@inproceedings{najafi-fyshe-2026-offline,
title = "Offline Preference Optimization via Maximum Marginal Likelihood Estimation",
author = "Najafi, Saeed and
Fyshe, Alona",
editor = "Demberg, Vera and
Inui, Kentaro and
Marquez, Llu{\'i}s",
booktitle = "Proceedings of the 19th Conference of the {E}uropean Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = mar,
year = "2026",
address = "Rabat, Morocco",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.eacl-long.318/",
pages = "6751--6764",
ISBN = "979-8-89176-380-7",
abstract = "Aligning Large Language Models (LLMs) with human preferences is crucial, but standard methods like Reinforcement Learning from Human Feedback (RLHF) are often complex and unstable. In this work, we propose a new, simpler approach that recasts alignment through the lens of Maximum Marginal Likelihood (MML) estimation. Our new MML-based Preference Optimization (MMPO) maximizes the marginal log-likelihood of a preferred text output, using the preference pair as samples for approximation, and forgoes the need for both an explicit reward model and entropy maximization. We theoretically demonstrate that MMPO implicitly performs preference optimization, producing a weighted gradient that naturally up-weights chosen responses over rejected ones. Across models ranging from 135M to 8B parameters, we empirically show that MMPO: 1) is more stable with respect to the hyperparameter compared to alternative baselines, and 2) achieves competitive or superior preference alignment while better preserving the base model{'}s general language capabilities. Through a series of ablation experiments, we show that this improved performance is indeed attributable to MMPO{'}s implicit preference optimization within the gradient updates."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="najafi-fyshe-2026-offline">
<titleInfo>
<title>Offline Preference Optimization via Maximum Marginal Likelihood Estimation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Saeed</namePart>
<namePart type="family">Najafi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alona</namePart>
<namePart type="family">Fyshe</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-03</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Vera</namePart>
<namePart type="family">Demberg</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kentaro</namePart>
<namePart type="family">Inui</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lluís</namePart>
<namePart type="family">Marquez</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Rabat, Morocco</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-380-7</identifier>
</relatedItem>
<abstract>Aligning Large Language Models (LLMs) with human preferences is crucial, but standard methods like Reinforcement Learning from Human Feedback (RLHF) are often complex and unstable. In this work, we propose a new, simpler approach that recasts alignment through the lens of Maximum Marginal Likelihood (MML) estimation. Our new MML-based Preference Optimization (MMPO) maximizes the marginal log-likelihood of a preferred text output, using the preference pair as samples for approximation, and forgoes the need for both an explicit reward model and entropy maximization. We theoretically demonstrate that MMPO implicitly performs preference optimization, producing a weighted gradient that naturally up-weights chosen responses over rejected ones. Across models ranging from 135M to 8B parameters, we empirically show that MMPO: 1) is more stable with respect to the hyperparameter compared to alternative baselines, and 2) achieves competitive or superior preference alignment while better preserving the base model’s general language capabilities. Through a series of ablation experiments, we show that this improved performance is indeed attributable to MMPO’s implicit preference optimization within the gradient updates.</abstract>
<identifier type="citekey">najafi-fyshe-2026-offline</identifier>
<location>
<url>https://aclanthology.org/2026.eacl-long.318/</url>
</location>
<part>
<date>2026-03</date>
<extent unit="page">
<start>6751</start>
<end>6764</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Offline Preference Optimization via Maximum Marginal Likelihood Estimation
%A Najafi, Saeed
%A Fyshe, Alona
%Y Demberg, Vera
%Y Inui, Kentaro
%Y Marquez, Lluís
%S Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 March
%I Association for Computational Linguistics
%C Rabat, Morocco
%@ 979-8-89176-380-7
%F najafi-fyshe-2026-offline
%X Aligning Large Language Models (LLMs) with human preferences is crucial, but standard methods like Reinforcement Learning from Human Feedback (RLHF) are often complex and unstable. In this work, we propose a new, simpler approach that recasts alignment through the lens of Maximum Marginal Likelihood (MML) estimation. Our new MML-based Preference Optimization (MMPO) maximizes the marginal log-likelihood of a preferred text output, using the preference pair as samples for approximation, and forgoes the need for both an explicit reward model and entropy maximization. We theoretically demonstrate that MMPO implicitly performs preference optimization, producing a weighted gradient that naturally up-weights chosen responses over rejected ones. Across models ranging from 135M to 8B parameters, we empirically show that MMPO: 1) is more stable with respect to the hyperparameter compared to alternative baselines, and 2) achieves competitive or superior preference alignment while better preserving the base model’s general language capabilities. Through a series of ablation experiments, we show that this improved performance is indeed attributable to MMPO’s implicit preference optimization within the gradient updates.
%U https://aclanthology.org/2026.eacl-long.318/
%P 6751-6764
Markdown (Informal)
[Offline Preference Optimization via Maximum Marginal Likelihood Estimation](https://aclanthology.org/2026.eacl-long.318/) (Najafi & Fyshe, EACL 2026)
ACL