New out now: VerbAtlas
VerbAtlas is a novel large-scale manually-crafted semantic resource for wide-coverage, intelligible & scalable Semantic Role Labeling, developed by the Sapienza NLP group and the multilingual Natural Language Processing research team at the Sapienza University of Rome.
The goal of VerbAtlas is to manually cluster WordNet synsets that share similar semantics into sets of semantically-coherent frames. The main features are:
- 466 semantically-coherent frames using 26 cross-frame VerbNet-inspired semantic roles for their argument structure.
- Available both for download and via RESTful API.
- Full coverage of WordNet 3.0 verb synsets (13,000+).
- Complete linkage to BabelNet 4.0, which supports 280+ languages (new version to come later this year!).
- Manual mapping to PropBank of all CoNLL-2009 and CoNLL-2012 dataset occurrences (5000+ mappings).
- Selectional preferences: the superconcept most probably associated with a semantic role in a frame (e.g. food for the patient role of the EAT frame).
- Default/shadow arguments: arguments logically implied or already incorporated into a verb.
- Implicit arguments: arguments that are implicit in the argument structure of a verb.
Andrea Di Fabio, Simone Conia and Roberto Navigli present VerbAtlas, a new, hand-crafted lexical-semantic resource whose goal is to bring together all verbal synsets from WordNet into semantically-coherent frames. The frames define a common, prototypical argument structure while at the same time providing new concept-specific information. In contrast to PropBank, which defines enumerative semantic roles, VerbAtlas comes with an explicit, cross-frame set of semantic roles linked to selectional preferences expressed in terms of WordNet synsets, and is the first resource enriched with semantic information about implicit, shadow, and default arguments.
The authors demonstrate the effectiveness of VerbAtlas in the task of dependency-based Semantic Role Labeling and show how its integration into a high-performance system leads to improvements on both the in-domain and out-of-domain test sets of CoNLL-2009.