Commonsense causal reasoning between short texts

摘要: Commonsense causal reasoning is the process of capturing and understanding dependencies amongst events actions. Such actions can be expressed in terms, phrases or sentences natural language text. Therefore, one possible way obtaining knowledge by extracting relations between terms from a large text corpus. However, are sparse, ambiguous, sometimes implicit, thus difficult to obtain. This paper attacks problem commonsense causality short texts (phrases sentences) using data driven approach. We propose framework that automatically harvests network causal-effect web Backed this network, we novel effective metric properly model strength terms. show these signals aggregated for reasonings texts, including phrases. In particular, our approach outperforms all previously reported results standard SE-MEVAL COPA task substantial margins.