@inproceedings{wu-etal-2026-knowledge-still,
title = "Knowledge is Still There: Why {RL} Beats {SFT} for Sustainable Continual Machine Translation",
author = "Wu, Junhong and
Zhao, Yang and
Xu, Yangyifan and
Zong, Chengqing",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.2340/",
pages = "50302--50320",
ISBN = "979-8-89176-462-0",
abstract = "Traditional research on Continual Machine Translation (CMT) has long been predicated on the assumption that catastrophic forgetting stems from the irreversible erasure of underlying linguistic knowledge. In this work, we challenge this conventional wisdom within the context of Large Language Models (LLMs). Through systematic behavioral and representational analyses, we demonstrate that the model{'}s latent knowledge often remains intact, yet its ability to align this knowledge with instructions becomes compromised. We attribute this '' expressive drift'' to the ubiquitous Supervised Fine-Tuning (SFT) paradigm. By minimizing token-level loss against fixed references, SFT neglects the inherent plurality of valid translations for a single source, inadvertently coercing the model toward narrow external distributions that conflict with its internal linguistic priors. To rectify this, we advocate for a transition from SFT to a Reinforcement Learning (RL) paradigm, which optimizes performance directly within the model{'}s own generative distribution. Our results show that RL significantly mitigates forgetting in both translation-specific and general tasks by preserving the model{'}s expressive stability. This work redefines the fundamental challenge of CMT and positions RL as a more principled approach for sustainable multilingual alignment in the era of LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wu-etal-2026-knowledge-still">
<titleInfo>
<title>Knowledge is Still There: Why RL Beats SFT for Sustainable Continual Machine Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Junhong</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yangyifan</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chengqing</namePart>
<namePart type="family">Zong</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Traditional research on Continual Machine Translation (CMT) has long been predicated on the assumption that catastrophic forgetting stems from the irreversible erasure of underlying linguistic knowledge. In this work, we challenge this conventional wisdom within the context of Large Language Models (LLMs). Through systematic behavioral and representational analyses, we demonstrate that the model’s latent knowledge often remains intact, yet its ability to align this knowledge with instructions becomes compromised. We attribute this ” expressive drift” to the ubiquitous Supervised Fine-Tuning (SFT) paradigm. By minimizing token-level loss against fixed references, SFT neglects the inherent plurality of valid translations for a single source, inadvertently coercing the model toward narrow external distributions that conflict with its internal linguistic priors. To rectify this, we advocate for a transition from SFT to a Reinforcement Learning (RL) paradigm, which optimizes performance directly within the model’s own generative distribution. Our results show that RL significantly mitigates forgetting in both translation-specific and general tasks by preserving the model’s expressive stability. This work redefines the fundamental challenge of CMT and positions RL as a more principled approach for sustainable multilingual alignment in the era of LLMs.</abstract>
<identifier type="citekey">wu-etal-2026-knowledge-still</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2340/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>50302</start>
<end>50320</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Knowledge is Still There: Why RL Beats SFT for Sustainable Continual Machine Translation
%A Wu, Junhong
%A Zhao, Yang
%A Xu, Yangyifan
%A Zong, Chengqing
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F wu-etal-2026-knowledge-still
%X Traditional research on Continual Machine Translation (CMT) has long been predicated on the assumption that catastrophic forgetting stems from the irreversible erasure of underlying linguistic knowledge. In this work, we challenge this conventional wisdom within the context of Large Language Models (LLMs). Through systematic behavioral and representational analyses, we demonstrate that the model’s latent knowledge often remains intact, yet its ability to align this knowledge with instructions becomes compromised. We attribute this ” expressive drift” to the ubiquitous Supervised Fine-Tuning (SFT) paradigm. By minimizing token-level loss against fixed references, SFT neglects the inherent plurality of valid translations for a single source, inadvertently coercing the model toward narrow external distributions that conflict with its internal linguistic priors. To rectify this, we advocate for a transition from SFT to a Reinforcement Learning (RL) paradigm, which optimizes performance directly within the model’s own generative distribution. Our results show that RL significantly mitigates forgetting in both translation-specific and general tasks by preserving the model’s expressive stability. This work redefines the fundamental challenge of CMT and positions RL as a more principled approach for sustainable multilingual alignment in the era of LLMs.
%U https://aclanthology.org/2026.findings-emnlp.2340/
%P 50302-50320
Markdown (Informal)
[Knowledge is Still There: Why RL Beats SFT for Sustainable Continual Machine Translation](https://aclanthology.org/2026.findings-emnlp.2340/) (Wu et al., Findings 2026)
ACL