Attentional weighting of different (non-spatial) dimensions of naturalistic face objects

Objects have previously been described as the unit of attentional selection: when attending to an object, all feature information belonging to that object would be selected as well. Here, we investigated whether this also holds true for naturalistic complex objects like faces. We asked subjects to attend to different non-spatial dimensions of face stimuli, i.e. to either attend to the face’s identity or the face’s emotional expression. While the face stimuli where rhythmically morphing into different identities (e.g., at 1.5Hz) and different emotional states (e.g., at 2.0Hz), subjects had to selectively attend to one or the other dimension and press a button upon the detection of a pre-defined target identity or emotional-state target. Cue-validity was 70%. In this project, we record MEG signals and behavioural measures of detection accuracy and reaction times, central eye-fixation was controlled by an eye-tracker. Comparing trials with valid, neutral, and invalid cues, we found strong top-down attentional weighting of feature dimensions in face stimuli. For both dimensions, reaction times were faster if the cue was valid, compared to trials with invalid or neutral cues. Also, detection accuracy was better in valid trials, with many targets in the invalidly-cued trials being often missed completely. A signal-detection analysis showed attentional effects on both perceptual sensitivity and the response bias. Analysing the enhancement versus inhibition of the respective frequency responses in the MEG signals (frequency-tagging), mirrors these behavioural results observed under sustained attention, and shows the interaction of top-down attention in dissociable neural networks representing the different feature dimensions of face stimuli. Our results show for the first time that top-down attention leads to a weighting of feature dimensions within a single object. Contrary to concurrent believes, attending to one dimension of a face does not imply the automatic selection of the un-attended dimension of the same face-object.