Automatic semantic role labeling in a Dutch corpus
Gerwert Stevens · Utrecht University Repository (Utrecht University) · 2006
In this thesis, an approach to automatic semantic role labeling (SRL) in Dutch corpora is presented. Although there has been an increasing interest in automatic SRL in recent years, previous research has focused mainly on English corpora. Adapting earlier research to the Dutch situation poses an interesting challenge. First and foremost, because the machine learning techniques used in previous research can not be applied to Dutch texts. This is due to the fact that there is no semantically annotated Dutch corpus available that can be used as training data. In order to solve this problem, a novel approach to rule-based tagging based on Alpino dependency trees is proposed in the first part of this thesis. This approach has been implemented in a rule-based semantic argument tagger, called XARA. After a corpus has been tagged automatically by XARA, manual annotation can be performed relatively fast, since annotators only need to correct XARA’s output instead of starting annotation from scratch. In the last part of this thesis, the training and evaluation of a learning system for SRL trained on such a manually corrected corpus is discussed. The evaluation shows that both XARA and the learning system achieve satisfactory results: the rule-based system achieves an F-score of 50,79, the learning system an F-score of 65,29.