Search papers, labs, and topics across Lattice.
This paper introduces Entropy-Valley (EV), a novel training-free length selector for masked diffusion language models (dLLMs) in machine translation, addressing the critical yet under-explored challenge of target length determination. By scoring candidate target canvases based on mean predictive entropy from all-mask forward passes, EV significantly improves translation adequacy, recovering up to 65.3% of the COMET-22 gain compared to traditional methods. Expert evaluations confirm that the approach enhances translation quality, particularly for Chinese-to-English translations, while demonstrating that the selection of target lengths is more impactful than the order of token unmasking.
Length selection in masked diffusion translation can recover over 65% of performance gains, shifting the focus from token order to canvas preparation.
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$\to$Zh, Zh$\to$En, and En$\to$De. Our diagnostics show that denoising-friendly lengths need not match reference lengths. Evaluation by three translation experts supports the En$\leftrightarrow$Zh adequacy gains, with stronger evidence on Zh$\to$En. Compared with a LLaMA-3-8B autoregressive (AR) model trained on the same fine-tuning data, the EV system ties on En$\to$Zh and leads on Zh$\to$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.
