ChineseErrorCorrector3-4B: State-of-the-Art Chinese Spelling and Grammar Corrector

TL;DR

The paper tackles Chinese spelling and grammatical error correction within a single, unified model. It introduces ChineseErrorCorrector3-4B, built on the Qwen3-4B backbone, and trains it via a two-stage supervised fine-tuning process: Stage I aligns broad linguistic knowledge using large, mixed-correction data, and Stage II performs joint fine-tuning on CSC and CGC data to foster cross-task transfer. Empirical results across CSC benchmarks (SIGHAN15, EC-LAW, MCSC) and the NaCGEC CGC benchmark demonstrate state-of-the-art performance, with particularly strong recall and a superior F0.5 score, underscoring robust performance in real-world, domain-diverse scenarios. The work shows that a single, well-prepared model can outperform specialized systems by leveraging data unification and joint optimization, offering practical implications for scalable, robust grammar and spelling correction in Chinese text. The findings highlight the value of cross-task synergy and a staged training paradigm for complex language-editing tasks.

Abstract

This paper introduces ChineseErrorCorrector3-4B, a unified model for Chinese spelling and grammatical error correction based on Qwen3-4B. The model demonstrates outstanding performance in general text correction tasks and achieves state-of-the-art results in both spelling correction (CSC) and grammatical correction (CGC). On several authoritative benchmark datasets -- including SIGHAN-2015, EC-LAW, MCSC, and NaCGEC -- the model's F1 and F0.5 scores significantly surpass existing publicly available models, ranking first in both spelling and grammatical error correction tasks.