The Impact of Face Inversion on Animacy Categorization

Abstract

Face animacy perception is categorical: Gradual changes in the real/artificial appearance of a face lead to nonlinear behavioral responses. Neural markers of face processing are also sensitive to face animacy, further suggesting that these are meaningful perceptual categories. Artificial faces also appear to be an “out-group” relative to real faces such that behavioral markers of expert-level processing are less evident with artificial faces than real ones. In the current study, we examined how categorical processing of real versus doll faces was impacted by the face inversion effect, which is one of the most robust markers of expert face processing. We examined how explicit categorization of faces drawn from a real/doll morph continuum was affected by face inversion (Experiment 1) and also how the response properties of the N170 were impacted by face animacy and inversion. We found that inversion does not change the position or steepness of the category boundary measured behaviorally. Further, neural markers of face processing are equally impacted by inversion regardless of whether they are elicited by real faces or doll faces. On balance, our results indicate that inversion has a limited impact on the categorical perception of face animacy.

Perceiving Animacy in Own-and Other-Species Faces

“…The results of Experiment 1 suggest that the orientation of morphed human/doll faces affects animacy categorization such that the point of subjective equality for inverted faces is shifted toward the “Real” end of the spectrum relative to upright faces. This is at odds with previous results (Balas et al, 2018) indicating that orientation generally does not impact psychometric curves for animacy categorization, so how do we explain the discrepancy? An important distinction between Experiment 1 and Balas et al (2018) is the inclusion of longer presentation times, which may indicate that orientation only affects animacy categorization when images are presented for sufficiently long periods of time.…”

“…Specifically, whether we consider artificial faces that are doll-like (Looser and Wheatley, 2010; Balas and Horski, 2012) or computer-generated (Cheetham et al, 2011; Balas, 2013; Balas and Tonsager, 2014), observers are capable of telling the difference between a real face and a synthetic one. The differences between real and artificial face appearance support basic category discrimination (Koldewyn et al, 2014; Balas et al, 2018) and also impact a number of other judgments regarding social characteristics and other more complex inferences based on physiognomy.…”

“…During the testing session, stimuli were presented in a pseudorandomized order that was generated for each participant using the Shuffle.m functions in the Matlab Psychophysics Toolbox v3. On each trial, participants were first presented with a fixation cross for 500 ms, followed by a single face presented either upright or inverted at the center of the screen for either 160, 320, or 640 ms. We chose these values based on the range of image durations reported in previous studies of animacy perception (Looser and Wheatley, 2010; Balas and Tonsager, 2014; Balas et al, 2018) and also guided by prior results demonstrating that more complex social variables (e.g., competence) can be measured at above chance levels from face images with an image duration of 100 ms (Willis and Todorov, 2006). This face was immediately followed by a noise mask depicting fractal noise with a 1/f power spectrum, which was displayed for 500 ms before it was erased (Figure 2).…”

“…Computer-generated artificial faces that are identity-matched to real individuals, thus guaranteeing close resemblance between images, are also not socially evaluated in the same way as real faces. Perceived trustworthiness is not generally preserved across real and artificial versions of the same person (Balas and Pacella, 2017), and more broadly the position of artificial faces in a social “Face Space” defined by trustworthiness and dominance (Oosterhof and Todorov, 2008) differs from that of real faces (Balas et al, 2018). Recognizing that a face is not real thus influences a range of judgments that govern how we evaluate social agents and how we may choose to interact with them (Hortensius and Cross, 2018).…”

“…Another approach to examining the role of expert-level processing in the context of real and artificial faces is to attempt to disrupt face perception via face inversion and examine the impact of this manipulation on the processing of face animacy itself, or the processing of other aspects of appearance as a function of face animacy. Face inversion appears to have relatively modest effects on animacy categorization in some contexts (Deska et al, 2017; Balas et al, 2018), but has a measurable impact on performance in lexical decision tasks (LDT) in which faces serve as primes for human- or machine-related words (Hugenberg et al, 2016). Critically, this latter result not only demonstrated the role of upright face appearance in priming LDT performance, but further demonstrated that human/animal judgments were differentially affected by inversion.…”