Toward Anthropomorphic Dialogue: A Closed-Loop Framework for Human-Like Chat Generation, Evaluation, and Preference Alignment

2026-07-19

AI Summary

This work proposes AnthroDial, a closed-loop framework for building human-level private dialogue systems that exhibit personality consistency, relational memory, knowledge boundaries, and multi-turn coherence. The framework innovatively integrates generation, evaluation, and alignment modules through shared behavioral-dimensional representations. Key components include role-conditioned runtime scheduling, long-term memory with a virtual time mechanism, L0 sparsity-gated effectiveness assessment, GRPO-based reinforcement learning, and a Zone of Proximal Development (ZPD)-inspired cognitive diagnostic reward. A Kalman filter dynamically estimates agent capabilities and focuses learning on diagnosable weaknesses. Evaluated on a benchmark of 55 personalities across 50 scenarios, Qwen3.6-27B achieves a strict accuracy of 39.00% (total score: 98.5) after SFT and RL fine-tuning. Notably, a smaller 9B model improves dramatically from 0% to 18.37%, demonstrating the approach’s efficacy and scalability.

Abstract

Human-like private chat requires more than fluent response generation: a system must preserve persona, relationship, memory, bounded knowledge, medium-specific timing, and a coherent multi-turn arc. We present AnthroDial, a closed-loop framework that formulates anthropomorphic dialogue as a joint problem of system architecture, executable evaluation, and diagnostic alignment. It combines (1) a role-conditioned scheduled dialogue runtime with persona and scenario cards, long-term memory, virtual time, and single-draft message decisions; (2) an executable benchmark with an L0 validity gate, five per-turn dimensions, and five dialogue-level dimensions; and (3) a post-training pipeline that filters 16,436 scheduled-decision examples for SFT and applies GRPO with a cognitive-diagnostic, ZPD-aware reward. The reward maintains Kalman-filtered capability estimates for each behavioral dimension, upweights dimensions with larger capability deficits, and uses rollout scores as task-level ZPD matches to focus optimization on learnable weak skills. On a benchmark with 55 personas, 50 scenarios, 50 persona-scenario bindings, and 100 role-conditioned cases per model, we evaluate 16 systems spanning frontier baselines, open models, thinking/no-think variants, and SFT/RL ablations. The strongest non-trained baseline reaches 32.00% strict ACC, while Qwen3.6-27B-SFT+RL reaches 39.00% strict ACC and a 98.5 overall score. In the 9B no-think setting, SFT and RL improve strict ACC from 0.00% to 13.00% and 18.37%. These results show that anthropomorphic dialogue benefits when generation, evaluation, and reward shaping share the same behavioral dimensions.

Problem

• anthropomorphic dialogue
• persona consistency
• multi-turn coherence
• human-like chat
• dialogue evaluation

Innovation

• anthropomorphic dialogue
• closed-loop framework
• cognitive-diagnostic reward
• ZPD-aware optimization
• executable evaluation