abstract
-
People form rapid first impressions when encountering novel faces or voices. A popular theory on first impressions suggests there are two dimensions on which voices and faces vary: valence and dominance. Studies using orthogonal factor rotation, consistently find a third dimension in face space. However, participants are always given the same questionnaire. Unconstrained descriptions of female voices and faces from free-form responses have been tested but rely on researchers’ opinions to categorize descriptions. To shift researcher degrees of freedom away from category membership, we used machine learning to categorize free responses to 50 female and 50 male voices saying ‘hi’. We found a rich set of features that people use to categorize voices that does not reduce to valence and dominance, suggesting that while these traits are important to researchers, they are not as important to participants.