Evaluation of large-scale synthetic data for Grammar Error Correction

TL;DR

This paper introduces three metrics—reliability, diversity, and distribution match—to evaluate the quality of large-scale synthetic data for Grammar Error Correction, aiming to improve data generation and system performance.

Contribution

It proposes novel metrics for assessing synthetic data quality in GEC and automates their evaluation to enhance data generation processes.

Findings

Metrics effectively differentiate data quality aspects.

Automated evaluation guides improvements in synthetic data generation.

Enhanced data quality leads to better GEC system performance.

Abstract

Grammar Error Correction(GEC) mainly relies on the availability of high quality of large amount of synthetic parallel data of grammatically correct and erroneous sentence pairs. The quality of the synthetic data is evaluated on how well the GEC system performs when pre-trained using it. But this does not provide much insight into what are the necessary factors which define the quality of these data. So this work aims to introduce 3 metrics - reliability, diversity and distribution match to provide more insight into the quality of large-scale synthetic data generated for the GEC task, as well as automatically evaluate them. Evaluating these three metrics automatically can also help in providing feedback to the data generation systems and thereby improve the quality of the synthetic data generated dynamically