3

I am a computational linguistics student, and I was thinking about defining a new framework, or a standard for translating or converting sentences into a conceptual or semantical intermediate language. Kind of like grabbing the essence or central idea or meaning of a sentence and then storing it in that intermediate language for later linguistics processing.

Now is IL a technique that the professionals in this field use at all nowadays?

Is this even a good idea, I mean defining a standard for representing meaning of the sentences in simplified intermediate language?

Thank you.

PS: This is different from the ILs that are mostly just syntactic and probably have the meaning distorted in their final output.

1

The idea of a comprehensive, cross-language/language-independent set of meanings is called linguistic universals. The task of identifying these universals is such a hugely daunting one that few have tried, and many even think it is not possible.

One attempt I personally think is fairly successful is the Natural Semantic Metalanguage project, begun over four decades ago by Anna Wierzbicka, and now worked on by a small but sizeable group of linguists. They have identified around 64 'semantic primes', the smallest indivisible concepts encoded in human languages, but it is very much a work in progress, and primes are added or removed every few years. NSM proposes not just the list of primes, but also proposes a universal syntax for how they combine together (semantically). If you were to develop an Intermediate Language I would strongly recommend building on the work of NSM.

Developing a NSM based computer format would probably not take too much work, but to develop a converter to take unparsed natural language and output an analysis in that format would be a multi-billion dollar project.

2
  • 2
    Yes, NSM is a good place to start. Cliff Goddard's work, especially, is very clear and perspicuous. My non-linguist partner attended a conference on generics where I chaired a session a few years ago, and the single talk she found interesting (and also the clearest) was Goddard's ("A Piece of Cheese, a Grain of Sand: The Semantics of Mass Nouns and Unitizers"), which is an NSM analysis. – jlawler Dec 7 '16 at 17:28
  • This is a very comprehensive and interesting answer @curiousdannii , thank you. You seem to have understood exactly what I had in mind, and what my idea was about. I want to pick it as the answer but I would like to wait a few more hours hoping new ideas or perspectives would start showing up, giving me more insight. Thank you – JackBixuis Dec 7 '16 at 20:40
2

The IL used in symbolic NLP is called first-order logic. There are various more or less differing notations but it all boils down to plain old FOL which can easily capture the meaning (literal or context-specific) of any well-formed sentence.

11
  • So, you basically mean trying to develop a framework to capture meaning is redundant work? Also, now that you have mentioned it, it rings a bell, considering our knowledge representation course. But, I was hoping that this IL could be used for both meaning extraction and also MT. Also FOL is too strict when it comes to constructing very complex concepts, it sometimes feels like it is beyond humans or totally impossible, especially for day to day human/natural language. – JackBixuis Dec 7 '16 at 14:22
  • 1
    Yes, it's redundant. There are event-based systems designed for universal use including MT. – Atamiri Dec 7 '16 at 18:39
  • 1
    "plain old FOL which can easily capture the meaning"? Plain old FOL without any additional extras is certainly not able to capture the precise meaning of even most simple declarative sentence. Or do you mean augmented FOL? – lemontree Dec 7 '16 at 19:35
  • @lemontree : Exactly what I wanted to say. FOL without extensions or the help of knowledge representation languages is totally unable to represent highly complex sentences. – JackBixuis Dec 7 '16 at 20:44
  • @lemontree Do you have an example of a simple declarative sentence whose meaning FOL can't capture? To be more specific, I meant "pure" FOL as used in the theory of Davidson, Parsons, Hobbs, etc. – Atamiri Dec 7 '16 at 21:28
1

Other answers refer to the classical paradigms. My answer will refer to a more recent, computation-centric paradigm. Distributional semantics may be taken as the observables sampled from a mixing process. The inferential semantics of such a scheme derive from the latent model structure, which is accessible by Monte Carlo , spectral, or constraint-satisfying inference, but in any case should be expressible as operator algebras over the space of factors underlying the conditional distributions of utterances.

In some application domains, e.g. machine translation, text generation, the Neural model has recently taken the stage by storm, demonstrating overwhelming out-performance on a preponderance of important benchmarks. Responsible professionals, therefore, are in process of reorienting themselves to align to the "deep learning" paradigm, which is consonant with an essentially distributional+spectral representation and inference schema. Rather than "standardizing" to an (imposed) crisp logical representation, such schemata let the data derive a representation - one more generally adequate to the capture of inherent ambiguities and other semantic phenomena which are particularly intractable in FoL decorated with simply typed lambda calculus, for example.

Which paradigm is most productive? Surely this will depend on application. If early disambiguation contributes strongly to mitigating loses, a combinatorial logical semantics (q.v. the Groningen Meaning Bank) is a good bet. If carrying ambiguities along through the chain of inference is helpful (even if only as a form of regularization), then a highly parametric vector space representation is likely to capture more of the underlying logical dynamics of agent linguistic processes.

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.