Two-Stage Method for Large-Scale Acquisition of Contradiction Pattern Pairs using Entailment

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 693–703, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Two-stage Method for Large-scale Acquisition of Contradiction Pattern Pairs using Entailment Julien Kloetzer∗ Stijn De Saeger† Kentaro Torisawa‡ Chikara Hashimoto§ Jong-Hoon Oh¶ Motoki Sano∥ Kiyonori Ohtake∗∗ Information Analysis Laboratory, National Institute of Information and Communications Technology (NICT), Kyoto, Japan {∗ julien, † stijn, ‡ torisawa, § ch, ¶rovellia, ∥ msano, ∗∗kiyonori.ohtake}@nict.go.jp Abstract In this paper we propose a two-stage method to acquire contradiction relations between typed lexico-syntactic patterns such as Xdrug prevents Ydisease and Ydisease caused by Xdrug. In the first stage, we train an SVM classifier to detect contradiction pattern pairs in a large web archive by exploiting the exci- tation polarity (Hashimoto et al., 2012) of the patterns. In the second stage, we enlarge the first stage classifier’s training data with new contradiction pairs obtained by combining the output of the first stage’s classifier and that of an entailment classifier. We acquired this way 750,000 typed Japanese contradiction pattern pairs with an estimated precision of 80%. We plan to release this resource to the NLP com- munity. 1 Introduction The ability to detect contradictory information in text has many practical applications. Among those, Murakami et al. (2009) pointed out that a contra- diction recognition system can detect conflicts and anomalies in large bodies of texts and flag them to help users identify unreliable information. For ex- ample, many Japanese web pages claim that agari- cus prevents cancer, where agaricus is a species of mushroom found in a variety of commercial prod- ucts. Although this has been accepted by many Japanese people, by Googling keywords ”agaricus”, ”promotes” and ”cancer”, we can find pages claim- ing that ”agaricus promotes cancer”, some of which point to a study authorized by the Japanese Min- istry of Health, Labour and Welfare1 reporting that 1 http://www.mhlw.go.jp/topics/bukyoku/iyaku/syoku- anzen/qa/060213-1.html a commercial product containing agaricus promoted cancer. Obviously, the existence of these pages casts serious doubt on the ability of agaricus to prevent cancer and encourages readers to dig more about this subject. The above example suggests that recognizing contradictory information can guide users to a true fact. Likewise, we believe that contradiction recog- nition is also useful when dealing with non-factual information that occupy most of our daily lives. For instance, there is a big controversy recently whether Japan should join an economic partnership agree- ment called the Trans Pacific Partnership (TPP), and quite serious but contradictory claims are plentiful in the mass media and on the web, e.g., TPP will wipe out Japan’s agricultural businesses and TPP will strengthen Japan’s agricultural businesses. Neither of these are facts; they are predictions that can only be realized or disputed after the underlying decision- making is done: joining or refusing the TPP. Furthermore, after reading documents including contradictory predictions, one should notice that each of them is supported by a convincing the- ory that has no obvious defect, e.g., “Exports of Japan’s agricultural …

Content truncated to 500 words