The Art of Socratic Inquiry: A Framework for Proactive Template-Guided Therapeutic Conversation Generation

TL;DR

The paper addresses the gap that psychological LLMs are largely reactive and fail to surface latent beliefs or guide behavior in CBT. It introduces the Socratic Inquiry Framework (SIF), a lightweight plug-and-play therapeutic intent planner that decouples when to ask from what to ask. SIF uses Strategy Anchoring and Template Retrieval to condition a conversation generator, trained with the Socratic-QA dataset. Experiments show that SIF increases proactive questioning, conversational depth, and therapeutic alignment, signaling a shift from passive responsiveness to active cognitive guidance.

Abstract

Proactive questioning, where therapists deliberately initiate structured, cognition-guiding inquiries, is a cornerstone of cognitive behavioral therapy (CBT). Yet, current psychological large language models (LLMs) remain overwhelmingly reactive, defaulting to empathetic but superficial responses that fail to surface latent beliefs or guide behavioral change. To bridge this gap, we propose the \textbf{Socratic Inquiry Framework (SIF)}, a lightweight, plug-and-play therapeutic intent planner that transforms LLMs from passive listeners into active cognitive guides. SIF decouples \textbf{when to ask} (via Strategy Anchoring) from \textbf{what to ask} (via Template Retrieval), enabling context-aware, theory-grounded questioning without end-to-end retraining. Complementing SIF, we introduce \textbf{Socratic-QA}, a high-quality dataset of strategy-aligned Socratic sequences that provides explicit supervision for proactive reasoning. Experiments show that SIF significantly enhances proactive questioning frequency, conversational depth, and therapeutic alignment, marking a clear shift from reactive comfort to proactive exploration. Our work establishes a new paradigm for psychologically informed LLMs: not just to respond, but to guide.

Figures (14)

• Figure 1: Comparison of dialogue control paradigms. (a) The standard Large Language Model paradigm: Operates reactively based on raw probabilities. (b) The soft-prompting paradigm: Relies on post-hoc rationale. (c) The decoupled planning paradigm (Ours): Introduces proactive planning.
• Figure 2: The proposed framework SIF integrates LPP module with CG module. Within LPP, the SA component first extracts conversational policies from seeker utterances; these policies then guide the TR module to select suitable question templates. The combined outputs of SA and TR condition the CG component, which is fine-tuned on our domain-specific Socratic-QA dataset to produce context-aware, goal-directed responses.
• Figure 3: An example of conversation generation. All conversations include predictions from both the SA and TR modules.
• Figure 4: Distribution statistics of Psy-Insight strategy.
• Figure 5: Distribution of multi-dimensional quality scores.