A Simple Recipe for Multilingual Grammatical Error Correction

TL;DR

This paper introduces a straightforward approach for training multilingual GEC models using synthetic data generation and large-scale language models, achieving state-of-the-art results across four languages and simplifying the training process.

Contribution

It proposes a language-agnostic synthetic data generation method and demonstrates that fine-tuning large multilingual models on this data improves GEC performance.

Findings

Achieved new state-of-the-art results in English, Czech, German, and Russian GEC benchmarks.

Created the cLang-8 dataset by cleaning lang-8 targets with the gT5 model.

Single-step fine-tuning on cLang-8 surpasses previous methods.

Abstract

This paper presents a simple recipe to train state-of-the-art multilingual Grammatical Error Correction (GEC) models. We achieve this by first proposing a language-agnostic method to generate a large number of synthetic examples. The second ingredient is to use large-scale multilingual language models (up to 11B parameters). Once fine-tuned on language-specific supervised sets we surpass the previous state-of-the-art results on GEC benchmarks in four languages: English, Czech, German and Russian. Having established a new set of baselines for GEC, we make our results easily reproducible and accessible by releasing a cLang-8 dataset. It is produced by using our best model, which we call gT5, to clean the targets of a widely used yet noisy lang-8 dataset. cLang-8 greatly simplifies typical GEC training pipelines composed of multiple fine-tuning stages -- we demonstrate that performing a single fine-tuning step on cLang-8 with the off-the-shelf language models yields further accuracy improvements over an already top-performing gT5 model for English.