| Paper Abstract and Keywords |
| Presentation |
2009-01-29 15:10
WFST-based Dialog Management Using Statistical Dialog Model Chiori Hori, Kiyonori Ohtake, Teruhisa Misu, Hideki Kashioka, Satoshi Nakamura (NICT-ATR) SP2008-130 |
| Abstract |
(in Japanese) |
(See Japanese page) |
| (in English) |
Abstract We have proposed an expandable and portable dialog scenario description and platform to manage dialog sys-tems using a Weighted Finite-State Transducer (WFST) in which user concept and system action tags are input and output of the transducer, respectively. This paper presents statistical dialog management acquired from an annotated corpus of hu-man-to-human conversation for hotel reservation is applied to WFST-based dialog systems. Interchange Format (IF) consisting of dialog acts and argument which is an interlingua for machine translation was used as user concept and system action tags. A scenario WFST for dialog management was automatically created by converting N-gram models of a tag sequence in the cor-pus. Additionally, a word-to-concept WFST for spoken language understanding (SLU) was obtained form the same corpus. The acquired scenario WFST and SLU WFST were composed together and then optimized. The composed WFST enables us to understand spoken language according to the state of the dialog. We have confirmed the automatically acquired dialog scenario can manage dialog reasonably. We evaluated statistic dialog management in terms of correctness to detect next system actions with and without detecting next user actions. The performance of detecting next systems actions were enhanced by the predic-tion of next user actions and the optimization of the scenario WFST and SLU WFST. |
| Keyword |
(in Japanese) |
(See Japanese page) |
| (in English) |
Weighted Finite State Transducer (WFST) / Spoken Dialog / Statistical Dialog Management / Tag N-gram / / / / |
| Reference Info. |
IEICE Tech. Rep., vol. 108, no. 422, SP2008-130, pp. 25-30, Jan. 2009. |
| Paper # |
SP2008-130 |
| Date of Issue |
2009-01-22 (SP) |
| ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| Download PDF |
SP2008-130 |