Efficient and Interpretable Grammatical Error Correction with Mixture of Experts

TL;DR

This paper introduces MoECE, a mixture-of-experts model for grammatical error correction that matches the performance of larger models while being more efficient and interpretable by identifying error types during inference.

Contribution

Proposes MoECE, a mixture-of-experts model that improves efficiency and interpretability in grammatical error correction by specializing sub-networks for different error types.

Findings

Achieves T5-XL performance with three times fewer parameters.

Provides interpretable corrections by identifying error types.

Reduces computational cost compared to system combination methods.

Abstract

Error type information has been widely used to improve the performance of grammatical error correction (GEC) models, whether for generating corrections, re-ranking them, or combining GEC models. Combining GEC models that have complementary strengths in correcting different error types is very effective in producing better corrections. However, system combination incurs a high computational cost due to the need to run inference on the base systems before running the combination method itself. Therefore, it would be more efficient to have a single model with multiple sub-networks that specialize in correcting different error types. In this paper, we propose a mixture-of-experts model, MoECE, for grammatical error correction. Our model successfully achieves the performance of T5-XL with three times fewer effective parameters. Additionally, our model produces interpretable corrections by also identifying the error type during inference.