Intermediary Semantic Representation through Proposition Structures

We propose an intermediary-level semanticrepresentation, providing a higher levelof abstraction than syntactic parse trees,while not committing to decisions in casessuch as quantification, grounding or verbspecificroles assignments. The proposalis centered around the proposition structureof the text, and includes also implicitpropositions which can be inferredfrom the syntax but are not transparent inparse trees, such as copular relations introducedby appositive constructions. Otherbenefits over dependency-trees are explicitmarking of logical relations betweenpropositions, explicit marking of multiwordpredicate such as light-verbs, and aconsistent representation for syntacticallydifferentbut semantically-similar structures.The representation is meant toserve as a useful input layer for semanticorientedapplications, as well as to providea better starting point for further levels ofsemantic analysis such as semantic-rolelabelingand semantic-parsing.