ALEE Datasets (AMR English Foils)

ALEE Datasets Three multilingual machine-translation benchmarks — FLORES-200, BOUQuET and WMT24++ — each extended with English “foil” (negative) columns: minimally edited, meaning-altered versions of the English text, produced by an Abstract Meaning Representation (AMR) round-trip and filtered with a Natural Language Inference (NLI) model so the foil is not entailment-equivalent to the original. The alee mt61 config is a merge of two datasets — google/wmt24pp (English + 55… See the full description on the dataset page: datasets.

Source: Hugging Face Hub (Psychias/alee_datasets). Metadata imported from the dataset’s Hub tags.