This dataset of minimal pairs is used as evaluation of pre-trained and fine-tuned models that we submitted to the BabyLM Challenge 2025. The chosen and rejected sentences are matched in terms of number of words they are composed of (whitespace tokenization).
