Face Information in Attribute Amnesia

Very Brief Introduction

How salient are identity, expression, and viewpoint in processing faces? What do we remember about them, what do we forget? Does this change as a function of face condition (i.e. are there interactions between face attributes?)

I will begin by investigating the effect of viewpoint on expression location and identification performance in an attribute amnesia paradigm. Do we report identities or expressions more accurately when faces are directed towards the viewer? Is there useful information available in face profiles that are beneficial for either or both tasks? We’ll see!

Methods

I am proposing an adaption of the paradigm established by Chen et al. (2019):

Model

I have chosen a 5 (viewpoint) x 2 (target valence category) between-subjects design.

• Viewpoint: frontal, 45- and 90-degrees, left and right (5 total)
• Target valence category: negative (angry, sad, fearful) vs. positive (happy)

Task Set-up

In the PRE-TRIALS, participants are asked to report the location of the face with the expression in the target valence category (negative vs. positive), from a sea of opposite valence category faces. This repeats for a total of 29 PRE-TRIALS.

On the 30th trial (the TEST-TRIAL), participants are suddenly asked to report the identity of the expression with the target valence category. Performance on the TEST-TRIAL is the sole measure of the attribute amnesia phenomenon.

The POST-TRIAL format is the same as the TEST-TRIAL. with an additional location component, as in the PRE_TRIALS. This repeats for 4 more trials, with a CATCH-TRIAL or two thrown in.

A CATCH-TRIAL is a task attention and understanding check. Participants will report what the task was for the majority of the experiment (correct response is something like: “locating the face with the target expression”).

Related Questions

What gets consolidated into working memory?

Typically, studies of identification performance across pose and expression conditions are aimed at the consistency of identification performance across such conditions. The goal of this study is to investigate how reliably we retain non-identity information in an identification task. The idea stemmed from how information is commingled and distributed in representations from DCNNs trained for face identification, such that: identity > viewpoint/pose > expression. Suppose this is the case in human perception. Could this be reflected in behavior?

Can we fuse this with familiarity studies? Does this hierarchy change with familiar vs. novel faces? Will other face information become more salient after we have experience with face identity? Novel stimuli does show an effect… attribute amnesia disappears for novel stimuli, is this differential for faces vs. other stimuli?