Elamur Research⌕
☰
home / papers / 2026 / Perceptual Validation of 3D Pose, Guided Sign Language Synthesis

Perceptual Validation of 3D Pose, Guided Sign Language Synthesis

Paper · 2026 · Ezekiel Maina, Lilian Wanzare and James Obuhuma

Summary

Sign language corpora are hard to share openly because video inevitably exposes the signer's face and body. Common fixes such as blurring also hide non-manual markers (NMMs) that carry grammar. The authors ask whether 3D pose data, rendered through avatars, can keep the linguistic content while removing identity. They worked with Kenyan Sign Language (KSL) material from the AI4KSL corpus. Sentence videos were cut into gloss-level clips in ELAN and filtered with a segment-integrity check. MediaPipe Holistic then extracted normalized 3D landmarks for the hands, torso and face, with facial features expressed as relative distances. A kinematically constrained mapper transferred these trajectories onto customizable, demographically diverse avatars. Evaluation combined geometric accuracy metrics with ratings from a large cohort of Deaf KSL users. Joint precision strongly predicted perceived gesture clarity, and varied avatar appearance made participants feel more included without reducing intelligibility. The work argues that pose abstraction can serve as a full corpus-level representation for privacy-respecting sign language resources, not just an anonymization trick.

Key points

  • Data: 11,000 KSL sentence videos from AI4KSL, recorded with 685 participants aged 15–24, were segmented in ELAN into 30,370 gloss-level clips. The annotation used three tiers: English Sentence, Gloss and Fingerspelling.
  • Data quality: The Segment Integrity Rate (SIR) check covered file size, frame count, H.264 compliance and timestamp drift. It reached 99.97%, rejecting only 8 clips.
  • Geometric fidelity: The pose pipeline achieved PCK = 92.7%, OKS = 0.88, PCP = 91.5% and PDJ > 85.3%.
  • Human evaluation: 384 Deaf KSL users rated anonymization, naturalness, gesture clarity and representativeness. Structural precision correlated with clarity at r = 0.76 (p < .01), and mean Gesture Clarity was 4.05.
  • Avatars vs. stick figures: Participants saw skeletal stick figures as technical artifacts but treated avatars as linguistic actors. Diverse avatar identities raised perceived inclusivity.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Maseno UniversityKisumu · Kenya2 works · 61 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksPsychology · 1,552 worksDevelopmental and Educational Psychology · 499 worksEngineering · 133 worksHuman-Computer Interaction · 141 worksControl and Systems Engineering · 45 works

Cite

Cite as Ezekiel Maina, Lilian Wanzare and James Obuhuma. 2026. Perceptual Validation of 3D Pose, Guided Sign Language Synthesis. Proceedings of the LREC 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/29qyih7xu7ym

@inproceedings{maina2026perceptual,
  title = {{Perceptual Validation of 3D Pose, Guided Sign Language Synthesis}},
  author = {Ezekiel Maina, Lilian Wanzare and James Obuhuma},
  booktitle = {Proceedings of the LREC 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion},
  year = {2026},
  doi = {10.63317/29qyih7xu7ym},
  url = {https://doi.org/10.63317/29qyih7xu7ym}
}

References

3 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Bragg, D., Koller, O., Bellard, M., Berke, L., Boudreault, P., Braffort, A., ... & Ringel Morris, M. (2019, October). Sign language recognition, generation, and translation: An interdisciplinary perspective. In Proceedings of the 21st international ACM SIGACCESS conference on computers and accessibi… — [doi:10.1145/3308561.3353](https://doi.org/10.1145/3308561.3353) find
  2. Cambridge University Press. Camgoz, N. C., Koller, O., Hadfield, S., & Bowden, R. (2020). Sign language transformers: Joint end-to-end sign language recognition and translation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 10023– 10033).https://doi.org/10… — [doi:10.48550/arxiv.2003](https://doi.org/10.48550/arxiv.2003) find
  3. Caselli, N. K., Sehyr, Z. S., Cohen-Goldberg, A. M., & Emmorey, K. (2017). ASL-LEX: A lexical database of American Sign Language. Behavior Research Methods, 49(2), 784– 801.https://doi.org/10.3758/s13428-016-0742-0 Chemnad, K., & Othman, A. (2025). Perception and monitoring of sign language acquisit… — [doi:10.3758/s13428-016-0742-0](https://doi.org/10.3758/s13428-016-0742-0) find

Updated

This page in Markdown → https://md.research.elamur.ai/free/en/papers/maina-2026-perceptual-validation-3d-pose-guided/