Improving semantic role labeling through semi-supervised learning
Speaker: Hagen Fürstenau
Institution: Saarland University
Abstract:
In this talk I am going to give an overview of my preliminary results and ongoing efforts of putting semi-supervised learning techniques to use in Frame Semantic role labeling. Different methods of assessing syntactic and semantic similarity of predicate-argument structures will be presented and compared as to their effectiveness in providing a given semantic role labeler with more training instances. After analysing results and errors, I will outline some ideas for future improvements. Feedback greatly appreciated!