Tracking evaluation in discourse

International audienceThis paper presents a study in the scope of discourse analysis, focussing on how to track evaluation and opinion in discourse. Our aim is to propose a method and a model for automatic analysis of students or teachers opinion in their textual production. A corpus of texts of opinion, collected from different domains, revealed recurrent patterns in the way the authors enunciate their evaluation : lexical, syntactic as well as semantic regularities. Figurative language is also a mean to reinforce the expression of opinion. We propose a NLP model exploiting the previous observations for detection and semantic analysis of the parts of discourse in which authors concentrate their evaluation. This model has been implemented and evaluated during the DEFT07 challenge. We comment the results and dis- cuss how to adapt these model and tool to the E-learning domain