User-Centered Analysis of Corpora Using Semantic Features Redundancy

Accessing textual information is still a complex activity when users have to browse through large corpora or long texts. In order to help users in such tasks, we propose a model dedicated to lexical representation of thematic domains as well as tools for personal corpora analysis. The lexical model is a differential one, inspired by Saussure's semiotics. It consists in structuring and describing lexical units by the way of semantic features which are differences between terms meanings. Each thematic domain is represented by a set of terms characterized by many semantic features. These are built by the user through an interactive tool developed by our team. Generally, domains include between 60 and 100 terms. Lexical resources are identified in the corpus with the ProxiDocs tool. It returns interactive maps and reports built from the distribution of domains terms in the corpus. Maps reveal proximities and links between texts or sets of texts. The most often repeated semantic features in texts and in sets of texts are pointed out on the maps. According to the Interpretative Semantics, we call such a redundancy “intertextual isotopies”. These intertextual isotopies can represent redundancies of global domains which reveal topics of the considered texts, or can indicate a local semantic property, such as an expression of violence for instance, shared by some texts of the corpus. In this paper, first we present the lexical model as well as the related tools for building personal lexical resources and interactively visualising them in a corpus. The second section deals with notions linked to the semantic features and particularly with the intertextual isotopies. We also propose in this section methods to detect them in corpus. Section 3 presents two experiments in order to illustrate how such a redundancy can be useful for two kinds of tasks: information retrieval in a Web pages corpus and semantic analysis of conceptual metaphors in a domain-specific corpus of newspapers. Finally, we conclude on the importance to take into account the intertextual isotopies, and more generaly the global context established by the corpus, in tasks of access to information.