Management of conversations

• US 20060080107A1
• Filed: 09/02/2005
• Published: 04/13/2006
• Est. Priority Date: 02/11/2003
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Unveil Technologies, Inc.
• Inventors: Zieman, Yuri, Hill, Jeffrey
• Granted Patent: US 7,487,095 B2
• Time in Patent Office: Days
• US Class Current: 704/275
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

In one aspect, an arbitrary natural language communication is received from a user. A concept recognition process is applied to automatically derive a representation of concepts embodied in the communication. The concept representation is used to provide to a human agent information useful in responding to the natural language communication.

Claims

1. A method comprising

receiving an arbitrary natural language communication from a user,

applying a concept recognition process to automatically derive a representation of concepts embodied in the communication, and

using the concept representation to provide to a human agent information useful in responding to the natural language communication.

2. The method of claim 1 in which the arbitrary natural language communication is expressed in speech.

3. The method of claim 2 in which the communication is expressed using a telephone or other voice instrument.

4. The method of claim 1 in which the communication is a method stored in a voice mailbox.

5. The method of claim 1 in which the arbitrary natural language communication is expressed in text.

6. The method of claim 5 in which the text is expressed electronically.

7. The method of claim 6 in which the text is expressed in an email.

8. The method of claim 7 in which the text is expressed through instant messaging.

9. The method of claim 5 in which the text is expressed in a manner associated with a web page.

10. The method of claim 1 in which the concept recognition process is universally applicable to any communication in a natural language.

11. The method of claim 1 in which the concept representation is expressed in a mark-up language.

12. The method of claim 1 in which the information provided to the human agent includes an audible playback of a recorded version of the natural language communication.

13. The method of claim 12 in which the playback is compressed in time relative to the communication.

14. The method of claim 1 in which the information provided to the human agent includes a display of a text corresponding to the communication.

15. The method of claim 1 in which the information provided to the human agent includes information about at least one prior communication or response that preceded the natural language communication.

16. The method of claim 15 in which the concept recognition process is used to determine how much information about prior communications to provide to the human agent.

17. The method of claim 1 in which the communication is part of a dialog between the user and a response system, the dialog including communications from the user and responses to the user, and the information provided to the human agent includes information about historical portions of the dialog.

18. The method of claim 17 in which a first mode of expression of the communications from the user is different from a second mode of expression of the responses to the user.

19. The method of claim 18 in which the first mode and second mode of expression comprise at least one of text or speech.

20. The method of claim 1 in which the information provided to the human agent includes information about possible responses to the user'"'"'s communication.

21. The method of claim 20 in which a first mode of expression of the communications from the user is different from a second mode of expression of the responses to the user.

22. The method of claim 20 in which the first mode and second mode of expression comprise at least one of text or speech.

23. The method of claim 20 in which the information about possible responses includes a text of a possible response.

24. The method of claim 20 in which the information about possible responses includes an indication of a level of confidence in the appropriateness of the response.

25. The method of claim 1 in which the communication comprises a question and the response comprises an answer to the question.

26. The method of claim 1 in which the communication comprises a question and the response comprises a request for additional information.

27. The method of claim 1 also including enabling the human agent to determine how the information useful in responding to the communication is selected.

28. The method of claim 27 in which the enabling of the human agent includes permitting the agent to use the communication from the user to control how the responsive information is selected.

29. The method of claim 27 in which the enabling of the human agent includes permitting the agent to enter a substitute communication to control how the responsive information is selected.

30. The method of claim 29 in which the substitute communication is a restatement by the human agent of the communication from the user.

31. The method of claim 1 in which the useful responding information is generated by applying the concept representation to a body of information representing other communications and their relationships to concepts.

32. The method of claim 31 in which applying the concept representation includes a matching process to determine a cluster of similar communications to which the user'"'"'s communication likely belongs.

33. The method of claim 1 in which a state is occupied prior to receipt of the communication, and also including selecting a transition to a next state based on the concept representation and on a set of possible transitions.

34. The method of claim 33 in which the transition includes an action to be taken in response to the communication.

35. The method of claim 34 in which the action to be taken comprises a reply communication.

36. The method of claim 34 in which the set of possible transitions is derived from examples of state-transition-state or stimulus-response sequences.

37. The method of claim 36 in which the examples include pre-run-time examples.

38. The method of claim 37 in which the pre-run-time examples comprise voice or text.

39. The method of claim 36 in which the examples occur at runtime.

40. The method of claim 1 also including providing a response to the communication from the user.

41. The method of claim 1 in which the response is selected by the human agent and delivered to the user automatically without the user knowing that it was a human agent who selected the response.

42. The method of claim 41 in which the response is generated by the human agent.

43. The method of claim 42 in which the response is spoken or typed by the human agent.

44. The method of claim 1 in which the response is selected without involvement of a human agent.

45. The method of claim 1 also including providing a graphical user interface for a workstation of the human agent, the information useful in responding being presented in the interface, the interface being presented as part of a user interface of a third party'"'"'s response system software.

46. The method of claim 45 in which the user interface provides conceptual context for a communication from a user.

47. The method of claim 1 also including providing a response to the communication.

48. The method of claim 47 in which the response is provided in real time relative to the communication.

49. The method of claim 47 in which the response is provided at a later time relative to the communication.

50. The method of claim 49 in which the communication is provided in speech and the response is provided in text.

51. The method of claim 1 also including selecting a human agent to handle a response to the communication.

52. The method of claim 51 in which the human agent is automatically selected by a work distribution process.

53. The method of claim 52 in which the work distribution process uses information deduced from the concept representation in automatically selecting the human agent.

54. A method comprising

receiving an arbitrary natural language communication from a user,

automatically deriving a representation of concepts embodied in the communication, and

using the concept representation, automatically providing a response to the communication in a different mode of expression than the mode of expression used for the communication.

55. The method of claim 54 in which the response is provided in other than real time relative to the communication.

56. The method of claim 54 in which the communication is provided in speech and the response is provided in text.

57. A method comprising

initiating a dialog with a user by sending a first natural language communication to the user,

in response to the first natural language communication to the user, receiving a second natural language communication from the user,

applying a concept recognition process to automatically derive a representation of concepts embodied in the second communication, and

using the concept representation to provide to a human agent information useful in responding to the second communication.

58. A method comprising

receiving a set of recordings or transcripts of dialogs between users and human agents,

recognizing the speech in the recordings,

separating each of the dialogs into communications each of which is made by either a user or a human agent,

applying a concept recognition process to derive a representation of concepts embodied in each of the communications, and

automatically creating a body of state-transition-state or stimulus-response information from the concept representations that enables automated determination of appropriate responses to natural language communications received from users.

59. A method comprising

receiving example dialogs each comprising a sequence of natural language communications between two parties,

applying a concept recognition process to automatically derive a representation of concepts embodied in each of the communications, and

using the sequences of communications to form a body of state-transition-state or stimulus-response information that enables a determination of an appropriate transition for any arbitrary communication that is received when in a particular one of the states.

60. The method of claim 59 in which the example dialogs comprise sound files or transcriptions of typed text.

61. The method of claim 60 also including using the concept representations to form clusters of communications that are related in the concepts that are embodied in them.

62. The method of claim 60 in which the example dialogs comprise historical dialogs.

63. The method of claim 60 in which the dialogs relate to contact center operation.

64. The method of claim 60 in which the dialogs comprise requests and responses to the requests.

65. The method of claim 60 in which the dialogs comprise real-time dialogs.

66. The method of claim 65 in which the dialogs comprise a string of voice messages.

67. The method of claim 60 in which the representations of concepts are expressed in a mark-up language.

68. The method of claim 61 in which the communications in the cluster comprise communications that represent different ways of expressing similar sets of concepts.

69. A method comprising

receiving an arbitrary natural language communication from a user,

applying business rules to a concept representation of the communication to determine whether or not to refer the communication to a human agent for response, and

if the business rules indicate that it is not necessary to refer the communication to the human agent, determining whether a confidence in an automatically generated response is sufficiently high to provide the response without referring the communication to the human agent.

70. A method comprising

receiving an arbitrary natural language communication from a user,

automatically selecting a level of response from among a set of different levels that differ in respect to the degree of involvement by the human agent in providing the response.

71. The method of claim 70 in which the selecting is based in part on an estimate of how long it would take the human agent to respond if the communication is referred to the human agent for response.

72. The method of claim 70 in which the level is selected based on a level of confidence in the appropriateness of an automatically generated response.

73. The method of claim 70 in which the level is selected based on business rules.

74. The method of claim 70 in which the levels include a level in which the response is provided automatically.

75. The method of claim 70 in which the levels include a level in which the response is generated by the human agent.

76. The method of claim 75 in which the response is entered as text or spoken.

77. The method of claim 75 in which the levels include a level in which the response is selected by the human agent.

78. The method of claim 77 in which the selected response is delivered automatically to the user.

79. The method of claim 78 in which the selected response is delivered to the user without the user knowing that the response had been selected by a human agent.

80. A method comprising

enabling a user to access a contact service facility,

receiving communications from the user at the contact service facility,

providing responses to the user'"'"'s communications, and

enhancing the user'"'"'s confidence in the contact service facility by causing at least one of the responses to be selected by a human agent based on the results of an automated concept matching process applied to the communications, the user being unaware that the human agent selected the response.

81. The method of claim 80 in which a first mode of expression of the communications from the user is different from a second mode of expression of the responses to the user.

82. The method of claim 81 in which the first mode and second mode of expression comprise at least one of text or speech.

83. A method comprising

maintaining a body of state-transition-state or stimulus-response information that represents possible sequences of natural language communications between a user and a response system, the information being generated automatically from historical sequences of communications, and

using selected ones of the sequences of communications to manage human agents who provide responses to user communications.

84. The method of claim 83 in which the selected ones are used to train the human agents.

85. The method of claim 83 in which the selected ones are used to evaluate the human agents.

86. The method of claim 83 in which the sequences are used to manage the human agents by providing the agents with communications that are part of the sequences and evaluating responses of the human agents against known appropriate responses.

87. A method comprising

maintaining a body of state-transition-state or stimulus-response information that represents possible sequences of natural language communications between a user and a response system, the information being generated automatically from historical sequences of communications, and

using the body of state-transition-state or stimulus-response information in connection with the operation of a user response system.

88. The method of claim 87 in which the body of information is used in connection with testing of the response system.

89. The method of claim 87 in which the body of information is used in connection with software processes used in the response system.

90. A method comprising

maintaining a body of state-transition-state or stimulus-response information that enables automated determination of appropriate responses to natural language communications received from users,

receiving other natural language communications from users for which appropriate responses cannot be determined,

tracking actions taken by a human agent in connection with responding to the other natural language communications, and

automatically inferring from the other natural language communications and the selected responses, information for inclusion in the body of state-transition-state or stimulus-response information.

91. The method of claim 90 in which the actions taken by the human agent include responses selected by the human agent for use in responding to the other natural language communications.

92. The method of claim 91 also including enabling an administrator to review the inferred information prior to including it in the body of state-transition-state or stimulus-response information.

93. The method of claim 90 in which the actions taken by the human agent include keystrokes or mouse actions.

94. The method of claim 90 also including providing the human agent with possible responses to the natural language communications, and in which the tracking of actions includes tracking which of the possible responses the human agent chooses and inferring that the chosen response is a correct response to one of the communications.

95. The method of claim 90 also including providing the human agent with possible responses to the natural language communications, and, if the human agent responds to the communication without choosing one of the possible responses, inferring that the possible responses are incorrect.

96. The method of claim 95 also including enabling the human user to indicate that one of the possible answers was correct, even though the human user is respond to the communication without making a choice among the possible responses.

97. A method comprising

maintaining a body of state-transition-state or stimulus-response information that enables automated determination of appropriate responses to natural language communications received from users, the state-transition-state or stimulus-response information being associated with a contact center of an enterprise,

updating the body of information based on communications received from users and responses provided by human agents of the contact center, and

analyzing the body of information to infer knowledge about the operation of the enterprise.

98. A method comprising

maintaining a body of state-transition-state or stimulus-response information that enables automated determination of appropriate responses to natural language communications received from users, the state-transition-state or stimulus-response information being based on concept representations derived from example natural language communications,

the example natural language communications being predominantly in one language, and

using the state-transition-state or stimulus-response information to provide appropriate responses to natural language communications received from users in a second language different from the one language.

99. A method comprising

displaying to a human agent a user interface containing concept representation-based information useful in responding to natural language communications from users,

the information including automatically generated possible natural language responses and indications of relative confidence levels associated with the responses.

100. The method of claim 99 also including

enabling the human agent to select one of the possible responses.

101. The method of claim 99 also including

enabling the human agent to enter a substitute of the user'"'"'s communication, and

generating the possible natural language responses from the substitute communication.

102. The method of claim 99 also including providing controls in the interface that enable the human agent to choose a level of response with respect to the degree of involvement of the human agent.

103. The method of claim 102 in which the level of response includes direct conversation with the user.

104. The method of claim 102 in which the level of response includes providing the response automatically.

105. A method comprising

maintaining a body of state-transition-state or stimulus-response information that enables automated determination of appropriate responses to natural language communications received from users, the state-transition-state or stimulus-response information being based on concept representations derived from example natural language communications, each of the states having possibly multiple transitions leading to a later state,

when in a predetermined one of the states, using information about the multiple transitions to improve the accuracy of recognition of a speech recognizer that is processing a spoken communication from a user.

106. The method of claim 105 also including

using the information about multiple transitions to improve the accuracy of discriminate matching of the concept representation of the spoken communication with clusters of concept representations in the body of information.

107. A method comprising

enabling two-way natural language communication between each pair of a user, a human agent, and an automated response system, and

facilitating the communication by representing the natural language communication as concepts and maintaining a body of state-transition-state or stimulus-response information about sequences of communications between at least two of the user, the human agent, and the response system.

108. A method comprising

receiving natural language communications from users,

automatically considering possible responses to the communications and confidence levels with respect to the responses,

providing automated responses to a portion of the users based on the confidence levels, and

refraining from providing automated responses to another portion of the users.

109. A method comprising

receiving natural language communications from users,

automatically recognizing concepts contained in the communications, and

distributing the communications to human agents for responding to the users, the distribution being based on the concepts recognized in the communications.

110. A medium bearing a body of information capable of configuring a machine to support an automated communication system, the body of information comprising

state-transition-state or stimulus-response information that represents possible sequences of natural language communications occurring back and forth between a user and a response system.

111. The medium of claim 110 in which the body of information also includes cluster information identifying clusters of variations of communications that express similar concepts, each of the transitions of the state-transition-state or stimulus-response information being associated with one of the clusters.

112. Apparatus comprising

a user interface for a human agent at a contact service facility, the user interface including

a window containing information provided by a contact service process, the

information including information about a user of the facility, and

window elements embedded in the window provided by the contact service process, the elements including a list of possible natural language responses based on concept representations for an active communication of a user, and indications of relative confidence that the respective responses are appropriate for the communication of the user.

113. The apparatus of claim 112 in which the window elements include a place for a human agent to view text corresponding to the communication of the user, and a place for the human-agent to enter a substitute text for the communication of the user.