Identifying Conflicting Information in Texts

Understanding relationships between text passages is key for information analysis. We focus here on the contradiction relationship, and build a system to detect conflicting statements. We show that such a system needs to make more fine-grained distinctions than the common systems for entailment. Also, we argue for the centrality of event coreference and therefore incorporate a component based on topicality. We propose a typology of contradictions that naturally arise in text, and give the first detailed breakdown of performance for the contradiction detection task. Although detecting some types of contradiction requires deeper inferential paths than our system is capable of, we achieve good performance on types arising from negation and antonymy.