This resource of predicative entailment rules contains three resources in two formats – shallow and syntactic. Resources are learned over the REVERB data set and using the local and algorithms described in Chapter 5 of Jonathan Berant’s thesis (which is part of the package). The local classifier was trained over a crowdsourced training set and with distributional similarity, lexicographic and string-similarity features. The global algorithm uses the local classifier and adds an additional transitivity constraint.
- Download the resource and documentation (0.5 GB rar file)
- readme file
- Contact: Jonathan Berant, jonatha6 located-at post.tau.ac.il