Output details

Evaluation of a hierarchical reinforcement learning spoken dialogue system

<22> Originality: In contrast to the spoken dialogue systems whose dialogue control strategies are either handcrafted by human designers or learnt automatically from data, the proposed hierarchical reinforcement learning approach enabled us to incorporate both approaches, which proved to outperform the conventional ones.

Significance: Developed and evaluated a heuristic simulation environment used to learn dialogue strategies in an automatic way, and developed and evaluated hierarchical spoken dialogue behaviours learnt using a Semi-Markov Decision Process (SMDP) to address the problem of scalable dialogue optimisation.

Rigour: Thorough experimental evaluations in terms of real and simulated users.