Toward a better understanding of causality between verbal events: Extraction and analysis of the causal power of verb-verb associations

The identification of causal relations between verbal events is important for achieving natural language understanding. However, the problem has proven notoriously difficult since it is not clear which types of knowledge are necessary to solve this challenging problem close to human level performance. Instead of employing a large set of features proved useful in other NLP tasks, we split the problem in smaller sub problems. Since verbs play a very important role in causal relations, in this paper we harness, explore, and evaluate the predictive power of causal associations of verb-verb pairs. More specifically, we propose a set of knowledge-rich metrics to learn the likelihood of causal relations between verbs. Employing these metrics, we automatically generate a knowledge base (KB_c) which identifies three categories of verb pairs: Strongly Causal, Ambiguous, and Strongly Non-causal. The knowledge base is evaluated empirically. The results show that our metrics perform significantly better than the state-of-the-art on the task of detecting causal verbal events.