Title:
RECOGNITION ERROR CORRECTION DEVICE AND CORRECTION MODEL
Document Type and Number:
WIPO Patent Application WO/2020/225999
Kind Code:
A1
Abstract:
The present invention addresses the problem of building an environment in which a process for correcting a recognition error in a recognition result for voice recognition or character recognition is presented. A recognition error correction device 1 is provided with: a pair data acquisition unit 21 that acquires pair data in which a sentence, of recognition results for voice recognition or character recognition, and a label string, configured from a process label which is a label indicating a process for correcting a recognition error regarding each word constituting the sentence, are associated; and a correction model generation unit 22 that implements machine learning by using the pair data acquired by the pair data acquisition unit 21 to thereby generate a correction model, which is a learned model for correcting a recognition error of the recognition results. The recognition error correction device 1 is further provided with a pair data creation unit 20 that creates pair data on the basis of a comparison between a sentence of a recognition result and correct answer data for the recognition result, and the pair data acquisition unit 21 may acquire pair data created by the pair data creation unit 20.
More Like This:
Inventors:
IKEDA TAISHI (JP)
Application Number:
PCT/JP2020/014584
Publication Date:
November 12, 2020
Filing Date:
March 30, 2020
Export Citation:
Assignee:
NTT DOCOMO INC (JP)
International Classes:
G06K9/03; G06N20/00; G10L15/22
Foreign References:
JPH0214000A | 1990-01-18 | |||
JP2011197410A | 2011-10-06 | |||
JP2014044363A | 2014-03-13 |
Other References:
SAWAI, YUICHIRO ET AL.: "Usage of the unlabeled corpus by pseudo error generation for grammatical error correction", PROCEEDINGS OF THE 23RD ANNUAL MEETING OF THE ASSOCIATION FOR NATURAL LANGUAGE PROCESSING, 6 March 2017 (2017-03-06), pages 714 - 717
IKEDA, TAISHI ET AL.: "Neural Sequence-Labelling Models for ASR Error Correction", THE 33RD ANNUAL CONFERENCE OF THE JAPANESE SOCIETY FOR ARTIFICIAL INTELLIGENCE, 4 June 2019 (2019-06-04)
IKEDA, TAISHI ET AL.: "Neural Sequence-Labelling Models for ASR Error Correction", THE 33RD ANNUAL CONFERENCE OF THE JAPANESE SOCIETY FOR ARTIFICIAL INTELLIGENCE, 4 June 2019 (2019-06-04)
Attorney, Agent or Firm:
HASEGAWA Yoshiki et al. (JP)
Download PDF: