Are there efforts to formalize and formally represent (e.g. as semantic network, as some kind of logic) of semantic of pragmatic knowledge. It is known, that every speaker/listener has two types of knowledge - linguistic/semantic knowledge (and pragmatic knowledge - sense about context) and everyday knowledge (commonsense knowledge). There are many efforts in computer science to formalize and represent commonsense knowledge. But what about efforts for formalize and represent linguistic/semantic/pragmatic knowledge?

If there are such efforts then combination of linguistic and commonsense knowledge processing can lead to really, really strong natural language processing and generation system.

Some examples of semantic knowledge (I know, that better list can be generated but I am only starting to learn semantics):

  • knowing that 'A above B' means the same as 'B under A'.
  • knowing popular metaphors, ability to differentiate between direct and indirect meaning of sentences
  • extraction of propositions from the sentences

Google is no help for me, because it return everything that is connected with the Web Semantics and description logics but it is completely unrelated to the semantics of natural language.

Computer science has managed to formalized 20 types of emotions and put human personality in 5 dimensional models, there are efforts of consciousness modeling, so - why not to formalize linguistic knowledge and use it for automation of natural language?

I am aware about formal semantics and categorial grammars but they are about technical parsing of the sentences and the knowledge represented by formal semantics of NL/combinatory categorial grammars (CCG) do not come near the completeness of the knowledge that is described in https://www.amazon.co.uk/Semantics-Introducing-Linguistics-John-Saeed/dp/1118430166/ and https://www.amazon.co.uk/Pragmatics-Oxford-Textbooks-Linguistics-Huang/dp/0199577765 So - my question is - are there efforts to formalize knowledge included in usual textbooks of NL semantics and pragmatics?

I am interested in logical methods of semantics and pragmatics, I have no belief in statistical methods.

  • 1
    They're big business. Semantics and Pragmatics have been formalized often and happily. – jlawler Nov 10 '17 at 0:03
  • I am very well aware of formal semantics of NL and categorial grammars, but they are only about parsing NL texts and formal semantics/categorial grammars say nothing about linguistic knowledge that is usually so required in selection of the best parse tree (unambiguation). Current CCG parsers use statistical metrics for disambiguation. So - my question is valid. – TomR Nov 10 '17 at 0:18
  • "Formal pragmatics" is good search term and but generally it does not refer to the formalization at the logic level, only good article ir link.springer.com/chapter/10.1007%2F978-94-017-8813-7_11 but it is also in the usual stule of categorial grammars. – TomR Nov 10 '17 at 0:26

There have been efforts to formalize pragmatic knowledge for quite some time now.

The main formalisms I know of that attempt to go beyond the sentence-level are things like RST (rhetorical structure theory) and its successors/competitors, like SDRT (Segmented Discourse Representation Theory) (cf Asher & Lascarides, 2007) or D-STAG (Discourse-Synchronous TAG, an attempt to integrate discourse analysis into the TAG syntactic formalism, cf Danlos 2008).

These formalisms are geared toward text coherence, discourse relations and inferred relations within discourse, and as such, they seem to fit especially the third problematic you give as an example. These formalisms are all about how propositions within a text are connected together to form a consistent whole.

As for your other two examples, knowing that "A above B" implies "B under A" seems to me to be more of a formal semantics question : are those two propositions strictly equivalent ? They seem to be, and as such they should be represented by the same logic formula. I think it's a very straightforward case, much less tricky than say, temporal relations, which require some kind of event based semantics. In this case, any kind of formal syntactico-semantic framework might be used (e.g HPSG).

And regarding "popular metaphors", I think it's pretty likely that if they are popular they are lexicalized to an extent and thus can be integrated seamlessly in syntactic paradigms as fixed expressions .

I think the bigger challenge is the case of metaphors which are not popular, which are incidental, and there are active efforts on metaphor detection, and efforts to bridge it with formal discourse analysis (for example this paper )

| improve this answer | |

I would say that the whole research area of Knowledge Representation, including Description Logics, Modal Logics, Temporal Logics, Fuzzy Logics, aims at formally representing semantics (and sometimes pragmatics) of knowledge that is usually represented in natural language.

However these research areas usually see the representation and reasoning independent from the question how to automatically translate natural language and natural language texts into formal representations.

Also representing knowledge is different from representing the structure of knowledge in text (RST) or the structure of natural language text (Syntax/Grammar).

| improve this answer | |

A well known approach in linguistics to formalizing semantics is Montague Semantics, see Stanford's Plato and jump to section 3.4 for a very brief summary and some references on pragmatics.

| improve this answer | |

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