Uncertainty interval content sensing within communications

• US 8,209,175 B2
• Filed: 06/08/2006
• Issued: 06/26/2012
• Est. Priority Date: 06/08/2006
• Status: Expired due to Fees

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Mukerjee, Kunal, Ballesteros, Rafael
• Primary Examiner(s): He, Jialong
• Application Number:
  US11/449,354
• Publication Number:
  US 20070288231A1
• Time in Patent Office: 2,210 Days
• Field of Search: 704/9, 704/242, 704/251, 704/270
• US Class Current: 704/251
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

Repetition of content words in a communication is used to increase the certainty, or, alternatively, reduce the uncertainty, that the content words were actual words from the communication. Reducing the uncertainty of a particular content word of a communication in turn increases the likelihood that the content word is relevant to the communication. Reliable, relevant content words mined from a communication can be used for, e.g., automatic internet searches for documents and/or web sites pertinent to the communication. Reliable, relevant content words mined from a communication can also, or alternatively, be used to automatically generate one or more documents from the communication, e.g., communication summaries, communication outlines, etc.

Claims

1. A method for content sensing a user'"'"'s communication, the method comprising:

generating based on speech recognition a first lattice for a first component of the user'"'"'s communication, the first lattice representing at least one hypothesis for the first component of the user'"'"'s communication, the first lattice comprising a first certainty value for a first content word of at least one hypothesis for the first component of the user'"'"'s communication;

generating a first uncertainty value for the first content word;

generating based on speech recognition a second lattice for a second component of the user'"'"'s communication, wherein the second component is a different segment of the user'"'"'s communication than the first component, the second lattice representing at least one hypothesis for the second component of the user'"'"'s communication, the second lattice comprising a second certainty value for the first content word, wherein the first content word is also a content word of at least one hypothesis for the second component of the user'"'"'s communication;

generating a second uncertainty value for the first content word based on the first uncertainty value for the first content word and the second certainty value for the first content word;

comparing the second uncertainty value for the first content word to an established confidence threshold; and

using the first content word to provide user information if the second uncertainty value of the first content word has a defined relationship to the established confidence threshold.

2. The method for content sensing a user'"'"'s communication of claim 1, wherein the first content word is used to provide the user information if the defined relationship is that the second uncertainty value of the first content word is less than the established confidence threshold.

3. The method for content sensing a user'"'"'s communication of claim 1, wherein the first content word is used to provide the user information if the defined relationship is that the second uncertainty value of the first content word is less than or equal to the established confidence threshold.

4. The method for content sensing a user'"'"'s communication of claim 1, further comprising:

generating the first lattice of one or more nodes for the first component of the user'"'"'s communication wherein the one or more nodes each represent at least one word of at least one hypothesis for the first component of the user'"'"'s communication and at least one of the one or more nodes of the first lattice represents the first content word; and

generating a second lattice of one or more nodes for the second component of the user'"'"'s communication wherein the one or more nodes each represent at least one word of at least one hypothesis for the second component of the user'"'"'s communication and at least one of the one or more nodes of the second lattice represents the first content word.

5. The method for content sensing a user'"'"'s communication of claim 4, further comprising:

identifying non-content words represented by nodes in the first lattice;

generating an uncertainty value for each content word that is represented by a node in the first lattice wherein a content word is a word represented by a node in the first lattice that is not a non-content word; and

storing the uncertainty value generated for each content word that is represented by a node in the first lattice in a table.

6. The method for content sensing a user'"'"'s communication of claim 1, wherein the information provided to the user is one or more web link(s) that are a result of a search on the internet using at least the first content word.

7. The method for content sensing a user'"'"'s communication of claim 1, wherein the information provided the user is at least one document that is generated using at least the first content word.

8. The method for content sensing a user'"'"'s communication of claim 1, wherein the first uncertainty value generated for the first content word is a result of subtracting the first certainty value generated for the first content word from one.

9. The method for content sensing a user'"'"'s communication of claim 8, wherein the second uncertainty value generated for the first content word is a result of multiplying the first uncertainty value generated for the first content word by one minus the second certainty value generated for the first content word.

10. The method for content sensing a user'"'"'s communication of claim 9, further comprising:

storing the first uncertainty value generated for the first content word in a table; and

replacing the first uncertainty value stored for the first content word in the table with the second uncertainty value generated for the first content word.

11. A method for uncertainty interval application in a system, the method comprising:

generating based on speech recognition a first lattice for a first component used by the system, the first lattice representing one or more hypotheses for the first component of the user'"'"'s communication;

generating based on the first lattice, by a processing unit, an uncertainty value for a portion of a first hypothesis of the one or more hypotheses;

storing the uncertainty value generated for the portion of the first hypothesis wherein the stored uncertainty value is a current accumulated uncertainty value;

generating based on speech recognition at least one additional lattice for each of one or more additional components used by the system, wherein the one or more additional components are distinct from the first component, each of the at least one additional lattices representing at least one hypothesis for one of the one or more additional components used by the system;

generating based on the at least one additional lattice an uncertainty value for a portion of each hypothesis generated for the one or more additional components used by the system;

multiplying the uncertainty value for one or more portions of the hypotheses generated for the one or more additional components used by the system by the current accumulated uncertainty value wherein after each multiplication of an uncertainty value by the current accumulated uncertainty value the result of the multiplication is stored as the current accumulated uncertainty value; and

comparing the current accumulated uncertainty value to a confidence threshold.

12. The method for the uncertainty interval application in a system of claim 11, further comprising:

generating a certainty value for the portion of the first hypothesis;

generating the uncertainty value for the portion of the first hypothesis by subtracting the certainty value for the portion of the first hypothesis from one;

generating a certainty value for a portion of each hypothesis generated for the one or more additional components used by the system; and

generating an uncertainty value for a portion of each hypothesis generated for the one or more additional components of the system by subtracting the certainty value for the portion of the hypothesis from one.

13. The method for uncertainty interval application in a system of claim 11, wherein the first component used by the system is a component of a monitored communication and wherein each of the one or more additional components used by the system are each a component of the monitored communication.

14. The method for uncertainty interval application in a system of claim 11, wherein the portion of the first hypothesis is the first hypothesis, the portion of each hypothesis generated for the one or more additional components of the system is the hypothesis generated for the one or more additional components of the system, and a current accumulated uncertainty value is the result of the multiplication of uncertainty values for the first hypothesis and at least one hypothesis generated for the one or more additional components of the system.

15. The method for uncertainty interval application in a system of claim 11, wherein the portion of the first hypothesis is a node of the first lattice for the first hypothesis, the portion of each hypothesis generated for the one or more additional components of the system is each a node of the at least one additional lattice for the at least one hypothesis generated for the one or more additional components of the system, and a current accumulated uncertainty value is the result of the multiplication of uncertainty values for at least two nodes which represent the same concept.

16. The method for the uncertainty interval application in a system of claim 15, wherein the same concept is the same word.

17. A method for increasing the reliability of word recognition in a user'"'"'s communication, the method comprising:

generating by a processing unit an initial uncertainty value for a first word that is hypothesized as being included in a first component of the user'"'"'s communication, the initial uncertainty value generated from a first lattice, the first lattice generated based on speech recognition;

generating a second uncertainty value for the first word when the first word is hypothesized as being included in a second component of the user'"'"'s communication, wherein the second uncertainty value is based at least in part on the initial uncertainty value for the first word and a second lattice, the second lattice generated based on speech recognition;

saving the second uncertainty value for the first word as the first word'"'"'s present uncertainty value;

generating a new uncertainty value for the first word when the first word is hypothesized as being included in a new component of the user'"'"'s communication, wherein a new component of the user'"'"'s communication is any component of the user'"'"'s communication other than the first component or the second component of the user'"'"'s communication, and wherein the new uncertainty value for the first word is based at least in part on the present uncertainty value of the first word;

saving the new uncertainty value for the first word as the first word'"'"'s present uncertainty value; and

designating the first word reliable if a present uncertainty value of the first word has a defined relationship to an established confidence threshold.

18. The method for increasing the reliability of word recognition in a user'"'"'s communication of claim 17, wherein the first word is designated reliable if the defined relationship is that the present uncertainty value of the first word is less than the established confidence threshold.

19. The method for increasing the reliability of word recognition in a user'"'"'s communication of claim 17, wherein the first word is designated reliable if the defined relationship is that the present uncertainty value of the first word is less than or equal to the established confidence threshold.

20. The method for increasing the reliability of word recognition in a user'"'"'s communication of claim 17, further comprising:

using a first certainty value assigned the first word to generate the initial uncertainty value for the first word, wherein the initial uncertainty value for the first word is the result of subtracting the first certainty value assigned the first word from one;

using a second certainty value assigned the first word to generate the second uncertainty value for the first word, wherein the second uncertainty value for the first word is the result of multiplying the initial uncertainty value for the first word by one minus the second certainty value assigned the first word; and

using a new certainty value assigned the first word when the first word is hypothesized as being included in a new component of the user'"'"'s communication to generate the new uncertainty value for the first word, wherein the new uncertainty value for the first word is the result of multiplying a present uncertainty value of the first word by one minus the new certainty value assigned the first word.