Defended on 14-12-2012, in Nice.
Résumé : Statistical natural language processing (SNLP) are quantitative methods of text comparison. With logometry, SNLP and recent linguistic theories are combined to produce new tools for textual observations. Our logometric practice is a statistical application of François Rastier’s semantic theories on a corpus of Pierre Mendès France’s discurses. Our statistical way determine particularity by generality : words by paragraph, paragraphs by text, texts by corpus. We want to describe how situation – historical or social practice – influe on textual structure. To study the distribution high frequency words in the corpus, our works use two innovent tools : the cotextual environnement and the asymmetric cooccurrence. The former produces "smallworlds" : description of lexical semantic networks in texts ; the latter describes rythms of lexical variations. In our corpus, these rythms are associated to distinct argumentative values : persuasive-explanatory vs informative.
View online : Thèses.fr