Dependency-Based Open Information Extraction

Abstract

Building shallow semantic representations from text corpora is the first step to perform more complex tasks such as text entailment, enrichment of knowledge bases, or question answering. Open Information Extraction (OIE) is a recent unsupervised strategy to extract billions of basic assertions from massive corpora, which can be considered as being a shallow semantic representation of those corpora. In this paper, we propose a new multilingual OIE system based on robust and fast rule-based dependency parsing. It permits to extract more precise assertions (verb-based triples) from text than state of the art OIE systems, keeping a crucial property of those systems: scaling to Web-size document collections. 1