DA-DLM: Explicitly Modeling Token Dependencies in Diffusion Language Models

Method and contribution

Adapts directed-acyclic dependency modeling to iterative diffusion using position-oriented candidate groups and fixed-token anchors.

A direct transfer of structured output modeling from non-autoregressive translation to diffusion denoising.

Read in context

Structured Decoding: Higher-Order CRFs, Diffusion Language Models, and Candidate Lattices →

Citation

Pengyu Ji, Zichen Zhang, Xiang Hu, Kewei Tu. DA-DLM: Explicitly Modeling Token Dependencies in Diffusion Language Models. EMNLP 2026.

@inproceedings{ji2026dadlm,
  title = {{DA-DLM: Explicitly Modeling Token Dependencies in Diffusion Language Models}},
  author = {Ji, Pengyu and Zhang, Zichen and Hu, Xiang and Tu, Kewei},
  year = {2026},
  booktitle = {EMNLP 2026},
  url = {https://aclanthology.org/2026.emnlp-main.2094/},
  eprint = {2609.15070},
  archivePrefix = {arXiv},
  doi = {10.18653/2026.emnlp-main.2094}
}