AraknionSRL is a semantic tagging system developed from the tools and data available to the Ancora corpus. AraknionSRL is based on two models: one for labeling the argument position and one for the thematic role.
For the construction, evaluation, use and classification of models, NaiveBayes methods of machine learning library WEKA have been used. The models are based on morphological and syntactic features of each candidate for a predicate argument (labeled with a syntactic function "func" in Ancora) and each predicate.
AraknionSRL corpus is able to tag both Ancora format as column format type SemEval-2010. May be used in catalan or spanish texts.
The evaluation with 10-fold cross-validation on the corpus Ancora, achieved an F1 of 0.83 for the argument position and 0.76 for the thematic role.
Download it aquí. (You must register and log in this website.)