Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
TRANSLATION MODEL TRAINING METHOD, TRANSLATION METHOD, AND DEVICE
Document Type and Number:
WIPO Patent Application WO/2022/127613
Kind Code:
A1
Abstract:
Provided is a translation model training method, relating to the field of artificial intelligence, comprising: obtaining a word vector sequence for a training statement; by means of an encoder of a first translation model, obtaining a first coding sequence of a word vector sequence, a unidirectional encoder being used for the encoder of the first translation model; by means of an encoder of a second translation model, obtaining a second coding sequence of a word vector sequence, a bidirectional encoder being used for the encoder of the second translation model; inputting the first coding sequence into a decoder of the first translation model to obtain a first prediction result; according to the first prediction result and a target translation result corresponding to the training statement, obtaining a first loss value; according to the first loss value and the distance between the first coding sequence and the second coding sequence, updating the first translation model. The method can improve the performance of a translation model.

Inventors:
ZHANG SHAOLEI (CN)
FENG YANG (CN)
LI LIANGYOU (CN)
Application Number:
PCT/CN2021/135238
Publication Date:
June 23, 2022
Filing Date:
December 03, 2021
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
HUAWEI TECH CO LTD (CN)
International Classes:
G06N3/04
Foreign References:
CN112597778A2021-04-02
CN109933809A2019-06-25
CN107729329A2018-02-23
CN109829172A2019-05-31
US20200034436A12020-01-30
Attorney, Agent or Firm:
SHENPAT INTELLECTUAL PROPERTY AGENCY (CN)
Download PDF: