Evaluating Prompting Strategies for Grammatical Error Correction Based on Language Proficiency

TL;DR

The paper investigates prompting strategies and fine-tuning for Grammatical Error Correction (GEC) using large language models, with a focus on how English language proficiency levels (A, B, C) modulate performance and overcorrection. It leverages zero-shot and few-shot prompting alongside fine-tuning, evaluated on the CEFR-labeled Cambridge W&I corpus and BEA2019 development data, using ERRANT-based metrics and $F$-scores to analyze precision/recall tradeoffs. Key findings show that few-shot prompting can boost performance for GPT-2 and GPT-3.5, but recall often dominates precision—particularly at higher proficiency (C)—and error-type analyses highlight punctuation and determiner/preposition issues as central; however, SOTA models still outperform prompting-based GEC across levels. The results underscore the value of proficiency-aware prompting design while reaffirming the superiority of extensively fine-tuned or SOTA approaches for robust GEC, with implications for tailoring GEC systems to language learners’ proficiency profiles.

Abstract

The writing examples of English language learners may be different from those of native speakers. Given that there is a significant differences in second language (L2) learners' error types by their proficiency levels, this paper attempts to reduce overcorrection by examining the interaction between LLM's performance and L2 language proficiency. Our method focuses on zero-shot and few-shot prompting and fine-tuning models for GEC for learners of English as a foreign language based on the different proficiency. We investigate GEC results and find that overcorrection happens primarily in advanced language learners' writing (proficiency C) rather than proficiency A (a beginner level) and proficiency B (an intermediate level). Fine-tuned LLMs, and even few-shot prompting with writing examples of English learners, actually tend to exhibit decreased recall measures. To make our claim concrete, we conduct a comprehensive examination of GEC outcomes and their evaluation results based on language proficiency.