@inproceedings{kachi-etal-2026-akibanlp,
title = "{A}kiba{NLP}-{TUT}: Improving Low-Resource Machine Translation via Language-Specific and Length-Adaptive Noise Injection",
author = "Kachi, Takumi and
Akiba, Tomoyosi and
Tsukada, Hajime",
editor = "Haddow, Barry and
Kocmi, Tom and
Koehn, Philipp and
Monz, Christof",
booktitle = "Proceedings of the Eleventh Conference on Machine Translation",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.wmt-main.151/",
pages = "2124--2134",
ISBN = "979-8-89176-474-3",
abstract = "We present a language-specific word-level noise injection method for low-resource machine translation that eliminates the need for an external monolingual corpus by constructing the low-resource language vocabulary directly from the source-side text of the target dataset. We also investigate two important hyperparameters of the original method: the maximum edit distance used for candidate word selection and the target noise ratio. Specifically, we propose a dynamic edit-distance constraint based on word length and evaluate multiple noise ratios from 5{\%} to 30{\%}. Experiments on Assamese-English translation show that the proposed dynamic constraint consistently outperforms the fixed edit-distance setting across all evaluated noise ratios, while a moderate noise ratio achieves the best translation performance. We further report our official results for the WMT 2026 Low-Resource Indic Language Translation Shared Task and discuss the effectiveness and limitations of the proposed approach under different resource conditions."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kachi-etal-2026-akibanlp">
<titleInfo>
<title>AkibaNLP-TUT: Improving Low-Resource Machine Translation via Language-Specific and Length-Adaptive Noise Injection</title>
</titleInfo>
<name type="personal">
<namePart type="given">Takumi</namePart>
<namePart type="family">Kachi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tomoyosi</namePart>
<namePart type="family">Akiba</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hajime</namePart>
<namePart type="family">Tsukada</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 Eleventh Conference on Machine Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Barry</namePart>
<namePart type="family">Haddow</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tom</namePart>
<namePart type="family">Kocmi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Philipp</namePart>
<namePart type="family">Koehn</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christof</namePart>
<namePart type="family">Monz</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">979-8-89176-474-3</identifier>
</relatedItem>
<abstract>We present a language-specific word-level noise injection method for low-resource machine translation that eliminates the need for an external monolingual corpus by constructing the low-resource language vocabulary directly from the source-side text of the target dataset. We also investigate two important hyperparameters of the original method: the maximum edit distance used for candidate word selection and the target noise ratio. Specifically, we propose a dynamic edit-distance constraint based on word length and evaluate multiple noise ratios from 5% to 30%. Experiments on Assamese-English translation show that the proposed dynamic constraint consistently outperforms the fixed edit-distance setting across all evaluated noise ratios, while a moderate noise ratio achieves the best translation performance. We further report our official results for the WMT 2026 Low-Resource Indic Language Translation Shared Task and discuss the effectiveness and limitations of the proposed approach under different resource conditions.</abstract>
<identifier type="citekey">kachi-etal-2026-akibanlp</identifier>
<location>
<url>https://aclanthology.org/2026.wmt-main.151/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>2124</start>
<end>2134</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T AkibaNLP-TUT: Improving Low-Resource Machine Translation via Language-Specific and Length-Adaptive Noise Injection
%A Kachi, Takumi
%A Akiba, Tomoyosi
%A Tsukada, Hajime
%Y Haddow, Barry
%Y Kocmi, Tom
%Y Koehn, Philipp
%Y Monz, Christof
%S Proceedings of the Eleventh Conference on Machine Translation
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-474-3
%F kachi-etal-2026-akibanlp
%X We present a language-specific word-level noise injection method for low-resource machine translation that eliminates the need for an external monolingual corpus by constructing the low-resource language vocabulary directly from the source-side text of the target dataset. We also investigate two important hyperparameters of the original method: the maximum edit distance used for candidate word selection and the target noise ratio. Specifically, we propose a dynamic edit-distance constraint based on word length and evaluate multiple noise ratios from 5% to 30%. Experiments on Assamese-English translation show that the proposed dynamic constraint consistently outperforms the fixed edit-distance setting across all evaluated noise ratios, while a moderate noise ratio achieves the best translation performance. We further report our official results for the WMT 2026 Low-Resource Indic Language Translation Shared Task and discuss the effectiveness and limitations of the proposed approach under different resource conditions.
%U https://aclanthology.org/2026.wmt-main.151/
%P 2124-2134
Markdown (Informal)
[AkibaNLP-TUT: Improving Low-Resource Machine Translation via Language-Specific and Length-Adaptive Noise Injection](https://aclanthology.org/2026.wmt-main.151/) (Kachi et al., WMT 2026)
ACL