Extracting Context-Rich Entailment Rules from Wikipedia Revision History

摘要

Recent work on Textual Entailment has shown a crucial role of knowledge to support entailment inferences. However, it has also been demonstrated that currently available entailment rules are still far from being optimal. We propose a methodology for the automatic acquisition of large scale context-rich entailment rules from Wikipedia revisions, taking advantage of the syntactic structure of entailment pairs to define the more appropriate linguistic constraints for the rule to be successfully applicable. We report on rule acquisition experiments on Wikipedia, showing that it enables the creation of an innovative (i.e. acquired rules are not present in other available resources) and good quality rule repository.