Abstract
According to a popular family of hypotheses, crossmodal matches between distinct features hold because they correspond to the same polarity on several conceptual dimensions (such as active-passive, good-bad, etc.) that can be identified using the semantic differential technique. The main problem here resides in turning this hypothesis into testable empirical predictions. In the present study, we outline a series of plausible consequences of the hypothesis and test a variety of well-established and previously untested crossmodal correspondences by means of a novel internet-based testing methodology. The results highlight that the semantic hypothesis cannot easily explain differences in the prevalence of crossmodal associations built on the same semantic pattern (fast lemons, slow prunes, sour boulders, heavy red); furthermore, the semantic hypothesis only minimally predicts what happens when the semantic dimensions and polarities that are supposed to drive such crossmodal associations are made more salient (e.g., by adding emotional cues that ought to make the good/bad dimension more salient); finally, the semantic hypothesis does not explain why reliable matches are no longer observed once intramodal dimensions with congruent connotations are presented (e.g., visually presented shapes and colour do not appear to correspond).