RELATIVE DELTA COMPUTATIONS FOR DETERMINING THE MEANING OF LANGUAGE INPUTS

• US 20080147390A1
• Filed: 02/22/2008
• Published: 06/19/2008
• Est. Priority Date: 10/01/2003
• Status: Active Grant

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Balchandran, Rajesh, Boyer, Linda M.
• Granted Patent: US 7,562,016 B2
• Time in Patent Office: Days
• US Class Current: 704/231
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

A method for processing language input can include the step of determining at least two possible meanings for a language input. For each possible meaning, a probability that the possible meaning is a correct interpretation of the language input can be determined. At least one relative data computation can be computed based at least in part upon the probabilities. At least one irregularity within the language input can be detected based upon the relative delta computation. The irregularity can include mumble, ambiguous input, and/or compound input. At least one programmatic action can be performed responsive to the detection of the irregularity.

Claims

1. A system for processing language input comprising:

a natural language processor configured to convert a language input into a plurality of possible meanings and to assign a confidence value to each of said possible meanings; and

an anomaly detector configured to detect the occurrence of at least one irregularity within said language input using at least one relative delta computation, wherein each value of a relative delta computation uniquely corresponds to a type of irregularity within said language input such that values of relative delta computations vary depending on the corresponding type of irregularity, and wherein the anomaly detector is further configured to determine the corresponding type of the at least one irregularity based upon said at least one relative delta computation.

2. The system of claim 1, further comprising:

a training engine configured to generate thresholds used by said anomaly detector to detect said irregularities.

3. The system of claim 1, wherein said anomaly detector further comprises a relative delta score calculator configured to calculate relative delta computations.

4. The system of claim 1, wherein said anomaly detector further comprises a relative delta score interpreter configured to compare relative delta computations to threshold values to detect said irregularities.

5. The system of claim 1, wherein said anomaly detector further comprises a learning module configured to adjust threshold values base upon operational behavior.

10. The machine-readable storage of claim 4, wherein said at least one irregularity detected by said method comprises two or more irregularities, said method further comprises the step of:

differentiating between different ones of said irregularities such that different programmatic actions are performed responsive to detecting different irregularities.

6. A machine-readable storage having stored thereon, a computer program having a plurality of code sections, said code sections executable by a machine for causing the machine to perform the steps of:

determining at least two possible meanings for a language input;

for each possible meaning, determining a probability that said possible meaning is a correct interpretation of said language input;

computing at least one relative delta computation based at least in part upon said probabilities, wherein each value of a relative delta computation uniquely corresponds to a type of irregularity within said language input such that values of relative delta computations vary depending on the corresponding type of irregularity;

detecting at least one irregularity within said language input and determining the corresponding type of the at least one irregularity based upon said relative delta computation; and

performing at least one programmatic action responsive to detecting said irregularity.

7. The machine-readable storage of claim 6, wherein said irregularity comprises mumble.

8. The machine-readable storage of claim 6, wherein said irregularity comprises ambiguous input.

9. The machine-readable storage of claim 6, wherein said irregularity comprises compound input.

11. The machine-readable storage of claim 6, wherein said programmatic action comprises the step of prompting a user for further input to resolve said irregularity.

12. The machine-readable storage of claim 6, said computing step further comprising the steps of:

computing a numerator by subtracting a probability of a second one of said meanings from the probability of a first one of said meanings;

setting a denominator to the probability of said first meaning; and

dividing said numerator by said denominator.

13. The machine-readable storage of claim 6, further comprising the step of receiving a top-n list of possible meanings for said language input, wherein said determining of possible meanings further comprises selecting at least two possible meanings contained within said top-n list.

14. The machine-readable storage of claim 13, wherein said possible meanings of said top-n list are each associated with a probability that said possible meaning is a correct interpretation of said language input, said selecting step further comprises the step of selecting said two possible meanings having the highest probabilities.

15. The machine-readable storage of claim 6, wherein said at least two possible meanings comprise at least three possible meanings, said method further comprises ordering said possible meanings according to said probabilities, wherein said computing step further comprises the steps of:

computing a first relative delta computation based upon two sequentially ordered ones of said possible meanings; and

computing a second relative delta computation based upon two different sequentially ordered ones of said ordered meanings, wherein said detecting of said irregularity is based upon said first relative delta computation and said second relative delta computation.

16. The machine-readable storage of claim 6, wherein said probability that each possible meaning is a correct interpretation is a confidence value;

said method further comprising the steps of;

receiving a plurality of language inputs;

for each language input, determining at least two possible meanings and associated confidence values;

plotting at least a portion of said confidence values on a graph; and

determining at least one threshold from said graph, wherein said relative delta computation is compared against said threshold when detecting said irregularity.

17. The machine-readable storage of claim 16, further comprising the step of:

presenting said graph within a graphical user interface; and

automatically adjusting said threshold using controls of said graphical user interface.

18. A system for processing language input comprising the steps of:

means for determining at least two possible meanings for a language input;

means for determining a probability that said possible meaning is a correct interpretation of said language input for each possible meaning;

means for computing at least one relative delta computation based at least in part upon said probabilities, wherein each value of a relative delta computation uniquely corresponds to a type of irregularity within said language input such that values of relative delta computations vary depending on the corresponding type of irregularity;

means for detecting at least one irregularity within said language input and determining the corresponding type of the at least one irregularity based upon said relative delta computation; and

means for performing at least one programmatic action responsive to detecting said irregularity.