CLMar 5, 2024

Revisiting Meta-evaluation for Grammatical Error Correction

arXiv:2403.02674v223 citationsh-index: 26TACL
Originality Synthesis-oriented
AI Analysis

This addresses challenges in evaluating GEC metrics for researchers, but it is incremental as it focuses on dataset improvements rather than a new method.

The paper tackled the problem of biases in meta-evaluation for grammatical error correction by introducing SEEDA, a new dataset with human ratings at edit and sentence granularities, showing that edit-based metrics may have been underestimated and traditional metrics perform poorly on neural systems.

Metrics are the foundation for automatic evaluation in grammatical error correction (GEC), with their evaluation of the metrics (meta-evaluation) relying on their correlation with human judgments. However, conventional meta-evaluations in English GEC encounter several challenges including biases caused by inconsistencies in evaluation granularity, and an outdated setup using classical systems. These problems can lead to misinterpretation of metrics and potentially hinder the applicability of GEC techniques. To address these issues, this paper proposes SEEDA, a new dataset for GEC meta-evaluation. SEEDA consists of corrections with human ratings along two different granularities: edit-based and sentence-based, covering 12 state-of-the-art systems including large language models (LLMs), and two human corrections with different focuses. The results of improved correlations by aligning the granularity in the sentence-level meta-evaluation, suggest that edit-based metrics may have been underestimated in existing studies. Furthermore, correlations of most metrics decrease when changing from classical to neural systems, indicating that traditional metrics are relatively poor at evaluating fluently corrected sentences with many edits.

Code Implementations2 repos
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes