Automatic rich annotation of large corpus of conversational transcribed speech

Jean-Yves Antoine (1) , Abdenour Mokrane (1) , Nathalie Friburger (1)

1 BDTLN - Bases de données et traitement des langues naturelles

This paper describes the use of the CasSys platform in order to achieve the chunking of conversational speech transcripts by means of cascades of Unitex transducers. Our system is involved in the EPAC project of the French National agency of Research (ANR). The aim of this project is to develop robust methods for the annotation of audio/multimedia document collections which contains conversational speech sequences such as TV or radio programs. At first, this paper presents the EPAC project and the adaptation of a former chunking system (Romus) which was developed in the restricted framework of dedicated spoken man-machine dialogue. Then, it describes the problems that are arising due to 1) spontaneous speech disfluencies and 2) errors for the previous stages of processing (automatic speech recognition and POS tagging).

Mots clés: language resources; speech corpus; chunking

Domaines: Informatique et langage [cs.CL]

Jean-Yves Antoine, Abdenour Mokrane, Nathalie Friburger. Automatic rich annotation of large corpus of conversational transcribed speech. sixth international conference on Language Resources and Evaluation, LREC 2008, May 2008, Marrakech, Morocco. pp.172. ⟨hal-00484046⟩