User intention based on N-best list of recognition hypotheses for utterances in a dialog

• US 8,140,328 B2
• Filed: 12/01/2008
• Issued: 03/20/2012
• Est. Priority Date: 12/01/2008
• Status: Active Grant

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: Williams, Jason
• Primary Examiner(s): Lerner, Martin
• Application Number:
  US12/325,786
• Publication Number:
  US 20100138215A1
• Time in Patent Office: 1,205 Days
• Field of Search: 704/236, 704/240, 704/252, 704/256, 704/256.1, 704/270, 704/275
• US Class Current: 704/236
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

Disclosed herein are systems, computer-implemented methods, and tangible computer-readable media for using alternate recognition hypotheses to improve whole-dialog understanding accuracy. The method includes receiving an utterance as part of a user dialog, generating an N-best list of recognition hypotheses for the user dialog turn, selecting an underlying user intention based on a belief distribution across the generated N-best list and at least one contextually similar N-best list, and responding to the user based on the selected underlying user intention. Selecting an intention can further be based on confidence scores associated with recognition hypotheses in the generated N-best lists, and also on the probability of a user'"'"'s action given their underlying intention. A belief or cumulative confidence score can be assigned to each inferred user intention.

Claims

1. A method of using alternate recognition hypotheses to improve whole-dialog understanding accuracy, the method comprising:

receiving an utterance as part of a user dialog;

generating an N-best list of recognition hypotheses for the user dialog to yield a generated N-best list;

selecting, via a processor, an underlying user intention based on a belief distribution across the generated N-best list and at least one contextually similar N-best list that is based on an analysis of confidence scores calculated by summing over all possible hidden states and all possible hidden user actions associated with the recognition hypotheses in the generated N-best list and the at least one contextually similar N-best list; and

responding to the user based on the selected underlying user intention.

2. The method of claim 1, the method further comprising assigning an individual confidence score to each recognition hypothesis in the generated N-best list and in the at least one contextually similar N-best list.

3. The method of claim 1, wherein confidence scores are operative over a whole dialog.

4. The method of claim 1, wherein recognition hypotheses in each of the plurality of contextually similar N-best lists with confidence scores below a threshold are not considered when selecting the underlying user intention.

5. The method of claim 1, wherein selecting the underlying user intention is further based on a probability of a user action.

6. The method of claim 5, the method further comprising calculating the probability of the user action by iterating over each user action to form a distribution over the user'"'"'s real intentions at each dialog turn and assimilating the distribution in each of the plurality of contextually similar N-best lists.

7. The method of claim 1, wherein selecting the underlying user intention is based on commonalities in the generated N-best list and at least on contextually similar N-best list.

8. A system for using alternate recognition hypotheses to improve whole-dialog understanding accuracy, the system comprising:

a processor;

a first module configured to control the processor to receive an utterance as part of a user dialog turn;

a second module configured to control the processor to generate an N-best list of recognition hypotheses for the user dialog turn and yield a generated N-best list;

a third module configured to control the processor to select an underlying user intention based on a belief distribution across the generated N-best list and at least one contextually similar N-best list that is based on an analysis of confidence scores calculated by summing over all possible hidden states and all possible hidden user actions associated with the recognition hypotheses in the generated N-best list and the at least one contextually similar N-best list; and

a fourth module configured to control the processor to respond to the user based on the selected underlying user intention.

9. The system of claim 8, the system further comprising a fifth module configured to control the processor to assign an individual confidence score to each recognition hypothesis in the generated N-best list and in the at least one contextually similar N-best list.

10. The system of claim 8, wherein confidence scores are operative over a whole dialog.

11. The system of claim 8, wherein recognition hypotheses in each of the plurality of contextually similar N-best lists with confidence scores below a threshold are not considered when selecting the underlying user intention.

12. A non-transitory computer-readable medium storing a computer program having instructions for using alternate recognition hypotheses to improve whole-dialog understanding accuracy, the instructions comprising:

receiving an utterance as part of a user dialog turn;

generating an N-best list of recognition hypotheses for the user dialog turn to yield a generated N-best list;

selecting an underlying user intention based on a belief distribution across the generated N-best list and at least one contextually similar N-best list that is based on an analysis of confidence scores calculated by summing over all possible hidden states and all possible hidden user actions associated with the recognition hypotheses in the generated N-best list and the at least one contextually similar N-best list; and

responding to the user based on the selected underlying user intention.

13. The non-transitory computer-readable medium of claim 12, the instructions further comprising assigning an individual confidence score to each recognition hypothesis in the generated N-best list and in the at least one contextually similar N-best list.

14. The non-transitory computer-readable medium of claim 12, wherein confidence scores are operative over a whole dialog.