Skip to Main content Skip to Navigation
Conference papers

Apprentissage en ligne interactif d’un générateur en langage naturel neuronal pour le dialogue homme-machine

Abstract : On-line Interactive Learning of Natural Language Neural Generation for Human-machine Dialogue. Recently, some propositions have emerged to handle natural language generation in spoken dialog systems with recurrent neural network models (Wen et al., 2016a). Those models require a huge amount of learning data, which complicates the data collection and annotation. To address this issue, we propose an online-learning protocol, based on a reinforcement learning approach. We learn a model on a reduced corpus produced with templates, and adapt it online. For this first experiment, we propose an approach using an adversarial bandit algorithm, studying its advantages and limitations.
Complete list of metadatas

Cited literature [15 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02021590
Contributor : Stéphane Huet <>
Submitted on : Saturday, February 16, 2019 - 12:45:52 PM
Last modification on : Tuesday, January 14, 2020 - 10:38:06 AM
Long-term archiving on: : Friday, May 17, 2019 - 3:24:29 PM

File

TALN17.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02021590, version 1

Collections

Citation

Matthieu Riou, Bassam Jabaian, Stéphane Huet, Fabrice Lefèvre. Apprentissage en ligne interactif d’un générateur en langage naturel neuronal pour le dialogue homme-machine. 24ème conférence sur le Traitement Automatique des Langues Naturelles (TALN), 2017, Orléans, France. ⟨hal-02021590⟩

Share

Metrics

Record views

57

Files downloads

46