Renforcement en-ligne pour l’apprentissage conjoint de l’analyseur sémantique et du gestionnaire de dialogue d’un système d’interaction vocale

Abstract : Design of dialogue systems has witnessed many advances lately, yet acquiring a huge dataset remains a hindrance to their fast development for a new task or language. On-line learning is pursued in this paper as a convenient way to alleviate these difficulties. After the system modules are initiated, a single process handles data collection, annotation and use in training algorithms. A new challenge is to control the cost of the on-line learning borne by the user. Our work focuses on learning the semantic parsing and dialogue management modules. In this context, we propose several variants of simultaneous learning which are tested in user trials to confirm that only a few hundred training dialogues allow us to achieve good performance and overstep a rule-based handcrafted system. The analysis of these experiments gives us some insights, discussed in the paper, about the difficulty for the system’s trainers to establish a coherent and constant behavioural strategy to enable a fast and good-quality training phase.
Complete list of metadatas

Cited literature [26 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02160317
Contributor : Florent Breuil <>
Submitted on : Friday, June 21, 2019 - 8:53:45 AM
Last modification on : Wednesday, June 26, 2019 - 2:59:32 PM

File

rjcia2019-04.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02160317, version 1

Collections

Citation

Matthieu Riou, Bassam Jabaian, Stéphane Huet, Fabrice Lefèvre. Renforcement en-ligne pour l’apprentissage conjoint de l’analyseur sémantique et du gestionnaire de dialogue d’un système d’interaction vocale. Rencontres des Jeunes Chercheurs en Intelligence Artificielle 2019, Jul 2019, Toulouse, France. pp.27-34. ⟨hal-02160317⟩

Share

Metrics

Record views

53

Files downloads

33