Arboreal Linguistics: Reactions to the SCK

[alice]

I haven't finished it yet, but these might still be interesting.

1. I have to redo all my conlangs. Doesn't Mark realise how many there are?
2. Page 169: HAHAHAHAHAHA!!!! You can never have too many words for insects. Who is this "Sam" supposed to be?
3. Syntactic theories seem to be a lot like programming languages: every few years someone comes along with a new one which attempts to fix all the problems with its predecessors, but introduces many more of its own.
4. What do syntax textbooks for people who don't speak English look like?
5. Has anyone tried to catalog all of at least the smallest possible syntax trees? There are five possible trees of three nodes, so surely each of them corresponds to something.
6. Come to think of it, this could be a very useful conlanging aid: a progression of increasingly complex syntax trees, and when you've implemented them all your conlang can be considered to be "finished" on some level. Never mind that the number of trees is likely to be infinite or at least very large.

[chris_notts]

3. Syntactic theories seem to be a lot like programming languages: every few years someone comes along with a new one which attempts to fix all the problems with its predecessors, but introduces many more of its own.

I have to admit I've never gotten along very well with formal syntactic theories. They mostly seem to be the linguistic equivalent of Ptolemaic astronomy. With enough bits bolted on they can accurately describe the languages for which they were developed (mostly English, e.g. for Chomsky's various theories), but they either over-generalise (predict language types that appear to be impossible) or more commonly under-generalise (claim that attested but not common language features are impossible).

They also often have the undesirable property of being prima facie implausible, e.g. Chomsky's Principles and Parameters. The parameter bit in particular seems to violate everything we know about how evolution and the brain works. In what other case do animals have finite, discrete sets of behaviour options built in which can be flipped on and off in their brains in such a binary manner? Given two theories each of which "fits" the facts, I'd always choose the one which is built on fewer, simpler, and more demonstrable general principles, and ones which fit with what else we know about the world.

I don't deny that there are strong tendencies, but I personally prefer theories that minimise the amount of language specific knowledge built-in to the minimum necessary, and to explain more in terms of general processing principles, limitations of the human brain when it comes to real-time processing, and the context in which languages are spoken. Given that people are as they are and have the desires and concerns they do (cooperation, social interaction, ...), there are limits to what you'd expect a language useful to human beings to look like. See for example Hawkins, who has demonstrated that many typological generalisations are derivable from simple processing constraints.

My favourite class of syntactic theory from casual reading: construction grammars, and in particular Radical Construction Grammar (see Croft's book the subject). But to be honest I don't use any formal theory in conlanging, I just read lots of typology books and grammars of real languages and then do what feels right given what I know about real, actual languages and what is attested. My model is my intuition and real-world examples.

4. What do syntax textbooks for people who don't speak English look like?

Good question. If formal linguistics had been developed by a speaker of a non-configurational language, would we really be talking about movement and scrambling rules? The problem is that the dominant developers of popular theories (e.g. Chomsky) have very little interest in typology or knowledge of more than a small number of languages. English seems to be the main evidence used for or against these theories.

[zompist]

3. Syntactic theories seem to be a lot like programming languages: every few years someone comes along with a new one which attempts to fix all the problems with its predecessors, but introduces many more of its own.

Yeah, that's a good analogy. Another similarity with programming is that any particular theory gets more and more filled up with cruft as the years go by, until someone decides to throw it out and start over...

4. What do syntax textbooks for people who don't speak English look like?

Depends on the language! You could do one for French or Irish or Japanese quite easily. The textbooks I consulted are full of non-English examples. I like the variety, but I don't like that approach because I want people to actually do syntax, and you have to have native speaker knowledge to do that.

For reasons described in the book, a syntactic theory that developed out of (say) Serbian or Latin would have to look somewhat different.

5. Has anyone tried to catalog all of at least the smallest possible syntax trees? There are five possible trees of three nodes, so surely each of them corresponds to something.

FWIW, a language need not fully explore the arboreal space, so to speak. E.g. if it's consistently head-final or head-initial, some types of trees shouldn't occur.

One really shouldn't reify the trees too much. I think surface structure is as real as phonemes, but there are puzzles in each area that the tools themselves can't solve. And the more extra nodes you add, the more unlikely the final theory...

[zompist]

They also often have the undesirable property of being prima facie implausible, e.g. Chomsky's Principles and Parameters. The parameter bit in particular seems to violate everything we know about how evolution and the brain works. In what other case do animals have finite, discrete sets of behaviour options built in which can be flipped on and off in their brains in such a binary manner?

I quite agree; in fact I'd put it even more strongly: P&P makes the theory unfalsifiable, since whatever phenomenon comes up, the theorist just adds a new parameter. Plus, the evidence from language acquisition is that there are simply no stages of acquisition that correspond to flipping switches.

But this particular hobbyhorse is mostly just Chomsky's, and honestly it's a side issue. It gets a lot of attention from outsiders because some people (e.g. Pinker) think it's the most exciting thing to tell non-linguists.

For conlangers, I think the most exciting thing about syntax is the loads of new data it offers. Just as learning phonology gave you plenty of sounds that were new to you, syntax gives you a bunch of new behaviors to think about and new decisions to make.

The problem is that the dominant developers of popular theories (e.g. Chomsky) have very little interest in typology or knowledge of more than a small number of languages. English seems to be the main evidence used for or against these theories.

In his latest books, it's hard to feel that Chomsky is even interested in English. But I do have to emphasize that an army of syntacticians has had fifty years to work on other languages, and it's not true that all work or most work is English-oriented. Textbooks, as opposed to Chomsky, are careful to provide multilingual data.

Some of this work, to my eyes, tries too hard to bend other languages into an English-based framework. (E.g. I'm not convinced that a T(ense) node or a Det(erminer) are general things.) But some syntactic theories try to be more universal; Robert Van Valin is the best example I'm aware of.

[akam chinjir]

How widely held is P&P these days? I was under the impression that a lot of times now syntactic variation gets derived from the lexicon instead---so that, e.g., a language will have DPs if it's got determiners to project them, and not otherwise. (But I'm definitely a tourist in this part of the academy.)

If formal linguistics had been developed by a speaker of a non-configurational language, would we really be talking about movement and scrambling rules?

I don't see why not, at least with movement. You presumably still get cases where one and the same constituent has to be interpreted at more than one place in a sentence's structure, or in a place incompatible with its overt position, and that's where you'd posit movement (or copying, or internal merge, or whatever).

The problem is that the dominant developers of popular theories (e.g. Chomsky) have very little interest in typology or knowledge of more than a small number of languages. English seems to be the main evidence used for or against these theories.

I agree with zompist about this.

Mark Baker and Maria Polinsky have both written methodological pieces you might find interesting (Baker, Formal Generative Typology; Polinsky, Linguistic Typology and Formal Grammar). They both do typological work from a formal perspective.

...Just laid virtual hands on the SCK, looking forward to it.

[Frislander]

3. Syntactic theories seem to be a lot like programming languages: every few years someone comes along with a new one which attempts to fix all the problems with its predecessors, but introduces many more of its own.

Yeah, that's a good analogy. Another similarity with programming is that any particular theory gets more and more filled up with cruft as the years go by, until someone decides to throw it out and start over...

Yeah but at the same time I doubt here in Cambridge they teach first year computer science students to program in bleeding FORTRAN or something like that, whereas us linguists in 1^st-year get taught XBar and only XBar , without the lecturers even admitting what everyone knows anyway, that the field has moved on, and that next year (if you even choose to do syntax) you get the joys of minimalism instead.

I think surface structure is as real as phonemes,

Well considering I don't agree that phonemes are a thing, to me that implies that I ought not to believe in surface structure as well. Surely the appropriate comparison is with deep structures?

They also often have the undesirable property of being prima facie implausible, e.g. Chomsky's Principles and Parameters. The parameter bit in particular seems to violate everything we know about how evolution and the brain works. In what other case do animals have finite, discrete sets of behaviour options built in which can be flipped on and off in their brains in such a binary manner?

I quite agree; in fact I'd put it even more strongly: P&P makes the theory unfalsifiable, since whatever phenomenon comes up, the theorist just adds a new parameter. Plus, the evidence from language acquisition is that there are simply no stages of acquisition that correspond to flipping switches.

Unfortunately in Cam we never get that, because our language acquisition professor is a hardline innateness hypothesis supporter, so the material we have in the libraries isn't very good for that sort of thing (though the supervisors might point us in helpful directions).

But this particular hobbyhorse is mostly just Chomsky's, and honestly it's a side issue. It gets a lot of attention from outsiders because some people (e.g. Pinker) think it's the most exciting thing to tell non-linguists.

...and because it's Chomsky's hobbyhorse, it's also practically assumed in most of Cambridge, heck even the historical linguistics paper is shackled to this, it's the one paper I'm seriously struggling on this year, mostly cause all the assignments have been shit questions about the nature of grammaticalisation and syntactic change, and the bloody supervisor asking "what are the implications of this for universal grammar" and I'm just struggling to hold back the urge to tell her I think UG is a pile of shit.

Sorry I've made this about me and my experience in this one university, but then again Cambridge is also supposed to be the best in the country for linguistics so...

[alice]

I forgot:

7. Presumably, if you're brave, there's nothing to stop you making up your own theory of syntax and basing your conlangs on that?

("reify"? I've only ever heard that word in distinctly Marxist contexts. Is this a subtle reference to Chomsky's politics?)

[Salmoneus]

("reify"? I've only ever heard that word in distinctly Marxist contexts. Is this a subtle reference to Chomsky's politics?)

I doubt it. Reification is just treating things as though they are real (i.e. with properties like haecceity and number and whatnot). I.e. things that make philosophers scream. Like, when akam chinjir says things like "You presumably still get cases where one and the same constituent has to be interpreted at more than one place in a sentence's structure, or in a place incompatible with its overt position, and that's where you'd posit movement" - where the metaphors are so mixed up and everything seems to be talking about things that don't exist as though they actually existed...


[but my only linguistics lectures were someone ranting about exactly this. Instead of the 'tree model' of analysis, he proposed what iirc he called the 'sausage' model: the structure of a sentence is that each word follows the next, like a chain of sausages. This obviously has the same predictive power, since any prediction in any theoretical model can be losslessly rewritten as a prediction in any other model, but has the advantage of ontological parsimony...]

[chris_notts]

Mark Baker and Maria Polinsky have both written methodological pieces you might find interesting (Baker, Formal Generative Typology; Polinsky, Linguistic Typology and Formal Grammar). They both do typological work from a formal perspective.

...Just laid virtual hands on the SCK, looking forward to it.

I'm not familiar with Maria Polinsky. I've seen bits of Mark Baker's work on polysynthetic languages and as an outsider found it a somewhat unconvincing and painful analysis of the topic.

I guess my issue with a lot of this stuff is it's not obvious to me why it's even useful. I don't dispute that the phenomena the theories are trying to explain are interesting, but I'm not sure what the formalism gets us that a written description of the surface phenomena doesn't. Even the theories I find less objectionable (e.g. construction grammar), are:

a) not useful for writing an actual language grammar, since the description in terms of the formal theory or its conventions is less clear than just writing grammars the way 99% of field linguists actually do it, using less theory specific terminology and more actual words. Certainly, all the grammars I've ever seen which actually attempt to consistently use a formal theory as the basis for their description are far, far, harder to read than the ones not wedded to the cult of formalism.

b) not that useful for predictive purposes in the general case, since all attempts either over-generate or under-generate compared to attested linguistic variation.

c) for the most part do not really explain the cognitive or biological basis of language, since most of the theories have been developed in isolation from any actual scientific study of how the brain processes language

d) In many cases are not even falsifiable, as mentioned above by someone else, since extra magic bits of machinery can always be inserted or reordered to explain almost any apparent violation of the theory's predictions

None of this is to say that a future theory based on more typological data necessarily couldn't be useful, but I've not seen one that made me think "gosh! this is a useful way to think/write about how languages work". Normally I think, "gosh, this is an over-complicated way of writing that makes me solve puzzles and/or learn dead theory X that was cool at the time to understand what on earth the author is going on about". That's why I tend to prefer reading typology books and grammars, where authors either tend to be less wedded to the latest craze in formalisms or at least keep it to themselves to maximise their audience.

Dixon calls the collection of common terminology and analysis used by typologists and field linguists "Basic Linguistic Theory", to try to give it a bit of gravitas it doesn't currently have. I'm not sure I'd go so far as to call this disparate collection of ideas a theory, but it does have the advantage of being clearer than most of the alternatives and good enough to describe attested variation. Clear description of almost all phenomena gives it at least 1 point that most of the alternatives lack.

[chris_notts]

I guess what I'm trying to say is that, for a theory to be useful, it needs to do at least one of the following:

1. Illuminate rather than obscure the phenomena it's trying to describe
2. Make non-trivial predictions that other theories don't

It's not obvious to me that any current formalism adds much value in either department compared to the less formal approach taken by most typologists and field linguists.

[Salmoneus]

Several of your objections may be obviated by recontextualising your understanding of chomsky.

The key point here is that Chomsky's objectives were primarily philosophical, rather than linguistic in the conventional sense (or, let us say, they were meta-linguistic, or concerned with the philosophy of language).

The problem is, various thinkers in the early to middle 20th century had raised a number of concerns about language, and in particular about the then-assumed idea of language as a rule-following activity. Central here are the concerns raised by Wittgenstein in his Philosophical Investigations regarding ostension, privacy and rule-following, although those who followed him (like Wisdom) or followed similar paths (like Ryle), or in parallel the american pragmatists (quine and davidson with the 'double indeterminacy of translation' and so on) raised similar concerns. Basically, we can imagine a naive metalinguistic model in which people a) learn what words refer to what things (eg by pointing and naming), b) learn how those words are arranged into sentences, and c) learn how the logical structure of propositions 'meant' by sentences relates systematically to the linguistic structure of the sentence, given the meanings of the words. So, you learn words, and then you learn syntactic rules to make well-ordered sentences, and then you learn what sentences mean given the words that are in them. This isn't necessarily intended as a chronological exact description of actual aquisitive processes, but it seems to outline conceptually an idea of how languages can be learnt. And if you can learn languages, there is something that is learnt...

The difficulty is, philosophical inquiry appears to show that all this is impossible. You can't learn what a word means unless you have at least some grammatical concepts (you can have a procedure of pointing and naming, for instance, but that has its own grammatical structure). So it seems that at least some innate (pre-)linguistic prejudices are necessary. And when you look closer, there are many issues where it seems as though learning syntactic rules from vocabulary and observation ought to be impossible, conceptually speaking.

Hence, you get behaviourism, both in philosophy and in linguistics - the idea that language (and other activities) should be thought of as patterns of physical behaviour, learnt as physical responses, the same way dogs learn to salivate when they hear a ringing bell. A dog is not exactly 'following a rule' when she salivates in response to a particular stimulus; it has just in some way 'evolved' an advantageous modification to its probabilisitically-determined response-selection. This is a very simple and solid way to understand language, but many people found it rather creepy. It doesn't allow for (and doesn't require) the existence of the old notion of "minds", or "free will", or the idea of humans being qualitatively (rather than just quantatively) different from dogs, or amoeba. It reduces people to physical cause and effect. And it raises big problems for the whole grammarian idea of "language" as something that can be right or wrong, and "meaning" as something inherent in speech and language. If language is an abstract rule-following procedure, then every utterance, even those never before encountered, has a clear and specific meaning - if I say "please genitally stimulate your MP's favourite ocelot", and you take this as an instruction to stack some orange duplo blocks upon one another, you are in some sense "wrong" in your understanding of my sentence. But if we see "understanding" language as nothing more than a probabilistically-biased instinctual physical response succesfully triggering the aquisition of more valuable goods, then you can't be "wrong" about what a new sentence "means" - at least, not until we see how people react. If understanding language is just being given the most doggy-treats, and I reward you for your block-stacking, then that's just what "please genitally stimulate your MP's favourite ocelot" happens to mean, logic and coherency be damned! If 'mean' even means anything anymore...


Chomsky, however, rejected behaviourism in favour of his own controversial notion of "cognitivism" - the idea that people have "minds", the contents of which include "thoughts", and that sentences "mean" things by virtue of being the way that we choose to, in some sense "encode" our thoughts. In other words, a return to the old, naive model. Which, he recognised, didn't work. But what he realised DID work excellently was if you turned everything upside down: instead of learning words and then puzzling out grammar, if we imagine that we all BEGIN by knowing grammar, everything makes sense. You can learn words easily once you know grammar. You can understand weird sentences easily if you know grammar and words. And you can translate relatively unproblematically from one language to another if they share the same grammar. The obvious model along these lines is to assume that we all share a common grammatical framework - a set of principles - and our overt linguistic behaviour is derived from that framework merely through the setting of a finite and manageable number of variables, comprising actual vocabulary and a small number of syntactic parameters. And lo and behold, in concluding this, you've set up a system of 'surface' and 'underlying' linguistic expressions, and if anyone's skeptical that you have a mind you can say, "look, if I don't have a mind, where do my underlying linguistic expressions live!?"... and if anyone's skeptical that you have thoughts, you can point out that your underlying linguistic expressions look a lot like what were traditionally names thoughts, and thus may be thoughts, or at least directly analogous to thoughts, or perhaps that underlying your underlying expressions we can conceive of an even more fundamentally underlying expression that constitutes thought.


So, the real idea was never about being able to write a language description efficiently. It wasn't even necessarily about making testable predictions about language acquisition, because of course all the above is a conceptual framework, and although it might make suggestions about real-world acquisition, it doesn't actually inherently make predictions about how this underlying framework is instantiated in actual processes. No, it's about rehabilitating the concept of mental phenomena, and about defending the idea of language as a rule-based, and hence predictable, or, as it were, pre-determined activity.

[an analogy: chess. "Cognitivists" in this analogy say that chess has rules - given a certain never-before-seen chess position, we can say which moves are legal, and sometimes even which moves are winning or losing, purely from knowledge of the pre-given rules of chess and the variables of the actual position. "Behaviourists" in this analogy say that the object of chess is for a certain person, the arbiter, to behave in a way that bestows on you the social status of 'winner'. We can make predictions about what moves might or might not cause the arbiter to act in this way, based on our prior observation of the behaviour of arbiters, but ultimately what move is legal or illegal, winning or losing, has to be a matter of experiment. Chomsky needs chess, and language, to have rules. But it doesn't seem the infinite sphere of linguistic potential is sufficiently objectively definable to allow objective rule-following - indeed, it seems as though the whole concept of rule-following by abstract, objective reasoning may be a myth. We cannot therefore objectively judge which of two dissenting people is actually following the rule correctly. Chomsky thinks that if everybody in fact does follow the same rules, then that universal behaviour provides the objective standard, and conflicts of judgement do not arise, except in marginal cases that are easy to recognise (self-recognised temporary error, insanity, etc).]


So to look at you points:

a) it wasn't designed to write language descriptions efficiently. It's the other way around: the fact that Chomsky was able to use his grammar to write language descriptions <i>at all</i> was a test of the plausibility of his philosophical theory. The more you can show that actual languages can be described as following certain rules, the more plausible the idea that language is rule-following (and hence the idea that the mind exists) appears. So when a historical linguist or a field linguist uses Chomsky's grammar (or that of his successors), it's not really that the theoreticians are helping out the linguist, but rather that the linguists are helping out the theoretician...

b) it wasn't intended to make predictions about actual languages; rather, data from actual languages was intended to refine the grammar (to show that it was possible to refine the grammar, to show that it might not be a coincidence that the grammar appeared to describe actual languages). Of course, Chomsky is arrogant, so he did make predictions, but getting them wrong just showed he didn't understand his grammar yet, not that his grammar didn't exist.

c) actual brain processes aren't particularly relevant, since Chomsky is specifically talking about cognitive processes - that is, mind processes, not brain processes. The mind-brain mapping problem is a whole separate philosophical issue. Again, it's sort of the opposite way around: if Chomsky can show what the mind is doing, then actual neurologists (etc) can show what the brain is doing at the same time, and then we can start to theorise about how one relates to the other.

d) the fact that it's not falsifiable is very much a feature, not a bug. It's not a scientific prediction - it's more like, say, a system of notation for use in making scientific predictions. Notation cannot be falsified. Indeed, since Chomsky could not know a priori how language worked, if his grammar HAD been falfisiable - i.e. only capable of describing certain possible languages and not others - then that would have been a major flaw! This means of course that his approach cannot strictly speaking be disproven empirically - but, as a philosophical theory, a way of interpreting the world, that's only to be expected. Philosophical theories can only be disproven - if ever - by reason, not by empirical evidence. [although, of course, enough known evidence can make a theory look too counterintuitive to be appealing to defend.



Chomsky's approach is thus perfectly solid and really quite clever, philosophically speaking (even if I happen to think it's completely wrongheaded). However, as interest in language faded in philosophy in the later 20th century, Chomsky came to have much more of a following about linguists than among philosophers. I am not a linguist... but it appears that many of those linguists, or at least popular descriptions of what those linguists have said, have rather gotten the wrong end of the stick about the cognitive project. And it's very valid to say - although of course Zompist will be along in a moment to chastise us, as this conversation usually goes - that it's not always clear what some linguists actually seem to think, on a metalinguistic level, the purpose of their endeavour really is, or at wholly clear about how the cognitive toolset is actually meant to further their aims.




[CAVEAT: all of the above is terribly over-simplified.]

[Salmoneus]

I guess what I'm trying to say is that, for a theory to be useful, it needs to do at least one of the following:

1. Illuminate rather than obscure the phenomena it's trying to describe
2. Make non-trivial predictions that other theories don't

It's not obvious to me that any current formalism adds much value in either department compared to the less formal approach taken by most typologists and field linguists.

I agree with you. But in the case of Chomsky, the phenomena he's trying to describe, roughly speaking, is how humans can learn, follow, and be judged to have followed objective rules, and hence how the conceptual category of the mental, as a non-physical place in which rules can be 'stored' and 'understood', can have at least a practical utility in describing linguistic phenomena. The non-trivial prediction, if there is one, is that to a surprising extent human languages can be described in common terms, with relatively few 'underlying' differences even where 'surface' differences (between parts of a language and between languages) appear extremely striking - to an extent that seems counterintuitive if we adopt a model in which these languages have developed independently, purely through a process of reward-reinforcement of simple, non-cognitive, physical responses to stimuli.

I'm not sure he ever really made his case, but I can admit that, in context, juxtaposing Chomsky and Skinner, it's not hard to see why his view was seen as the more illuminating at the time. It's probably become less appealing over time as work has been done (in many scientific and mathematical fields) on the spontaneous development of complex orders from simple but asymmetrical starting conditions (which is where your 'convergence due to common processing demands' comes in).

[Frislander]

I once had a chat with a Cambridge non-linguist alumnus (I think an engineer?), who commented that Chomsky's work "was a rather clever bit of mathematics but wasn't really terribly relevant to actual language", and that's kind of stuck with me.

[zompist]

Chomsky's approach is thus perfectly solid and really quite clever, philosophically speaking (even if I happen to think it's completely wrongheaded). However, as interest in language faded in philosophy in the later 20th century, Chomsky came to have much more of a following about linguists than among philosophers. I am not a linguist... but it appears that many of those linguists, or at least popular descriptions of what those linguists have said, have rather gotten the wrong end of the stick about the cognitive project. And it's very valid to say - although of course Zompist will be along in a moment to chastise us, as this conversation usually goes - that it's not always clear what some linguists actually seem to think, on a metalinguistic level, the purpose of their endeavour really is, or at wholly clear about how the cognitive toolset is actually meant to further their aims.

Eh, I have no objection to your explanation of why Chomsky might appeal to philosophers.

As to what syntactians think they're doing-- well, obviously it depends on the syntactician. Chomsky's innatism is shared by some, such as Fodor and Jackendoff and Pinker. Others, such as the cognitive linguistics school, are much more interested in psychology and language acquisition-- though honestly, philosophy is something that they'd see as bringing up interesting questions, but no answers, and advances are seen as more likely to come from studying children or the brain than reading Quine one more time.

[zompist]

I guess my issue with a lot of this stuff is it's not obvious to me why it's even useful. I don't dispute that the phenomena the theories are trying to explain are interesting, but I'm not sure what the formalism gets us that a written description of the surface phenomena doesn't. Even the theories I find less objectionable (e.g. construction grammar), are:

a) not useful for writing an actual language grammar, since the description in terms of the formal theory or its conventions is less clear than just writing grammars the way 99% of field linguists actually do it, using less theory specific terminology and more actual words.

Just to be a little contrarian: the grammars those field linguists write would not satisfy a syntactician, because they leave out the syntax. I mean, yes, they'll tell you if it's SVO or something else, but if you don't know what modern syntax has discovered, you won't ask your informants how it goes in the language.

Back when I was researching numbers, I looked at hundreds of grammars, and of course I was severely disappointed if a grammar didn't bother to list the numbers! Similarly, if a linguist don't know modern syntax, their grammar is likely to simply not answer a bunch of questions. E.g. it may explain basic relative clauses, but not the limits on what can be relativized, or pronoun behavior in very complex relatives, or in what contexts restrictive and nonrestrictive clauses don't behave the same. Or they define cases or pronouns or subjects or valence in ways that are simply wrong, because they're not trained to look for more difficult sentences.

I agree with you that an over-reliance on a particular syntactic theory is likely to make a grammar impenetrable. But I find some notion of transformations highly useful in writing grammars, just as I find modern morphosyntax useful.

I've sometimes said that you don't need to create a single syntactic tree, but the other day I was looking at whimemsz's Ojibwe blog, and his intro page has... some syntactic trees. And they help!

[chris_notts]

I'm not sure he ever really made his case, but I can admit that, in context, juxtaposing Chomsky and Skinner, it's not hard to see why his view was seen as the more illuminating at the time. It's probably become less appealing over time as work has been done (in many scientific and mathematical fields) on the spontaneous development of complex orders from simple but asymmetrical starting conditions (which is where your 'convergence due to common processing demands' comes in).

Sal, I think I'd struggle to reply to everything you wrote above, since my brain has been marinating in Christmas alcohol for a few days. I think the main point I take away from this it is that:

1. To the extent that your characterisation of Chomsky's work as a philosophical pursuit is correct, I think the questions he's trying to answer are not that interesting to me
2. They are also not, I don't believe, even science, since a basic criteria for something to be science is that it is potentially falsifiable and testable
3. To the extent that linguistics is a science (and I think it should be), it's also not what I would consider to be linguistics
4. It's not obvious to me that most linguists themselves understand where Chomsky was coming from, since many of them seem to take his work as a provable claim about the nature of human language and thought, not just an unprovable thought experiment

[chris_notts]

Just to be a little contrarian: the grammars those field linguists write would not satisfy a syntactician, because they leave out the syntax. I mean, yes, they'll tell you if it's SVO or something else, but if you don't know what modern syntax has discovered, you won't ask your informants how it goes in the language.

I think this is a bit unfair. It's true that lots of grammars neglect syntax in favour or morphology, and I've complained about this myself. But I think there are grammars out there which deal with aspects of syntax in a lot of detail but which require one of the various formal theories to do so. There is a difference between being familiar with facts of syntax in e.g. English and testing things like how/if island constraints occur in whatever language you're documenting, and deciding that the best way to present those facts is in formalism X.

Back when I was researching numbers, I looked at hundreds of grammars, and of course I was severely disappointed if a grammar didn't bother to list the numbers! Similarly, if a linguist don't know modern syntax, their grammar is likely to simply not answer a bunch of questions. E.g. it may explain basic relative clauses, but not the limits on what can be relativized, or pronoun behavior in very complex relatives, or in what contexts restrictive and nonrestrictive clauses don't behave the same. Or they define cases or pronouns or subjects or valence in ways that are simply wrong, because they're not trained to look for more difficult sentences.

And I completely agree that the surface facts that fed into some of these syntactic theories are interesting and linguists should know about them. What I dispute is whether the formalism is independently of use once you understand the phenomena they're trying to explain, or whether they're just an inefficient notation which obscures the basic facts at hand.

I agree with you that an over-reliance on a particular syntactic theory is likely to make a grammar impenetrable. But I find some notion of transformations highly useful in writing grammars, just as I find modern morphosyntax useful.

I've sometimes said that you don't need to create a single syntactic tree, but the other day I was looking at whimemsz's Ojibwe blog, and his intro page has... some syntactic trees. And they help!

Maybe Whimemsz is that rare linguistic who's managed to illuminate using a syntax tree. As I said, I usually find that use of these formalisms obscures rather than illuminates what's actually happening. And it could be just me... but I've read a lot of linguistics papers over the years so it's not like I haven't spent quite a bit of time fighting to extract meaning from them.

[zompist]

Just to be a little contrarian: the grammars those field linguists write would not satisfy a syntactician, because they leave out the syntax. I mean, yes, they'll tell you if it's SVO or something else, but if you don't know what modern syntax has discovered, you won't ask your informants how it goes in the language.

I think this is a bit unfair. It's true that lots of grammars neglect syntax in favour or morphology, and I've complained about this myself. But I think there are grammars out there which deal with aspects of syntax in a lot of detail but which require one of the various formal theories to do so. There is a difference between being familiar with facts of syntax in e.g. English and testing things like how/if island constraints occur in whatever language you're documenting, and deciding that the best way to present those facts is in formalism X.

This is a little like saying "The results of double slit experiments are fascinating, but nobody needs this quantum mechanics nonsense." How do you think John Ross came up with island constraints? By carefully working out details of Formalism X, in this case 1960s transformational grammar.

I do sympathize with the feeling that many linguists are too wedded to a particular formalism. And you don't have to be interested in formal approaches to syntax. But those formal approaches are what led to the interesting syntactic facts whose value you recognize.