Kemtc: Knowledge-Enhanced Multilingual Text Correction with Large Language Model

Text correction is an important task in the field of natural language processing. Traditional research primarily focuses on monolingual correction, which is not convenient for correcting multilingual texts. This paper introduces a unified multilingual text correction framework that uses a large language model as the base model and employs a multi-task learning approach to simultaneously learn text correction and correction explanations. This enables the model to master diverse grammatical structures and expressions. Additionally, to mitigate the issue of “overcorrection” often caused by large language models striving for sentence fluency, overcorrection information is incorporated as auxiliary data to help the model reduce excessive corrections. Comprehensive experiments show that our approach simplifies the deployment process and achieves promising results in Chinese, English, German, and Russian.