Summary
Most sign language understanding systems operate at the isolated-gesture level, limiting their usefulness in natural communication. This paper by Thanh-Hoang Nguyen Doan (arXiv:2607.09611) studies sentence-level sign language translation (SLT) with a focus on real-time deployment rather than novel architectures. The authors fine-tune a SHuBERT-ByT5 translation stack on a 9,872-example subset of How2Sign, achieving BLEU 16.7 on the validation set, BLEU 15.9 and BLEURT 44.7 on the test set. The main contribution is a hardware-aware streaming system: a Raspberry Pi 4B reference client handles camera capture, local text display, and speech output, while computationally intensive perception and translation run on CPU/GPU backends. Chunked ingestion, bounded queues, parallel perception, temporal re-ranking, and a sentence-boundary state machine reduce average post-finalization latency from 1.873 seconds to 1.354 seconds, a 27.71% improvement.
Paper Overview
Field: NLP
Author: Thanh-Hoang Nguyen Doan
Released: 2026-07-10
arXiv: 2607.09611
Abstract
Most sign language understanding systems operate at the isolated-gesture level, which limits their practical utility in natural communication. This paper investigates sentence-level sign language translation (SLT), with the primary goal of real-time deployment rather than proposing new architectures.
The authors fine-tune a SHuBERT-ByT5 translation stack on a 9,872-example subset of How2Sign, achieving:
System-level techniques include chunked ingestion, bounded queues, parallel perception, temporal re-ranking, and a sentence-boundary state machine. Together these reduce average post-finalization latency from 1.873 seconds to 1.354 seconds — a
27.71% improvement.
---
*Auto-collected on 2026-07-14.*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178395122