We explore the behaviour of language models on adjectival scales in connection with negation when prompted with material used in human experiments. We propose several metrics extracted from the model predictions and analyze those metrics in relation to human data as well as use them to propose new items to be tested in human experiments.