Identifier les « singletons » dans des corpus français annotés en coréférence : peut-on prévoir l’absence de reprise coréférentielle ?

Abstract

Finding coreferences in corpora is a difficult task for which the identification of singletons is an important issue. Solving this issue would allow for improving the process of corpus annotation and the identification of referential chains. To achieve this, it is important to determine whether or not singletons have linguistic properties of their own. After an overview of the question, the article presents a corpus study. Based on the results of the study, it is possible to “profile” the mentions of a referent remaining in the singleton state. A thousand mentions were studied in different genres and types of texts. The results suggest that the genre/text type and the ontological category of the referent predict the repetition or the absence of repetition of a referent in a text.

Keywords

• annotation
• corpus
• coreference
• text genre
• reference chain
• singletons