Towards Robust Chinese Spelling Check Systems: Multi-round Error Correction with Ensemble Enhancement

Xiang Li, Hanyue Du, Yike Zhao, Yunshi Lan

Abstract: Chinese Spelling Check requires a system to automatically correct spelling errors in a sentence. There are diverse methods proposed to solve this task. A few methods improve the robustness of the model through data augmentation, but they have some weaknesses. Errors inserted randomly might disturb the real distribution of data. Moreover, different models may produce different results when predicting the same error sentence. Based on these intuitions, we develop a multi-round error correction method with ensemble enhancement, which is robust in solving Chinese Spelling Check challenges. Specifically, multi-round error correction follows an iterative correction pipeline, where a single error is corrected at each round, and the subsequent correction is conducted based on the previous results. Furthermore, we proposed two strategies of ensemble enhancement. For each predicted correction, results of multiple models are mutually authenticated by weighted voting and dominate voting. Experiments have proved the effectiveness of our system. It achieves the best performance on NLPCC 2023 CSC shared tasks. More analyses verify that both multi-round error correction and ensemble enhancement contribute to its good results. Our code is publicly available on GitHub.