New framework automates sign language dataset creation for SASL

Researchers have developed a novel framework for automatically collecting and preparing sign language datasets with minimal human intervention, using South African Sign Language (SASL) as a case study. The Maswao dataset was created by gathering video content from public resources, transcribing accompanying audio using ASR tools, and identifying sentence boundaries through gradient descent. Benchmarking this dataset for Sign-to-text Translation revealed that Transformer models outperformed recurrent neural networks, achieving specific BLEU, ChrF, and ROUGE-L scores, indicating the task's potential for further research. AI

IMPACT This work could accelerate research and development in sign language processing by providing a scalable method for dataset creation.

RANK_REASON The item is an academic paper detailing a new dataset and methodology for sign language processing. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• American Sign Language
• British Sign Language
• German Sign Language
• Maswao
• Recurrent Neural Networks
• South African Sign Language
• Transformer Models

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →