From Raw Text to Linear λ-Terms
July 6, 2019
Recent advances in distributional-compositional semantics propose the use of vectorial word representations as a means of constructing meaning for larger linguistic units. The spaces occupied by these representations and the manner in which they combine are dictated by linear λ-terms which encode the sentence semantics. Despite their crucial role in the composition process, the terms themselves are given little attention and are assumed to be known in advance. This work will present a general framework for converting unprocessed text into a Type-Logical syntactic derivation. The Curry-Howard correspondence between proofs and programs identifies the derivation with a unique corresponding term, which can then be used as a generic structural skeleton for any semantic interpretation.