Title:
TEXT ERROR CORRECTION METHOD, SYSTEM AND DEVICE, AND READABLE STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2021/189851
Kind Code:
A1
Abstract:
A text error correction method, system and device, and a computer-readable storage medium, which relate to the technical field of artificial intelligence. The method comprises: acquiring a sequence of text to be subjected to error correction, and identifying the sequence of said text by means of a Bert-based mask language model to determine a target word, on which error correction needs to be performed, from the sequence of said text; generating a candidate word set of the target word according to the target word and the sequence of said text; and screening the candidate word set of the target word according to a preset screening rule, determining a target replacement word of the target word, and generating a replacement text sequence according to the target replacement word and the sequence of said text. By using the Bert-based mask language model, the problem of over-fitting caused by insufficient parallel corpora for Chinese text error correction can be avoided; and by means of dynamically generating candidate words on the basis of the context of the target word, the problem in the prior art of inflexible generation of the candidate words caused by the use of a confusion set is avoided.
Inventors:
HUI YANFEI (CN)
WANG JIANZONG (CN)
CHENG NING (CN)
WANG JIANZONG (CN)
CHENG NING (CN)
Application Number:
PCT/CN2020/125011
Publication Date:
September 30, 2021
Filing Date:
October 30, 2020
Export Citation:
Assignee:
PING AN TECH SHENZHEN CO LTD (CN)
International Classes:
G06F40/279; G06F40/166; G06F40/226
Foreign References:
| CN111310443A | 2020-06-19 | |||
| CN110807319A | 2020-02-18 | |||
| CN110196894A | 2019-09-03 | |||
| CN110852087A | 2020-02-28 | |||
| US20200192983A1 | 2020-06-18 |
Attorney, Agent or Firm:
SHENZHEN WORLD INTELLECTUAL PROPERTY AGENCY (GENERAL PARTNERSHIP) (CN)
Download PDF: