In this article, Manning discussed about deep learning's impact and contribution to NLP and NLP's future. He quoted Michael Jordan's AMA reply on the R website,
Although current deep learning research tends to claim to encompass NLP, I'm (1) much less convinced about the strength of the results, compared to the results in, say, vision; (2) much less convinced in the case of NLP than, say, vision, the way to go is to couple huge amounts of data with black-box learning architectures.
and said he agrees with point (1) but feels reserved towards point (2).
The problem is, I am not sure about point (2)'s position.
Later in the article, Manning pointed out that domain knowledge, e.g. linguistics & cognitive psychology, etc., is the center of NLP's recent advancement, and thus, in some sense, downplayed the role of deep learning in advancing NLP.
Therefore, was Manning arguing against mono-focusing on "huge amounts of data" and "black-box learning architectures" since they are often attributed to deep learning? Was Jordan arguing for them as the "way to go" for future NLP?
I was wondering if Computational Semantics is actually used in any real-world system? (Simple examples here and here). I would like to see how an actual system works.
It seems like there are a bunch of issues with actually using Computational Semantics in any real world system:
It seems just labeling sentences with part-of-speech tags is error prone.
But you also need a reliable parse tree which is error prone and there can be many valid trees for one sentence.
Finding what pronouns are referring to what entities is error prone.
Word disambiguation is also another source of errors and multiple meanings could be valid in the same context.
Any context-free-grammar of English I can find seems to be incomplete.
Finally, after all these sources of error are dodged, we can finally convert the sentence to FOL with Computation Semantics!
Also, I can't seem to figure out how to deal with prepositions in Computation Semantics.
Is this really just an academic exercise or is Computational Semantics actually useful?
There are several better aproaches to natural language than simple lambda calculus and context free grammars, ie. HPSG, Montague Grammar, TAG, ...
Word disambiguation can be handled by Markov chains, for example.
Siri, Google Now, Cortana and IBM Watson are some examples for real world systems.
Google Translate is another application that uses Computational Semantics.
I believe (bu't don't quote me on this) that the technology spun out of the now-defunct Natural Language Theory and Technology group at the Palo Alto Research Center (PARC, formerly Xerox PARC), utilizes the lambda calculus to provide inferences about textual entailments. idk i only worked there a summer (freshman, so was wonderfully igorant of most of the goings-on there).
Anyway, that 'technology' was developed over 30 years and then Powerset bought the right to all of it for $15 million, attempting to disrupt smart search in general. Then Bing's fatass came along, gobbled it up nom nom nom, then continued devouring the entire research group as whole. The principal core investigators now work solely as adjuct profs at Stanford. Sad.
I've tried looking around but I can't seem to find an answer on what math I need before jumping into NLP. I was hoping to get a solid foundation in math before jumping into NLP.
From what I've gathered it's mostly:
Probability,
some Statistic,
Discrete Math
Thank you for your time.
As in most fields, you'll find once you dive in that the title "NLP" covers a fairly broad range of sub-fields. The math requirements vary widely depending on what you're trying to accomplish. So a little more detail about your goals would help.
That said, I can address parsing and the related fields I have some experience in, and offer very general comments on a few others.
You'll find discrete math and automata theory useful in any computer science discipline, so you can't go wrong there.
Some NLP work is closer to linguistics or psychology than computer science. So some linguistic theory might be helpful if that's where your interests lie, and some background in statistical hypothesis testing (applied statistics of the sort you might find in a social science department, although the more rigorous the better).
For morphology, tagging, parsing, and related fields, some probability theory is helpful (as is experience thinking about dynamic programming, although that's not really math background). If you're doing anything involving machine learning (which is most of NLP), it helps to understand some linear algebra.
That said, if your goals are more applied, you can accomplish quite a lot by applying existing tools, without detailed knowledge of the underlying math (it doesn't require any linear algebra to train an SVM, if all you need is a classifier).
For comprehension, there is no mathematics to model language. The 'model' means that a function maps language expression to number or utility. The 'comprehension' comes from representation and composition, which have exceeded the nature science.
As it currently stands, this question is not a good fit for our Q&A format. We expect answers to be supported by facts, references, or expertise, but this question will likely solicit debate, arguments, polling, or extended discussion. If you feel that this question can be improved and possibly reopened, visit the help center for guidance.
Closed 9 years ago.
Question
So I've recently came up with some new possible projects that would have to deal with deriving 'meaning' from text submitted and generated by users.
Natural language processing is the field that deals with these kinds of issues, and after some initial research I found the OpenNLP Hub and university collaborations like the attempto project. And stackoverflow has this.
If anyone could link me to some good resources, from reseach papers and introductionary texts to apis, I'd be happier than a 6 year-old kid opening his christmas presents!
Update
Through one of your recommendations I've found opencyc ('the world's largest and most complete general knowledge base and commonsense reasoning engine'). Even more amazing still, there's a project that is a distilled version of opencyc called UMBEL. It features semantic data in rdf/owl/skos n3 syntax.
I've also stumbled upon antlr, a parser generator for 'constructing recognizers, interpreters, compilers, and translators from grammatical descriptions'.
And there's a question on here by me, that lists tons of free and open data.
Thanks stackoverflow community!
Tough call, NLP is a much wider field than most people think it is. Basically, language can be split up into several categories, which will require you to learn totally different things.
Before I start, let me tell you that I doubt you'll have any notable success (as a professional, at least) without having a degree in some (closely related) field. There is a lot of theory involved, most of it is dry stuff and hard to learn. You'll need a lot of endurance and most of all: time.
If you're interested in the meaning of text, well, that's the Next Big Thing. Semantic search engines are predicted as initiating Web 3.0, but we're far from 'there' yet. Extracting logic from a text is dependant on several steps:
Tokenization, Chunking
Disambiguation on a lexical level (Time flies like an arrow, but fruit flies like a banana.)
Syntactic Parsing
Morphological analysis (tense, aspect, case, number, whatnot)
A small list, off the top of my head. There's more :-), and many more details to each point. For example, when I say "parsing", what is this? There are many different parsing algorithms, and there are just as many parsing formalisms. Among the most powerful are Tree-adjoining grammar and Head-driven phrase structure grammar. But both of them are hardly used in the field (for now). Usually, you'll be dealing with some half-baked generative approach, and will have to conduct morphological analysis yourself.
Going from there to semantics is a big step. A Syntax/Semantics interface is dependant both, on the syntactic and semantic framework employed, and there is no single working solution yet. On the semantic side, there's classic generative semantics, then there is Discourse Representation Theory, dynamic semantics, and many more. Even the logical formalism everything is based on is still not well-defined. Some say one should use first-order logic, but that hardly seems sufficient; then there is intensional logic, as used by Montague, but that seems overly complex, and computationally unfeasible. There also is dynamic logic (Groenendijk and Stokhof have pioneered this stuff. Great stuff!) and very recently, this summer actually, Jeroen Groenendijk presented a new formalism, Inquisitive Semantics, also very interesting.
If you want to get started on a very simple level, read Blackburn and Bos (2005), it's great stuff, and the de-facto introduction to Computational Semantics! I recently extended their system to cover the partition-theory of questions (question answering is a beast!), as proposed by Groenendijk and Stokhof (1982), but unfortunately, the theory has a complexity of O(n²) over the domain of individuals. While doing so, I found B&B's implementation to be a bit, erhm… hackish, at places. Still, it is going to really, really help you dive into computational semantics, and it is still a very impressive showcase of what can be done. Also, they deserve extra cool-points for implementing a grammar that is settled in Pulp Fiction (the movie).
And while I'm at it, pick up Prolog. A lot of research in computational semantics is based on Prolog. Learn Prolog Now! is a good intro. I can also recommend "The Art of Prolog" and Covington's "Prolog Programming in Depth" and "Natural Language Processing for Prolog Programmers", the former of which is available for free online.
Chomsky is totally the wrong source to look to for NLP (and he'd say as much himself, emphatically)--see: "Statistical Methods and Linguistics" by Abney.
Jurafsky and Martin, mentioned above, is a standard reference, but I myself prefer Manning and Schütze. If you're serious about NLP you'll probably want to read both. There are videos of one of Manning's courses available online.
If you get through Prolog until the DCG chapter in Learn Prolog Now! mentioned by Mr. Dimitrov above, you'll have a good beginning at getting some semantics into your system, since Prolog gives you a very simple way of maintaining a database of knowledge and belief, which can be updated through question-answering.
As regards the literature, I have one major recommendation for you: run out and buy Speech and Language Processing by Jurafsky & Martin. It is pretty much the book on NLP (the first chapter is available online); used in a frillion university courses but also very readable for the non-linguist and practically oriented, while at the same time going fairly deep into the linguistics problems. I really cannot recommend it enough. Chapters 17, 18 and 21 seem to be what you're looking for (14, 15 and 18 in the first edition); they show you simple lambda notation which translates pretty well to Prolog DCG's with features.
Oh, btw, on getting the masters in linguistics; if NL semantics is what you're into, I'd rather recommend taking all the AI-related courses you can find (although any courses on "plain" linguistic semantics, logic, logical semantics, DRT, LFG/HPSG/CCG, NL parsing, formal linguistic theory, etc. wouldn't hurt...)
Reading Chomsky's original literature is not really useful; as far as I know there are no current implementations that directly correspond to his theories, all the useful stuff of his is pretty much subsumed by other theories (and anyone who stays near linguists for any matter of time will absorb knowledge of Chomsky by osmosis).
I'd highly recommend playing around with the NLTK and reading the NLTK Book. The NLTK is very powerful and easy to get into.
You could try reading up a bit on phrase structured grammers, which is basically the mathematics behind much language processessing. It's actually not that heavy, being largely based on set and graph theory. I studied it many moons ago as part of a discrete math course, and I guess there are many good references available at this stage.
Edit:Not as much as I expected on google, although this one looks like a good learning source.
One of the early explorers into NLP is Noam Chomsky; he wrote small books on the subject in the 50s through the 70s. You may find that engaging reading.
Cycorp have a short description of how their Cyc knowledge base derives meaning from sentences.
By utilising a massive knowledge base of common facts, the system can determine the most logical parse of a sentence.
A simpler place to begin with the building blocks is the look at the documentation for a package that attempts to do it. I'd recommend the Python [Natural Language Toolkit (NLTK)1, particularly because of their well-written, free book, which is filled with examples. It won't get you all the way to what you want (which is an AI-hard problem), but it will give you a good footing. NLTK has parsers, chunkers, context-free grammars, and more.
This is really hard stuff. I'd start off by getting at least a Masters in Linguistics, and then work towards my PhD in computer science, concentrating on NLP.
The problem is that most of us don't have the understanding of what language is. And without that understanding, it's bloody tough to implement a solution.
Other comments give some readings, which are probably fine if you want to get started playing around with a small subset of the problem, but in order to come up with a really robust solution, then there are no shortcuts. You need the academic background in both disciplines.
A very enjoyable readable introduction is The Language Instinct by Steven Pinker. It goes into the Chomsky stuff and also tells interesting stories from the evolutionary biology angle. Might be worth starting with something like that before diving into Chomsky's papers and related work, if you're new to the subject.
Does anyone known of a a good reference for canonical CS problems?
I'm thinking of things like "the sorting problem", "the bin packing problem", "the travailing salesman problem" and what not.
edit: websites preferred
You can probably find the best in an algorithms textbook like Introduction to Algorithms. Though I've never read that particular book, it's quite renowned for being thorough and would probably contain most of the problems you're likely to encounter.
"Computers and Intractability: A guide to the theory of NP-Completeness" by Garey and Johnson is a great reference for this sort of thing, although the "solved" problems (in P) are obviously not given much attention in the book.
I'm not aware of any good on-line resources, but Karp's seminal paper Reducibility among Combinatorial Problems (1972) on reductions and complexity is probably the "canonical" reference for Hard Problems.
Have you looked at Wikipedia's Category:Computational problems and Category:NP Complete Problems pages? It's probably not complete, but they look like good starting points. Wikipedia seems to do pretty well in CS topics.
I don't think you'll find the answers to all those problems in only one book. I've never seen any decent, comprehensive website on algorithms, so I'd recommend you to stick to the books. That said, you can always get some introductory material on canonical algorithm texts (there are always three I usually recommend: CLRS, Manber, Aho, Hopcroft and Ullman (this one is a bit out of date in some key topics, but it's so formal and well-written that it's a must-read). All of them contain important combinatorial problems that are, in some sense, canonical problems in computer science. After learning some fundamentals in graph theory you'll be able to move to Network Flows and Linear Programming. These comprise a set of techniques that will ultimately solve most problems you'll encounter (linear programming with the variables restricted to integer values is NP-hard). Network flows deals with problems defined on graphs (with weighted/capacitated edges) with very interesting applications in fields that seemingly have no relationship to graph theory whatsoever. THE textbook on this is Ahuja, Magnanti and Orlin's. Linear programming is some kind of superset of network flows, and deals with optimizing a linear function on variables subject to restrictions in the form of a linear system of equations. A book that emphasizes the relationship to network flows is Bazaraa's. Then you can move on to integer programming, a very valuable tool that presents many natural techniques for modelling problems like bin packing, task scheduling, the knapsack problem, and so on. A good reference would be L. Wolsey's book.
You definitely want to look at NIST's Dictionary of Algorithms and Data Structures. It's got the traveling salesman problem, the Byzantine generals problem, the dining philosophers' problem, the knapsack problem (= your "bin packing problem", I think), the cutting stock problem, the eight queens problem, the knight's tour problem, the busy beaver problem, the halting problem, etc. etc.
It doesn't have the firing squad synchronization problem (I'm surprised about that omission) or the Jeep problem (more logistics than computer science).
Interestingly enough there's a blog on codinghorror.com which talks about some of these in puzzle form. (I can't remember whether I've read Smullyan's book cited in the blog, but he is a good compiler of puzzles & philosophical musings. Martin Gardner and Douglas Hofstadter and H.E. Dudeney are others.)
Also maybe check out the Stony Brook Algorithm Repository.
(Or look up "combinatorial problems" on google, or search for "problem" in Wolfram Mathworld or look at Hilbert's problems, but in all these links many of them are more pure-mathematics than computer science.)
#rcreswick those sound like good references but fall a bit shy of what I'm thinking of. (However, for all I know, it's the best there is)
I'm going to not mark anything as accepted in hopes people might find a better reference.
Meanwhile, I'm going to list a few problems here, fell free to add more
The sorting problem Find an order for a set that is monotonic in a given way
The bin packing problem partition a set into a minimum number of sets where each subset is "smaller" than some limit
The travailing salesman problem Find a Hamiltonian cycle in a weighted graph with the minimum total weight