the "language instinct"

Zwaarddijk · **#81** by **Zwaarddijk** » Nov 24, 2010 4:06 pm

It would be sort of testable in that a constructed language whose grammar is not permitted by UG would not be possible to produce or parse the way we can produce and parse linguistic utterances. (Someone could, of course, sit down with the rules in mind and produce correct statements much like one'd solve a maths problem).

Going about that is kind of difficult, though.

palindnilap · **#82** by **palindnilap** » Nov 24, 2010 4:24 pm

katja z wrote:Thanks palindnilap, I've been hoping you would join the fray.

Nice of you, but I am afraid that I have already shot all my bullets. I am a (former) mathematician, not a linguist !

I wonder if you've come across LFT? I understand it has good claims as well, and I think it has been used in the development of automated translation programmes.

No I haven't. If I understand well, it has the meaning of the sentences embedded while GG has not. In that sense, it sounds a more realistic model about how people (if not computer) function. I don't think even Piaget would have believed that at stage one children learned grammar without meaning, and at stage two they put meaning on top of the cake. :grin:

Has anybody been taking UG in this sense? Variations certainly don't disprove UG, nobody has said that, but if Chomsky was correct, 1) we should be able to account for ALL of this variation in terms of a limited number of universal principles and a limited set of parametres (switches) (in other words, we should be able to fit all known ways of structuring linguistic material into a single, coherent framework), and 2) we should be able to show these were innate, which is where it all gets sticky.

The problem is that the math behind the UG really don't say more than "a finite set of possible grammars", so your 1) is already saying more than I can defend. Gold's theorem, which proves the existence of UG under certain learning conditions, doesn't say that there are any unifying principles or generic frameworks with parameters. Any two grammars members of the UG could well have nothing in common. In that sense, Seeker is right to call UG empirically unfalsifiable.

Now maybe there are further commonsense arguments saying that given that we must have evolved the information about an UG in order for all children to be able to learn their language, it is much more likely that that UG has some coherence. But as I said, this is already more than I am able to defend.

If I go with your wording, how can "a final set of possible grammars" be given in advance (which logically it should be if UG is an innate set of principles for structuring language)? I'm ok with this as a descriptor for the status quo (as long as it captures enough of the variation to be useful, which it apparently does), but not as an evolutionary story.

How to make a good evolutionary story out of UG ? I haven't really think about that, but I guess some more competent people have.

palindnilap · **#83** by **palindnilap** » Nov 24, 2010 4:28 pm

Zwaarddijk wrote:It would be sort of testable in that a constructed language whose grammar is not permitted by UG would not be possible to produce or parse the way we can produce and parse linguistic utterances. (Someone could, of course, sit down with the rules in mind and produce correct statements much like one'd solve a maths problem).

Going about that is kind of difficult, though.

Oh right, thank you, I had thought about that some time ago, but had forgotten about it and was ready to concede untestability. :cheers:

So yes, if Chomsky is correct, then it is possible to construct a language such as children won't be able to learn it (but Gold's theorem is not constructive, so we don't have any sure recipe to construct such a language). How impressive would that that given that they have a limited working memory anyway, I am not sure.

seeker · **#84** by **seeker** » Nov 24, 2010 4:29 pm

Zwaarddijk wrote:It would be sort of testable in that a constructed language whose grammar is not permitted by UG would not be possible to produce or parse the way we can produce and parse linguistic utterances. (Someone could, of course, sit down with the rules in mind and produce correct statements much like one'd solve a maths problem).

Going about that is kind of difficult, though.

I don´t think that would be an accurate test if Chomsky is understood in palindnilap´s way, because the UG proponent would say that the tested grammar is also included in the "set of possible grammars", just that it´s not found in natural cultures because of some extrinsic reasons. That´s what makes it a non-empirical claim.
For making an empirical claim, there must be something more explicit about it, and therefore we must change palindnilap´s wide definition of UG.

palindnilap · **#85** by **palindnilap** » Nov 24, 2010 4:48 pm

seeker wrote:I´m not so sure of this. First, I don´t think that “Chomsky's theory is the way of modeling computer languages”: here you´re merely talking about “formal modelling” which is wider than “Chomsky's theory”. There´s nothing specifically "chomskyan" in doing formal models (for example, Elman has proposed formal models of grammar, but he rejects Chomsky´s theory).

I am not enough of a computer scientist to be sure of that, but I really think that generative grammar is the way to go when writing a compiler. Maybe someone here knows more about it.

Second, I don´t think it´s “the most parsimonious way”. Parsimony is a complex issue, because it depends on a background of evidence that we assume for assessing it. For example, Elman would argue that his connectionist models are more parsimonious, because they are more biologically plausible and they don´t postulate mechanisms beyond our evidence (Elman´s models are based on what we know about how our brains work).

I strongly disagree here. Unless I have used the wrong word, there is nothing mysterious about the kind of parsimony usually referred to within the realm of computer science (and I would have thought, more and more in general epistemology as well). Kolmogorov complexity is the minimal size of a program generating the desired output, and up to an additive constant it doesn't depend on the programming language. It is then rather obvious that one can't make simpler (in that sense) than a chomskyan generative grammar in order to generate all possible grammatically correct sentences. But that is, without any meaning attached to them !

The problem is that, if we accept your definition, Chomsky has proposed a non-empirical claim: “all existent grammars are included in the very large but finite set of all possible grammars”.

In some sense, yes it is non-empirical. The only way I see to test it was given by Zwaarddijk and it can prove the existence of UG, but cannot disprove it (since it is impossible to exhibit children able to learn an infinity of different grammars). Now, the existence of UG is a mathematical result, so I didn't expect it to be empirically testable anyway.

If that´s the way we should understand Chomsky, then he didn´t propose an empirical theory, and his relevance for linguistics is null.

Maybe. Chomsky's UG doesn't prove anything about natural languages, I tend to agree. Yet it suggests some stronger claims - as I told Katja, some small, coherent set of grammars sounds more probable (more "parsimonious"?) than a disconnected set. How stronger a claim, that would require to know more about linguistics than I do.

seeker · **#86** by **seeker** » Nov 24, 2010 5:16 pm

palindnilap wrote:I am not enough of a computer scientist to be sure of that, but I really think that generative grammar is the way to go when writing a compiler. Maybe someone here knows more about it.

Then, how do you explain the fact that non-chomskyans (e.g. Elman) can also do formal models?

palindnilap wrote:I strongly disagree here. Unless I have used the wrong word, there is nothing mysterious about the kind of parsimony usually referred to within the realm of computer science (and I would have thought, more and more in general epistemology as well).

I´ve never said “mysterious”, there´s nothing mysterious in the concept of parsimony. There´s something relative to the researcher´s goal, though. Artificial intelligence researchers can design optical recognition devices that are much simpler than the biological organs, using mechanisms that have never appeared in the evolution of biological species: those examples would be parsimonious for the AI researcher whose goal is constructing a useful device, but they´re unparsimonious for a biologist whose goal is modelling real biological organs.

palindnilap wrote:In some sense, yes it is non-empirical. The only way I see to test it was given by Zwaarddijk and it can prove the existence of UG, but cannot disprove it (since it is impossible to exhibit children able to learn an infinity of different grammars). Now, the existence of UG is a mathematical result, so I didn't expect it to be empirically testable anyway.

We agree on untestability, then.

palindnilap wrote:
seeker wrote:If that´s the way we should understand Chomsky, then he didn´t propose an empirical theory, and his relevance for linguistics is null.

Maybe. Chomsky's UG doesn't prove anything about natural languages, I tend to agree. Yet it suggests some stronger claims - as I told Katja, some small, coherent set of grammars sounds more probable (more "parsimonious"?) than a disconnected set. How stronger a claim, that would require to know more about linguistics than I do.

I still don´t see any relevance for the empirical sciences. I´d agree that formal modelling is useful, but as I´ve argued, there´s nothing specifically “chomskyan” in that.

katja z · **#87** by **katja z** » Dec 03, 2010 2:28 pm

I've just come across the concept of emergent grammar proposed by Paul Hopper:

postulates that rules for grammar and syntactic structure emerge as language is used. It is distinguished from what Hopper calls the A Priori Grammar Postulate, which posits that grammar is a set of rules existing in the mind before anything else, and is exemplified by the school of generative grammar and the concept of Universal Grammar.

This seems to provide an excellent alternative model to UG, in which structural features of language emerge from language use in speaking communities.

The notion of emergent grammar is the basis for Fluid Construction Grammar:

FCG is a fully operational formalism for construction grammars and proposes a uniform mechanism for parsing and production. It integrates many notions from contemporary computational linguistics such as feature structure and unification-based language processing. Rules are considered bi-directional and hence usable both for parsing and production. Processing is flexible in the sense that it can even cope with partially ungrammatical or incomplete sentences. FCG is called 'fluid' because it acknowledges the premise that language users constantly change and update their grammars.

A quick search on google scholar showed a wealth of articles on FCG. I can't believe I've never heard of it! Where to begin reading ... does anyone have any suggestions?

(A machine for multiplying the number of hours in a day would come in handy as well. :whine:

)

seeker · **#88** by **seeker** » Dec 05, 2010 7:37 pm

katja z wrote:I've just come across the concept of emergent grammar proposed by Paul Hopper...
The notion of emergent grammar is the basis for Fluid Construction Grammar...

This is more evidence for my criticism of palindnilap´s argument: there´s nothing specifically “chomskyan” in formal modelling. And UG is still an untestable non-empirical claim, because any learnable grammar will obviously be included in "the very large but finite set of all possible grammars". Even if a new grammar (different from all known natural languages) shows to be learnable, it wouldn´t falsify the claim, because the UG proponent would say that the new grammar is included in the "set of possible grammars", but it was not found in natural cultures because of some extrinsic reasons.
I still don´t see any scientific value in Chomsky´s speculations.

katja z · **#89** by **katja z** » Dec 05, 2010 7:55 pm

seeker wrote:
katja z wrote:I've just come across the concept of emergent grammar proposed by Paul Hopper...
The notion of emergent grammar is the basis for Fluid Construction Grammar...

This is more evidence for my criticism of palindnilap´s argument: there´s nothing specifically “chomskyan” in formal modelling. And UG is still an untestable non-empirical claim, because any learnable grammar will obviously be included in "the very large but finite set of all possible grammars". Even if a new grammar (different from all known natural languages) shows to be learnable, it wouldn´t falsify the claim, because the UG proponent would say that the new grammar is included in the "set of possible grammars", but it was not found in natural cultures because of some extrinsic reasons.
I still don´t see any scientific value in Chomsky´s speculations.

Agreed, I posted this to supplement my previous comment that generative/transformational grammar was only one of the possible contemporary approaches to grammar. That was a general comment, but it seems to be true of formal modelling as well.

I've just downloaded a few papers on FCG. Here are two of them in pdf, I haven't read them yet so I can't comment, but they are by its "inventor" so I suppose they should give you a fair idea.
A (very) brief introduction to FCG
Unify and merge in FCG

katja z · **#90** by **katja z** » Dec 06, 2010 7:53 pm

On a completely different note, and to continue a bit with the creole argument: I'm just reading up on Salikoko Mufwene, who tells a very different story from Bickerton. He argues that "creoles have developed by the same restructuring processes that mark the evolutions of noncreole languages" and that "contact is an important factor in all such developments" (i.e., he treats all language change as driven by contact and competition - be it among same-language sociolects or among varieties of different languages).
Source: S. Mufwene, The ecology of language evolution, Cambridge University Press, 2001, p. 1
(I haven't read the book yet, but I'm reading a paper by him on population movements and contacts in language evolution, and quite like what I've read so far.)

So even within creolistics, there are perfectly good rival accounts of the formation of creoles that do not pressupose anything like an UG. I suspected as much when it started to dawn on me that the story of creole formation was probably not quite as simple and clear-cut as Bickerton would have it, but it's nice to see there is actual research that bears out at least part of my ramblings. :grin:

ETA: Mufwene argues that the notion of an abrupt development of creoles from pidgins is a itself a myth (so much for my confident assertion at the beginning of the thread that this is well-established ... :oops:

):

"History suggests, instead, a gradual development from the colonial koiné ancestors spoken as vernaculars by the Creole populations of both European and non-European descent in the homestead communities that preceded the plantation communities."
("Population movements and contacts in language evolution", Journal of language contact, 2007, p. 5; link to pdf here)

edited for grammar :doh:

Gentleheart · **#91** by **Gentleheart** » Dec 13, 2010 5:48 pm

Mr.Samsa wrote:God I hate the creole argument.. Can you explain to me how it's supposed to demonstrate universal grammar?

I hope you don't mind me focusing on this because I thought that the Creoles were supposed to demonstrate a universal grammar (according to the language instinct) because they develop in a single generation through the children of "pidgeon" language speaking parents. The argument being that the children fill in the grammatical blanks left by the pidgeon language because during a crucial phase in early childhood (18 months - 3yrs) hardware in the brain switches on and het presto grammar appears. I'm afraid I don't really know what I am talking about but I did think that this was the reasoning.

katja z · **#92** by **katja z** » Dec 13, 2010 6:10 pm

Gentleheart wrote:
Mr.Samsa wrote:God I hate the creole argument.. Can you explain to me how it's supposed to demonstrate universal grammar?

I hope you don't mind me focusing on this because I thought that the Creoles were supposed to demonstrate a universal grammar (according to the language instinct) because they develop in a single generation through the children of "pidgeon" language speaking parents. The argument being that the children fill in the grammatical blanks left by the pidgeon language because during a crucial phase in early childhood (18 months - 3yrs) hardware in the brain switches on and het presto grammar appears. I'm afraid I don't really know what I am talking about but I did think that this was the reasoning.

Hey Gentleheart, please check my post just above yours, you will see that the hypothesis about the development of creoles from pidgins that you have presented (correctly) is at least highly questionable, something I've only become aware of recently myself. As creoles do not develop quite so suddenly, courtesy to one generation of children, as Bickerton would have it, they obviously do not prove anything about a supposed UG (and even if they did develop this way, it is not clear to me that this would necessarily constitute hard evidence for an innate grammar).

edited for grammar :doh:

Mr.Samsa · **#93** by **Mr.Samsa** » Dec 13, 2010 11:05 pm

Gentleheart wrote:
Mr.Samsa wrote:God I hate the creole argument.. Can you explain to me how it's supposed to demonstrate universal grammar?

I hope you don't mind me focusing on this because I thought that the Creoles were supposed to demonstrate a universal grammar (according to the language instinct) because they develop in a single generation through the children of "pidgeon" language speaking parents. The argument being that the children fill in the grammatical blanks left by the pidgeon language because during a crucial phase in early childhood (18 months - 3yrs) hardware in the brain switches on and het presto grammar appears. I'm afraid I don't really know what I am talking about but I did think that this was the reasoning.

Thanks Gentleheart :cheers:

I do understand what the argument is though, I'm more concerned about how it's supposed to demonstrate the existence of a universal grammar, even if the specifics weren't questionable (as pointed out by Katja). What I mean is that how do these researchers attempt to separate an innate propensity for grammar from a progression of efficiency in language use.

Gentleheart · **#94** by **Gentleheart** » Dec 14, 2010 6:15 pm

I thought it sounded a bit too good to be true! No worries, at some point I might investigate the whole thing more carefully myself...are Pinkers' books a good place to start? :cheers:

katja z · **#95** by **katja z** » Dec 14, 2010 6:54 pm

Gentleheart wrote:I thought it sounded a bit too good to be true! No worries, at some point I might investigate the whole thing more carefully myself...are Pinkers' books a good place to start?

I suppose so, if you want to hear arguments in favour of a "language instinct", and as long as you are aware that his views are far from being the be-all and end-all on the subject ... um ... I'd say just the reverse, in fact, but don't take my word for it. :grin:

the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Re: the "language instinct"

Who is online