@inproceedings{zhong-etal-2026-parallelism,
    title = "Parallelism and Generation Order in Masked Diffusion Language Models: Limits Today, Potential Tomorrow",
    author = "Zhong, Yangyang  and
      Gu, Yanmei  and
      Zang, Zhengqing  and
      Li, Xiaomeng  and
      Ding, Yuqi  and
      Jia, Xibei  and
      Shen, Yuting  and
      Lan, Zhenzhong  and
      Zhu, Liwang  and
      Liu, Weiping  and
      Zhou, Junlin  and
      Liu, Haisheng  and
      Yu, Zhong Xin  and
      Luo, Pengxin  and
      Qi, Donglian  and
      Yan, Yunfeng  and
      Zhao, Junbo",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.357/",
    pages = "7178--7218",
    ISBN = "979-8-89176-395-1",
    abstract = "Masked Diffusion Language Models (MDLMs) promise parallel token generation and arbitrary-order decoding, yet it remains unclear to what extent current models truly realize these capabilities. We characterize MDLM behavior along two dimensions{---}parallelism strength and generation order{---}using Average Finalization Parallelism (AFP) and Kendall{'}s {\ensuremath{\tau}}. We evaluate eight mainstream MDLMs (up to 100B parameters) on 58 benchmarks spanning knowledge, reasoning, and programming. The results show that MDLMs still lag behind comparably sized autoregressive models, mainly because parallel probabilistic modeling weakens inter-token dependencies. Meanwhile, MDLMs exhibit adaptive decoding behavior: their parallelism and generation order vary significantly with the task domain, the stage of reasoning, and whether the output is correct. On tasks that require ``backward information'' (e.g., Sudoku), MDLMs adopt a solution order that tends to fill easier Sudoku blanks first, highlighting their advantages. Finally, we provide theoretical motivation and design insights supporting a Generate-then-Edit paradigm, which mitigates dependency loss while retaining the efficiency of parallel decoding."
}