Leveraging Linguistic Structure For Open Domain Information Extraction

Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, pages 344–354, Beijing, China, July 26-31, 2015. c⃝2015 Association for Computational Linguistics Leveraging Linguistic Structure For Open Domain Information Extraction Gabor Angeli Melvin Johnson Premkumar Department of Computer Science Stanford University {angeli, melvinj, manning}@cs.stanford.edu Christopher D. Manning Abstract Relation triples produced by open domain information extraction (open IE) systems are useful for question answering, infer- ence, and other IE tasks. Traditionally these are extracted using a large set of pat- terns; however, this approach is brittle on out-of-domain text and long-range depen- dencies, and gives no insight into the sub- structure of the arguments. We replace this large pattern set with a few patterns for canonically structured sentences, and shift the focus to a classifier which learns to extract self-contained clauses from longer sentences. We then run natural logic infer- ence over these short clauses to determine the maximally specific arguments for each candidate triple. We show that our ap- proach outperforms a state-of-the-art open IE system on the end-to-end TAC-KBP 2013 Slot Filling task. 1 Introduction Open information extraction (open IE) has been shown to be useful in a number of NLP tasks, such as question answering (Fader et al., 2014), rela- tion extraction (Soderland et al., 2010), and infor- mation retrieval (Etzioni, 2011). Conventionally, open IE systems search a collection of patterns over either the surface form or dependency tree of a sentence. Although a small set of patterns covers most simple sentences (e.g., subject verb object constructions), relevant relations are often spread across clauses (see Figure 1) or presented in a non-canonical form. Systems like Ollie (Mausam et al., 2012) ap- proach this problem by using a bootstrapping method to create a large corpus of broad-coverage partially lexicalized patterns. Although this is effective at capturing many of these patterns, it Born in Honolulu, Hawaii, Obama is a US Citizen. Our System Ollie (Obama; is; US citizen) (Obama; is; a US citizen) (Obama; born in; (Obama; be born in; Honolulu) Honolulu, Hawaii) (Honolulu; be born in; Hawaii) (Obama; is citizen of; US) Friends give true praise. Enemies give fake praise. Our System Ollie (friends; give; true praise) (friends; give; true praise) (friends; give; praise) (enemies; give; fake praise) (enemies; give; fake praise) Heinz Fischer of Austria visits the US Our System Ollie (Heinz Fischer; visits; US) (Heinz Fischer of Austria; visits; the US) Figure 1: Open IE extractions produced by the system, alongside extractions from the state- of-the-art Ollie system. Generating coherent clauses before applying patterns helps reduce false matches such as (Honolulu; be born in; Hawaii). Inference over the sub-structure of arguments, in turn, allows us to drop unnecessary information (e.g., of Austria), but only when it is warranted (e.g., keep fake in fake praise). can lead to unintuitive behavior on out-of-domain text. For instance, while Obama is president is extracted correctly by Ollie as (Obama; is; pres- ident), replacing is with are in cats …

Content truncated to 500 words