Qualia structures collapse for geometric shapes, but not faces, when spatial attention is withdrawn

Top-down attentional amplification is often assumed to affect 'what' we see, that is, the contents of conscious experience. Previously, this claim has been examined by studies that manipulated attention and characterized conscious perception in binary categorical labels (e.g. seen versus unseen). However, these categorical judgments are not powerful enough to characterize the quality of conscious perception, or 'how' we see, or qualia, for short. To address this, we introduce a similarity rating paradigm to consciousness research that tries to characterize the attentional effects on the structure of the quality of experience, or qualia structures for short. Under the dual-task paradigm, participants rated the similarity of stimulus pairs in the periphery. We used three stimulus sets, the rotated letters 'L' and 'T' (<i>N</i> = 14), rotated red/green bisected disks (<i>N</i> = 14) or greyscale faces (<i>N</i> = 13). The similarity ratings of all the pairs described the phenomenological relationships between the stimuli, and served as a proxy for the qualia structure of conscious experience of the stimuli; which we characterized with dimension reduction and an unsupervised optimal transport alignment technique. We found that alignment accuracy remained high for face qualia structures under both full and poor attention. Withdrawal of attention collapsed qualia structures for letters and disks. Extending previous dual-task approaches from binary categorizations to relational judgments, our approach establishes a novel pathway to elucidate qualia structures.