Saved in:
Bibliographic Details
Main Authors: Chen, Kuanwei, Tsai, Mengfeng
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.09554
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910207147245568
author Chen, Kuanwei
Tsai, Mengfeng
author_facet Chen, Kuanwei
Tsai, Mengfeng
contents Sign Language Translation (SLT) converts sign language videos into spoken-language text, bridging communication between Deaf and hearing communities. Current gloss-free approaches rely on large encoder-decoder models, limiting deployment. We propose a compact 77M-parameter pipeline that couples MMPose skeletal pose extraction with a single linear projection into T5-small. By varying the input frame rate, we expose a practical efficiency trade-off: at 12 fps the model halves its sequence length, achieving a 75% reduction in encoder quadratic self-attention computational complexity while incurring only a modest BLEU-4 drop (9.53 vs. 10.06 at 24 fps on How2Sign). Our system is roughly 3x smaller than prior T5-base systems, demonstrating that a lightweight architecture can remain competitive without hierarchical encoders or large-scale models.
format Preprint
id arxiv_https___arxiv_org_abs_2605_09554
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Towards Compact Sign Language Translation: Frame Rate and Model Size Trade-offs
Chen, Kuanwei
Tsai, Mengfeng
Computation and Language
Computer Vision and Pattern Recognition
Sign Language Translation (SLT) converts sign language videos into spoken-language text, bridging communication between Deaf and hearing communities. Current gloss-free approaches rely on large encoder-decoder models, limiting deployment. We propose a compact 77M-parameter pipeline that couples MMPose skeletal pose extraction with a single linear projection into T5-small. By varying the input frame rate, we expose a practical efficiency trade-off: at 12 fps the model halves its sequence length, achieving a 75% reduction in encoder quadratic self-attention computational complexity while incurring only a modest BLEU-4 drop (9.53 vs. 10.06 at 24 fps on How2Sign). Our system is roughly 3x smaller than prior T5-base systems, demonstrating that a lightweight architecture can remain competitive without hierarchical encoders or large-scale models.
title Towards Compact Sign Language Translation: Frame Rate and Model Size Trade-offs
topic Computation and Language
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2605.09554