SYNTHESIS NOTE
TopicsPsychology Therapy Practicethis note

Can we measure therapist-patient alliance from dialogue turns in real time?

Explores whether computational methods can detect working alliance quality at turn-level resolution during therapy sessions, enabling immediate feedback on whether the therapeutic relationship is strengthening.

Synthesis note · 2026-02-23 · sourced from Psychology Therapy Practice
What makes therapeutic chatbots actually work in clinical practice? How do you build domain expertise into general AI models?

COMPASS uses sentence embeddings (SentenceBERT, 384-dimensional) to project each dialogue turn onto representations of the 36-item Working Alliance Inventory. The result: a 36-dimensional working alliance score for every patient and therapist turn, decomposable into three subscales — task (collaborative nature), bond (affective connection), and goal (agreement on objectives). Combined with Temporal Topic Modeling using the Embedded Topic Model (ETM), this produces turn-resolution topic scores that track conversation focus over time.

Analyzing 950+ sessions across anxiety, depression, schizophrenia, and suicidality reveals condition-specific dynamics. Anxiety and depression sessions show convergence in bond and task scales as therapy progresses — a positive signal of alliance formation. Schizophrenia and suicidality sessions do not show this convergence. Suicidality trajectories are notably more spread out in bond and task scales, indicating significant patient-therapist misalignment.

The interpretable output identifies actionable patterns: discussing "Emotional States and Mental Health" increases task and bond scales for depression but decreases them for suicidality. Topic-to-alliance mapping enables therapists to identify which conversational strategies are working or failing for each condition — something previously requiring clinical intuition.

Since Can conversation structure predict dialogue success better than content?, alliance trajectories may represent a domain-specific instance of a general phenomenon: the shape of the conversation carries diagnostic information independent of content. The therapeutic application — real-time feedback on whether alliance is forming or deteriorating — is more clinically mature than general conversational geometry, because it maps onto a validated clinical construct (WAI).

Inquiring lines that read this note 41

This note is a source for these research framings, grouped by the broader line of inquiry each explores. Scan the bold lines of inquiry; follow any specific question forward.

How can real-time alliance measurement improve therapy outcomes? Why do LLM chatbots fail as independent therapeutic agents? How do evaluation biases undermine LLM quality assessment systems? Why should disagreement be treated as signal in collaborative reasoning? How should conversational agents balance goal-driven initiative with user control? How can language models sustain linguistic synchrony and intersubjectivity during dialogue? Can single-axis benchmarks accurately predict agent deployment success? How do LLMs distinguish causal reasoning from temporal and semantic associations? How do we evaluate AI systems when user perception misleads actual performance? How should dialogue recommender systems manage conversation history and state?

Related concepts in this collection 3

This note in its neighbourhood — explore the map, then jump to a related concept in the list below.

Concept map
14 direct connections · 71 in 2-hop network ·medium cluster Open in graph ↗

Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph

your link semantically near linked from elsewhere

Related papers in this collection 8

Papers most semantically related to this note, ranked by cosine similarity in the embedding space.

Original note title

working alliance can be computationally inferred from session transcripts at turn-level resolution — enabling real-time therapist feedback