Computational phenomenology of self and time in borderline and narcissistic personality disorders: cross-group supervised semantic differential

Bridging computational phenomenology with phenomenological psychopathology requires methods that preserve the lived, contextual texture of clinical narratives while still enabling statistical inference. The Supervised Semantic Differential (SSD) supports this goal by measuring and explaining how speakers construe target concepts from language, aggregating pretrained word embeddings from local contexts around lexically anchored concept references to derive interpretable semantic directions. Here, we adapt SSD for cross-group comparison by combining permutation-based inference with centroid-contrast interpretation of personal concept vectors. We apply the method to Polish life-story clinical interviews from in-patients with Borderline Personality Disorder (BPD), Narcissistic Personality Disorder (NPD), and a non-clinical control group ($N=62$), focusing on the concepts of self and time. Self-concept representations show robust omnibus and pairwise separations across all groups, indicating broad cross-diagnostic differences in narrative self-construal. In contrast, time-concept differences are driven primarily by BPD-related contrasts, with no reliable separation between NPD and controls, suggesting BPD-specific disruption of lived temporality. Qualitative cluster interpretation indicates that BPD narratives organize experience through relationally contextualized affective episodes and potentially crisis-linked temporal scaffolding (explicit dates, durations, and waiting), whereas NPD self-construal is comparatively decontextualized and self-referential, emphasizing regulation, stability/variability talk, and low-arousal negative affect. We relate these findings to existing phenomenological accounts of disturbed self-experience and temporality in personality disorders.