Data processing: speech signal processing – linguistics – language – Speech signal processing – Application
Reexamination Certificate
1999-10-25
2003-01-07
{haeck over (S)}mits, T{overscore (a)}livaldis Ivars (Department: 2641)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Application
C704S270000
Reexamination Certificate
active
06505162
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates generally to spoken dialogue systems, and more particularly to a method and apparatus for portable dialogue management in a spoken dialogue system.
BACKGROUND OF THE INVENTION
Rapid progress in computer technology makes it possible for a human user to communicate with a computer using spoken dialogue. The function of a spoken dialogue system is to provide a method that allows a human user to communicate with a computer by means of natural language words. A spoken dialogue system can be used in many different applications such as conversational systems for weather inquiring, railroad information access or city guide, . . . , etc.
FIG. 1
illustrates the general architecture of a typical spoken dialogue system. The typical spoken dialogue system includes five modules, i.e., speech recognition module
101
, language understanding module
102
, dialogue management module
103
, language generation module
104
, and speech synthesis module
105
.
The spoken dialogue system first converts the speech input into sentences using speech recognition module
101
. Secondly, the language understanding module
102
makes use of a vocabulary set, grammar rules and semantic knowledge on the language to represent the semantic meaning of the sentences. Based on the semantic representation, the dialogue management module
103
takes appropriate actions and passes a response semantic frame to the language generation module
104
. The language generation module
104
generates appropriate sentences in the target language from the semantic frame. According to the generated sentences from the language generation module
104
, the speech synthesis module
105
finally synthesizes speech and provides appropriate responses to the user.
The dialogue management module is the kernel for controlling the dialogue flow between a user and a spoken dialogue system. In different applications, the system may be engaged in different dialogues between a user and the system. Therefore, a dialogue management module is the module that has to deal with many domain-dependent characteristics in a spoken dialogue system. In developing a spoken dialogue system, it is very important to have a portable dialogue manager that can be easily ported to a different domain.
There have been several approaches to designing a dialogue manager embodied in dialogue systems. For example, Glass et al. presented “Multilingual Spoken-language Understanding in the MIT Voyager System” in Speech Communication, Vol. 17, No. 1-18, 1995. A spoken dialogue system can be quickly constructed using Glass's approach. However, in the design it is necessary to modify the dialogue manager when the interaction over the dialogue is changed. In other words, the dialogue manager is not portable.
An alternative approach is to develop a dialogue manager based on a finite state network model. The following are some of the arts on the subject:
Kita K. et al., “Automatic Acquisition of Probabilistic Dialogue Models”, Proceedings of ICSLP'96, pp. 196-199, Philadelphia, USA, 1996.
Levin et al., “Using Markov Decision Process for Learning Dialogue Strategies”, Proceedings of ICASSP'98, pp. 201-203, Seattle, USA, 1998.
Colton et al. “A Laboratory Course for Designing and Testing Spoken Dialogue Systems”, Proceedings of ICASSP'96, pp. 1129-1132, Atlanta, Calif., 1996.
In the above arts, a dialogue management technology is used to collect all possible dialogues and messages from other system resources through the whole interactions with the user. Then, a network having plurality of nodes is formed based on the dialogues and messages. The network connection is constructed according to the relationship among the nodes. The complete control over the dialogue is then handled directly by the network.
Such a dialogue manager is portable. However, the dialogue manager can only manage dialogue that is highly structured so that it can define all possible states over the dialogue and the connections among the states. In addition, the complete control over the dialogue is directed along the path defined in the network. Therefore, the dialogue manager is more suited in a system in which the dialogue flow is system-initiated.
Another approach is to develop a dialogue manager based on a form-based model. For example, Goddeau et al. disclosed a method that designs a form for needed information in “A Formed-Based Dialogue Manager for Spoken Language Applications”, Proceedings of ICSLP'96, pp. 701-704, Philadelphia, USA, 1996. The user input is used to fill in corresponding fields of the form. Once a field in the form is filled in, the system responds with a corresponding action. For example, requesting the user for more information to fill in other fields. The user may fill in the form with the information according to the system's prompt or by any order. This type of mixed-initiative dialogue is limited to a goal-specific spoken dialogue system, such as accessing some information from a large database for a user's interest.
An alternative approach is to develop a dialogue manager based on a tree-structured model. The following are some of the arts on the subject:
U.S. Pat. No. 5,694,558 granted to Sparks et al. entitled “Method and System for Interactive Object-Oriented Dialogue Management”.
Camineo-Gil et al., “Data-Driven Discourse Modeling for Semantic Interpretation”, Proceedings of ICASSP'96, pp. 401-404, Atlanta, Calif. 1996.
Masahiro et al., “A Cooperative Man-Machine Dialogue Model for Problem Solving”, Proceedings of ICSLP'94, pp. 883-886, Yokohama, JP, 1994.
In the above art, the dialogue management technology is based on a task-oriented dialogue model. In the approach, the system task includes a set of subtasks. Below each subtask is a set of smaller tasks. The dialogue plan is implemented as a tree structure. Controlling over the dialogue is like searching nodes of the tree, and can not be switched among the subtasks at the same level. Therefore, the dialogue plan tends to be inflexible.
Another approach is to develop a dialogue manager based on a table-driven model. For example, Stephanie Seneff discloses a method that provides a set of variables representing dialogue states in “Discourse and Dialogue Modeling in the GALAXY Systems”, Seminar of Spoken Dialogue System and Discourse Analysis, pp. 12-24, Taipei, Taiwan, ROC, 1997. The expressions performed on the variables trigger the system actions based on default rules such as Boolean operations, arithmetic operations or string comparisons. A table consisting of the well-defined variables, rules and system actions represents the complete flow over the dialogue. The values of the variables may vary during the dialogue execution. This will cause different rules to trigger the corresponding actions. Therefore, it is a dialogue system of a mixed-initiative type.
In the method, the complete flow over the dialogue system is described in a table. It can not provide a structural description for the dialogue system. When the dialogue system tries to achieve multiple subjects, all the variables on the subjects should be specified in the same table. In practice, it may happen that the variables used by some subjects be not used by other subjects. In such a situation, the degree of complexity becomes too much for one single table, and it is thus hard to maintain the table.
A similar approach to the table-driven model to make the dialogue manager portable is based on a task description table (TDT)
106
as shown in FIG.
1
.
Other approach is to develop a dialogue manager based on a dialogue-state or a stack model. For example, Mark-Jan Nederhof et al. presented “Grammatical Analysis in the OVIS Spoken Dialogue System” based on dialogue states in Proceedings Workshop sponsored by the Association for Computational Linguistics, pp. 66-73, Madrid, Spain, 1997. Emiel Krahmer et al. presented “How to Obey the 7 Commands for Spoken Dialogue?” based on stacks in Proceedings Workshop Sponsored by the Association for Computational Linguis
Chiang Tung-Hui
Lin Yi-Chung
Wang Huei-Ming
Industrial Technology Research Institute
{haeck over (S)}mits T{overscore (a)}livaldis Ivars
LandOfFree
Apparatus and method for portable dialogue management using... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Apparatus and method for portable dialogue management using..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Apparatus and method for portable dialogue management using... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3045936