MultiBLiMP

MultiBLiMP MultiBLiMP is a massively Multilingual Benchmark for Linguistic Minimal Pairs. The dataset is composed of synthetic pairs generated using Universal Dependencies and UniMorph. The paper can be found here. We split the data set by language: each language consists of a single .tsv file. The rows contain many attributes for a particular pair, most important are the sen and wrong sen fields, which we use for evaluating the language models. Using MultiBLiMP To… See the full description on the dataset page:

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