Computational measurement of perceived pointiness from pronunciation

Abstract Sound symbolism is a well-researched topic of psycholinguistics, which tries to comprehend the connection between sound word and its meanings. The Bouba-Kiki effect , one form symbolism, claims that people perceive pronunciation “Kiki” as pointier than “Bouba.” There no research focuses on modeling such perception, i.e., how pointy sounds humans, through computational data-driven approaches. To address this, this paper first proposes novel concept “phonetic pointiness” defined shape humans are most likely associate with given pronunciation. We then model phonetic pointiness from approaches calculate score for an arbitrary three proposed models: referential model, expressive combined integrates previous two. idea comes existing psycholinguistic classification two types symbolisms: where former relates vocabulary knowledge, while latter based pure human intuition. models constructed only image language data available Web, therefore not requiring task-specific annotations. evaluate these crowd-sourced user study, finding promising correlation perception calculated by models. results indicate can be modeled better combining both symbolisms. Furthermore, observing behaviors models, we show several possible use-cases, product naming research, useful insight further studies applications.