LocalDoc/

en_az_translate_benchmark

High-quality English-Azerbaijani parallel benchmark for evaluating machine translation systems.

Source: FLORES-200 devtest English sentences Reference translations: Generated by GPT-5.4 (EN→AZ direction) Size: 1012 sentence pairs Purpose: Evaluation benchmark (NOT for training)

Original FLORES-200 Azerbaijani references contain systematic errors:

Semantic inaccuracies ("per unit of GDP" translated… See the full description on the dataset page: https://huggingface.co/datasets/LocalDoc/en_az_translate_benchmark.