IME-MTCSC: A Multi-Typo Chinese Spelling Correction Model by Input Method Editor
Ran Li, Lifen Jiang, Fengbo Zheng, Xiu Zhang, Haoyu Guo, Bo‐Wei Chen · 2024
Chinese Spelling Checking (CSC) task is to detect and correct spelling typos in Chinese texts to improve the accuracy and readability. Unlike English text, Chinese text relies on input methods, among which Pinyin IME method is the most widely used. Based on the real input method editor, we correct the incorrect Pinyin and selected candidate words. The Chinese typo correction model based on Bert has certain limitations and performs poorly in correcting text with multiple typos. In this paper, we propose a multi-typo Chinese Spelling Checking model by input method editor (IME-MTCSC), which is based on Bert for real application scenarios. Our model is robust to noise caused by typos.