Investigating Data Variance in Evaluations of Automatic Machine Translation Metrics
This highlights a critical methodological issue for researchers in machine translation evaluation, cautioning against overreliance on single-dataset results.
The paper demonstrates that automatic machine translation metric evaluations are sensitive to data variance, with metric rankings changing across different datasets, and identifies potential causes like insignificant data points and deviations from i.i.d. assumptions.
Current practices in metric evaluation focus on one single dataset, e.g., Newstest dataset in each year's WMT Metrics Shared Task. However, in this paper, we qualitatively and quantitatively show that the performances of metrics are sensitive to data. The ranking of metrics varies when the evaluation is conducted on different datasets. Then this paper further investigates two potential hypotheses, i.e., insignificant data points and the deviation of Independent and Identically Distributed (i.i.d) assumption, which may take responsibility for the issue of data variance. In conclusion, our findings suggest that when evaluating automatic translation metrics, researchers should take data variance into account and be cautious to claim the result on a single dataset, because it may leads to inconsistent results with most of other datasets.