E-CONAN Benchmarks for Arabic Natural Language Inference
September 11, 2026
E-CONAN introduces two new Arabic textual entailment datasets, E-CONAN-2 (RTE) and E-CONAN-3 (NLI), utilizing machine-translated, human-validated, and hand-crafted sentence pairs. The benchmark enables zero-shot evaluation of multilingual pretrained models on Arabic semantic relations.
HOW THIS AFFECTS YOU
●
researcherYou can use these datasets to evaluate multilingual model performance specifically for Arabic NLI tasks.