Few-Shot Domain Adaptation for Grammatical Error Correction via Meta-Learning

TL;DR

This paper introduces a meta-learning approach for few-shot domain adaptation in grammatical error correction, enabling effective model adaptation to new language domains with minimal data without relying on pseudo data.

Contribution

It extends meta-learning to GEC domain adaptation, leveraging source domains for rapid adaptation to resource-poor target domains without pseudo data.

Findings

Outperforms multi-task transfer learning baseline by 0.50 F0.5 score on average

Achieves effective adaptation with only 200 parallel sentences

Demonstrates successful adaptation across nine source and five target L1 domains

Abstract

Most existing Grammatical Error Correction (GEC) methods based on sequence-to-sequence mainly focus on how to generate more pseudo data to obtain better performance. Few work addresses few-shot GEC domain adaptation. In this paper, we treat different GEC domains as different GEC tasks and propose to extend meta-learning to few-shot GEC domain adaptation without using any pseudo data. We exploit a set of data-rich source domains to learn the initialization of model parameters that facilitates fast adaptation on new resource-poor target domains. We adapt GEC model to the first language (L1) of the second language learner. To evaluate the proposed method, we use nine L1s as source domains and five L1s as target domains. Experiment results on the L1 GEC domain adaptation dataset demonstrate that the proposed approach outperforms the multi-task transfer learning baseline by 0.50 $F_{0.5}$ score on average and enables us to effectively adapt to a new L1 domain with only 200 parallel sentences.