@inproceedings{preeti-etal-2026-tap,
title = "{TAP}: Policy Interfaces for Diffusion Language Models",
author = "Preeti and
Kushwaha, Ankita and
Ravish, Kiran and
Kumar, Pawan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2623/",
pages = "59616--59657",
ISBN = "979-8-89176-459-0",
abstract = "Diffusion language models denoise a masked canvas, making them a natural setting for selective generation decisions such as answering, waiting, clarifying, or refusing. We introduce TAP-DiLM, which adds a typed policy variable to masked diffusion LMs, and study how this policy should be attached to the denoising process. Across BERT-base, MDLM-OWT, LLaDA-8B, and Dream-7B, we compare shared-vocabulary slots, dedicated slot heads, and CLS-based external heads. The results reveal a clear interface-backbone interaction: dedicated heads provide robust policy learning across scales, while shared-vocabulary heads can be effective when the LM-head geometry is favorable. A simple LM-head sharpness probe predicts when shared-head policy internalization is reliable, giving a practical diagnostic for interface selection. Causal clamp and slot-injection tests show that dedicated policy slots can influence body denoising through feedback, especially at 8B scale. Safety-refusal experiments further show that TAP-DiLM supports competitive typed refusal behavior when trained with diverse unsafe prompts."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="preeti-etal-2026-tap">
<titleInfo>
<title>TAP: Policy Interfaces for Diffusion Language Models</title>
</titleInfo>
<name>
<namePart>Preeti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ankita</namePart>
<namePart type="family">Kushwaha</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kiran</namePart>
<namePart type="family">Ravish</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pawan</namePart>
<namePart type="family">Kumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Diffusion language models denoise a masked canvas, making them a natural setting for selective generation decisions such as answering, waiting, clarifying, or refusing. We introduce TAP-DiLM, which adds a typed policy variable to masked diffusion LMs, and study how this policy should be attached to the denoising process. Across BERT-base, MDLM-OWT, LLaDA-8B, and Dream-7B, we compare shared-vocabulary slots, dedicated slot heads, and CLS-based external heads. The results reveal a clear interface-backbone interaction: dedicated heads provide robust policy learning across scales, while shared-vocabulary heads can be effective when the LM-head geometry is favorable. A simple LM-head sharpness probe predicts when shared-head policy internalization is reliable, giving a practical diagnostic for interface selection. Causal clamp and slot-injection tests show that dedicated policy slots can influence body denoising through feedback, especially at 8B scale. Safety-refusal experiments further show that TAP-DiLM supports competitive typed refusal behavior when trained with diverse unsafe prompts.</abstract>
<identifier type="citekey">preeti-etal-2026-tap</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2623/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>59616</start>
<end>59657</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T TAP: Policy Interfaces for Diffusion Language Models
%A Kushwaha, Ankita
%A Ravish, Kiran
%A Kumar, Pawan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%A Preeti
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F preeti-etal-2026-tap
%X Diffusion language models denoise a masked canvas, making them a natural setting for selective generation decisions such as answering, waiting, clarifying, or refusing. We introduce TAP-DiLM, which adds a typed policy variable to masked diffusion LMs, and study how this policy should be attached to the denoising process. Across BERT-base, MDLM-OWT, LLaDA-8B, and Dream-7B, we compare shared-vocabulary slots, dedicated slot heads, and CLS-based external heads. The results reveal a clear interface-backbone interaction: dedicated heads provide robust policy learning across scales, while shared-vocabulary heads can be effective when the LM-head geometry is favorable. A simple LM-head sharpness probe predicts when shared-head policy internalization is reliable, giving a practical diagnostic for interface selection. Causal clamp and slot-injection tests show that dedicated policy slots can influence body denoising through feedback, especially at 8B scale. Safety-refusal experiments further show that TAP-DiLM supports competitive typed refusal behavior when trained with diverse unsafe prompts.
%U https://aclanthology.org/2026.emnlp-main.2623/
%P 59616-59657
Markdown (Informal)
[TAP: Policy Interfaces for Diffusion Language Models](https://aclanthology.org/2026.emnlp-main.2623/) (Preeti et al., EMNLP 2026)
ACL
- Preeti, Ankita Kushwaha, Kiran Ravish, and Pawan Kumar. 2026. TAP: Policy Interfaces for Diffusion Language Models. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 59616–59657, Budapest, Hungary. Association for Computational Linguistics.