2
$\begingroup$

My question is - is there a semantic natural language processing that tries to understand the meaning of the texts and that tries to derive the consequences of the understood meaning? Is there a universal knowledge base that can be used for the "grounding" of the texts?

I have heard a lot about statistical NLP and NLP with neural networks but those approaches are not scalable, are not exact and are not satisfying. Is there (and if not - then why) a semantic NLP and semantic natural language understanding that tries to translate the texts into logical formulas? Today we have a vast array of logical formals - both rigorous and both nonmontonic, adaptable, fuzzy, probabilistic and so on? So - if we have those logics why don't we translate texts into them?

And if texts are translated into logical formulas then the universal knowledge base (KB) can be built. This KB should be used as reference KB. I.e. if some text contains the phrase "logical continuum" then this KB should contain all the possible definitions (expressed as logical formulas) of this phrase (according to the different texts, authors) and one should be able to reason over those definitions, to use definitions for applying this phrase into the new text (possibly - computer generated texts), to use definitions for deriving more general or more special notions with relevant definitions, to form completely new concepts. There can be two types of terms in this KB: abstract ideas and concrete real world facts, like country Belgium and so on.

I am aware of existence of some KBs like ConceptNet and WordNet (not really useful) and Cyc/OpenCyc (formal KB with reasoning capabilities) and there is also very promising framework OpenCog which (fortunately) has interesting reasoning engine, but (unfortunately) lacks the public knowledge base for experiments. OpenCog is really interesting, because it unifies probabilitist and rigorous reasoning: each concept (Atom in their terms) have strength/probability value and if those values converge to 1, then proababilistic reasoning tends to rigorous classical logic style reasoning. But that's all.

So - are there such notion as semantic NLP and are there endeavours to create universal knowledge base for semantic interpretation of any text? Are there ongoing projects in this field?

$\endgroup$
  • $\begingroup$ There are such notions as symbolic / subsymbolic knowledge. Subsymbolic knowledge being encoded in statistical relations and neural networks. And then there is connectionist approach to bridge those approaches. I have found one artical about symbolic NLP and it is partially the direction in what I am looking. $\endgroup$ – TomR Jan 7 '17 at 20:26
1
$\begingroup$

There are several implementations of semantic parsers that convert natural-language texts into formal logical representations of their meanings. Natural language understanding systems can also be based on discourse representation theories that represent the meanings of English texts using first-order logical predicates.

I have found at least one system that is able to generate a knowlege base from statements that are given in a natural language. The ACE Wiki is based on Attempto Controlled English, which is a semantically unambiguous subset of the English language.

$\endgroup$

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.