/sci/ - Science & Math

>>>/lit/

>>>/lit/

I bet you're a filthy prescriptivist.
Fuck off to >>>/lit/

>syntax trees

Ahahaha, have fun with scrambling languages with that approach, chumps.

The easiest way to model a grammar is through a context-free grammar: https://en.wikipedia.org/wiki/Context_free_grammars

Note that this only allow for very simple grammars and ususally has problems with gender/number concordance, although that's not a huge problem if you plan on using english only.

For more advanced grammars that allow for some context, I sugest that you pick a book on Natural Language Processing (and in Automata Theory, if you don't feel confortable with regular expressions/context free grammars).

ha ha ha

>>6305914
Parse trees are good for a specific parse, but it's often times possible to parse a sentence in lots of different ways (draw lots of different trees) all with different meanings. Mainly, thee take-away I got out of parse trees while learning NLP is that languages are such shit that NLP is more or less impossible except through probabilistic means.

>>6305942
>parse tree
>prescriptivism
Do you even linguistics, bro?

Because human languages are terrible at following rules.
Seriously terrible.
Just like they're terrible at not being ambiguous.
Probably a symptom of them evolving slowly over time rather than being something that was designed exactly

>>6306409
>Mainly, thee take-away I got out of parse trees while learning NLP is that languages are such shit that NLP is more or less impossible except through probabilistic means.
You say that like probabilistic means are bad.

>>6305930
>>6305939
Fuck off back to your consciousness and 0.999... threads. Wikipedia:
>Linguistics is the scientific[1] study of language.[2]

>>6305914
>teach me the math behind grammar and sentence structure
https://en.wikipedia.org/wiki/Context_free_grammars#Formal_definitions

>>6306449
What about Esperanto?

>>6306550
Which countries uses it? That was an interesting idea but in practice it became as useful as elfic.

>>6306550
Esperanto's arguably WORSE.

http://www.xibalba.demon.co.uk/jbr/ranto/

>>6305944
The details of scrambling languages actually have a lot of interesting information to contribute to linguistics, and need to be explored more deeply. Turns out that when you get into the details of it, scrambling generally has certain restrictions, and these things can be played with to tease out really neat stuff.

>>6305944
This is just one of the reasons why HPSG's are a God-tier formalism.

>>6305914
It is difficult to get into modern linguistics because many of the arguments are spread out over many different articles at different times.

For contemporary Chomskyan linguistics the reader Minimalist Syntax: Essential Readings has most of the main modern parts of generative grammar, along with The Minimalist Program itself.
Ed Stabler has written quite a bit about about programming them/putting them in logic.

Heim and Kratzner is a good book for getting into semantics and a sense of how it relates to the syntax. Also the work of Montague. There are many many theories in linguistics, but these have a lot of the technology used in "mainstream" syntax and semantics.

Aravind Joshi has been working on another computational approach TAG. Coming from type theory and Lambek calculus are other grammars related more directly to mathematics. These overlap with Chomsky grammars a lot. These come out of the work of Grishin, Lambek, Ajdukiewicz, Steedman, Chris Barker, Gentzen, Moortgat and others, which are in many ways related to the CS work of Curry and the Church lambda-calculus, as well as model theory, and categorial logic.

These are only a handful of related theories, and there are many others. Many probabilistic theories make use of parts of some of these structures in many ways, while stripping away other aspects to be taken care of by n-grams or other kinds of "intelligent" pattern-seeking put to probabilistic parameters. Almost no theory of syntax and semantics totally avoids the tools found in what I've mentioned, though there's a lot more because it is a very new and diverse field. I am also just restricting to what you have asked, which is for syntax/semantics.

>>6306611
>minimalist program
>not a bunch of bullshit that's only taken seriously because of Chomsky's earlier work

>semantics
>not highly speculative

I bet you believe in null constituents too

>>6306618
I'm not saying what I think is or isn't speculative, or even that the minimalist program is right*, but these books have some of the earlier work, and regularly reference back to them. I recommended them because they contain a lot of current stuff in context of old stuff.

(*I don't. But it has a very interesting relationship with many other theories of grammar that I think are compelling. All of this is the beginning of what will hopefully be a much broader theory, which will probably take at least a generation or two of revision.)

The other computational theories** cover some more abstract principles, and show relationships to lots of other theories of grammar.

(**These are related to very general mathematical statements, and most elementary formalisms of grammar.)

Null constituents might be a problem of notation..

Are there any constructed languages that are structured to have almost no ambiguity?

learn a case-system language with no word order. it's fun. Hungarian is probably the easiest, but I suggest Russian for the number of speakers.

>>6306629
http://attempto.ifi.uzh.ch/site/
http://en.wikipedia.org/wiki/Attempto_Controlled_English
>Attempto Controlled English (ACE) is a controlled natural language, i.e. a subset of standard English with a restricted syntax and a restricted semantics described by a small set of construction and interpretation rules.

>>6306627
>Null constituents might be a problem of notation..
It's a problem with conceptualisation. Syntacticians are overstepping their ground into semantics. They do it with wh-movement too.

>>6306467
It means that at best we can translate from one language to another probabilistically but we can never derive any real meaning from it.

doubleplusungood

>>6306984
>mfw

>>6306629
http://en.wikipedia.org/wiki/Lojban

>>6305914
>please take me back to when I was never taught this
why.avi

>>6307006
CS/NLP nerds are as dogmatic about their hopeless "formalism" as us "rule-based" linguists. Give up the ghost man, and let's go forward into this century.

>>6308169
also, probability models are essentially "meaningless" I don't know how you get around this; run statistics on some very surface distinct languages like English and Yoruba or one of the Kichaga languages and tell me how much of their semantics system you've learned.

bumpan

bump

bump

bump

>>6306409
You basically get a fuzzy set of correct meanings which make sense without context, then use context to narrow it down.

Then you realize that the most correct meaning won't always be correct for any finitely complex parser, which is why even humans make errors.

>>6306550
It was deliberately designed and therefore doesn't count as a natural language.

Many NLP-parsing machines use a modified version of Esperanto as an intermediary language.

bump

bump

bump

bump

.

>not using Combinatory categorial grammars
>2014

>>6306629
http://www.ithkuil.net/
With the added advantage of a script that looks like it was made by The Predator.

>>6312064
if ccg's are better, why are cfg's more common?

>>6306629
yes
http://en.wikipedia.org/wiki/Mathematics

>>6306611
Nobody sensible believes in Chomsky anymore though. GG as a whole has gone off the deep end somewhere in the late 80s.

>>6306641
Russian isn't a word order free language.

>>6306606
Enjoy your slow-as-shit parses!

>>6306984
>It means that at best we can translate from one language to another probabilistically but we can never derive any real meaning from it.
How so? Meaning may be probabilistic too.

>>6312064
Yes! (And I'm the 'Chomskyan'). Although generalizations found in categorial logic/MTLG have been more interesting to me recently, and fix a lot of the shortcomings of CCG in the most natural way (they also fit neatly in with MGs).

>>6312568
Doesn't know about equivalence proofs.

Also, again not that I wholeheartedly believe minimalism, but MG is just a MCFG, and most of the Chomsky-bashing I've read completely mischaracterizes modern syntax. None of it is that far fetched or even distinct from what almost all linguists use to some extent. There are gentzen presentations of MG which emulate it completely and are nearly identical to any mainstream formalism you can cook up.

>>6312568
To be more clear, cfg's are more common cause those are the straightforward scraps that trickled down to CS people. many linguistic phenomena seem to go beyond CFGs, and 'typed' combinators as in CCG, or 'multiple' dimensions for the CFG to work out over as in MCFGs are two of the most natural ways to expand CFGs. Depending on choice of combinators, CCGs usually push a CFG a little into the "mildly sensitive" region, while every MG is equivalent to some MCFG. Basically, the reason is CFGs are simpler so CS people like them, but linguistics realized it is better to move past them than try to brute force them to work, which may just be impossible.

>>6305944
>hasn't seen a syntax paper from the last 15 years
scrambling, at least not of the slavic and japanese flavors, seems to be unrestricted by any reasonable definition. In fact, the limitations come out from a very rich interaction of minimality principles (or particular mixing properties, etc.)

>>6313062
*does NOT seem to be unrestricted

>>6305914
I allays had trouble understanding grammar. Am I retarded?

>>6313065
>does NOT seem to be unrestricted
I'm going to need some citation on this.

>>6313148
You are retarded even among retards. People with William's syndrome, who average an IQ of roughly 70, have no problem speaking, and speaking fluently at that! You, sir, have no business here and I encourage you to promptly leave. Good day!

>>6306611
>Aravind Joshi has been working on another computational approach TAG. Coming from type theory and Lambek calculus are other grammars related more directly to mathematics. These overlap with Chomsky grammars a lot. These come out of the work of Grishin, Lambek, Ajdukiewicz, Steedman, Chris Barker, Gentzen, Moortgat and others, which are in many ways related to the CS work of Curry and the Church lambda-calculus, as well as model theory, and categorial logic.
Anon, no. Type theory is everywhere. Why here too?

>>6305914
>implying that isn't actually useful

>>6306984
>humans can do it
>machines can't
>humans are special

>>6314045
What's wrong with type theory?

>>6314654
Nothing, it's actually cool as fuck. I'm just waiting for the coming of typist type II.

>>6305914
... but I don't believe it.

>>6314026
Hm. I tried to dig a few up, but it turns out many of the papers which we've been reading in lab are not out yet, since it's a relatively recent research topic. The "seminal papers" which most work references are Saito 1992 and Saito 2006. There are a number of binding and quantificational (and I think some hyperraising? I don't study Japanese) restrictions. Recent work has gone into making the A/A' distinction come out of other principles, but they seem to not have the same (non-)restrictions as each other anyway.

>>6314045
>>6314654
I'm also confused. I'm not trying to convince anyone of type theory; it's a fact that it simply directly inspired some work in syntax and semantics which I listed there, seeing as OP was asking about math approaches to grammar. "MTLG" stands for multimodal type logical grammar - it straightforwardly is type logic as a grammar, and people like Oehrle and Moortgat have been working on it and relating it to the CCGs of Steedman.

>>6314813
>not believing in phrase structure grammars
>2014

>>6314816
>we've been reading in lab are not out yet
Who's "we?" What have you been reading?

bump

>mfw pseudoscience philistines are still trying to model natural language without any insight from the systems and evolutionary neuroscience of language

>>6316529
>pseudoscience philistines
Depends on the subfield in linguistics.

>still trying to model natural language
That's about right. Phrase structure grammars seem pretty close, but not quite perfect.

> without any insight from the systems and evolutionary neuroscience of language
0/10, apply yourself.

>>6316165
Would risk anon-iminity. I do something too specific in too small a field, and you could google me very easily. I'll say that some of it was Boskovic's more recent/inchoate work, which is very interesting. My advisor is a first generation Chomsky student (though I'm not Chomskyan, exactly) at one of the 10 major US ling schools.

>>6316529
0/10 We read 50+ pages of evolutionary science for my syntax course week one (not a ton, but a lot considering it's not course content), not to mention how much neuro/psych even us formalists have to take (1-2 years min + colloquia).

>>6305914

shit, we had a test on this today

>>6316827
> I do something too specific in too small a field, and you could google me very easily.
is this a may may, or do you just copy and paste the exact same this every time?

bump

bump