Automatic Discourse Parsing of Arabic Texts: The Case of Attachments

Abstract

The study of the structure of the Arabic texts is considered as a modern concern. Its importance lies in its ability to determine the semantic and rhetorical meaning of the discourse. Through a coherent structural graph consisting of text units and rhetorical relations linking them. It also highlights its importance by employing it in several applications from the natural language processing field, for example the question/answer system, the automatic translation and the automated text summary system and the Acquisition of the Arabic terminology. The rhetorical analysis is based on three important pillars. The first pillar is to divide the text into text units. The second pillar is to look for structural links between different text units. The third pillar connects these units to each other through rhetorical relations with semantic meanings. In this context, our task of the automatic construction of discourse structure: the case of attachments falls within the third pillar of rhetorical analysis. This approach of rhetorical analysis is based on the segmented discourse representation theory (SDRT) within our proposed method and on the classifier RandomForest. Our method was tested on the corpus of test, where the Fmeasure was 73%.