A cloud-based personalized recursive dialogue game system for computer-assisted language learning

In this paper we present the design and experimental results of a cloud-based personalized recursive dialogue game system for computer-assisted language learning. A number of tree-structured sub-dialogues are used sequentially and recursively as the script for the game. The dialogue policy at each dialogue turn is optimized to offer the most appropriate training sentence for every individual learner considering the learning status, such that the learner can have the scores for all selected pronunciation units exceeding a pre-defined threshold in minimum number of turns. The policy is modeled as a Markov Decision Process (MDP) with high-dimensional continuous state space and trained with a huge number of simulated learners generated from a corpus of real learner data. A real cloud-based system is implemented and the experimental results demonstrate promising outcomes.