Romy Beauté: Mapping of Subjective Accounts into Interpreted Clusters (MOSAIC)

This talk is part of the Paradigms in Consciousness Science Workshop of the Bamberg Mathematical Consciousness Science Initiative (BAMΞ): https://www.uni-bamberg.de/en/bamxi/r...

Abstract: To quantify subjective experience, consciousness science often relies on predefined measurement axes (standard questionnaires, canonical categories, or a small set of agreed-upon constructs). While these tools enable comparison and cumulative progress, they can also impose an ontology on the phenomenological territory, leaving potential novel or idiosyncratic aspects of experience uncaptured. In this talk, I present MOSAIC (Mapping Of Subjective Accounts into Interpreted Clusters), a computational pipeline that maps the latent structure of experience directly from free-text reports. By combining unsupervised topic modelling with Large Language Models (LLMs) for interpretable topic representations, MOSAIC supports a “wide-angle” discovery of experiential dimensions with reduced a priori constraints on expected phenomenological structure.

I illustrate the utility of this framework through two experimental studies. First, applied to the Dreamachine dataset (N=349 reports) - an art-science installation using music and stroboscopic lighting to induce visual hallucinations - MOSAIC suggests that stroboscopic stimulation induced not only simple geometric visuals, as traditionally reported in previous studies, but also richer themes such as complex imagery, narrative-like content, and broader altered-state descriptors. Second, analysing data from the Minimal Phenomenal Experience (MPE) project (n=830 reports), I use MOSAIC as a coverage diagnostic: mapping free reports of “pure awareness” to the MPE-M92 factor space reveals partial phenomenological overlap alongside substantial residue, with many descriptions remaining unclassified by the predefined dimensions.

Overall, the talk argues that computational phenomenology can serve as a rigorous complement to traditional psychometrics: it can test construct coverage, refine measurement ontologies, and help surface “unknown unknowns” in reported experience. I will close by briefly noting how the same discovery-first logic could inform future neurophysiological work, motivating wide-angle feature extraction and more cautious marker selection when linking brain dynamics to newly mapped experiential structure.