ZORE: A Syntax-based System for Chinese Open Relation Extraction

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1870–1880, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics ZORE: A Syntax-based System for Chinese Open Relation Extraction Likun Qiu and Yue Zhang Singapore University of Technology and Design, Singapore qiulikun@gmail.com, yue zhang@sutd.edu.sg Abstract Open Relation Extraction (ORE) over- comes the limitations of traditional IE techniques, which train individual extrac- tors for every single relation type. Sys- tems such as ReVerb, PATTY, OLLIE, and Exemplar have attracted much attention on English ORE. However, few studies have been reported on ORE for languages beyond English. This paper presents a syntax-based Chinese (Zh) ORE system, ZORE, for extracting relations and seman- tic patterns from Chinese text. ZORE identifies relation candidates from auto- matically parsed dependency trees, and then extracts relations with their semantic patterns iteratively through a novel double propagation algorithm. Empirical results on two data sets show the effectiveness of the proposed system. 1 Introduction Traditional Information Extraction (IE) system- s train extractors for pre-specified relations (Kim and Moldovan, 1993). This approach cannot scale to the web, where target relations are not defined in advance. Open Relation Extraction (ORE) at- tempts to solve this problem by shallow-parsing- based, syntax-based or semantic-role-based pat- tern matching without pre-defined relation types, and has achieved great success on open-domain corpora ranging from news to Wikipedia (Banko et al., 2007; Wu and Weld, 2010; Nakashole et al., 2012; Etzioni et al., 2011; Moro and Nav- igli, 2013). Many NLP and IR applications, in- cluding selectional preference learning, common- sense knowledge and entailment rule mining, have benefited from ORE (Ritter et al., 2010). Howev- er, most existing ORE systems focus on English, and little research has been reported on other lan- guages. In addition, existing ORE techniques are mainly concerned with the extraction of textual re- lations, without trying to give semantic analysis, which is the advantage of traditional IE. Our goal in this paper is to present a syntax- based Chinese (Zh) ORE system, ZORE, which extracts relations by using syntactic dependen- cy patterns, while associating them with explic- it semantic information. An example is shown is Figure 1, where the relation (cnê (Oba- ma) oÚ (President) , Pred[.’ (graduate)], M à (Harvard) {Æ� (Law School)) is extract- ed from the given sentence “cnê (Obama) o Ú (President) .’ (graduate) u (from) Mà (Harvard) {Æ� (Law School)”, and general- ized into the syntactic-semantic pattern {nsubj- NR(Af) Pred[. ’ (graduate)] prep-u (from) pobj-NN(Di)}. Here, Af and Di stand for human and institution, respectively, according to a Chi- nese taxonomy Extended Cilin (Che et al., 2010). Rather than extracting binary relations and then generalizing them into semantic patterns, which most previous work does (Mausam et al., 2012; Nakashole et al., 2012; Moro and Navigli, 2012; Moro and Navigli, 2013), we develop a novel method that extracts relations and patterns simul- taneously. A double propagation algorithm is used to make relation and pattern information reinforce each other, so that negative effects from automatic …

Content truncated to 500 words