A Dataset Development for Fake News Detection in Low-Resource Romanized Manipuri

In today’s digital landscape, the rapid spread of fake news presents significant challenges, particularly in multilingual nations like India. Manipuri, the primary language of Manipur in Northeast India, has seen a surge in online usage and digital adaptation, making it easier for Manipuri-speaking communities to engage with each other. However, despite this growth, there is a lack of resources for identifying fake news in Manipuri, especially in its Romanized form. This study focuses on developing a dataset to detect fake news in Romanized Manipuri, especially in the sports domain. Since there were no existing datasets for this language, the dataset was compiled by collecting articles from credible Manipuri news sources, transliterating them with contributors, and labelling each as either ‘fake’ or ‘real’ to aid in effective detection. Every step, from collecting data to labelling it, was done carefully by hand to ensure accuracy. Statistical tests were also performed to find important language patterns that help distinguish between real and fake news. Our findings show the unique language features of Romanized Manipuri and provide a solid foundation for future work in developing machine-learning models for detecting misinformation. By developing this dataset, we aim to improve the availability of reliable information for Manipuri-speaking communities.