Investigating the Performance of Transformer-Based NLI Models on Presuppositional Inferences