Évaluation de l'adaptation par renforcement d'un générateur en langage naturel neuronal pour le dialogue homme-machine

Abstract : Evaluation of the reinforcement adaptation of a neural natural language generation system for human-machine dialogue. Traditional systems for natural language generation in spoken dialogue systems use patterns and rules to generate system answers. Recently, systems based on recurrent neural network models have been proposed (Wen et al., 2016a). Those systems require a large amount of data to be learned, which can be difficult to collect and annotate. Therefore we proposed a framework to adapt the NLG module online through direct interactions with the users (Riou et al., 2017). In this paper, we study the practical interest of the approach with real data collected as automatic speech recognition of users' suggestions and having humans assessing the system's outputs.
Complete list of metadatas

Cited literature [10 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02021596
Contributor : Stéphane Huet <>
Submitted on : Saturday, February 16, 2019 - 12:52:25 PM
Last modification on : Wednesday, May 15, 2019 - 10:12:03 AM
Long-term archiving on : Friday, May 17, 2019 - 4:10:05 PM

File

JEP18.pdf
Files produced by the author(s)

Identifiers

Collections

Citation

Matthieu Riou, Bassam Jabaian, Stéphane Huet, Fabrice Lefèvre. Évaluation de l'adaptation par renforcement d'un générateur en langage naturel neuronal pour le dialogue homme-machine. XXXIIe Journées d’Études sur la Parole (JEP), 2018, Aix-en-Provence, France. pp.347-355, ⟨10.21437/JEP.2018-40⟩. ⟨hal-02021596⟩

Share

Metrics

Record views

32

Files downloads

21