Semantic distance and creativity in linguistic synaesthesia

In this work, we aim at quantitatively assessing the creativity of linguistic combinations in terms of semantic distance, using synaesthetic metaphors (e.g. bitter voice) as a case study. We created an evaluation dataset containing examples of synaesthesia that are actually occurring in corpora and automatically generated synaesthetic metaphors, together with a control set of non-synaesthetic adjective-noun combinations. Then, we tested on the dataset three quantitative models of linguistic creativity that have been proposed in the NLP and in the cognitive science literature, and we compared their performance in discriminating between creative and non-creative, directional and non-directional synaesthetic metaphors, and between synaesthetic metaphors and non-synaesthetic phrases.