A 48-class handwritten dataset for the endangered chakma language

Area of Science:

• Natural Language Processing
• Computer Vision
• Digital Humanities

Background:

• The Chakma language, spoken in Bangladesh, lacks sufficient digital resources.
• Handwritten Character Recognition (HCR) is crucial for preserving and digitizing endangered scripts.

Purpose of the Study:

• To create a comprehensive, open-access handwritten character dataset for the Chakma language.
• To facilitate the development of digital tools and linguistic research for Chakma.

Main Methods:

• Collected 37,708 handwritten samples from native Chakma speakers across diverse demographics.
• Digitized samples via high-resolution scanning, followed by automated cropping and resizing to 40x40 pixels.
• Organized data into 48 classes, including 38 letters and 10 numerals.

Main Results:

• Developed a standardized, machine-learning-ready dataset of Chakma handwritten characters.
• The dataset captures authentic stylistic variations in handwriting.
• The resource is now available for public use.

Conclusions:

• This dataset addresses a critical gap in digital resources for the Chakma language.
• It enables advancements in HCR, synthetic font generation, and linguistic analysis.
• Promotes the preservation and accessibility of indigenous scripts through technology.