Length-Adaptive Decoding for Masked Diffusion Machine Translation
On the map: Diffusion models → Discrete and text diffusion
Abstract (arXiv)
Machine translation tests masked diffusion language models (dLLMs) because every source token must be rendered faithfully, while fixed canvas decoding must choose target length before denoising. Existing masked diffusion decoding work mainly studies token unmasking order, leaving this length decision under-explored despite its direct effect on coverage and redundancy. We introduce Entropy-Valley (EV), a training-free length selector that scores candidate target canvases by mean predictive entropy from all-mask forward passes and selects the canvas the backbone is most prepared to fill. Relative to a baseline using training corpus length statistics, EV recovers 64.9%, 65.3%, and 33.0% of the COMET-22 gain from reference target lengths on En→Zh, Zh→En, and En→De. Our diagnostics show that denoising-friendly lengths need not match reference lengths. Evaluation by three translation experts supports the EnleftrightarrowZh adequacy gains, with stronger evidence on Zh→En. Compared with a LLaMA-3-8B autoregressive (AR) model trained on the same fine-tuning data, the EV system ties on En→Zh and leads on Zh→En; an oracle-length diagnostic further shows that, in this masked diffusion MT setting, deciding which tokens to reveal first matters less than how the target length is supplied.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Discrete and text diffusion, one of 14 directions in Diffusion models. Builds on 17 of its 27 references that are in the map. First posted on arXiv 23 Aug 2026.
Builds on
17 of 27 references.
Closest ancestors · 9
Same direction, newest first
- 2026Diffusion Language Models Are Natively Length-Aware6 Mar
- 2026Improving Variable-Length Generation in Diffusion Language Models via Length Regularization7 Feb
- 2026Unifying Masked Diffusion Models with Various Generation Orders and Beyond2 Feb
- 2026Diffusion LMs Can Approximate Optimal Infilling Lengths Implicitly31 Jan
- 2026ρ-textttEOS: Training-free Bidirectional Variable-Length Control for Masked Diffusion LLMs30 Jan
- 2025Any-Order Flexible Length Masked Diffusion31 Aug
- 2024Scaling Diffusion Language Models via Adaptation from Autoregressive Models23 Oct
- 2023Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution25 Oct
- 2022DiffuSeq: Sequence to Sequence Text Generation with Diffusion Models17 Oct
Other references · 8
- 2018A Call for Clarity in Reporting BLEU Scorescited by 1,799
- 2021Structured Denoising Diffusion Models in Discrete State-Spacescited by 1,719
- 2022Diffusion-LM Improves Controllable Text Generationcited by 1,027
- 2017Non-Autoregressive Neural Machine Translationcited by 463
- 2024Contrastive Preference Optimization: Pushing the Boundaries of LLM Performance in Machine Translationcited by 394
- 2024Tower: An Open Multilingual Large Language Model for Translation-Related Taskscited by 206
- 2020Glancing Transformer for Non-Autoregressive Neural Machine Translationcited by 103
- 2025Diffusion LLM with Native Variable Generation Lengths: Let [EOS] Lead the Waycited by 22
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026CARVE: Verified Expansion for Variable-Length Generation in Diffusion Language Models
- 2026Survival-Guided Length Control for Efficient Diffusion Language Models
- 2025Soft-Masked Diffusion Language Models
- 2026Re-evaluating Confidence Remasking in Masked Diffusion Language Models
- 2026Confidence-Based Decoding is Provably Efficient for Diffusion Language Models
- 2026Predict, Don't Iterate: Efficient Adaptive-Length Infilling for Diffusion Language Models
- 2025Beyond Static Cutoffs: One-Shot Dynamic Thresholding for Diffusion Language Models
- 2026Swordsman: Entropy-Driven Adaptive Block Partition for Efficient Diffusion Language Models