MSLC25: Metric Performance on Low-Quality Machine Translation, Empty Strings, and Language Variants

AbstractIn this challenge set, we examine how automatic metrics for machine translation perform on a wide variety of machine translation output, covering a wider range of quality than the WMT submissions. We also explore metric results on specific types of corner cases, such as empty strings, wrong- or mixed-language text, and more. We primarily focus on Japanese–Chinese data, with some work on English and Czech.