Skip to Main content Skip to Navigation
Book sections

La saillance référentielle pour la détection des thèmes

Abstract : The purpose of this paper is to describe an automatic topic detection system for French, using two types of cohesion markers: lexical markers and referential markers, as reference chains. Our study focuses on the automatic identification of reference chains in order to determine the most salient referent in a textual segment. We propose a new symbolic method for reference computation, based on the salience of referential expressions. For each candidate, the salience is fixed. The reference computation method combines three elements: a hierarchy of referential expressions based on the Accessibility Theory, some genre-based parameters and the syntactic function of each expression. In addition, antecedent-anaphora candidates verify some lexical, morpho-syntactic and semantic constraints. The automatic topic identification is then performed by combining the reference chains with a similar lexical head of the first mention.
Complete list of metadata

https://hal.inria.fr/hal-00870259
Contributor : Amalia Todirascu <>
Submitted on : Sunday, October 6, 2013 - 5:23:15 PM
Last modification on : Thursday, February 27, 2020 - 2:16:02 PM

Identifiers

  • HAL Id : hal-00870259, version 1

Collections

Citation

Laurence Longo, Amalia Todirascu. La saillance référentielle pour la détection des thèmes. Boisseau Maryvonne and Hamm Albert. Saillance (2) : La saillance en langue et en discours, Presses Universitaires de Franche-Comté, 2015. ⟨hal-00870259⟩

Share

Metrics

Record views

273