People who look at the FrameNet annotation work frequently ask, "Can't you automate this?". In fact, a number of people have used machine learning techniques to build systems which can be trained on FrameNet annotation data and automatically produce similar annotation on new (previously unseen) texts. This process can be called (automatic) frame semantic role labeling (ASRL), or sometimes, semantic parsing. This should be distinguished from other systems for semantic role labeling, which are not based on Fillmore's concept of semantic frames, such as those based on PropBank ( Palmer et al. 2005)
Several systems for doing FrameNet-based ASRL have been freely distributed for anyone to use:
- SEMAFOR, created by Dipanjan Das and other members of Noah Smith's NLP group at Carnegie Mellon University (Download: http://www.ark.cs.cmu.edu/SEMAFOR)
- Framat, Michael Roth and Mirella Lapata (2015 TACL) Download: https://github.com/microth/mateplus
- Open-SESAME, Swabha Swayamdipta et al. 2017 paper on arXiv: https://arxiv.org/abs/1706.09528v1 Download: https://github.com/Noahs-ARK/open-sesame
- Older systems
- Shalmaneser, created by Katrin Erk and Sebastian Padó at Saarland University, Saarbrücken (http://www.coli.uni-saarland.de/projects/salsa/shal)
- LTH, created by Richard Johannson at Lund University (http://nlp.cs.lth.se/software/semantic-parsing-framenet-frames )