CASL-W60: A word-level dataset for central African sign language recognition

Area of Science:

• Linguistics
• Computer Science
• Artificial Intelligence

Background:

• Sign languages are vital non-verbal communication systems, carrying cultural and regional significance.
• Central African sign languages are linguistically distinct but lack sufficient digital resources for research.
• A gap exists in word-level datasets for machine learning, hindering AI development in this area.

Purpose of the Study:

• To introduce the Central African Sign Language Word-level 60 (CASL-W60) dataset.
• To address the underrepresentation of Central African sign languages in scientific literature and machine learning datasets.
• To facilitate research and development of AI applications for sign language.

Main Methods:

• Collected 60 word-level Central African sign language (CASL) signs from 19 volunteers.
• Recorded 10-12 video samples per word per signer, adhering to African sign language video standards.
• Organized MP4 video files into an online repository for accessibility.

Main Results:

• The CASL-W60 dataset contains 60 distinct word-level signs.
• Demonstrated the dataset's utility via successful word-level classification of the 60 signs.
• The dataset is available online for public use.

Conclusions:

• The CASL-W60 dataset is a valuable resource for advancing research in Central African sign languages.
• It supports the development of sign language translation, sentence recognition, and sign gloss detection tools.
• This initiative helps bridge the digital divide for underrepresented sign languages.