LingglePolish: Elevating Writing Proficiency Through Comprehensive Grammar and Lexical Refinement

Many existing writing enhancement tools excel at correcting syntactic errors but fall short in addressing lexical inaccuracies or providing valuable word choice suggestions for learning purposes. We present LingglePolish, an innovative interactive tool designed to assist language learners in rectifying both syntactic and lexical errors, thereby improving learners’ overall writing proficiency. Utilizing masked language models, we augment an existing Grammar Error Correction (GEC) corpus to encapsulate comprehensive editorial insights concerning syntactic and lexical inaccuracies. We fine-tune a pre-trained generative model on this augmented dataset, culminating in the development of LingglePolish. Our evaluation of LingglePolish entails a comparative analysis with Grammarly, employing a corpus of authentic essays penned by language learners at the university level. Preliminary findings reveal that although LingglePolish demonstrates weaker performance concerning syntactic errors, it also exhibits remarkable proficiency in providing word choice suggestions. Notably, when addressing word-choice errors, our system significantly outperforms Grammarly, boasting a coverage rate that is three times higher. This enhanced capability underscores the potential of LingglePolish to mark a significant advancement in the domain of language learning tools.