Improving Persian Dependency-Based Semantic Role Labeling using Semantic and structural Relations
Soghra Lazemi, Hossein Ebrahimpour-Komleh, Nasser Noroozi · 2019
In the automatic processing of the natural language, understanding the meaning of the sentence is accomplished by the semantic role labeler. The semantic role labeler does this task by examining the semantic connections between words (often verbs and their dependents) and imposing semantic roles on each of them (dependents) according to the occurred event in the sentence. This paper provides a dependency-based semantic role labeler with the help of clustering algorithms for Persian. We have examined the semantic role labeling problem as a classification problem. In our proposed method, for each verbal predicate, the candidate arguments are identified with the help of dependency relationships, and then the feature vector for them is extracted using the information in the dependency trees. In the next step, using the classification algorithms, the appropriate label for each of the arguments is determined. Finally, to improve the results, the sentences are projected to semantic and structural vector spaces and clustering is performed on them. The resulted information from clustering is used to correct semantic role labels. Experiments have been done on the first semantic role corpus in Persian language and the corpus provided by the authors. The achieved Macro-average F1-measure is 74.87 for the first corpus and 73.62 for the second one.