Cause-Effect Relation Learning.

To be able to answer the question What causes tumors to shrink? , one would require a large cause-effect relation repository. Many efforts have been payed on is-a and part-of relation leaning, however few have focused on cause-effect learning. This paper describes an automated bootstrapping procedure which can learn and produce with minimal effort a cause-effect term repository. To filter out the erroneously extracted information, we incorporate graph-based methods. To evaluate the performance of the acquired cause-effect terms, we conduct three evaluations: (1) human-based, (2) comparison with existing knowledge bases and (3) application driven (SemEval-1 Task 4) in which the goal is to identify the relation between pairs of nominals. The results show that the extractions at rank 1500 are 89% accurate, they comprise 61% from the terms used in the SemEval-1 Task 4 dataset and can be used in the future to produce additional training examples for the same task.