Tatiana Bladier (Heinrich Heine University of Dusseldorf)
In my talk I will present ongoing work on data-driven frame-semantic parsing for French with Lexicalized Tree Adjoining Grammars [LTAG; 4] using French FrameNet data [1, 2]. I am using a supertagging approach combined with deep learning to facilitate the task assigning the core semantic roles and a subsequent frame recognition. In my talk, I will present the results of two experiments with joint prediction of semantic roles , with and without using the supertag features:
- predicting the full spans of semantic roles;
- predicting semantic heads in the first step (i.e. 1-token spans) and using dependency graphs for a subsequent reconstruction of the full span of the semantic role.
I will also present a hybrid approach to full semantic parsing with French FrameNet enriched with proto-frames built from deep syntactic annotations and discuss the possible strengths and the weaknesses of the approach.
-  Candito, M., Amsili, P., Barque, L., Benamara Zitoune, F., De Chalendar, G., Djemaa, M., Haas, P., Huyghe, R., Yannick Mathieu, Y., Muller, P., et al. (2014). Developing a french framenet: Methodology and first results.
-  Djemaa, M., Candito, M., Muller, P., and Vieu, L. (2016). Corpus annotation within the french framenet: a domain-by-domain methodology.
-  He, L., Lee, K., Levy, O., and Zettlemoyer, L. (2018). Jointly predicting predicates and arguments in neural semantic role labeling. arXiv preprint arXiv:1805.04787.
-  Joshi, A. K. and Schabes, Y. (1997). Tree-adjoining grammars. In Handbook of formal languages, pages 69–123. Springer.