Chinese spelling error correction by multi-task learning with pronunciation gap predictor

Hao Pan; Junmin Wu

doi:10.1117/12.2674737

2 May 2023 Chinese spelling error correction by multi-task learning with pronunciation gap predictor

Hao Pan, Junmin Wu

Proceedings Volume 12642, Second International Conference on Electronic Information Engineering, Big Data, and Computer Technology (EIBDCT 2023); 126421J (2023) https://doi.org/10.1117/12.2674737
Event: Second International Conference on Electronic Information Engineering, Big Data and Computer Technology (EIBDCT 2023), 2023, Xishuangbanna, China

Abstract

Chinese Spell Check (CSC) aims to detect and correct spelling errors in Chinese text, almost all of which are related to phonetic or visual similarity. Large-scale pre-trained models (PLMs) are currently making substantial progress on the CSC task. However, when correcting errors, PLMs tend to select those words that are semantically sound or expressively fluent, sometimes ignoring pronunciation similarities. Meanwhile, the models lack knowledge of pronunciation differences. To address this problem, we propose a multi-task learning model to help enhance the CSC task. The auxiliary task is to estimate the degree of pronunciation gap between the original input and the corresponding correct text from the granularity of each word. Specifically, we use the edit distance of Pinyin to measure the degree of pronunciation discrepancy. The edit distance scheme we use is modified, due to the specificity of the Pinyin structure. Experiments on a open available benchmark dataset demonstrate the effectiveness of our strategy.

Citation Download Citation

Hao Pan and Junmin Wu "Chinese spelling error correction by multi-task learning with pronunciation gap predictor", Proc. SPIE 12642, Second International Conference on Electronic Information Engineering, Big Data, and Computer Technology (EIBDCT 2023), 126421J (2 May 2023); https://doi.org/10.1117/12.2674737

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available