge of language] is called the grammaticality-judgment task. The judgment task involves asking a native speaker to read a sentence, and judge whether it is well-formed (grammatical), marginally well-formed, or ill-formed (unacceptable or ungrammatical)." (Carnie, Syntax, p. 14)

      "A [...] reason for using grammaticality judgments is to obtain a form of information that scarcely exists within normal language use at all—namely, negative information, in the form of strings that are not part of the language." (Schütze 1996/2016, p. 2)

      I don't think these are just innocent, convenient textbook simplifications. They feed directly into a lot of what's going on in syntactic theory, as I explain above in my response to Tim and Norbert.

    9. Addendum, even though I'm already taking up much more airtime here than I should and intended to: I want to emphasize, as I did above, that I don't mean to denigrate in any way the works/authors cited above. They're just particularly succinct examples of a particular view of grammar (one which Omer very eloquently spelled out and defended in another post in this forum), which is not as innocent as is often presupposed, and this should be acknowledged and discussed.

      Just to be clear!

    10. I don't think that the Core Syntax quote backs up your assertion that work in syntax "conflates grammaticality and acceptability into some notion of well-formedness that defines the class of expressions to be characterized by the grammar." That quote is preceded by a discussion of acceptability being influenced by semantic, pragmatic and processing factors and immediately followed by this: "The acceptability of a sentence will also often depend upon whether the sentence expresses the intended meaning. Here the judgement doesn’t state whether the string is acceptable or not, but rather whether the string is assigned the meaning that is specified. So sometimes you might see the claim that a sequence of words is ‘starred (*) under the intended interpretation’. What this means is that the sentence is unacceptable as an expression of a particular proposition."

      My own assessment is that that kind of data (acceptability under an interpretation/pairings of external forms and meanings) is the major source of data for the structure of FL over most of the recent history of Syntax.

    11. @David: In terms of our view of the prevailing attitude in the field, we'll have to agree to disagree. You're right that my citation from Core Syntax was taken out of context, but I don't think that this affects the general point.


  4. This comment has been removed by the author.

  5. (When I saw that I had misspelled Pirahã throughout my post, I couldn't resist editing it, and once I opened that can of worms I couldn't resist altering the content too. Mostly I have just shortened it.)

    I have heard Dan Everett present his work on Pirahã and I have read some of it, and as far as I can tell, he has made an important contribution. The combination of serious and detailed investigation of an understudied language with modern theoretical tools, carried all the way to publication, is something we see all too little of.

    Like Nevins et al., I'm skeptical of his analytic claims since 2005.

    I understand the *irrelevance* charge to be very specific: The absence in Pirahã of like-category embedding (XP-over-XP recursion; Everett 2005) is claimed (by Nevins et al 2009) to be irrelevant to the proposal that Merge is the core property allowing syntax and therefore language (Hauser et al. 2002).

    I think that if the temperature of the controversy were to come down a few degrees there might be an interesting discussion to be had about what the implication could be of the systematic absence in a language of like-category embedding, but right now the perceived stakes are too high for the engaged parties to be dispassionate.

    Let's keep this in perspective and not use Dan Everett as a punching bag for whatever angst we are experiencing about the foundations of the field.

    1. I would be somewhat more sympathetic if E were not such a glutton for fame and fortune. It's not me that's pushing the "big" consequences of his research. This is partially self generated, and he seems to not have stepped back from it. So, I really don't think that the pushback is undeserved. It's not the hotheads on MY side that are the problem.

      However, I am really interested in how you think what E says MIGHT be interesting. There are many languages that don't allow multiple genitive specifiers, for example, yet I don't think that this absence tells us very much (but correct me if I am wrong). What do you think the absence of sentence embedding in a language (say it existed) would reveal about FL or UG or whatever. Having unbounded embedding tells us something (that we need some kinds of recursion). However the absence might not tell us much. BTW, this abstracts away from the question of what rules there might be other than Merge. If there is ONLY merge then it looks like even Piraha requires recursive application of the rule just to get a simple sentence. In fact, just to get ANY hierarchy will require multiple applications of Merge. But, like I say, I am abstracting away from this. N

    2. I might have gone out on a limb there but I was reacting to the apparently spurious invocation of Everett in the above discussion. I would prefer to stay focused on scientifically grounded disputes rather than on personality issues. I think Nevins et al. did an admirable job of that, and of distinguishing among the inconclusive, the unsupported, the irrelevant, and the wrong. It's important to debunk bad analyses (I don't care so much about calling out bad taste), and as far as I know nobody has accused Everett of scientific misconduct such as fabricating data. (Wolfe is another matter; he's not even play-acting at science but creating slanderous propaganda, and I have nothing to say in his defense.)

      For the sake of argument let's suppose there's a language Near-Pirahã, or NP for short, which lacks category-self-recursion (XP over XP).

      Correct, if there is only Merge then NP can't be very different from other languages. Just in subcategorization and/or the existence of things like subordinating complementizers. But it could be interesting to work that out in detail.

      Lots of people don't even think there is subcategorization (Borer, for example), in which case you can't say that NP is special in lacking verbs that subcategorize for a clausal complement. You can try to say that no verb in NP has the right semantics to embed a clause, but then you need a theory of that semantics that is different from subcategorization, and you have to say how it is acquired.

      An alternative would be that NP lacks the appropriate kind of complementizer. But in a language like English, a variety of clausal projections can be embedded, and on the usual analysis, many of those are subparts of a CP, without any complementizer. NP lacks ECM (there are plenty such languages) which could mean that it lacks verbs with the appropriate semantic content, but that story gets less satisfying when NP also lacks any verbs taking finite complements --- and lacks any other self-embedding categories.

      So at the very least NP could provide an argument for subcategorization.

      If Merge isn't all there is then things could get more interesting.

    3. Not all of the heat here is from Everett ... Noam has been quite nasty too. And that recent interview with the Italian journalist is a real shocker. Chomsky should probably just stop giving interviews.

    4. Are you talking about this one?

      If so, what's shocking about it?

    5. @Avery: Is the interview the one Dennis links to? If so, maybe you could tell us what is so shocking. Chomsky makes the same basic points: E's argument is irrelevant even if correct AND it is likely to be incorrect. The only new twist I saw is Ken Hale being tagged as the person who did its of work on Piraha. Is this the nastiness? Otherwise, I cannot find it. If anything, he stayed away from any personal name calling and concentrated on the work. So, IF this is the piece, there is not much there there.

    6. I don't see what's nasty but the part about the Pirahã speaking Portuguese appears to be wrong.

      We don't know much about the effects of bilingualism on i-grammars, but I don't think it is beyond the pale to think that clausal embedding is something that a bilingual speaker could "transfer" from one language to another. There are many cases where languages have borrowed clausal embedding along with a complementizer.

      If some features are regularly transferred by bilinguals, then those features might be misrepresented in descriptions which used bilingual informants, as many do. I mean that certain languages lacking trait X might be falsely reported as having trait X because the consultants were bilingual and also spoke a language with X.

      If feature X were randomly distributed across the world's languages, this might not matter in the long run, but it has also been argued that some traits are correlated with social and industrial aspects of society, for example some traits are only found in languages spoken by small closed communities and other traits are only found in languages with an established literary tradition -- call them literate languages. I'm somewhat skeptical of most of these claims but if they are even partly right, and X is a trait which is only found in literate languages, and the fieldworkers tend to use a literate language as their contact language, then this could lead to us thinking that literate-language trait X is much more ubiquitous than it really is.

      In that case, it would be really important to pay close attention to data coming from monolingual speakers of a language which hasn't been in much contact with any literate languages to see whether there are in fact any differences.

    7. @Peter, I think you're missing the point of Chomsky's Portuguese statement. It shows that those people don't have a different UG, as Everett claims they do (if you take him literally). And as he says, that just ends the discussion. No need to get into details about bilingualism, as far as I can see. What matters is what people qua linguistic creatures can acquire, not whatever quirk some corpus of data may suggest about particular grammars.

      I agree with Norbert. Everett is just a blowhard, and he deserves to be called that.

    8. Yes, that interview .. but I didn't say it was nasty, but rather that it was 'a shocker', in fact on the basis of sloppiness and inaccuracy. But on review, I can see why people thought I was attributing nastiness -- there has been nastiness elsewhere.

      1. Ken did not work on Pirahã, nor was he really Everett's mentor, he talked about it with Everett when Everett was there, just like everybody else who visits MIT with a language in tow. And, like so many other people, Everett found talking with Ken to be very rewarding.

      2. The Pirahã do not all speak fluent Portuguese, the men mostly command some kind of limited Portuguese sufficient for their commercial activities

      3. The fact that humans have a genetic endowment that enables them to learn language but bricks don't has nothing to do with any interesting version of the UG idea; the interesting idea is I think that linguistic structure manifests some distinctive genetic properties of humans as opposed to other animals. I agree with presumed majority here that this is very plausible, but not that it is adequately argued for (I think there are lots almost-arguments lying around, but they're not quite set up right, and typically too intertwined with lo level arbitrary implementation decisions to really get off the ground in a convincing way).

      4. It's probably wrong that chimpanzees exposed to normal language would learn no words ... even some dogs learn a few words such as especially 'walk' when exposed to no special training different from what children get (eg human appears, brandishing the leash, saying in a hi pitched voice with final rise at the end "doyouwantogoforaWALK"). What animals really learn from exposure to human speech is not clear; it is clearly not human language grammars, but it is often more than just noise.

      5. And then there is this bizarre statement "As a matter of simple logic, it would be impossible for the language to contradict any theory of mine, even if the claims about the language were true. The reason is simple. ..." I don't know what to make of that. Perhaps a not very careful way of saying that it usually takes more than a single observation to falsify a nontrivial theory?

      NPR vs Everett is harder to evaluate, but I find it hard to think that they have a knockdown argument given what we know about modal subordination. So I would not be completely astonished to learn that there was a dialect somewhere in English where the correct way to see "I think that that is too big" was. "I am thinking like this. That be too big"

      But the interview does end well with the two sentences about Trump.

    9. @Dennis: I don't know of any evidence that the primary sources of Everett's data spoke fluent Portuguese and there is evidence to the contrary. See Nevins et al. 2009 (the first article, not the reply), section 5.2 for a critical assessment. The reports I've heard from the expedition that Uli Sauerland was part of are consistent with them having very limited command of Portuguese. So as far as I can tell, "They speak fluent Portuguese" is false.

      Everett has been clear from the start that he does not believe there is any genetic difference between the Pirahã and other people, only a cultural one. So for *some* Pirahã to have become fluent in Portuguese is not a challenge to anything he has said (that I know of). I don't believe his cultural explanation but I do think that the bilingualism issue is important in a different way.

      There is a range of possibilities for bilingualism, but two poles would be: Bilinguals have two grammars which are isolated from each other, or bilinguals have two grammars which are intertwined (or one grammar with a set of language-specific alternatives, which I take to amount to the same thing).

      Suppose the truth is somewhere in between. Consider in this light a fairly standard method of doing fieldwork on an undocumented language. The fieldworker finds an articulate, maybe even educated speaker of the undocumented language who also speaks a contact language that the consultant and the fieldworker can use together to discuss the language. So a substantial amount of our typological data is based on the intuitions of people who are fluent in English, or Spanish, or Portuguese, or French, or Russian, or some other regional or colonial contact language. If Everett's grammar of Pirahã is not based on that method, as he claims, then it has a quality that many grammar lack, just in case the concern I mentioned about bilingualism turns out to be founded -- i.e. that *some* kinds of intuitions could be affected by the other language.

      That's why I'm interested in the bilingualism issue, not because I believe Everett's cultural theory. "That just ends the discussion" sounds a lot like "I'm tired of hearing about this" when it's not preceded by a valid argument.

  6. Thanks both for clarifying. I guess you're right about him being sloppy on that point. I don't think it matters, though, whether they "all" speak Portuguese "fluently" or not; what matters is that they have the capacity to acquire it, or other grammars that show signs of recursion in their E-language. I took this to be Chomsky's point.

  7. In case you missed it