Abstract
Humans regularly describe bodies using global and local feature terms (e.g., pear-shaped, fit, broad shoulders). Remarkably little is known about how these features map onto complex body shapes and how bodies compare perceptually to one another. Here, we "reverse engineered" a human body space from participants' ratings of male and female bodies. Participants (n = 31) rated 60 male and 165 female bodies using 27 common descriptor terms that captured variations in global shape (e.g., rectangular, curvy), local features (e.g., short torso, long legs), physical health (e.g., fit, muscular), and gender-related attributes (e.g., masculine). The rating choices for each descriptor were: "1: applies perfectly; 2: applies somewhat; 3: does not apply". Ratings were compiled across participants and submitted to a correspondence analysis—a multivariate technique that placed the bodies and feature descriptor terms in a common space. We generated separate multidimensional spaces for female and male bodies. For both spaces, axis 1 captured weight variation, axis 2 captured height variation, and axis 3 described sex-specific global shape variations. The sex-specific shape variations in the female body space contrasted "curvy" and "pear-shaped" with "skinny" and "lean"; for males, it contrasted "muscular" and "built" with "skinny" and "small." An additional common axis (4th for males, 5th for females) captured variation in the ratio of torso and leg length, contrasting bodies having short torsos and long legs with bodies having long torsos and short legs. Combined, the described axes explained 62% of the variance for the male space and 59% for the female space. These spaces offer a first look at the perceptual structure of human body variability and give insight into how individual bodies align with the labels used to describe them. The results suggest that relatively stable perceptual representations of bodies can be reverse engineered from linguistic labels.
Meeting abstract presented at VSS 2014