Transcription and annotation of a apprenticeship corpus: application to the correction and self-correction strategies

Jean-Léon Bouraoui (1) , Gwenael Bothorel (2, 3) , Nadine Vigouroux (4, 5, 6)

1 IRIT-IHCS - Interacting Humans with Computing Systems
2 DGAC - Direction Générale de l'Aviation Civile
3 IRIT-APO - Algorithmes Parallèles et Optimisation
4 IRIT-ELIPSE - Etude de L’Interaction Personne SystèmE
5 CNRS - Centre National de la Recherche Scientifique
6 GREYC - Groupe de Recherche en Informatique, Image et Instrumentation de Caen

This paper presents a corpus-based study devoted to corrections, self-corrections strategies to recover errors during dialogues. The corpus used results from exercises where air-traffic controllers being formed interacts with people simulating pilots in practice. Considering correction strategies as an answer to errors, we present a fine-grained typology of these strategies, the errors they aims to compensate, and their markers. We also give various statistics about their distribution in the corpus, and comment them, in regards with their application to spoken dialog systems.

Domaines: Interface homme-machine [cs.HC]

Jean-Léon Bouraoui, Gwenael Bothorel, Nadine Vigouroux. Transcription and annotation of a apprenticeship corpus: application to the correction and self-correction strategies. ISCA Tutorial and Research Workshop on Error Handling in Spoken Dialogue Systems 2003, Aug 2003, Vaud, Switzerland. pp.35-40. ⟨hal-03636743⟩