Man-machine dialogue system, controls dialogue between system and user using dialogue specification employing augmented transition networks propagating tokens

• US 7,024,368 B1
• Filed: 02/11/2000
• Issued: 04/04/2006
• Est. Priority Date: 03/12/1999
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Matheson, Caroline
• Primary Examiner(s): Azad, Abul K.
• Application Number:
  US09/936,623
• Time in Patent Office: 2,244 Days
• Field of Search: 704/257, 704/270, 704/275
• US Class Current: 704/275
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

A man-machine dialogue system employing an interactive computer system comprising an input device for receiving input from the user. An output device generating output to the user. There is provided an object system which is an information source or store, or a command and control device. The system has a dialogue manager, which orchestrates the dialogue between the object system and user dependent upon a dialogue specification. The specification employs a set of augmented transition networks (ATNs), through which are propagated one or more tokens, the token comprising a set of fields which together define the current state of belief of the dialogue controller, dependent upon the preceding user-system interchanges and information obtained from the object system.

Claims

1. A man-machine dialogue system comprising an interactive computer system with:

an input device for receiving speech input from a user;

an output device for generating output to a user;

an object system which is an information source or store, or a command and control device, the object of the dialogue being to interface between this system and a user, wherein the object system further comprises knowledge sources to assist recognition;

a dialogue manager, which controls the dialogue between the object system and a user comprising;

a dialogue specification applied to the speech input from the user during recognition comprising;

a set of augmented transition networks (ATNs), having a set of states, wherein each state is one of a question/answer cycle, data processing, a simple junction, and a sub-dialogue state;

one or more tokens propagated through the states, each token comprising a set of fields which together define a current state of belief of a dialogue controller, dependent upon the preceding user-system interchanges and information obtained from the object system.

2. A system according to claim 1, further comprising means for copying and updating the token through each state in an ATN as the dialogue controller passes therethrough.

3. A system according to claim 2, wherein each token is linked to its predecessor, so enabling the dialogue controller to regain a previous state of data maintained at some point during the history of the dialogue.

4. A system according to claim 1, wherein a token or tokens can be received, modified, and transmitted at each state.

5. A system according to claim 1, employing an ATN to specify the course of the dialogue, and wherein each state may represent a junction, a system action, such as a user interaction, or an embedded dialogue represented by a further augmented transition network.

6. A system according to claim 1, where the input device is a speech recognizer.

7. A system according to claim 1, where the input device is constrained by a set of statistical grammars which may be defined using an ATN.

8. A system according to claim 7, wherein each of the states may represent a junction, a terminal state, such as a word or other acoustical phenomenon, or an embedded statistical grammar represented by a further ATN.

9. A system according to claim 7, wherein tokens are propagated for the user-interaction dialogue state to the start of the input network, and through a best matching path of the input network back to the user-interaction dialogue state.

10. A system according to claim 7, wherein a token is propagated for each alternative input hypothesis considered by the input device, and the score assigned to the hypothesis is a modifiable field of the token.

11. A system according to claim 1, wherein the output device is a speech generator.

12. A system according to claim 1, wherein user output is represented by an augmented transition network.

13. A system according to claim 11, where the states represent an output word or other acoustical or linguistic phenomenon.

14. A system according to claim 12, wherein tokens are propagated from the dialogue user-interaction state to the first user output state, and from each user output state to the start of the input network.

15. A system comprising:

an output device for generating output to a user;

an input device for receiving input from a user;

a tool for generating an object system which is an information source or store, or a command and control device, the object of the dialogue being to interface between the system and the user; and

a dialogue manager, which controls the dialogue between the object system and the user based upon;

a dialogue specification, comprising a set of augmented transition networks (ATNs), having a set of states through which are propagated one or more tokens, wherein each state is one of a question/answer cycle, data processing, a simple Junction, and a sub-dialog state, and wherein each token comprises a set of fields which together define a current state of belief of a dialogue controller, dependent upon the preceding user-system interchanges and information obtained from the object system.

16. A method for controlling dialogue between a computer system and a human user comprising the steps of:

outputting at least one speech signal to the user;

inputting a speech signal from the user in response to each of the at least one speech signal output to the user;

propagating at least one token over a set of augmented transition networks comprising a set of states, wherein each state is one of a question/answer cycle, data processing, a simple junction, and a sub-dialogue state, wherein propagating comprises;

applying a dialogue specification to the speech signal input from the user; and

updating the at least one token over the set of augmented transition networks, each token comprising a set of fields which together define a current state of belief of a dialogue controller, dependent upon the preceding user-system interchanges and information obtained from the object system.

17. The method of claim 16, wherein generating output to the user comprises asking the user a question, and wherein receiving input from the user comprises receiving the user'"'"'s answer to the question.

18. The method of claim 16, wherein propagating comprises propagating a token for each alternative hypothesis being considered, and further comprising scoring each alternative hypothesis based on a best matching path through the augmented transition network.

19. A computer-readable medium including instructions readable by a computer which, when implemented perform steps comprising:

generating a speech-based output to the user;

receiving a speech-based input from the user in response to the generated speech-based output;

propagating at least one token over a set of augmented transition networks, wherein propagating comprises;

applying a dialogue specification to the speech signal input from the user; and

updating the at least one token over the set of augmented transition networks comprising a set of states, wherein each state is one of a question/answer cycle, data processing, a simple junction, and a sub-dialogue state, and wherein each token comprises a set of fields which together define a current state of belief of a dialogue controller, dependent upon the preceding user-system interchanges and information obtained from the object system.

20. The computer readable medium of claim 19, wherein each token further comprises a link to its predecessor, from which the history of the dialogue is obtainable.