@inproceedings{vandendriessche-etal-2026-grounding,
title = "Grounding Sign Language Representation Learning in Phonology",
author = "Vandendriessche, Toon and
De Coster, Mathieu and
Dambre, Joni",
editor = "Efthimiou, Eleni and
Fotinea, Stavroula-Evita and
Hanke, Thomas and
Hochgesang, Julie A. and
Mesch, Johanna and
Schulder, Marc",
booktitle = "Proceedings of the {LREC} 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion",
month = may,
year = "2026",
address = "Palma, Mallorca (Spain)",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.signlang-1.48/",
doi = "10.63317/2kyvv756bfmz",
pages = "468--476",
abstract = "Sign language recognition systems are commonly trained using gloss-level supervision, treating signs as holistic lexical units. While effective for classification, such approaches entangle sub-lexical structure and fail to capture the phonological parameters that govern sign formation, limiting interpretability, robustness, and cross-lingual transfer. In this work, we propose a phonologically informed representation learning architecture that explicitly structures the latent space according to linguistic principles. Grounded in the Dependency Model {--} a phonological model used to describe Flemish Sign Language (VGT) {--} our hierarchical architecture disentangles parameter-specific subspaces for handshape and location and is trained with multi-label phoneme supervision. To evaluate whether phonological information is directly encoded in the geometry of the embedding space, we introduce a non-parametric probing method that measures neighbourhood consistency across increasing scales. We show that conventional gloss-based networks achieve reasonable performance only for very small neighbourhoods, reflecting incidental visual similarity. In contrast, our disentangled representations maintain stable performance for larger neighbourhoods. This behaviour indicates that phonological structure is preserved across broader regions of the space, yielding more coherent and robust embeddings. Together, our results show that explicit phonological supervision {--} and crucially, disentangled representation learning {--} provides a principled foundation for interpretable and transferable sign language representations. Keywords: Sign Language, Machine Learning"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="vandendriessche-etal-2026-grounding">
<titleInfo>
<title>Grounding Sign Language Representation Learning in Phonology</title>
</titleInfo>
<name type="personal">
<namePart type="given">Toon</namePart>
<namePart type="family">Vandendriessche</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mathieu</namePart>
<namePart type="family">De Coster</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joni</namePart>
<namePart type="family">Dambre</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the LREC 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion</title>
</titleInfo>
<name type="personal">
<namePart type="given">Eleni</namePart>
<namePart type="family">Efthimiou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stavroula-Evita</namePart>
<namePart type="family">Fotinea</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thomas</namePart>
<namePart type="family">Hanke</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Julie</namePart>
<namePart type="given">A</namePart>
<namePart type="family">Hochgesang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Johanna</namePart>
<namePart type="family">Mesch</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marc</namePart>
<namePart type="family">Schulder</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resources Association (ELRA)</publisher>
<place>
<placeTerm type="text">Palma, Mallorca (Spain)</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Sign language recognition systems are commonly trained using gloss-level supervision, treating signs as holistic lexical units. While effective for classification, such approaches entangle sub-lexical structure and fail to capture the phonological parameters that govern sign formation, limiting interpretability, robustness, and cross-lingual transfer. In this work, we propose a phonologically informed representation learning architecture that explicitly structures the latent space according to linguistic principles. Grounded in the Dependency Model – a phonological model used to describe Flemish Sign Language (VGT) – our hierarchical architecture disentangles parameter-specific subspaces for handshape and location and is trained with multi-label phoneme supervision. To evaluate whether phonological information is directly encoded in the geometry of the embedding space, we introduce a non-parametric probing method that measures neighbourhood consistency across increasing scales. We show that conventional gloss-based networks achieve reasonable performance only for very small neighbourhoods, reflecting incidental visual similarity. In contrast, our disentangled representations maintain stable performance for larger neighbourhoods. This behaviour indicates that phonological structure is preserved across broader regions of the space, yielding more coherent and robust embeddings. Together, our results show that explicit phonological supervision – and crucially, disentangled representation learning – provides a principled foundation for interpretable and transferable sign language representations. Keywords: Sign Language, Machine Learning</abstract>
<identifier type="citekey">vandendriessche-etal-2026-grounding</identifier>
<identifier type="doi">10.63317/2kyvv756bfmz</identifier>
<location>
<url>https://aclanthology.org/2026.signlang-1.48/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>468</start>
<end>476</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Grounding Sign Language Representation Learning in Phonology
%A Vandendriessche, Toon
%A De Coster, Mathieu
%A Dambre, Joni
%Y Efthimiou, Eleni
%Y Fotinea, Stavroula-Evita
%Y Hanke, Thomas
%Y Hochgesang, Julie A.
%Y Mesch, Johanna
%Y Schulder, Marc
%S Proceedings of the LREC 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca (Spain)
%F vandendriessche-etal-2026-grounding
%X Sign language recognition systems are commonly trained using gloss-level supervision, treating signs as holistic lexical units. While effective for classification, such approaches entangle sub-lexical structure and fail to capture the phonological parameters that govern sign formation, limiting interpretability, robustness, and cross-lingual transfer. In this work, we propose a phonologically informed representation learning architecture that explicitly structures the latent space according to linguistic principles. Grounded in the Dependency Model – a phonological model used to describe Flemish Sign Language (VGT) – our hierarchical architecture disentangles parameter-specific subspaces for handshape and location and is trained with multi-label phoneme supervision. To evaluate whether phonological information is directly encoded in the geometry of the embedding space, we introduce a non-parametric probing method that measures neighbourhood consistency across increasing scales. We show that conventional gloss-based networks achieve reasonable performance only for very small neighbourhoods, reflecting incidental visual similarity. In contrast, our disentangled representations maintain stable performance for larger neighbourhoods. This behaviour indicates that phonological structure is preserved across broader regions of the space, yielding more coherent and robust embeddings. Together, our results show that explicit phonological supervision – and crucially, disentangled representation learning – provides a principled foundation for interpretable and transferable sign language representations. Keywords: Sign Language, Machine Learning
%R 10.63317/2kyvv756bfmz
%U https://aclanthology.org/2026.signlang-1.48/
%U https://doi.org/10.63317/2kyvv756bfmz
%P 468-476
Markdown (Informal)
[Grounding Sign Language Representation Learning in Phonology](https://aclanthology.org/2026.signlang-1.48/) (Vandendriessche et al., SignLang 2026)
ACL
- Toon Vandendriessche, Mathieu De Coster, and Joni Dambre. 2026. Grounding Sign Language Representation Learning in Phonology. In Proceedings of the LREC 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion, pages 468–476, Palma, Mallorca (Spain). ELRA Language Resources Association (ELRA).