[passionate11]

This PR adds our ACL 2026 paper at the top of the Natural Language section (newest first), using the same 4-line entry format (title / authors / venue + links / arXiv date), so website/convert.py can parse it as usual.

• Empirical Analysis of Decoding Biases in Masked Diffusion Models (UNCODE), ACL 2026 Main, Oral (tagged (Oral) like other entries in the list)
• arXiv: https://arxiv.org/abs/2508.13021 (v1/v2 were titled "PC-Sampler: Position-Aware Calibration of Decoding Bias in Masked Diffusion Models")
• ACL Anthology: https://aclanthology.org/2026.acl-long.311/
• Project page: https://passionate11.github.io/Uncode-project-page/
• Code: https://github.com/NEUIR/Uncode

Gist: uncertainty-based decoding in masked diffusion language models has a rigid boundary bias and a trivial token bias; UNCODE is a training-free decoding calibration (positional-trajectory + content-informativeness priors) that improves over existing decoding strategies by more than 7% on reasoning/planning benchmarks across three MDMs.

I am the first author. Checked that the paper is not already listed.

[jxjhsk]

收到啦！