A dataset for Moroccan sign language recognition and translation

TL;DR

This paper introduces a new dataset for Moroccan Sign Language to help improve sign language recognition and translation technologies.

Contribution

The novelty lies in creating a systematically annotated and organized dataset for a low-resource sign language.

Findings

The dataset includes 2199 word-level signs recorded by nine native signers.

It covers a wide range of lexical categories including letters, numbers, and common words.

The dataset is publicly available to support inclusive communication tools for the deaf and hard-of-hearing.

Abstract

This paper introduces a new dataset of Moroccan Sign Language (MoSL) that aims for use in sign language recognition and translation research. The dataset has been created by recording 2199 word-level isolated signs by nine native signers in MP4 video format. It covers a broad spectrum of lexical categories like letters, numbers, pronouns, as well as frequently used day-to- day words. Sign language being a visual-gestural means of communication holds not just linguistic importance but also a cultural as well as regional identity. A language like MoSL that holds such significance constitutes a low-resource language with digital representation that can be scraped barely on currently available linguistic datasets. A solution towards this shortage in publicly available resources was found by offering systematically annotated as well as video file-organized datasets by means of an online repository. MoSL dataset thus counts as a useful dataset that can further computer vision as well as natural language processing task-oriented apps in sign language research leading towards more inclusive communication tools for deaf as well as hard-of-hearing populations.

Figures