Arabic Natural Language Inference Triplet

Arabic NLI Triplet is a dataset derived from the Arabic versions of the SNLI and MultiNLI corpora, curated as a triplet subset. While it was originally created for Natural Language Inference (NLI) tasks, this version is structured to support training and fine-tuning embedding models for semantic textual similarity. Each data point is represented as a triplet consisting of an anchor, a semantically related positive, and a semantically unrelated negative sentence, with all columns stored as strings. For example, an anchor sentence describing a person on a horse jumping over a broken airplane is paired with a positive sentence that preserves the same semantic context and a negative sentence that introduces an unrelated scenario.