Skip to Main content Skip to Navigation
Theses

Modélisation des attentes en dialogue oral

Abstract : This thesis aims at modelling speaker's expectations in human dialogue and in man-machine dialogue. Speaker's expectations are the expected results of its acts, assuming that someone in finalized dialogue (oral or not, human or man-machine) has certain expectations from its addressee following his statement.
The theory developed draws its sources from analytic philosophy and its speech acts, from ethnomethodology through the faces theory, from cognisciences and the theory of relevance, and from linguistics. A notation of dialog acts has been made to represent knowledge, roles and speech acts of participants. The model considers rules - associating acts and expectations - supplied with probabilities, strategies and dialogic goals.
For rules and probabilities evaluation, two corpora were collected: first with human dialogue (within a framework of speech translation), second with man-machine dialogue (to develop a virtual assistant). For the collection of the second one, a wizard of Oz based platform has been made in order to simulate the final system. The corpora collected were transcribed, annotated and analysed in order to show the interest of considering expectations rather than making simple speech acts prediction.
The results are interesting for both human and man-machine dialogue: considering the expectations improves identification of interlocutor dialogue acts, and dialogue management.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00095971
Contributor : Yannick Fouquet <>
Submitted on : Monday, September 18, 2006 - 4:00:24 PM
Last modification on : Friday, November 6, 2020 - 3:51:44 AM
Long-term archiving on: : Friday, November 25, 2016 - 11:22:17 AM

Identifiers

  • HAL Id : tel-00095971, version 1

Collections

UJF | IMAG | CNRS | UGA

Citation

Yannick Fouquet. Modélisation des attentes en dialogue oral. Autre [cs.OH]. Université Joseph-Fourier - Grenoble I, 2004. Français. ⟨tel-00095971⟩

Share