Speech and Language Technology in Education (SLaTE 2013)
We present and demonstrate a cloud-based personalized dialogue game for
computer-assisted learning of Mandarin Chinese. A sequence of tree-structured sub-dialogues
in restaurant scenario are linked recursively and used as the script for the game. Based on
NTU Chinese, a Mandarin Chinese pronunciation evaluation software
(http://chinese.ntu.edu.tw/), the user can get immediate evaluation on pronunciation, pitch,
timing and emphasis and corresponding corrective feedback on each syllable as well as on
sentence level for each utterance produced. The system policy is optimized to offer
personalized dialogue path planning for each individual learner such that more practice
opportunities are given along the dialogue path to poorly produced pronunciation units. When
using the system, the learner can practice the sub-dialogues in either sequential or random
order; at each dialogue turn, the learner also can choose to pronounce an arbitrary candidate
sentence or following the recommended sentences by the system policy. Following the system
recommendation along the sub-dialogues sequentially offers the fastest learning though. The
above evaluation and learning records are displayed and stored in personal profile.
The system framework is modeled as a Markov Decision Process (MDP) with highdimensional continuous state space considering the learning status of the learner. The dialogue policy is trained using a huge number of simulated learners generated from a corpus recorded by 278 real Mandarin Chinese learners from 36 countries with various mother tongues. The detailed principles of this system are presented in a companion paper also submitted to SLaTe 2013 . This is a joint work with the International Chinese Language Program of National Taiwan University.
Bibliographic reference. Su, Pei-hao / Yu, Tien-han / Su, Ya-Yunn / Lee, Lin-shan (2013): "NTU Chinese 2.0: a personalized recursive dialogue game for computer-assisted learning of Mandarin Chinese", In SLaTE-2013, 104.