Languages Still Left Behind: Toward a Better Multilingual Machine Translation Benchmark
This work highlights a problem for researchers and practitioners in machine translation by exposing vulnerabilities in a widely used benchmark, advocating for more robust and culturally neutral evaluation methods.
The study identified critical shortcomings in the FLORES+ multilingual machine translation benchmark, revealing that translations in four languages often fall below the claimed 90% quality standard and that the benchmark is biased toward English-speaking domains, leading to poor performance of models on it despite gains on domain-relevant data.
Multilingual machine translation (MT) benchmarks play a central role in evaluating the capabilities of modern MT systems. Among them, the FLORES+ benchmark is widely used, offering English-to-many translation data for over 200 languages, curated with strict quality control protocols. However, we study data in four languages (Asante Twi, Japanese, Jinghpaw, and South Azerbaijani) and uncover critical shortcomings in the benchmark's suitability for truly multilingual evaluation. Human assessments reveal that many translations fall below the claimed 90% quality standard, and the annotators report that source sentences are often too domain-specific and culturally biased toward the English-speaking world. We further demonstrate that simple heuristics, such as copying named entities, can yield non-trivial BLEU scores, suggesting vulnerabilities in the evaluation protocol. Notably, we show that MT models trained on high-quality, naturalistic data perform poorly on FLORES+ while achieving significant gains on our domain-relevant evaluation set. Based on these findings, we advocate for multilingual MT benchmarks that use domain-general and culturally neutral source texts rely less on named entities, in order to better reflect real-world translation challenges.