A layman's question: Given a sentence whose meaning is unique, is there any standard algorithm to uniquely determine/describe its grammatical structure such that it can be classified in a standard way (i.e. like, if I don't err, what exists in fields like organic chemistry etc.)? If yes, please give some good references. Thanks in advance.

  • What do you call classification? How does it apply to proteins and DNA? Do you assume that domains of meaning are finite or discrete, or dense over some continuous domain? – babou Jun 12 '14 at 16:45
  • As foreigner I learned decades ago English in school where I remember to have done excercies requiring one to find from a given sentence subject, verb, object and determine noun, adjective, adverb etc. etc. and subsequently to draw a tree diagram representing the grammatical structure of that sentence. I suppose that sentences having the same tree representation could well be considered to be of the same class. (BTW, I failed (with my limited searches) to find any books systematically dealing with such diagramming techniques. Could someone help in giving references? Thanks in advance.) – Mok-Kong Shen Jun 12 '14 at 18:45

There are multiple competing procedures.

Linguistics is still quite far from proving true even fractions of the grammar of any specific natural language. Even the most basic phenomena are interpreted differently in various paradigms. For the most obvious example: one important fraction within theoretical syntax, and at least historically the most important one, is Generative Grammar in a Chomskian tradition. For example, GG relies on strictly binary branching trees and assumes every noun phrase carries "case". This approach is all but marginal in computational linguistics, where algorithms are created to parse sentences; instead, computational linguistics favors dependency grammar and constructionist approaches. So at the very least, syntactical theory and computational practice are so far anything but unified, or in agreement about a standard.

What does exist are some subfield standards which cover, if at all, only aspects of what you're asking for. For example, WALS employs a standard coding scheme, and constituency tests are a near-standardized syntactical toolkit for establishing what things should be bracketed together. There are also powerful computational tools for classifying sentences according to one specific grammar, and some parsers count as quasi-gold standards to compare other algorithms to. Yet, a GG syntactican would reject many of these analyses as being all but irrelevant for "real" syntax.
However, none of these are uncontroversial even within these subfields, and none of them come close to providing a unified standard for exhaustively describing the syntactic structure of any sentence.

Also see this question.

  • The gulf between the computational linguists with their dependency-based parsing schemes and the theoretical syntacticians with their Chomskyan complexity is indeed extreme. My work is in theoretical DG (not computatonal DG), and I find that too many of the computational people lack linguistic insight with their parsing schemes. The best scenario is for the computational people to sit down with the theoreticians to try build bridges across the fields. – Tim Osborne Jun 12 '14 at 16:48
  • Much non-GG theoretical work with as much depth as the Chomskian tradition is much easier to combine with computational work. Construction grammar is not dumbed down in any way, fantastical syntacticians are doing CxG. I'm speaking as a cognitive linguistic here, not a computational linguist, but I'll still say: @Tim Osborne, do not try to connect with a tradition hostile to implementations and data simply because they have big mouthes and big names. – user3503 Jun 12 '14 at 17:03
  • 1
    The Stanford parsing scheme illustrates my point well. That scheme takes auxiliary verbs as heads over content verbs. Almost all work in theoretical syntax, constitency- or dependency-based, does it the other way around. This is a prominent example of where a group of computational people are getting it wrong. There are of course other computational people who are getting it right, in line with work in theoretical syntax. For the record, I am a fan of CxG, and I think much of Chomskyan syntax is nonsense. – Tim Osborne Jun 12 '14 at 17:24
  • I see. I hope you didn't feel talked down to Tim. – user3503 Jun 12 '14 at 17:29
  • No worries. I have however noticed that some computational people are quick to discount most work in theoretical syntax, regardless of whether it is associated with Chomskyan syntax or not. Many people doing theoretical DG, including myself, do not like being lumped in with the Chomskyans. – Tim Osborne Jun 12 '14 at 18:52

I do not know whether there is any sentence in the world that has a unique meaning. I am not sure I can identify all the things that contribute to the meaning of a sentence, since it depends on context, on the way it is stated or written, on the way people interpret various signs that go with it. Even single words seldom have a unique meaning. Meaning may also depend on the mood of whoever produces or receives the sentence.

Regarding algorithms to determine grammatical structure, they are dependent on how you define the grammatical structure of a language, and that is very far from being standardized (or to put it differently, there are competing ones). It is that definition that matters, not whatever algorithm you may produce to recognize the structure.

It is actually unlikely that any single description will explain all structures of a language. It is probably as unlikely that two people would agree on what belong or does not belong to a language. Furthermore, any language you consider is constantly evolving in time, and may vary geographically (and dead languages changed over time, so that remaining documents do not necessarily represent the same time or space variant).

In other word, it is a lot fuzzier than chemistry, and even organic chemistry, because the human brain is a very big and complicated machine. This fuzzyness is somewhat underscored by the fact that a good part of the work on natural language processing is based on statistical techniques.

For the time being, the best we can hope for is analyzing a variety of linguistic phenomena, and try to condense them into theories that will account for most, expecially the more frequent.

Grammatical structures may be more complex than trees, depending on what you wish to include. But even ignoring that, I would not really see the point of considering together sentences with the same tree structure as a special class. What matters in the theoretical support that assign such structure to sentences.

I would guess that most textbooks on natural language processing will have a chapter about describing the structure of sentences, though it may change some, depending on the book. This does not mean in any way that the knowledge is random, or totally disorganized. There are many bridges between the various approaches, and various ways to preserve coherence. As I said, the human brain is big and complicated.

If you are interested in formalization (as you seem to be), there is a considerable amount of formal results in computational linguistics. But this does not mean it is able to accurately describe all aspects of natural language.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.