Numeric weighting of error recovery prompts for transfer to a human agent from an automated speech response system

• US 8,566,104 B2
• Filed: 11/01/2011
• Issued: 10/22/2013
• Est. Priority Date: 08/16/2005
• Status: Expired due to Fees

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Michelini, Vanessa V., Polkosky, Melanie D.
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): ADESANYA, OLUJIMI A
• Application Number:
  US13/287,004
• Publication Number:
  US 20120046951A1
• Time in Patent Office: 721 Days
• Field of Search: 704/270, 704/270.1, 704/275
• US Class Current: 704/270.1
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

A method for a speech response system to automatically transfer users to human agents. The method can establish an interactive dialog session between a user and an automated speech response system. An error score can be established when the interactive dialog session is initiated. During the interactive dialog session, responses to dialog prompts can be received. Error weights can be assigned to receive responses determined to be non-valid responses. Different non-valid responses can be assigned different error weights. For each non-valid response, the assigned error weight can be added to the error score. When a value of the error score exceeds a previously established error threshold, a user can be automatically transferred from the automated speech response system to a human agent.

Claims

1. A computer-implemented method used in conjunction with an automated speech response system, the method comprising:

during an interactive dialog session between a user and an automated speech response system, determining a plurality of user responses to dialog prompts;

detecting non-valid user responses from the plurality of user responses;

assigning error weights to the non-valid user responses, wherein different non-valid user responses are assigned different error weights;

for each one of the non-valid user responses, adjusting an error score for the interactive dialog session based upon the error weight assigned to the one of the non-valid user responses; and

when a value of said error score exceeds a previously established error threshold, automatically transferring said user from the automated speech response system to a human agent.

2. The method of claim 1, further comprising:

when the value of the error score does not exceed the previously established error threshold, continuing the interactive dialog session between the user and the automated speech response system.

3. The method of claim 1, further comprising:

determining a severity of each of said non-valid user responses, wherein the assigned error weights are based at least in part upon said determined severity.

4. The method of claim 1, the assigning further comprising:

for each non-valid user response, categorizing said non-valid user response into one of a plurality of categories, wherein the error weight associated with each non-valid user response is based at least in part upon a category of the plurality of categories assigned to the non-valid user response.

5. The method of claim 4, wherein the plurality of categories comprises a misrecognition category and a time out category.

6. The method of claim 4, wherein said plurality of categories comprises at least one category selected from the group consisting of an acoustic error category, a lexical error category, a syntactic error category, and a conceptual error category.

7. The method of claim 1, further comprising:

for each one of the non-valid user responses for which the value of the error score does not exceed the previously established error threshold, subsequent to the adjusting, presenting an error prompt associated with the one of the non-valid user responses, wherein different non-valid user responses are associated with different error prompts.

8. The method of claim 1, wherein said dialog prompts are audible prompts including automatically generated speech, and wherein said user responses to the dialog prompts comprise speech utterances that are speech-to-text processed by the automated speech response system.

9. The method of claim 1, wherein the automated speech response system is an interface for a plurality of applications configured to be utilized during the interactive dialog session, wherein a value of the error score is maintained as the automated speech response system utilizes different ones of the plurality of applications.

10. The method of claim 1, further comprising:

during the interactive dialog session, detecting a score adjustment event; and

responsive to detecting the score adjustment event, decreasing the error score.

11. The method of claim 1, wherein the method is performed by a service agent or a computing device manipulated by the service agent, the method being performed in response to a service request.

12. The method of claim 1, wherein the method is performed by a machine in accordance with a plurality of machine-readable instructions that are executed by the machine, and wherein a service agent responding to a service request specifically configures said machine to perform the method.

13. A computer-implemented method for automatically transferring a user from a speech response system to a human agent, the method comprising:

during a dialog session, presenting a prompt to a user;

receiving and processing a response from the user to the prompt; and

when the response is a non-valid user response,

assigning an error weight to the non-valid user response, wherein different non-valid user responses are assigned different error weights;

adding the error weight to an error score for the dialog session;

comparing the error score to an error threshold; and

automatically transferring the user to a human agent when the error score exceeds the error threshold.

14. The method of claim 13, further comprising:

when the response is a valid user response, performing an appropriate programmatic action for the response; and

repeating the presenting and receiving until a method-terminating event occurs, wherein the method-terminating event comprises detecting a user-initiated session termination event or detecting a system-initiated transfer to the human agent.

15. The method of claim 13, further comprising:

when the response is a non-valid user response, categorizing the response into one of several categories of error responses, wherein the assigning step assigns the error weight to the non-valid user response based at least in part upon the category of the non-valid user response.

16. The method of claim 13, further comprising:

when the response is a non-valid user response, determining an error severity level of the non-valid user response, wherein the assigning step assigns the error weight to the non-valid user response based at least in part upon the determined error severity level of the non-valid user response.

17. The method of claim 13, wherein the method is performed by a service agent or a computing device manipulated by the service agent, the method being performed in response to a service request.

18. A speech response system comprising at least one processor programmed to:

detect user error responses during an interactive session;

for each detected user error response, add to an error score for the interactive session an error weight associated with the user error response, wherein different error weights are assigned to different user error responses; and

automatically transfer a user to a human agent when the error score for the interactive session exceeds an error threshold.