MDCSpell: A Multi-task Detector-Corrector Framework for Chinese Spelling Correction

Zhu, Chenxi; Ying, Ziqiang; Zhang, Boyu; Mao, Feng

doi:10.18653/v1/2022.findings-acl.98

Cited by 20 publications

(11 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Zhang et al (2020) identify the potential error characters by a detection network and then leverage the soft masking strategy to enhance the eventual correction decision. Zhu et al (2022) use a multi-task network to minimize the misleading impact of the misspelled characters (Cheng et al 2020). There is also a line of work that incorporates phonological and morphological knowledge through data augmentation and enhances the BERT-based encoder to assist mapping the error to the 1 https://github.com/Claude-Liu/ReLM 2 https://github.com/gingasan/lemon correct one (Guo et al 2021;Li et al 2021;Liu et al 2021a;Cheng et al 2020;Huang et al 2021;Zhang et al 2021).…”

Section: Related Workmentioning

confidence: 99%

Chinese Spelling Correction as Rephrasing Language Model

Liu,

Wu,

Zhao

2024

AAAI

View full text Add to dashboard Cite

This paper studies Chinese Spelling Correction (CSC), which aims to detect and correct potential spelling errors in a given sentence. Current state-of-the-art methods regard CSC as a sequence tagging task and fine-tune BERT-based models on sentence pairs. However, we note a critical flaw in the process of tagging one character to another, that the correction is excessively conditioned on the error. This is opposite from human mindset, where individuals rephrase the complete sentence based on its semantics, rather than solely on the error patterns memorized before. Such a counter-intuitive learning process results in the bottleneck of generalizability and transferability of machine spelling correction. To address this, we propose Rephrasing Language Modeling (ReLM), where the model is trained to rephrase the entire sentence by infilling additional slots, instead of character-to-character tagging. This novel training paradigm achieves the new state-of-theart results across fine-tuned and zero-shot CSC benchmarks, outperforming previous counterparts by a large margin. Our method also learns transferable language representation when CSC is jointly trained with other tasks.

show abstract

Section: Related Workmentioning

confidence: 99%

Chinese Spelling Correction as Rephrasing Language Model

Liu,

Wu,

Zhao

2024

AAAI

View full text Add to dashboard Cite

show abstract

“…A two-stage distant supervision approach which is based on the T5 transformer and using retrieved evidence was employed to achieve better results than existing works. Zhu [5] proposed a cloze-style detector-corrector framework that firstly detects whether a character is erroneous before correcting it to solve the problem of spelling error correction. There are also some error correction toolkits (e.g.…”

Section: Related Workmentioning

confidence: 99%

“…Unfortunately, although there are some researches focusing on correcting one or several specific kinds of noise or errors (e.g. grammatical errors [3], translation errors [4], and spelling errors [5]), few of them focus on the robustness of a KBQA system to various noisy questions.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A More Robust Model to Answer Noisy Questions in KBQA

Wang

Xü

et al. 2023

IEEE Access

View full text Add to dashboard Cite

In practical applications, the raw input to a Knowledge Based Question Answering (KBQA) system may vary in forms, expressions, sources, etc. As a result, the actual input to the system may contain various errors caused by various noise in raw data and processes of transmission, transformation, translation, etc. As a result, it is significant to evaluate and enhance the robustness of a KBQA model to various noisy questions. In this paper, we generate 29 datasets of various noisy questions based on the original SimpleQuestions dataset to evaluate and enhance the robustness of a KBQA model, and propose a model which is more robust to various noisy questions. Compared with traditional methods, the main contribution in this paper is that we propose a method of generating datasets of different noisy questions to evaluate the robustness of a KBQA model, and propose a KBQA model which contains incremental learning and Mask Language Model (MLM) in the question answering process, so that our model is less affected by different kinds of noise in questions and achieves higher accuracies on datasets of different noisy questions, which shows its robustness. Experimental results show that our model achieves an average accuracy of 78.1% on these datasets and outperforms the baseline BERT-based model by an average margin of 5.0% with the similar training cost. In addition, further experiments show that our model is compatible with other pre-trained models such as ALBERT and ELECTRA.

show abstract

“…CSC models: In recent years, BERT-style (Devlin et al, 2018) models have dominated the research of the CSC task (Hong et al, 2019;Liu et al, 2022;Zhu et al, 2022;Zhang et al, 2020;Bao et al, 2020;. However, due to the lack of large-scale and high-quality datasets, the performance of these models is greatly limited.…”

Section: Related Workmentioning

confidence: 99%

CSCD-IME: Correcting Spelling Errors Generated by Pinyin IME

Hu¹,

Meng²,

Zhou³

2022

Preprint

View full text Add to dashboard Cite

Chinese Spelling Correction (CSC) is a task to detect and correct spelling mistakes in texts. In fact, most of Chinese input is based on pinyin input method, so the study of spelling errors in this process is more practical and valuable. However, there is still no research dedicated to this essential scenario. In this paper, we first present a Chinese Spelling Correction Dataset for errors generated by pinyin IME (CSCD-IME), including 40,000 annotated sentences from real posts of official media on Sina Weibo. Furthermore, we propose a novel method to automatically construct large-scale and high-quality pseudo data by simulating the input through pinyin IME. A series of analyses and experiments on CSCD-IME show that spelling errors produced by pinyin IME hold a particular distribution at pinyin level and semantic level and are challenging enough. Meanwhile, our proposed pseudo-data construction method can better fit this error distribution and improve the performance of CSC systems. Finally, we provide a useful guide to using pseudo data, including the data scale, the data source, and the training strategy 1 .

show abstract

MDCSpell: A Multi-task Detector-Corrector Framework for Chinese Spelling Correction

Cited by 20 publications

References 0 publications

Chinese Spelling Correction as Rephrasing Language Model

Chinese Spelling Correction as Rephrasing Language Model

A More Robust Model to Answer Noisy Questions in KBQA

CSCD-IME: Correcting Spelling Errors Generated by Pinyin IME

Contact Info

Product

Resources

About