Semantic Sensitivities and Inconsistent Predictions: Measuring the Fragility of NLI Models

Open in new window