A computational signature of self-other mergence in Borderline Personality Disorder.

Abstract

A tendency to merge mental representations of self and other is thought to underpin the intense and unstable relationships that feature in Borderline Personality Disorder (BPD). However, clinical theories of BPD do not specify, in computational terms, how the perspectives of self and other might become confused. To address this question, we used a probabilistic false belief task (p-FBT) to examine how individuals with BPD (N = 38) and matched controls from the general population (N = 74) selectively assigned beliefs to self or other. The p-FBT requires participants to track a gradually changing quantity, whilst also predicting another person's belief about that quantity. We found that BPD participants showed less selectivity in belief assignment compared with controls (Cohen's d = 0.64). Behaviourally, participants with BPD tended to predict that others' beliefs resembled their own. Modelling analysis revealed that BPD participants were prone to generalise their own learning signals to others. Furthermore, this generalising tendency correlated with BPD symptomatology across participants, even when controlling for demographic factors and affective psychopathology. Our results support a computational account of self-other mergence, based on a generalisation of learning across agents. Self-other generalisation in learning purports to explain key clinical features of BPD, and suggests a potential transdiagnostic marker of mentalising capability.