Preprint
Is ChatGPT a Highly Fluent Grammatical Error Correction System? A Comprehensive Evaluation
Tao Fang et al., arXiv, 2023
Read the originalRead on October 17, 2026
- What it is
- An evaluation of ChatGPT, built on the GPT-3.5 model, as a grammar correction system on English benchmarks, with human judges scoring its corrections. The authors report strong error detection alongside a tendency to over-correct.
- How much weight it carries
- A 2023 preprint on an older model. Newer chat models may behave differently. It covers sentence-level grammar correction rather than spoken transcripts, but the over-correction it describes is the behavior a cleanup prompt can trigger.
Facts taken from this source
- In a 2023 evaluation of ChatGPT as a grammar corrector, its very fluent corrections were possibly due to over-correction and not following the principle of minimal edits.
“possibly due to its over-correction tendencies and not adhering to the principle of minimal edits” (Abstract)