Linguistics

1. All science and math related topics welcome.

File: treelanguages.600.529.s.png (262KB, 600x529px) Image search: [Google]

262KB, 600x529px

>>7892990
Linguistics is a science. It's not even a soft science; it's fucking science.

>>7892988
I think linguistics is an extremely interesting and promising scientific endeavor, but it's unfortunate that there's so much disagreement among professional linguists. almost every subfield has a few divisions. in historical linguistics you have strict neogrammarians vs lexical-diffusion people. in morphology you have word-based vs morpheme-based approaches. in semantics it seems like nobody agrees on anything. in phonology you have mostly optimality theory people but there are definitely dissenters out there. then of course there's syntax, where there is substantial agreement among a plurality of linguists around minimalist/biolinguistic/chomskyan syntax, but there are a ton of (vehemently) non-mainstream proposals like the different kinds of categorial grammar, construction grammar, hpsg, lfg, gpsg, and other things.

File: 1418687469786.gif (31KB, 255x144px) Image search: [Google]

31KB, 255x144px

>>7892999
>linguistics
>""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""science""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""

>>7893163
in your opinion what makes linguistics unscientific?

>>7893169
It is really difficult to control variables for diverse studies.

linguisticfags, generative grammar or generative syntax and why?

>>7893185
what kind of studies are you thinking of? I can see how that concern would apply to psycholinguistics or sociolinguistics

>>7893169
It's supposed to study the evolution and future of languages as well as other shit in languages such as pronounciation, word etymology, etc, etc, I know...yet it can't prove that [math]animu\; = chinese\;gril\;cartoons[/math]

I like machine learning and sometimes I train recurrent neural networks on books/ rap lyrics for shits and gigs.
I'm interested, what do linguisticfags think about probabilistic language modeling/text generation with things like markov chains and RNNs? Do they shed any light onto human language, or are they just a clever trick that doesn't yield any deeper insight?


>>7893185
do you have any idea what linguists actually do or is that just your knee jerk criticism of everything related to social science?

>>7893232
I don't know too much about machine learning, but I know computational linguists have used it for certain language acquisition problems. There's not really any doubt that domain-general probabilistic learning plays a role in language acquisition, and there are definitely some linguists (mostly people whose formal training is in computer science) who think that's all there is to language acquisition and language in general. that's not the view among mainstream linguists though, I think most linguists would say that some innate knowledge of language plays a role as well.

>>7893203
Those studies do very poorly in reproducibility.

Im just criticizing the term science being thrown around senselessly

>>7893163
>being a highschooler

>>7893244
ok. I guess I wouldn't be surprised if sociolinguistic studies for example had that problem. In the more serious theoretical subfields like syntax and phonology though I think you might be surprised how different the situation is.

>>7893251
I'm not against the fields l, but the scientific method is far more complex than just doing an experiment and report. And historically saying X is a science has caused major issues in academia.

>>7893261
yeah, you're probably right. linguistics will get there before too long though.

>>7892988
What did the one with English included look like

I think that grammars existed in mammalian brains before language ever did.

It's literally theoretical computer science for idiots. And that means something.

>>7893637
Explain this opinion

I don't see how linguistics relates to CS

>>7893416
t. Noam

>>7893251
Sounds more like math than science

What can you actually do with a linguistics degree?

Languagefag here, studied languages (with some linguistics modules) for undergrad, before being a retard and doing a shitty business masters to try and make myself employable. But language is what I'm good at.

I was thinking of looking more into NLP, but I have no stats background, though my programming is coming along. Considering doing a masters in linguistics down the line, but just now sure where I could actually go with it

>>7892988
How do you read this? Say if I looked at Chinese, does that mean Chinese derives from Tamil or Tamil came from Chinese?

Lately, I have been researching the etymology of adjectives specifically with an end to evaluate any effect the phonetics that make up those adjectives might have on its meaning (as opposed to definition) or sentiment of the word.

Doing this with action words, verbs, I found that, for example, the verb "to kill" in all languages have plosives, sibilants, guttural diphthongs and very long or very short, truncated vowels.
This results in an interesting apparent phenomena in that even if the language the verb being said in is unknown to the listener, and the verb is spoken in a neutral tone with no associated "body-language", the listener when asked on their impression on the word "understand" that whatever the verb being said means, it does not sound like something good or pleasant.
With adjectives it is much harder....

>>7893261
>but the scientific method is far more complex than just doing an experiment and report.
what is the scientific method then ?

>>7892988
Two questions.

Do any of you guys do mathematical linguistics? I found some lecture notes online some time ago. It looked crazy hard.

The gist I got is that you end up working in an axiomatic system with thousands of axioms that you don't know.

As a pure math and comp sci fag my main approach to linguistics has always been through formal languages but I don't know dick about how natural languages are dealt with (besides simple stuff I've seen in natural language processing). Do you guys have any intro books on linguistics that you would recommend.

It's shit. Whatever these guys can do, any monkey with machine learning can do way better

File: contrarian.jpg (35KB, 386x600px) Image search: [Google]

35KB, 386x600px

>>7893823
> all languages
.... that "apparent phenomena", complete BS

you are either an amateur and therefor an idiot for pretending or just very young and inexperienced

>>7893194
Aren't both the same theory?

>>7893794
the number means that this amount of sites with language X (in this case Tamil) point to sites with language Y (in this case Chinese)
So around 5% of Tamil sites point to chinese sites.

What's more interesting is how many point to german sites

>still denying that Germany is kebabcountry
even the science backs it up

>>7893861
I know a good amount of categorial grammar. basically all the words in your lexicon are axioms and you have a couple logical implication rules for composing them. the words all have 3 parts, a prosodic form (either some string or a function that gives you a string), a semantics (just a predicate logical statement), and a syntactic type, which just tells you what kind of function the semantics is and how to put string-type prosodies together.

>>7893823
>the verb "to kill" in all languages have plosives, sibilants, gutteral diphtongs, and very long or very short, truncated vowels.
do you mean to say that the word for "kill" in every language has all these features?

>>7892988
can someone explain the link between belorussian and armenian? These languages aren't closely related either (unlike the other slavic languages pointing to russian for example)

>>7892988
My smart cousin (out of like fifteen) has a PhD in it.
He specializes in medieval european languages or some shit

>>7893775

>What can you actually do with a linguistics degree?

I did an undergraduate degree in linguistics and am now doing a PhD. My impression was that most undergraduates who do linguistics don't know what they want to do, but take linguistics because they enjoy languages, or at least found language classes tolerable and thought linguistics would be more of the same. Which it isn't. I don't know what these people do, on the whole. Some go into ESL, which is a mediocre career.

Some people are more serious and go on to graduate school. If you're a "computer person", you could go somewhere with that angle. My impression is that a lot of the people who do linguistics but don't want to go into academia go this route, related to NLP or maybe something more CSy and tangentially language-y. You'll make more money than the academics this way.

For the subset that want to linguistics as an academic career, the problem is that there's a pretty small set of schools that consistently pump their PhDs out into decent academic positions. It varies a bit based on subfield, but there's not many in any case. So if you don't get into a good PhD program, it seems like your statistical odds are bad for academia.

Be aware that there's a difference in being good at languages, and being good at the kind of analysis that is necessary for (theoretical) linguistics. Knowing a lot about languages and having a knack for getting your head around them can contribute to working with the theory, but isn't enough in of itself.

>>7893194
Any theory of language that seeks to accurately predict the attested patterns, and accurately not predict the unattested ones, is a generative theory of grammar. Syntax is just one sub-section of the larger theory of grammar.

>>7893823
This goofery isn't linguistics.

>>7893861
>thousands of axioms
what?

>>7894892
http://people.mokk.bme.hu/~kornai/termeszetes/ml.pdf

>Physicists speak of mesoscopic systems when these contain, say, fifty atoms, too large to be given a microscopic quantum-mechanical description but too small for the classical macroscopic properties to dominate the behavior of the system. Linguistic systems are mesoscopic in the same broad sense: they have thousands of rules and axioms compared with the handful of axioms used in most branches of mathematics. Group theory explores the implications of five axioms, arithmetic and set theory get along with five and twelve axioms respectively (not counting members of axiom schemes separately), and the most complex axiom system in common use, that of geometry, has less than thirty axioms.

>It comes as no surprise that with such a large number of axioms, linguistic systems are never pursued microscopically to yield implications in the same depth as group theory or even less well-developed branches of mathematics. What is perhaps more surprising is that we can get reasonable approximations of the behavior at the macroscopic level using the statistical techniques pioneered by A. A. Markov (see Chapters 7 and 8).

I'm in my 2nd year as a CS/Ling undegrad and I feel like an undergraduate degree in linguistics isn't very useful. Most undergrad material is easy enough to be picked up pretty quickly by a non-linguist wanting to do research work. So what's the point?

>>7894363
I've never heard of categorical grammar. Is this categorical in the same sense as category theory? If so I'm very interested as I have a category theory background.

>>7895010
Never mind, I should have googled. This is not what I expected but is interesting nonetheless.

https://en.wikipedia.org/wiki/Categorial_grammar

File: linguine-plain-300x213.png (128KB, 300x213px) Image search: [Google]

128KB, 300x213px

>>7892999
linguine

>>7895048
there are all kinds of different versions of categorial grammar. it's pretty cool, but it's super super flexible, which I'd consider to be both an advantage and a drawback.

>>7895006
How much coursework have you done? It gets harder, but it depends what kind of things you do, and how deep your instructors go.

>>7892988
Thread might be dead... but AMA, etc.

I'm a PhD student at a highly ranked ling school in the Chomskyan tradition.

I focus in syntax, working mainly in mathematical linguistics. My research is a little outside the mainstream, but my advisor was a prominent figure in the development of the modern theories, so I'm pretty familiar with most of generative syntax, as well as some CCG, TLG, TAG, and traditional and intensional logic, and have finished my coursework, so also a bit of phonology, etc.

>>7895048
It's closer to type theory plus combinators. The type theory does the semantics/syntactic category work, the combinators are pretty much there just to get things in the right word order.

>>7893151
I agree that linguistics is fragmented, but I think this wouldn't be a bad thing if people weren't so dogmatic. CCG, HPSG, LFG, and all Chomskyan theories are really all forms of generative grammar, and at the CFG level nearly or exactly equivalent, diverging really only where they have to hack to get the right empirical results anyway. I wish there was more actual formal work closer to mathematics, instead of arguments about notation and totally informal, relative arguments about simplicity.

OT does seem to be hot right now, but there are many reasons to doubt some of its claims. Not only are there technical problems, like doing opacity without fairly complicated constraints, but it makes some very strange (bad) predictions once you get into OT which uses probabilities. I'm not an expert in OT, but I have a friend who is running some experiments that show a lot of its (clearly preposterous) predictions are in fact false.

>>7895310
I'm not aware of any big linguistics departments that do mathematical linguistics, but then again I'm not plugged into that sort of thing so I probably wouldn't know.

>>7895310
talking to myself:

the frustrating things about all these formalism feuds is that almost all of them are (a) entirely couched in computational terms and (b) the kinds of computational differences between them seems like it would be nearly irrelevant for most (human) parsers anyway on the order of most sentences formed and uttered (especially considering that we have _no_ idea how grammar is actually organized neurologically).

my slightly polemic view is that computation, while interesting, and probably resembling some component of the language faculty, is largely a red herring. Working at the level of algorithm, and not really developing any other line of inquiry (formally) at the expense of other mathematical properties the system might have, seems short-sighted to me.

You saw people asking more "structural" questions in the late 80s and 90s, and then this whole minimalism thing came along and everyone went right back to saying "every constraint in the grammar better come from a fact about the algorithm/combinatorics and only that and wait something vague about the interfaces that we can't describe because we haven't defined the mathematical category(s) we are working in" which is t b h kinda dumb.

I think it's partially why you see a proliferation of "linking theories" - because people only study ways to "glue on" algorithmic ad hocs they need, instead of studying abstract structure. The arguments are usually "2 is more complicated than 1" cheap application of occam's razor, even when more elegant options might be waiting if we just gave up this - totally idiosyncratic to linguistics - method of argumentation.

Also, something to be concerned with with OT (and rule theories), and even most statistical methods, is very few of them actually tell you how SOUND and SIGNAL - you know, the things actually being perceived, are turned into mental representations. Almost no version of OT even acknowledges that it is sound waves that are being perceived.

>>7895339
They don't. I split my work between my passion (math ling) and the more mainstream work being done there. Had to bite the bullet if I wanted to go to a good school.

>>7895479
Oh, makes sense.

>>7895466

>we haven't defined the mathematical category(s) we are working in" which is t b h kinda dumb.

As a non-math-savvy person, I don't entirely understand this. I don't see how mathematizing what we're doing necessarily adds anything, beyond perhaps a translation of things into a different formalism. Or am I interpreting "math" here in some naive way?

>Almost no version of OT even acknowledges that it is sound waves that are being perceived.

At least at my school, I know for a fact that there are people doing OT phonology taking perceptual cues and so on in into account.

>>7895466
what do you think of the claim that language can be simplified to just domain general learning mechanisms? there are faculty in my linguistics department who legitimately believe this and deny that POS is a legitimate argument.

do you buy into the idea that the main reason so many people do minimalism is chomsky's personal authority? there are many faculty at my university who adamantly believe this as well.

>>7895547
>math
We make a lot of arguments about "structure" in linguistics - domains of dependencies, licensing, boundaries for movement, arguments about something being interpretable in one configuration and not another - however, the notion of abstract structure requires that you know, well, what that structure is.

(cont'd)

>>7895748
Instead, what is often done is "I have a rule for building objects, and whatever I get is what I get." Without knowing what category you are working in, you don't even know what structure is inherent in the object (i.e. holds up to isomorphism), and which parts are `incidental' because of the manner in which you wrote them down. This makes it incredibly difficult to prove things abstractly about the structures you get, or even reason about which parts are relevant, because what you end up with is a theory about diacritical marks you used to keep track of something instead of a theory about structure. This is more of a problem in Chomksyan linguistics than the formal models, though I have found some problems there as well.

An example - I could have the open subsets of a topological space, arranged as a lattice. Now, I know that this lattice actually has both finite and infinite intersections and meets, but I only SHOULD consider it with finite meets, but infinite intersections, and not, say, the other way around. If I just wrote down the objects, and said "well that's it!" I wouldn't really know what I'm saying because I would only have a theory of writing down a structure, not what that structure is, i.e. the category of frames≠category of lattices≠category of complete lattices.

>OT
Of course you can put in perceptual cues. You can glue on data of any kind in any formal system if you add enough bells and whistles. The problem is similar to in syntax - you end up having a very good description of the facts, but you haven't actually explained the explicit relationship between sound and phoneme. In fact, even from an ASR perspective, it is nontrivial to "reverse engineer" a sound from a bunch of numbers about VOT and formant transitions.

>>7895618
I think there is most definitely a cult of personality around Chomsky. I also (heterodoxically) think that there is a lot more to be said about the role of domain-general features in language.

However, the people who argue against Chomsky are not very good at it often. Take or leave POS, there are a lot of highly abstract properties of language that seem to hold well across language boundaries that don't seem to just be about, say, convenience, or functionality.

Looking at enough data, it is hard to argue against the idea that there is a grammar. Cyclicity is a nice argument we use often- it was claimed that phrases move in sort of "sentence-sized" cycles for theory-internal reasons explaining why extraction can occur out of some clauses, but not others, and subsequently various kinds of morphological evidence were found - each taking a totally different form, from languages as diverse as Irish to English to Dinka.

>>7895763

>Without knowing what category you are working in, you don't even know what structure is inherent in the object

I honestly don't understand what exactly this is meant to mean, but probably because I don't have the background to appreciate where you're coming from, so I don't necessarily have the ability to see what problems you're pointing out. Since a lot of theoregical linguists don't have much of a mathematical background, having bumbled into linguistics from a variety of angles, I suspect that this is not unusual.

>>7895829
have you ever seen work that's been coming out recently that tries to reduce island effects to processing? that kind of leads into this general confusion that I might have about minimalism. at my school everybody's strictly anti-chomskyan. faculty here are constantly trying to say that chomsky was alright for his time, but now he overcomplicates things and this and that. but isn't it true that a reduction of a lot of these language anomalies to domain-general features fits squarely within the minimalist enterprise? the goal is to see just how closely language resembles a "perfect" system, right?

>>7895829
>there's a cult of personality around chomsky
I find it a little strange that people think there's any truth to this. that's the kind of argument you get from conspiracy theorists. what are your thoughts?

>>7895913
Ok, here's an example. Say I want to design a bunch of sets, where the particular iterated elementhood relation might be important. Say we stipulate that our sets will be of the form {...xi...,{...yi...}}, used to denote two sets S={...xi...} and T={...yi...}, where S is considered "more outside" than T.

Suppose I have the set {x,{x}}. Now I ask you - is the fact that an element of this set is also an element of an element of this set part of the structure?

What if this is just a special case of sets of the form {a,{b}}, and it just so happens that a=b? Then {x,{x}} might be isomorphic to {a,{b}}, if all the notation is doing is really giving me two sets, an "outer" and "inner" one, since {x} is in bijection with {a}, and similarly {x} and {b}.

But really, you have no idea, because you don't know the category you are working in. If "having elements in the inner set that are also in the outer set" is part of the structure of the category we have decided we are working in, then clearly {a,{b}} and {x,{x}} are two different structures, because for the first there is no element of T in S, but in the second there is.

Knowing how to write down and manipulate your objects is only half of the structure, since you must also specify what about the object is important.

While the above structure may seem ludicrous from a math perspective, it is unfortunately these kinds of objects that have been proposed for syntax, with just as little information as I have given you now.

>>7895983
it seems like you're getting at the difference between internal and external merge, but what are you suggesting is ludicrous here?

>>7895983
While I'd say I understand what you've said here, it is hard to bring it in focus without getting a sense of what empirical predictions might hinge on these issues. Personally, my approach is generally to think about the data, and see how abstract theory should be optimized to accommodate the data. Because of that habit, if you throw a chunk of theory at me, I don't know what to think, unless it is framed in relation to a particular empirical puzzle. Though, this is evocative to me of the issues of the determination of syntactic category and labeling, which are not really well fleshed out aspects of the theory.

>>7896013
>note, i'm excluding "computational" theories which hard-code this data

Sure. I'm not saying empirical research should be on halt while you are developing a theory. It might just be the case that there are some formal things we have to say that have little or no empirical import, but are necessary just to have a coherent formal theory. People should be developing these seriously in parallel.

Especially since we debate about things like "object has to be labeled in such and such way for interface to interpret it", and "all structures are endocentric" etc., it is important that somebody - admittedly not most people - actually sort out the details of what that can mean formally if we are to *really* understand it. I think most minimalist theories are interesting, but leave a lot to be desired in terms of spelling out their abstract structure, while they simultaneously claim that some of these problems have been resolved. Having read Chomsky's late 90's-00's work, I don't think that many of his "solutions" actually work to fully describe the structure we usually need.

Even think about copying - a lower copy of an NP may not have case or have agreed yet, but a higher copy might. In what sense are they "copies" then, if they are not actually identical? We might write a diacritic indexing them as in some sense "the same", but is that really a set-theoretical structure? Did bare phrase structure solve this? No. What is it? It might be trivial, but eventually must be worked out. Same with describing where features actually fit into the structure.

I wasn't trying to claim that minimalism was ludicrous. I am a Chomskyan. But that is what I meant by most Chomskyan theories failing to actually describe the structure of the objects they manipulate, even though they specify how to build objects. They just build objects that don't actually have many of the properties we ascribe to them.

However, for the record, these are incredibly odd mathematical objects.

>>7896064
what's odd about them? I don't have a background in math or anything, I just wouldn't have thought they were that weird. is it the idea that an element of the set is also an element of an element of the set?

>>7896064
To some extent, I don't see why I should necessarily care about the mathematical well-formedness of these things, but I see no harm in seeking that, and the general intuition you express is something I agree with.

>it is important that somebody - admittedly not most people - actually sort out the details of what that can mean formally if we are to *really* understand it.

I'm not convinced that translating and making one formalism digestible to mathematical formalism necessarily does anything that makes us *really understand* that thing. It may well gives us more perspective and lead to better integration and refinement in respect to other areas of inquiry, sure.

>Even think about copying ... In what sense are they "copies" then, if they are not actually identical? We might write a diacritic indexing them as in some sense "the same", but is that really a set-theoretical structure?

I don't necessarily care if they are set-theoretic structures, but I would be interested to what interesting predictions set theory might make or how it might inform how we should understand the question you raise here, because it is indeed mysterious how we should understand things like copying and coindexing. Most slap indices on things and don't think about it much more, which is of course not the end of the story.

I have nothing against developing a mathematically coherent picture in tandem with the empirically focused side of theorybuilding, but I'm unconvinced that mathematizing a formalism brings *real* understanding in any profound way. Just, more understanding, which is good.

>>7896080
Outside of a few novelty constructions like the von Neumann hierarchy, and a few constructions used in foundations of set theory, embedded-set structure is almost never used, and even more rarely looking "between" the embedded sets as part of the structure. That's not to say that we don't allow sets to be elements of other sets, rather this is always the case in a pure set theory, it just doesn't usually play a role in the constructions you carry out with the set.

Chomsky effectively is using elementhood as a partial order, and is assuming some sort of equality predicate, but thinks its simpler to not admit it. Also, those constructions were usually used in a "pure" set theory, where there are only sets, not sets and non-sets. Clearly this is also not what Chomsky has in mind. A category theorist even joked about this: Set is one of the only categories where the relevant structures are NOT preserved under morphisms.

That is, no mathematical category I know of uses the particular embedding of elements in a set, and which non-set elements are equal at different embedding "levels", in the definition of structure. Not spaces, rings, groups, vector spaces, not even set-theory like constructions like toposes, or even the category of sets and set-functions itself. I only have ever seen similar considerations in foundations of set theory itself, when people were worried about well-foundedness.

>>7896090
I guess. I didn't mean to say that we will *really understand* language - I meant that we'd actually understand what it is we are proposing, as opposed to...not knowing exactly what it is we are proposing.

If we think command, domains, feature relations, labels, merge, movement etc. are fundamental, and hopefully, related, properties of language, it will be hard to show in what sense they arise out of the structure-building procedures if we don't have a grasp of what "structure" is. And it puts arguments about constraints based off of the structure of derived objects (nearly all of them) on shaky ground until it is sorted out.

Placeholder definitions are fine - I am probably more attached to this topic than I should be, if only because a lot of people think these problems have been resolved when they haven't, and it's what my research actually handles very well.

>>7895920
islands are a bitch and nobody knows what's going on. I am familiar with both the subjacency/ECP account and phase accounts, and some others; they are horrendously messy.

There is no theory I know of that successfully models them all, almost all formal theories are incredibly stipulative, and even experimental data seems all over the place.

An interesting topic for sure, but probably not a place to battle theories, as no one really has a great story outside of the few facts they tackle that day.

>>7895967
There is definitely a cult of personality around Chomsky. It's not even covert. It's not all bad, he has a lot of good ideas and most people work in his framework, but many of the devotees are loud about it and I have been witness to it at many presentations by visiting scholars/professors.

>>7896124
there may be people who really like chomsky personally, but that's not a good explanation for their choice of formalism

>>7896141
It exists with good reason. He's insanely prolific. For every shitty idea he spouts out at least another good one, and constantly. I think people accept a lot of his framework if only because they can't think of anything faster, and he's at least cooked the books to get a lot of empirical facts right right off the bat. Later work then goes into finding out why his stipulations worked to describe the data, and how to clean them up, and how they can be extended to capture new data.

However, in the mix, there are things he's said that (at least I think) are probably technically incorrect (not to mention experimentally/philosophically), that get ignored and never repaired.

>>7896090
>>7896109
part of my issue is not just that we could mathematize things but haven't, its that there have been mathematizations by Chomsky and others that I think actually fail to capture the structure claimed, and further arguments are based off of it. that is, some of the more modern models don't actually have the structure needed to make the arguments we'd like to.

>>7896148
so because chomsky can come up with good ideas faster than most professionals, people are happy to just ignore the problems (some of which are purportedly quite serious) and accept the good parts? why do they ignore the problems? are they just too lazy to think things through?

>>7896164
first, I should admit that the formal problems I've been talking about are entirely my own observations. I've had a hard time convincing most linguists outside of my advisors that they are even problems, mostly because linguists aren't versed in math outside of probability, automata theory, logic, and the lambda calculus, if at all. Think of it like the way a lot of young students think that vectors actually *are* an array of numbers (in fact, I've met linguists who use vector models who still think this). It's just confusion about the difference between an abstract structure and a representation, and they don't have enough examples of structure to ground themselves in these kinds of arguments.

Second, Chomsky is not a clear writer. It can be hard to even figure out exactly what it is that is problematic often times. Also, there are so many differing views, and the theory changes often enough, that many of the problems aren't really his fault - an analysis that used to work doesn't go through in the updated model anymore, and people have to come up with a new analysis, or forgot about the old analysis or didn't realize it got messed up.

There are a lot of moving parts to language (obviously), and since we don't have as well-understood a formalism/model like in mathematics, physics, or chemistry, it's really easy to lose track.

That said, Chomsky sometimes contradicts himself within a single book, maybe not in the body, but between the footnotes and body, clearly just trying to reckon with sometimes horribly unwieldy empirical facts.

A lot of syntactic research is taking an analysis that made a bunch of good predictions, but a few bad ones, and seeing how much you have to tweak it to get the new data without messing up the old. It's tough, data is very messy, and its not always clear when its even your problem to explain, and not someone who works in, say, processing.

And of course, with all these moving parts, it's easy to slip into laziness.

>>7896183
I see. a lot of linguists at my university feel very strongly about this cult of personality around chomsky, but they have a very hard time explaining themselves. thanks for taking the time to elaborate as much as you have.

>>7896195
He also really spearheaded a conversation that needed to happen anyway - learning and knowing language is not just statistical pattern learning. It has a lot of internal structure that people are sensitive to. People really didn't think this way before Chomsky or at least take this view very seriously, and a whole lot of cognitive scientists are grateful to just have had someone say it to make room for their work.

>>7893888
last I checked, I had a friend doing PhD NLP with a guy from google who told me the state of the art techniques were basic vector models, and fancy probabilities on PSG trees...ie where there is literally no semantics...not even modeling the fact that "what did you eat" is a question about what was eaten.

troll
harder
into
the
future

>ps I'm only exaggerating a little. this ain't far from the truth

>>7894485
>>Any theory of language that seeks to accurately predict
what does predict means ? from what things and what rules do you predict what you seek to predict ?

>>7896560
if you think of a language as a set of all possible and valid ("grammatical") sentences in a language, then one way of measuring a theory is to see how often it correctly labels valid sentences as valid and how it labels invalid sentences as invalid. it is in this sense that a theory "predicts".

>>7896585
so it is on the side of the syntax so far ?

and what is the ontological claim behind the predictions ? why do people seek theories ? is it like in science, if a theory predicts according to a few people, these people will say that '''the theory is real, the theory describe the reality, the languages'' ?

>>7895310
What type of math do you need to know to get into Chomskian lingustics? I'm a second year undegrad and, while I probably won't major or minor in it, I do find the field of theoretical linguistics pretty interesting in its own right.

>>7892988

If physics weren't a thing, I'd probably go into linguistics. It seems really interesting, unfortunately physics and its related maths stole my heart.

>>7896183
>Think of it like the way a lot of young students think that vectors actually *are* an array of numbers (in fact, I've met linguists who use vector models who still think this)
what vectors are supposed to be ?

>>7896638
None to start with, but if you do formal semantics you'll have to learn basic set theory, predicate logic, and lambda calculus, functions etc.

>>7896614
Doesn't have to be syntax. It can apply to phonology or semantics or whatever too. Is a hypothetical sentence of a phonological form that a speaker would accept, or not? Is a given semantic interpretation for an utterance something speakers accept, or not? Does our model capture these facts?

Most wouldn't try to claim that these models tell us what is literally happening in people's brains when they use and evaluate instances of language based on their internal grammar. We are definitely modeling something that is real, because grammars exist in people's brains in some sense, and give us empirical facts about what a given language does and does not do. The model, though, it seems to me, is just a model that isn't representative of the literal mental processes. Linguistics is not united enough with brain-studies for that, yet.

>>7892988
My main interests are in chemistry and astrophysics/cosmology, but I also find linguistics and languages pretty fascinating. I will often find myself browsing wikipedia articles on languages for hours (many times when I should actually be doing something more productive.) But learning about things like language families and especially isolates is pretty cool to me. E.g. the fact that Finnish is actually substantially different than say Swedish or Norwegian, at least IIRC.

That's my layman's opinion anyway.
Btw, have they ever concluded where Basque came from?

>>7892988
well i construct mathematical models of biological phenomena.

In a way, language is a biological phenomena, different phonemes make up words in language and different graphemes make up words in writing. It isnt a stretch to say you could create a robust system that could generate neographs and constructed languages.

very interesting to say the least

>>7896638
none; Carnie, Adger, and Lasnik/Uriagareka all have intro books. Many syntacticians never learn much math. I indicated my research interests are more fringe - you can be great at analyzing languages without having to know any more formalism than the basic GB theory, and most people do that quite successfully.

>>7897056
>>7897323
Weird, cause I read about a few theorems derived in part by Chomsky, which pretty were related to abstract algebra. As a math major that sounded pretty interesting. Also, if you don't mind me asking how did you guys get into linguistics?

>>7897371
There was early stuff linking very simple grammars to certain rings of power sets, I think. It is virtually unimportant to using the grammars themselves. I think these results were also largely proven by Schützenberger, but I could be wrong. Also, at a very basic level, Turing machines are closely related to semi thue system, themselves closely related to monoids, but the algebra is so boring the comparison is barely worth making. formal language theory is kinda boring imo in general, as you mostly get implicational results telling you when something is too complicated to fall into a class of languages which are in some sense simple, usually by some brute force counting method.

Much more interesting work linking math + language has come out of the type logical grammar side, and the model-theoretic perspective imo (sort of mathematical extensions of CCG and GB theory).

I minored in linguistics, got into math too late to ever be able to do that, but had a great advisor in undergrad who guided me toward the "deep questions" really fast. I got into a very good PhD dept with someone I wanted to work with, so decided to go with it, even if temporary, cause I felt like I had something to say that I needed time to work out. It's ok.

>>7897082
>Btw, have they ever concluded where Basque came from?

Not that 'mysterious' I don't think: last surviving language from before the Indo-European family and its speakers came into Europe.