Mapping the Jingle–Jangle Jungle of Social-Learning Constructs: A Semantic–Psychometric Study of Mentalising, Epistemic Trust and Emotion Regulation

Contemporary models of developmental psychopathology increasingly view vulnerability to mental health problems as emerging from disruptions in social learning processes rather than from isolated symptoms or static traits. Within this framework, social learning is the interpersonal process through which individuals acquire, update, and generalise knowledge about themselves, others, and the social world via communication with trusted others. This process relies on the coordinated functioning of mentalising, understood as the capacity to interpret behaviour in terms of intentional mental states; epistemic trust, defined as openness to socially conveyed information as relevant, reliable, and generalisable across contexts; and emotion regulation, which maintains affective arousal within bounds that enable reflective processing rather than defensive or automatic responding. From this perspective, disruptions in these processes hinder social learning and increase vulnerability to psychopathology (Fonagy et al., 2019; Luyten et al., 2020; Fonagy et al., 2024). Despite their shared theoretical grounding, these constructs have largely been operationalised through self-report questionnaires developed within partially independent research traditions. As a result, the field faces ongoing risks of conceptual and measurement ambiguity, including jingle–jangle fallacies, in which distinct constructs are assessed with semantically overlapping items, or, conversely, similar psychological phenomena are labelled and measured as separate constructs (Hanfstingl et al., 2024; Weidman et al., 2017). This problem has been explicitly noted in recent “state-of-the-field” discussions in developmental psychopathology, which call for greater conceptual clarity and methodological integration across levels of analysis (Weidman et al., 2017; Hanfstingl et al., 2024; Fonagy et al., 2024; Fonagy et al., 2025). At the same time, advances in computational and network-based approaches offer new tools for examining psychological constructs beyond traditional latent variable models, particularly in contexts where constructs are theoretically interdependent and empirically overlapping (Borsboom et al., 2021; Zavlis et al., 2025). Recent work has highlighted the potential of such approaches, including Large Language Models (LLMs), to clarify construct boundaries and relations, while also emphasising the importance of transparency and preregistration to mitigate analytic flexibility and post hoc interpretation (Demszky et al., 2023; Wulff & Mata, 2025). Study Aims and Method The present study addresses these challenges by integrating psychometric and computational semantic methods to examine the correspondence between how social-learning constructs are described linguistically and how they are expressed empirically in participant responses. Focusing on three widely used self-report instruments assessing mentalising (Reflective Functioning Questionnaire for Youth [RFQY-13]; Martin-Gagnon et al., 2023), epistemic trust (Epistemic Trust, Mistrust and Credulity Questionnaire [ETMCQ]; Campbell et al., 2021), and emotion regulation (Difficulties in Emotion Regulation Scale [DERS-18]; Victor & Klonsky, 2016), the study investigates these constructs at the item level in a sample of emerging adults. Using multilingual sentence-embedding models (Hussain et al., 2024; Wulff & Mata, 2025), the study first quantifies semantic similarity between questionnaire item wordings, constructing a semantic similarity space that captures linguistic proximity between items. In parallel, a psychometric similarity space is derived from empirical item-level associations based on participant responses. The primary aim is to evaluate the degree of alignment between these two spaces—referred to as structural fidelity—and to identify systematic mismatches between semantic and psychometric similarity. Such mismatches are interpreted as potential indicators of jingle or jangle fallacies at the measurement level, where linguistic similarity does not align with empirical similarity, or vice versa (Hanfstingl et al., 2024). By formally testing the association between semantic and psychometric similarity matrices, the study provides an empirical assessment of the extent to which current questionnaire-based operationalisations of social-learning constructs align with their linguistic content. In addition, the study examines the clinical relevance of semantic–psychometric mismatch by testing whether items or item configurations characterised by greater mismatch are differentially associated with impairments in self and interpersonal functioning, assessed via the Level of Personality Functioning Scale – Brief Form (LPFS-BF; Hutsebaut et al., 2016; Wu et al., 2024). This step is informed by transdiagnostic models that conceptualise personality functioning as a core dimension of psychopathology and a downstream manifestation of disrupted social learning (Fonagy et al., 2024). Rather than testing predictions about specific effect sizes or directional outcomes, the study is designed to clarify the measurement structure of key social-learning constructs. Expected outcomes are framed in terms of (a) mapping semantic overlap and distinctions across instruments, (b) empirically assessing the correspondence between linguistic and behavioural similarity, and (c) identifying regions of conceptual redundancy or divergence with implications for theory, measurement refinement, and future process-based research. All analytic decisions relevant to hypothesis testing are specified a priori to ensure transparency and reproducibility.