Cross-Corpora Evaluation and Analysis of Grammatical Error Correction Models --- Is Single-Corpus Evaluation Enough?

TL;DR

This paper demonstrates that evaluating grammatical error correction models on a single corpus is inadequate, as model performance varies significantly across different learner corpora, highlighting the need for cross-corpora evaluation.

Contribution

It provides a comprehensive cross-corpora evaluation of GEC models, revealing variability in performance and model rankings across different datasets.

Findings

Model rankings vary across different corpora.

Single-corpus evaluation may misrepresent model effectiveness.

Cross-corpora evaluation offers a more reliable assessment.

Abstract

This study explores the necessity of performing cross-corpora evaluation for grammatical error correction (GEC) models. GEC models have been previously evaluated based on a single commonly applied corpus: the CoNLL-2014 benchmark. However, the evaluation remains incomplete because the task difficulty varies depending on the test corpus and conditions such as the proficiency levels of the writers and essay topics. To overcome this limitation, we evaluate the performance of several GEC models, including NMT-based (LSTM, CNN, and transformer) and an SMT-based model, against various learner corpora (CoNLL-2013, CoNLL-2014, FCE, JFLEG, ICNALE, and KJ). Evaluation results reveal that the models' rankings considerably vary depending on the corpus, indicating that single-corpus evaluation is insufficient for GEC models.