Device and method for understanding user intent

• US 10,037,758 B2
• Filed: 03/31/2014
• Issued: 07/31/2018
• Est. Priority Date: 03/31/2014
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Corporation
• Original Assignee: Mitsubishi Electric Corporation
• Inventors: Jing, Yi, Fujii, Yoichi, Ishii, Jun
• Primary Examiner(s): Zhu, Richard
• Application Number:
  US15/120,539
• Publication Number:
  US 20170011742A1
• Time in Patent Office: 1,583 Days
• Field of Search: None
• CPC Class Codes: G01C 21/3608 using speech input, e.g. using speech recognition
• CPC Class Codes: G06F 40/268 Morphological analysis
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/088 Word spotting
• CPC Class Codes: G10L 2015/223 Execution procedure of a spoken command

Abstract

A voice recognizer 3 generates plural voice recognition results from one input speech 2. For each of the voice recognition results, an intent understanding processor 7 estimates an intent to thereby output one or more candidates of intent understanding results and scores of them. A weight calculator 11 calculates standby weights using setting information 9 of a control target apparatus. An intent understanding corrector 12 corrects the scores of the candidates of intent understanding result, using the standby weights, to thereby calculate their final scores, and then selects one from among the candidates of intent understanding result, as an intent understanding result 13, on the basis of the final scores.

Claims

1. An intent understanding device, comprising:

a voice recognizer that recognizes one speech spoken in a natural language by a user, to thereby generate plural voice recognition results of highly ranked recognition scores;

a morphological analyzer that converts the respective voice recognition results into morpheme strings;

an intent understanding processor that estimates an intent about the speech by the user on the basis of each of the morpheme strings, to thereby output from each one of the morpheme strings, one or more candidates of intent understanding result and scores indicative of degrees of likelihood of the candidates and generate the candidates of intent understanding result in descending order of likelihoods of the plural voice recognition results;

a weight calculator that calculates respective weights for the candidates of intent understanding result using setting information previously set by a user of a control target apparatus that operates based on the intent understanding result selected by the intent understanding corrector; and

an intent understanding corrector that corrects the scores of the candidates of intent understanding result, using the weights, to thereby calculate their final scores, and then selects the candidate of intent understanding result with the final score that satisfies a preset condition first, as the intent understanding result,

wherein the weight calculator has a table in which constrained conditions and the weights to be used when the respective constrained conditions are satisfied, are defined, and determines whether or not the constrained condition is satisfied, based on the setting information previously set by the user of the control target apparatus, to thereby select each of the weights.

2. The intent understanding device of claim 1, wherein the weight calculator performs weighting so that the candidate of intent understanding result corresponding to an intent expected from a flow of a dialogue user, is likely to be selected by the intent understanding corrector.

3. The intent understanding device of claim 2, including a hierarchical tree in which intents of the user are expressed in a tree structure so that the closer the intent to its root, the more abstract the intent becomes, and the closer the intent to its leaf, the more specific the intent becomes,

wherein the weight calculator performs weighting on the basis of the hierarchical tree so that the candidate of intent understanding result that is placed in a side toward a branch/leaf, of the intent corresponding to the intent understanding result just previously selected, is likely to be selected.

4. The intent understanding device of claim 3, wherein the intent understanding processor estimates the intent of the user using, in addition to the morpheme string generated from the speech currently made, the morpheme string generated from a speech previously made.

5. The intent understanding device of claim 3, further comprising a keyword search processor that searches, from among keywords in a keyword table in which correspondence relationships between intents and the keywords are defined;

a keyword matched to the morpheme string, to thereby acquire the intent corresponding to the searched keyword;

wherein the weight calculator calculates each of the weights using the hierarchical tree and the intent acquired by the keyword search processor.

6. The intent understanding device of claim 1, further comprising an intent confirmation processor that, when the intent understanding corrector excludes the candidate of intent understanding result with a highest likelihood and selects as the intent understanding result, the candidate of intent understanding result other than the excluded one, determines acceptance or non-acceptance of said intent understanding result after making confirmation with the user on whether or not to accept that result.

7. An intent understanding method, comprising:

recognizing one speech spoken in a natural language by a user, to thereby generate plural voice recognition results of highly ranked recognition scores;

converting the respective voice recognition results into morpheme strings;

estimating an intent about the speech by the user on the basis of each of the morpheme strings, to thereby output from each one of the morpheme strings, one or more candidates of intent understanding result and scores indicative of degrees of likelihood of the candidates and generate the candidates of intent understanding result in descending order of likelihoods of the plural voice recognition results;

calculating respective weights for the candidates of intent understanding result using setting information previously set by a user of a control target apparatus that operates based on the intent understanding result selected by the intent understanding corrector; and

correcting the scores of the candidates of intent understanding result, using the weights, to thereby calculate their final scores, and then select the candidate of intent understanding result with the final score that satisfies a preset condition first, as the intent understanding result,

wherein the weight calculator has a table in which constrained conditions and the weights to be used when the respective constrained conditions are satisfied, are defined, and determines whether or not the constrained condition is satisfied, based on the setting information previously set by the user of the control target apparatus, to thereby select each of the weights.