Frame-Semantic Structure Prediction

Abstract

Frame semantics (Fillmore 1982) is a linguistic theory that has been instantiated for English in the FrameNet lexicon (Fillmore, Johnson, and Petruck 2003). We formalize frame-semantic parsing as a structure prediction problem. We present a two-stage statistical model that takes lexical targets (i.e., content words and phrases) in their sentential context and predicts frame-semantic structures. Given a target in context, the first stage disambiguates it to a semantic frame. This model employs latent variables and semi-supervised learning to improve frame disambiguation for targets unseen at training time. The second stage finds the target’s locally expressed semantic arguments. A fast exact dual decomposition algorithm collectively predicts all the arguments of a frame at once in order to respect declaratively stated linguistic constraints at inference time, resulting in qualitatively better structures than naïve local predictors. Both components are feature-based and discriminatively trained on a small set of annotated frame-semantic parses. On a benchmark dataset, the approach, along with a heuristic identifier of frame-evoking targets, outperforms prior state of the art by significant margins. Additionally, we present experiments on a much larger recent dataset and have released our accurate frame-semantic parser as open-source software.