IsharaKotha: Avatar-based Bangla Sign Language Corpus

Read full text Summarise with VIRAT
Authors: MD. Ashikul Islam, Prato Dewan, Md Fuadul Islam, Md Ataullha Saim, M Shahidur Rahman
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

We present a dataset of Bangla Sign Language (BDSL) used to develop a translation system into a 3D avatar. We used the Hamburg Notation System (HamNoSys) to encode 3,823 BDSL lexical signs. The data collection process involved collaboration with BDSL experts and the Bangladesh Standards and Testing Institution (BSTI) to ensure the consistency of sign representations. Each sign was transcribed capturing essential linguistic features such as hand shape, hand orientation, movement, location, and palm position. The resulting dataset was converted to SiGML (Signing Gesture Markup Language) files for a real-time translation system that converts written Bangla words into sign language represented by a 3D avatar. The corpus covers 31 semantic categories organized into named folders including কাজ (Work), মানুষের চারিত্রিক বৈশিষ্ট্য (Human Characteristics), রাজনীতি (Politics), খাবার ও পানীয় (Food & Drinks), শিক্ষা (Education), and others. The dataset was validated by two certified sign language interpreters and one native deaf signer, achieving an average quality score of 3.14 out of 4.00. Applications for the dataset include sign language education and assistive technologies for the deaf community. The structured and annotated BDSL lexicon helps with the ongoing development of technologies that allow everyone to communicate and with making more progress in recognizing and synthesizing sign language.

Fields of Study & Subjects

Hand Gesture Recognition Systems Hearing Impairment and Communication Natural Language Processing Techniques

Publication details

Journal / source
Mendeley Data
Publication date
30 March 2026
Language
EN
Document type
Dataset