@inproceedings{de-mijolla-etal-2025-waste,
title = "Waste Not, Want Not; Recycled {G}umbel Noise Improves Consistency in Natural Language Generation",
author = "De Mijolla, Damien and
Saddiq, Hannan and
Moore, Kim",
editor = "Chiruzzo, Luis and
Ritter, Alan and
Wang, Lu",
booktitle = "Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)",
month = apr,
year = "2025",
address = "Albuquerque, New Mexico",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.naacl-long.292/",
doi = "10.18653/v1/2025.naacl-long.292",
pages = "5662--5686",
ISBN = "979-8-89176-189-6",
abstract = "Consistency in the output of language models is critical for their reliability and practical utility. Due to their training objective, language models learn to model the full space of possible continuations, leading to outputs that can vary significantly in style, content, and tone, even for similar inputs. To address this, we propose a novel decoding algorithm that enhances response consistency across different prompts with no degradation in response quality. By incorporating a latent variable into the next-token sampling process based on the Gumbel reparametrisation trick, our method outperforms standard sampling by up to 10{\%} across semantic and stylistic consistency benchmarks. Additionally, our approach integrates seamlessly with existing sampling methods with negligible computational overhead, providing a practical solution for improving the reliability of language model outputs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="de-mijolla-etal-2025-waste">
<titleInfo>
<title>Waste Not, Want Not; Recycled Gumbel Noise Improves Consistency in Natural Language Generation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Damien</namePart>
<namePart type="family">De Mijolla</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hannan</namePart>
<namePart type="family">Saddiq</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kim</namePart>
<namePart type="family">Moore</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-04</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Luis</namePart>
<namePart type="family">Chiruzzo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alan</namePart>
<namePart type="family">Ritter</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lu</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Albuquerque, New Mexico</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-189-6</identifier>
</relatedItem>
<abstract>Consistency in the output of language models is critical for their reliability and practical utility. Due to their training objective, language models learn to model the full space of possible continuations, leading to outputs that can vary significantly in style, content, and tone, even for similar inputs. To address this, we propose a novel decoding algorithm that enhances response consistency across different prompts with no degradation in response quality. By incorporating a latent variable into the next-token sampling process based on the Gumbel reparametrisation trick, our method outperforms standard sampling by up to 10% across semantic and stylistic consistency benchmarks. Additionally, our approach integrates seamlessly with existing sampling methods with negligible computational overhead, providing a practical solution for improving the reliability of language model outputs.</abstract>
<identifier type="citekey">de-mijolla-etal-2025-waste</identifier>
<identifier type="doi">10.18653/v1/2025.naacl-long.292</identifier>
<location>
<url>https://aclanthology.org/2025.naacl-long.292/</url>
</location>
<part>
<date>2025-04</date>
<extent unit="page">
<start>5662</start>
<end>5686</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Waste Not, Want Not; Recycled Gumbel Noise Improves Consistency in Natural Language Generation
%A De Mijolla, Damien
%A Saddiq, Hannan
%A Moore, Kim
%Y Chiruzzo, Luis
%Y Ritter, Alan
%Y Wang, Lu
%S Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)
%D 2025
%8 April
%I Association for Computational Linguistics
%C Albuquerque, New Mexico
%@ 979-8-89176-189-6
%F de-mijolla-etal-2025-waste
%X Consistency in the output of language models is critical for their reliability and practical utility. Due to their training objective, language models learn to model the full space of possible continuations, leading to outputs that can vary significantly in style, content, and tone, even for similar inputs. To address this, we propose a novel decoding algorithm that enhances response consistency across different prompts with no degradation in response quality. By incorporating a latent variable into the next-token sampling process based on the Gumbel reparametrisation trick, our method outperforms standard sampling by up to 10% across semantic and stylistic consistency benchmarks. Additionally, our approach integrates seamlessly with existing sampling methods with negligible computational overhead, providing a practical solution for improving the reliability of language model outputs.
%R 10.18653/v1/2025.naacl-long.292
%U https://aclanthology.org/2025.naacl-long.292/
%U https://doi.org/10.18653/v1/2025.naacl-long.292
%P 5662-5686
Markdown (Informal)
[Waste Not, Want Not; Recycled Gumbel Noise Improves Consistency in Natural Language Generation](https://aclanthology.org/2025.naacl-long.292/) (De Mijolla et al., NAACL 2025)
ACL