Extracting N-ary Cross-sentence Relations Using Constrained Subsequence Kernel

Abstract

Most of the past work in relation extraction dealswith relations occurring within a sentence and having onlytwo entity arguments. We propose a new formulation of therelation extraction task where the relations are more generalthan intra-sentence relations in the sense that they may spanmultiple sentences and may have more than two arguments.Moreover, the relations are more specific than corpus-levelrelations in the sense that their scope is limited only withina document and not valid globally throughout the corpus. Wepropose a novel sequence representation to characterize instancesof such relations. We then explore various classifiers whosefeatures are derived from this sequence representation. For SVMclassifier, we design a Constrained Subsequence Kernel which isa variant of Generalized Subsequence Kernel. We evaluate ourapproach on three datasets across two domains: biomedical andgeneral domain.