Identifying Constant and Unique Relations by using Time-Series Text

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 883–892, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Identifying Constant and Unique Relations by using Time-Series Text Yohei Takaku∗ Toyo Keizai Inc. Chuo-ku, Tokyo 103-8345, Japan takaku.yohei@gmail.com Nobuhiro Kaji and Naoki Yoshinaga Institute of Industrial Science, University of Tokyo Meguro-ku, Tokyo 153-8505, Japan {kaji,ynaga}@tkl.iis.u-tokyo.ac.jp Masashi Toyoda Institute of Industrial Science, University of Tokyo Meguro-ku, Tokyo 153-8505, Japan toyoda@tkl.iis.u-tokyo.ac.jp Abstract Because the real world evolves over time, nu- merous relations between entities written in presently available texts are already obsolete or will potentially evolve in the future. This study aims at resolving the intricacy in con- sistently compiling relations extracted from text, and presents a method for identifying constancy and uniqueness of the relations in the context of supervised learning. We ex- ploit massive time-series web texts to induce features on the basis of time-series frequency and linguistic cues. Experimental results con- firmed that the time-series frequency distribu- tions contributed much to the recall of con- stancy identification and the precision of the uniqueness identification. 1 Introduction We have witnessed a number of success stories in acquiring semantic relations between entities from ever-increasing text on the web (Pantel and Pennac- chiotti, 2006; Banko et al., 2007; Suchanek et al., 2007; Wu et al., 2008; Zhu et al., 2009; Mintz et al., 2009; Wu and Weld, 2010). These studies have suc- cessfully revealed to us millions of relations between real-world entities, which have been proven to be beneficial in solving knowledge-rich problems such as question answering and textual entailment (Fer- rucci et al., 2010). ∗This work was conducted while the first author was a grad- uate student at University of Tokyo. There exists, however, a great challenge to com- pile consistently relations extracted from text by these methods, because they assume a simplifying assumption that relations are time-invariant. In other words, they implicitly disregard the fact that state- ments in texts actually reflect the state of the world at the time when they were written, which follows that relations extracted from such texts eventually become outdated as the real world evolves over time. Let us consider that relations are extracted from the following sentences:1 (1) a. 1Q84 is written by Haruki Murakami. b. Moselle river flows through Germany. c. U.S.’s president is George Bush. d. Pentax sells K-5, a digital SLR. Here, italicized predicates represent the relations, while underlined entities are their arguments. The relations in statements 1a and 1b are true across time, so we can simply accumulate all the relation instances. The relations in 1c and 1d in contrast evolve over time. The relation written in 1c be- comes outdated when the other person takes the position, so we need to supersede it when a new relation is extracted from text (e.g., U.S’s president is Barack Obama). For the relation in 1d, we do not always need to supersede it with a new relation. This study is …

Content truncated to 500 words