@inproceedings{peng-etal-2026-actually,
title = "What Actually Transfers? Diagnosing Output-Projection Initialization in Low-Resource Sign Language Translation",
author = "Peng, Kai-Yang and
Lim, Chunki and
Hsieh, Shu-Kai",
editor = "Singh, Sanjeet and
Joshi, Abhinav and
Dattatreya, Pokala and
Artiaga, Keren and
Pujara, Khushi Jaydeep and
{\v{Z}}elezn{\'y}, Tom{\'a}{\v{s}} and
Gruber, Ivan and
Honz{\'i}k, Jakub and
Kamila, Sabyasachi and
Quiroga, Facundo Manuel and
Hr{\'u}z, Marek and
Hasanuzzaman, Mohammed and
Modi, Ashutosh",
booktitle = "Proceedings of the Second Workshop on Sign Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.wslp-1.5/",
pages = "43--53",
ISBN = "979-8-89176-488-0",
abstract = "Released sign-to-text checkpoints combine visual and text-side parameters, so intact-checkpoint comparisons cannot identify what drives transfer. We study this confound in low-resource Taiwan Sign Language (TSL) by separating and recombining the visual and mT5 components of Uni-Sign checkpoints trained on Chinese Sign Language (CSL)-to-Chinese and American Sign Language (ASL)-to-English data, then fine-tuning the combinations identically on TSL. The CSL-derived visual initialization improves content-word accuracy by 15.0 points with its released mT5 parameters, but changes accuracy by only $-0.5$ points with untouched mT5-base, showing that the measured benefit of visual pretraining depends on the text-side initialization. We assess video dependence by replacing target clips with duration-matched held-out clips at evaluation. Further interventions localize much of the accuracy and video-dependence effects to the untied output projection: replacing an ASL-to-English model{'}s projection with mT5-base{'}s improves accuracy by 25.0 points and increases sensitivity to clip replacement. Permutation and timing controls show that the effect depends on pretrained row-to-token correspondence and on the projection being present during fine-tuning. Absolute translation quality remains low, so we interpret these experiments diagnostically. The results show that released checkpoints should be analyzed as structured initializations whose components can interact, rather than treated as indivisible evidence of cross-language transfer."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="peng-etal-2026-actually">
<titleInfo>
<title>What Actually Transfers? Diagnosing Output-Projection Initialization in Low-Resource Sign Language Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kai-Yang</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chunki</namePart>
<namePart type="family">Lim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shu-Kai</namePart>
<namePart type="family">Hsieh</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 Second Workshop on Sign Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sanjeet</namePart>
<namePart type="family">Singh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Abhinav</namePart>
<namePart type="family">Joshi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pokala</namePart>
<namePart type="family">Dattatreya</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Keren</namePart>
<namePart type="family">Artiaga</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Khushi</namePart>
<namePart type="given">Jaydeep</namePart>
<namePart type="family">Pujara</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tomáš</namePart>
<namePart type="family">Železný</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ivan</namePart>
<namePart type="family">Gruber</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jakub</namePart>
<namePart type="family">Honzík</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sabyasachi</namePart>
<namePart type="family">Kamila</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Facundo</namePart>
<namePart type="given">Manuel</namePart>
<namePart type="family">Quiroga</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marek</namePart>
<namePart type="family">Hrúz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammed</namePart>
<namePart type="family">Hasanuzzaman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ashutosh</namePart>
<namePart type="family">Modi</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">979-8-89176-488-0</identifier>
</relatedItem>
<abstract>Released sign-to-text checkpoints combine visual and text-side parameters, so intact-checkpoint comparisons cannot identify what drives transfer. We study this confound in low-resource Taiwan Sign Language (TSL) by separating and recombining the visual and mT5 components of Uni-Sign checkpoints trained on Chinese Sign Language (CSL)-to-Chinese and American Sign Language (ASL)-to-English data, then fine-tuning the combinations identically on TSL. The CSL-derived visual initialization improves content-word accuracy by 15.0 points with its released mT5 parameters, but changes accuracy by only -0.5 points with untouched mT5-base, showing that the measured benefit of visual pretraining depends on the text-side initialization. We assess video dependence by replacing target clips with duration-matched held-out clips at evaluation. Further interventions localize much of the accuracy and video-dependence effects to the untied output projection: replacing an ASL-to-English model’s projection with mT5-base’s improves accuracy by 25.0 points and increases sensitivity to clip replacement. Permutation and timing controls show that the effect depends on pretrained row-to-token correspondence and on the projection being present during fine-tuning. Absolute translation quality remains low, so we interpret these experiments diagnostically. The results show that released checkpoints should be analyzed as structured initializations whose components can interact, rather than treated as indivisible evidence of cross-language transfer.</abstract>
<identifier type="citekey">peng-etal-2026-actually</identifier>
<location>
<url>https://aclanthology.org/2026.wslp-1.5/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>43</start>
<end>53</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T What Actually Transfers? Diagnosing Output-Projection Initialization in Low-Resource Sign Language Translation
%A Peng, Kai-Yang
%A Lim, Chunki
%A Hsieh, Shu-Kai
%Y Singh, Sanjeet
%Y Joshi, Abhinav
%Y Dattatreya, Pokala
%Y Artiaga, Keren
%Y Pujara, Khushi Jaydeep
%Y Železný, Tomáš
%Y Gruber, Ivan
%Y Honzík, Jakub
%Y Kamila, Sabyasachi
%Y Quiroga, Facundo Manuel
%Y Hrúz, Marek
%Y Hasanuzzaman, Mohammed
%Y Modi, Ashutosh
%S Proceedings of the Second Workshop on Sign Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-488-0
%F peng-etal-2026-actually
%X Released sign-to-text checkpoints combine visual and text-side parameters, so intact-checkpoint comparisons cannot identify what drives transfer. We study this confound in low-resource Taiwan Sign Language (TSL) by separating and recombining the visual and mT5 components of Uni-Sign checkpoints trained on Chinese Sign Language (CSL)-to-Chinese and American Sign Language (ASL)-to-English data, then fine-tuning the combinations identically on TSL. The CSL-derived visual initialization improves content-word accuracy by 15.0 points with its released mT5 parameters, but changes accuracy by only -0.5 points with untouched mT5-base, showing that the measured benefit of visual pretraining depends on the text-side initialization. We assess video dependence by replacing target clips with duration-matched held-out clips at evaluation. Further interventions localize much of the accuracy and video-dependence effects to the untied output projection: replacing an ASL-to-English model’s projection with mT5-base’s improves accuracy by 25.0 points and increases sensitivity to clip replacement. Permutation and timing controls show that the effect depends on pretrained row-to-token correspondence and on the projection being present during fine-tuning. Absolute translation quality remains low, so we interpret these experiments diagnostically. The results show that released checkpoints should be analyzed as structured initializations whose components can interact, rather than treated as indivisible evidence of cross-language transfer.
%U https://aclanthology.org/2026.wslp-1.5/
%P 43-53
Markdown (Informal)
[What Actually Transfers? Diagnosing Output-Projection Initialization in Low-Resource Sign Language Translation](https://aclanthology.org/2026.wslp-1.5/) (Peng et al., WSLP 2026)
ACL