Comparison of the number of the translated tokens and BLEU scores in different sizes of data subsets.

Subset-L, M, and S denote three randomly divided subsets of data with different sizes. PET consistently shows the best performance in all data subsets of varying sizes.子集L、M、S分别指代三组经随机划分所得的不同规模数据子集。PET在各类不同规模的数据子集上始终展现出最优性能。