Replacing frames by Intersective Levin Classes
Giulea and Moschitti (2006) propose to use ILC as feature in semantic role labeling instead of frame information. Since there is vastly more data (e.g. PropBank) to train a classifier for ILC than FrameNet frames, this approach helps increase coverage.
Their experiments show that the use of ILC results in a negligible drop in performance, i.e. ILC carries pretty much the same information as FrameNet frames.
The mapping between frames and ILC can be problematic though:
- frames can be defined over adjectives and nouns while ILC are for verbs only
- one frame may correspond to many ILC
- one ILC may correspond to many frames
- a frame and a ILC can form overlapping verb groups
TODO: how big are those problems?
Drawback: complicated, waste (don't use) FrameNet annotations.
Generating new examples
Bauer and Rambow (2011) use FrameNet corpus and VerbNet equivalent classes to generate new examples.
- Giuglea, A. M., & Moschitti, A. (2006, July). Semantic role labeling via framenet, verbnet and propbank. In Proceedings of the 21st International Conference on Computational Linguistics and the 44th annual meeting of the Association for Computational Linguistics (pp. 929-936). Association for Computational Linguistics.
- Daniel Bauer and Owen Rambow, 2011. Increasing Coverage of Syntactic Subcategorization Patterns in FrameNet Using Verbnet. PDF