Learning apparatus and method

• US 10,446,146 B2
• Filed: 02/08/2017
• Issued: 10/15/2019
• Est. Priority Date: 03/20/2015
• Status: Active Grant

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Fujii, Hiroko
• Primary Examiner(s): Neway, Samuel G
• Application Number:
  US15/427,347
• Publication Number:
  US 20170148441A1
• Time in Patent Office: 979 Days
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates
• 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
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: G10L 2015/0635 updating or merging of old and new templates; Mean values; Weighting

Abstract

According to one embodiment, a learning apparatus includes a first storage and a processing circuitry. The first storage stores a plurality of similarity information items, each including at least a second text, intention candidates and certainty degrees representing a certainty with which the intention candidates are regarded as being identical to intentions of the second text. The processing circuitry detects corresponding similarity information item including the second text identical to a first text from the plurality of similarity information items, corrects a utterance intention to an intention candidate included in the corresponding similarity information item where a certainty degree included in the corresponding similarity information item is not less than a threshold.

Claims

1. A learning apparatus that uses utterance intention of a user, the utterance intention being inferred from a first text representing results of speech recognition of the user'"'"'s utterances, the learning apparatus comprising:

a first storage which stores a plurality of similarity information items, each including at least (i) a second text representing results of speech recognition of a set of similar utterances included in a dialog history, (ii) intention candidates inferred based on utterances included in the similar utterances and considered to result in successful dialogs, and (iii) certainty degrees representing a certainty with which the intention candidates are regarded as being identical to intentions of the second text; and

a processing circuitry configured to;

detect a similarity information item as a corresponding similarity information item from the plurality of similarity information items, the corresponding similarity information item including a second text identical to the first text; and

correct the utterance intention to an intention candidate included in the corresponding similarity information item when a certainty degree included in the corresponding similarity information item is not less than a threshold.

2. The apparatus according to claim 1, wherein the processing circuitry is further configured to register the first text and the corrected utterance intention as utterance intention information.

3. The apparatus according to claim 1, further comprising:

a second storage which stores the dialog history, wherein the processing circuitry is further configured to;

extract, from the second storage, a second text regarding utterances of a single user, an intention candidate inferred based on utterance of the single user, a third text representing results of speech recognition of the utterances determined to result in the successful dialogs; and

update the similarity information item including the second text identical to the third text stored in the first storage, based on the extracted second text, the extracted intention candidate and the extracted third text.

4. The apparatus according to claim 3, wherein the processing circuitry extracts a first utterance and a second utterance as the similar utterances when the first utterance and the second utterance are similar, and generates a similarity utterance group including at least one similar utterance, the second utterance being given before or after the first utterance.

5. The apparatus according to claim 4, wherein the processing circuitry selects a latest utterance from the similarity utterance group as an utterance determined to result in the successful dialogs, and determines an utterance intention regarding the latest utterance as the intention candidate inferred based on utterance of the single user.

6. The apparatus according to claim 1, wherein the processing circuitry is further configured to cause the user to confirm a pair of the first text and the corrected utterance intention.

7. The apparatus according to claim 3, wherein the processing circuitry registers the first text and the corrected utterance intention in the first storage when the user approves the corrected utterance intention as an intention of the first text.

8. The apparatus according to claim 3, wherein the similarity information item including the second text identical to the third text further includes date information representing when the similarity information is generated, and

wherein the processing circuitry deletes from the first storage similarity information after elapse of a first time duration from a date indicated by the date information.

9. The apparatus according to claim 1, wherein the certainty degrees represent a frequency of appearance representing how many times a pair of (i) an intention candidate inferred from an utterance determined to result in a successful dialog and (ii) a second text of utterances other than the utterance determined to result in the successful dialog appear.

10. The apparatus according to claim 1, wherein the processing circuitry is further configured to send, to an administer, a message indicating that an utterance regarding the first text is an ambiguous utterance, when there are a plurality of corresponding similarity information items and certainty degrees of the corresponding similarity information items are not less than the threshold.

11. The apparatus according to claim 1, wherein the processing circuitry is further configured to send, to an administer, a message indicating that an utterance regarding the first text is an ambiguous utterance, when the utterance intention differs from the intention candidate and a certainty degree of a combination of the utterance intention and the intention candidate is not less than the threshold.

12. The apparatus according to claim 2, wherein the processing circuitry is further configured to update intention analysis information for inferring the utterance intention, using the utterance intention information.

13. The apparatus according to claim 12, wherein the processing circuitry deletes the corresponding similarity information item from the first storage when a correction is performed and the correction is reflected in the intention analysis information.

14. A learning method that uses utterance intention of a user, the utterance intention being inferred from a first text representing results of speech recognition of the user'"'"'s utterances, the learning method comprising:

storing, in a first storage, a plurality of similarity information items, each including at least (i) a second text representing results of speech recognition of a set of similar utterances included in a dialog history, (ii) intention candidates inferred based on utterances included in the similar utterances and considered to result in successful dialogs, and (iii) certainty degrees representing a certainty with which the intention candidates are regarded as being identical to intentions of the second text;

detecting a similarity information item as a corresponding similarity information item from the plurality of similarity information items, the corresponding similarity information item including a second text identical to the first text; and

correcting the utterance intention to an intention candidate included in the corresponding similarity information item when a certainty degree included in the corresponding similarity information item is not less than a threshold.

15. A non-transitory computer readable medium including computer executable instructions, wherein the instructions, when executed by a processor, cause the processor to perform a method that uses utterance intention of a user, the utterance intention being inferred from a first text representing results of speech recognition of the user'"'"'s utterances, the method comprising:

storing, in a first storage, a plurality of similarity information items, each including at least (i) a second text representing results of speech recognition of a set of similar utterances included in a dialog history, (ii) intention candidates inferred based on utterances included in the similar utterances and considered to result in successful dialogs, and (iii) certainty degrees representing a certainty with which the intention candidates are regarded as being identical to intentions of the second text;

detecting a similarity information item as a corresponding similarity information item from the plurality of similarity information items, the corresponding similarity information item including a second text identical to the first text; and

correcting the utterance intention to an intention candidate included in the corresponding similarity information item when a certainty degree included in the corresponding similarity information item is not less than a threshold.