Variance-Aware Machine Translation Test Sets
This work addresses the need for more reliable and efficient test sets in machine translation evaluation, though it is incremental as it builds on existing WMT test sets.
The authors tackled the problem of indiscriminative test instances in machine translation evaluation by releasing 70 variance-aware test sets (VAT) that outperform original WMT test sets in correlation with human judgment across mainstream language pairs.
We release 70 small and discriminative test sets for machine translation (MT) evaluation called variance-aware test sets (VAT), covering 35 translation directions from WMT16 to WMT20 competitions. VAT is automatically created by a novel variance-aware filtering method that filters the indiscriminative test instances of the current MT test sets without any human labor. Experimental results show that VAT outperforms the original WMT test sets in terms of the correlation with human judgement across mainstream language pairs and test sets. Further analysis on the properties of VAT reveals the challenging linguistic features (e.g., translation of low-frequency words and proper nouns) for competitive MT systems, providing guidance for constructing future MT test sets. The test sets and the code for preparing variance-aware MT test sets are freely available at https://github.com/NLP2CT/Variance-Aware-MT-Test-Sets .