Framenet And Natural Language Processing

Miriam R. L. Petruck, International Computer Science Institute, UC Berkeley

Stevenson Hall 1300
12:00 PM - 12:50 PM

This talk presents an overview of FrameNet, a research project in corpus-based computational lexicography, based on the principles of Frame Semantics (e.g. Fillmore 1985, inter alia). FrameNet's initial goals included providing information about the valences, i.e., the semantic and syntactic combinatorial possibilities for the vocabulary of contemporary English, and documented by corpus findings. Initially developed primarily as a lexicographic effort, FrameNet data are used in a range of Natural Language Processing (NLP) tasks, including Semantic Role Labeling (e.g. Das et al. 2014, Roth & Lapata, 2015, Roth 2016), Question Answering (e.g. Sinha 2008, Ofoghi 2009, He et al. 2015) Information Extraction (e.g. Harabagiu & Bejan 2010, Zhang et al. 2010, Søgaard et al. 2015), and Sentiment Analysis (Ruppenhoffer & Rehbien 2012, Ruppenhoffer 2013 ), to name a few. In addition to reporting on recent developments in FrameNet, this presentation will address the implications of these developments for NLP.