TMU-GFM-Dataset

A dataset for GEC metrics with manual evaluations of grammaticality, fluency, and meaning preservation for system outputs.
More detail about the creation of the dataset can be found in Yoshimura et al. (2020).

File format

The are 9 columns in the tmu-gfm-dataset.

1. source: source sentence.
2. output: system output sentence.
3. grammer: Grammaticaliry annotations by 5 annotators.
4. fluency: Fluency annotations by 5 annotators.
5. meaning: Meaning Preservation annotations by 5 annotators.
6. system: Which system the output sentence is from.
7. ave_g: Average grammer score.
8. ave_f: Average fluency score.
9. ave_m: Average meaning score.