Multimodal Affective Coherence: A Moment-Level Signal for Attunement in Clinical Interaction

Clinical assessment during emotionally loaded interactions relies on moment-to-moment judgments of attunement rather than static symptom counts or explicit emotional labels. Clinicians attend to whether what a person says aligns with how it is expressed across language, voice, facial behavior, timing, and engagement, particularly under conversationalpressure. These judgments are process-oriented, local in time, and rarely formalized computationally. We introduce Multimodal Affective Coherence (MAC), a moment-level signal that captures short-term coordination across heterogeneous affective channels within a single interaction. MAC defines a formally specified class of computable signals that quantify cross-modal affective alignment within short temporal windows. The formulation is method-flexible yet structurally constrained, allowing multiple instantiations while preserving a common alignment-based definition. By operationalizing attunement as relative coordination across modalities over time, MAC yields a time series that reflects regulation stability, strain, and recovery as interaction unfolds. This paper remains at the level of signal definition and does not specify implementation details. The formulation is agnostic to modalities, representations, similarity metrics, and architectures. We formalize MAC at the level of signal definition, specify its structural constraints, and describe an evaluation strategy using clinically aligned multimodal interview data. By focusing on within-session dynamics rather than long-horizon trajectories or diagnostic outcomes, MAC establishes a process-level representation of attunement aligned with clinical reasoning and extensible across research.