This paper discusses how to extract causal knowledge from a large text collection. Our goal is to develop an automatic method of identifying the causal knowledge that supports the underlying coherence of a given causal expression. We report our ongoing attempts to design the overall problem setting and to explore the feasibility of pursue the goal, forcusing on a class of target sentences including a resultative conjunction "tame". While the recall was not very high, the precision we achieved in an experiment empirically proved that our approach is promising.
展开▼