Proceedings of the Annual Meeting of the Cognitive Science Society

Everyday decisions about quantity, size, and duration require people to make judgments about magnitude. Perceived energy/intensity influences such judgments, and here we report evaluative meaning is independently predictive. For example, ''wonderful" is reliably judged to be more than ''terrible." Results on antonym pairs show robust agreement across participants on two tasks: an explicit rating task (which is more?) and a task where pairs of adjectives are positioned on a line extending from an origin toward infinity. In both, energy ratings strongly predicted responses and valence predicted additional variance. We further find both energy and valence are reflected in distributional semantic models: FastText embeddings projected onto a direction for magnitude correlate strongly with human judgments, with comparable effects of valence and energy on the model's predictions. Thus the evaluative component of magnitude is recoverable from linguistic statistics. We discuss potential mechanisms underlying this association, including metaphorical mediation, linguistic markedness asymmetries, and cognitive biases toward simulating improvement.