Computational Formal Semantics: English Fragment 2 and Parsing Preview

Sun 19 May 2013 by Adrian Brasoveanu

Plan for the May 21 (11 am-12 pm) meeting: we will finish up the discussion of EF1 syntax and semantics and we will introduce our second English fragment EF2. The discussion and code will be based on ch. 7 of the “Computational Semantics” textbook by van Eijck & Unger. Please download the files below and place them in the same folder:

  • EF1 syntax: EF1syn.hs; we keep the same syntax as before, our previous semantics did not interpret all the determiners listed in the EF1 syntax module
  • EF2 semantics: EF2sem.hs; the EF2 semantics module provides direct, compositional interpretations for all Eng. expressions in our fragment, including the non-first-order-definable generalized quantifier most
  • the model we already used for both FOL and EF1: Model.hs — the GHCI script EF2ghci.hs we will use to introduce and interact with the above modules

After we wrap up EF2, the rest of our meetings this quarter will be dedicated to parsing. Parsing is perhaps not obviously related to our current main topic (computational modeling of formal semantics). But parsing is an essential component if we want to embed our semantic models into psychologically realistic models of the human sentence/discourse comprehension process.

Building models of sentence/discourse processing with semantically explicit commitments is useful for formal semanticists for at least two reasons. First, this can further constrain our theories of human semantic competence: semantic theories should be embeddable in broader theories of human linguistic performance and it is largely unexplored how we can achieve this for the detailed theories currently available in the formal semantics literature.

Secondly, building processing models that take formal semantics seriously, i.e., that incorporate the highly structured, hierarchical and abstract representations and operations common in formal semantics, will enable us to connect in a very direct and formally explicit manner (i) our semantic theories and (ii) the kinds of behavioral measures that are common in psycholinguistics (e.g., reaction times in self-paced reading or eye-tracking tasks).

Although they do not include any (systematic) discussion of formal semantics, the three papers below provide detailed discussions of parsing in the context of developing psychologically realistic models of human sentence comprehension:

The hope is that the frameworks proposed in these papers could be enriched with the kind of hypotheses, representations and operations we use in formal semantics analyses of natural language phenomena.