ALEE Dataset (AMR English Foils)

ALEE: Any-Language Evaluation of Embeddings ALEE evaluates text embeddings across a wide range of languages using English-centric minimal pairs. Each English sentence is paired with a set of altered (negative) variants and translated into many languages, so an embedding model can be tested on whether it captures small but meaning-changing differences in any language. Dataset structure The dataset has three configurations, each with a single test split: alee bq275… See the full description on the dataset page: datasets.

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