@inproceedings{zhang-etal-2026-layer,
title = "Layer-aware Dual-directional Modulation for Low-resource Machine Translation",
author = "Zhang, Siqi and
Song, Ran and
Jiang, Shuting and
Huang, Yuxin and
Yu, Zhengtao",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.1054/",
pages = "20988--21000",
ISBN = "979-8-89176-395-1",
abstract = "Although Large Language Models (LLMs) have achieved remarkable success in Machine Translation (MT), a significant performance gap persists between high-resource and low-resource languages due to imbalanced pre-training data. In this paper, we first investigate the internal mechanisms driving this performance disparity from a layer-wise perspective.We propose a metric termed Activation Disparity ($\Delta R$) to quantify the activation divergence between high- and low-resource MT. Based on this metric, we distinguish between Task-Adaptive Layers (TAL, $\Delta R > 0$) that encode task-specific signals and Legacy-Inert Layers (LIL, $\Delta R < 0$) dominated by pre-trained bias. Leveraging this finding, we propose the \textbf{L}ayer-\textbf{a}ware \textbf{D}ual-directional \textbf{M}odulation (\textbf{LaDM}). Integrated with Low-Rank Adaptation (LoRA), LaDM employs a sparse strategy to bidirectionally modulate optimization dynamics. Specifically, it amplifies contributions from TAL to accelerate feature consolidation while inhibiting LIL to dampen misaligned legacy biases. Extensive experiments on Chinese-to-seven low-resource language translation using Llama-3.1, Qwen2.5, and Gemma-2 demonstrate that LaDM significantly outperforms standard LoRA fine-tuning, achieving an average improvement of 1.73 spBLEU.Code is available at \url{https://github.com/zzssqqq/LaDM}."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-layer">
<titleInfo>
<title>Layer-aware Dual-directional Modulation for Low-resource Machine Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Siqi</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ran</namePart>
<namePart type="family">Song</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuting</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuxin</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhengtao</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Although Large Language Models (LLMs) have achieved remarkable success in Machine Translation (MT), a significant performance gap persists between high-resource and low-resource languages due to imbalanced pre-training data. In this paper, we first investigate the internal mechanisms driving this performance disparity from a layer-wise perspective.We propose a metric termed Activation Disparity (Δ R) to quantify the activation divergence between high- and low-resource MT. Based on this metric, we distinguish between Task-Adaptive Layers (TAL, Δ R > 0) that encode task-specific signals and Legacy-Inert Layers (LIL, Δ R < 0) dominated by pre-trained bias. Leveraging this finding, we propose the Layer-aware Dual-directional Modulation (LaDM). Integrated with Low-Rank Adaptation (LoRA), LaDM employs a sparse strategy to bidirectionally modulate optimization dynamics. Specifically, it amplifies contributions from TAL to accelerate feature consolidation while inhibiting LIL to dampen misaligned legacy biases. Extensive experiments on Chinese-to-seven low-resource language translation using Llama-3.1, Qwen2.5, and Gemma-2 demonstrate that LaDM significantly outperforms standard LoRA fine-tuning, achieving an average improvement of 1.73 spBLEU.Code is available at https://github.com/zzssqqq/LaDM.</abstract>
<identifier type="citekey">zhang-etal-2026-layer</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.1054/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>20988</start>
<end>21000</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Layer-aware Dual-directional Modulation for Low-resource Machine Translation
%A Zhang, Siqi
%A Song, Ran
%A Jiang, Shuting
%A Huang, Yuxin
%A Yu, Zhengtao
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F zhang-etal-2026-layer
%X Although Large Language Models (LLMs) have achieved remarkable success in Machine Translation (MT), a significant performance gap persists between high-resource and low-resource languages due to imbalanced pre-training data. In this paper, we first investigate the internal mechanisms driving this performance disparity from a layer-wise perspective.We propose a metric termed Activation Disparity (Δ R) to quantify the activation divergence between high- and low-resource MT. Based on this metric, we distinguish between Task-Adaptive Layers (TAL, Δ R > 0) that encode task-specific signals and Legacy-Inert Layers (LIL, Δ R < 0) dominated by pre-trained bias. Leveraging this finding, we propose the Layer-aware Dual-directional Modulation (LaDM). Integrated with Low-Rank Adaptation (LoRA), LaDM employs a sparse strategy to bidirectionally modulate optimization dynamics. Specifically, it amplifies contributions from TAL to accelerate feature consolidation while inhibiting LIL to dampen misaligned legacy biases. Extensive experiments on Chinese-to-seven low-resource language translation using Llama-3.1, Qwen2.5, and Gemma-2 demonstrate that LaDM significantly outperforms standard LoRA fine-tuning, achieving an average improvement of 1.73 spBLEU.Code is available at https://github.com/zzssqqq/LaDM.
%U https://aclanthology.org/2026.findings-acl.1054/
%P 20988-21000
Markdown (Informal)
[Layer-aware Dual-directional Modulation for Low-resource Machine Translation](https://aclanthology.org/2026.findings-acl.1054/) (Zhang et al., Findings 2026)
ACL