TY - GEN
T1 - Learning adaptive referring expression generation policies for spoken dialogue systems
AU - Chandrasekaran Janarthanam, Srinivasan
AU - Lemon, Oliver
PY - 2010
Y1 - 2010
N2 - We address the problem that different users have different lexical knowledge about problem domains, so that automated dialogue systems need to adapt their generation choices online to the users' domain knowledge as it encounters them. We approach this problem using Reinforcement Learning in Markov Decision Processes (MDP). We present a reinforcement learning framework to learn adaptive referring expression generation (REG) policies that can adapt dynamically to users with different domain knowledge levels. In contrast to related work we also propose a new statistical user model which incorporates the lexical knowledge of different users. We evaluate this framework by showing that it allows us to learn dialogue policies that automatically adapt their choice of referring expressions online to different users, and that these policies are significantly better than hand-coded adaptive policies for this problem. The learned policies are consistently between 2 and 8 turns shorter than a range of different hand-coded but adaptive baseline REG policies. © 2010 Springer-Verlag Berlin Heidelberg.
AB - We address the problem that different users have different lexical knowledge about problem domains, so that automated dialogue systems need to adapt their generation choices online to the users' domain knowledge as it encounters them. We approach this problem using Reinforcement Learning in Markov Decision Processes (MDP). We present a reinforcement learning framework to learn adaptive referring expression generation (REG) policies that can adapt dynamically to users with different domain knowledge levels. In contrast to related work we also propose a new statistical user model which incorporates the lexical knowledge of different users. We evaluate this framework by showing that it allows us to learn dialogue policies that automatically adapt their choice of referring expressions online to different users, and that these policies are significantly better than hand-coded adaptive policies for this problem. The learned policies are consistently between 2 and 8 turns shorter than a range of different hand-coded but adaptive baseline REG policies. © 2010 Springer-Verlag Berlin Heidelberg.
KW - Referring Expression Generation
KW - Reinforcement Learning
KW - Spoken Dialogue System
UR - http://www.scopus.com/inward/record.url?scp=77956314853&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-15573-4_4
DO - 10.1007/978-3-642-15573-4_4
M3 - Conference contribution
SN - 3642155723
SN - 9783642155727
VL - 5790 LNAI
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 67
EP - 84
BT - Empirical Methods in Natural Language Generation - Data-Oriented Methods and Empirical Evaluation
T2 - 12th European Workshop on Natural Language Generation
Y2 - 30 March 2009 through 3 April 2009
ER -