Computer Science > Computation and Language
arXiv:2607.26178 (cs)
Submitted on 28 Jul 2026
Title: DuplexGen: Adaptive Synthesis of Human-AI Turn-Taking Dialogues
Authors: Takyoung Kim, Kang-wook Kim, Sang Hoon Woo, Julia Hirschberg, Gunhee Kim, Dilek Hakkani-Tür
Abstract
Turn-taking is a fundamental component of full-duplex interaction. The appropriateness of specific turn-taking behaviors depends heavily on the interaction scenario, yet existing models apply a uniform norm across diverse contexts. This limitation stems from their training data: human-human speech corpora capture natural timing phenomena but offer little role grounding or scenario-specific norms. Conversely, heuristic or prompted synthesis methods inject turn-taking behaviors without grounding them in human preferences.
We introduce DuplexGen, a framework that generates dialogues with scenario-adaptive turn-taking by calibrating large language model (LLM) predictions against a small set of slot-level human preference annotations. Across six cooperative and competitive tasks, human turn-taking preferences differ systematically. DuplexGen aligns substantially more closely with these preferences than uncalibrated prompting or training solely on generic human-human data. A full-duplex model trained on DuplexGen-generated data exhibits distinctive, human-preferred turn-taking behaviors.
These results demonstrate that human calibration—rather than corpus scale or prompt design alone—is the key to enabling scenario-specific turn-taking synthesis.
Comments: Manuscript under review
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2607.26178 [cs.CL] (or arXiv:2607.26178v1 [cs.CL] for this version)
DOI: https://doi.org/10.48550/arXiv.2607.26178 (pending registration)
via ArXiv CL