LMSpell

LMSpell is an open-source research project focused on neural spell correction for low-resource languages, with a particular focus on Sinhala. The project explores the use of pre-trained language models for spell correction and provides resources for researchers and developers working in multilingual and low-resource NLP. Our research was published at MERCon 2026 and presents an extensive evaluation of pre-trained language models for neural spell correction across multiple languages with special focus on Sinhala.

• 📄 Read the Research Paper on IEEE Xplore
• 📑 Read the Extended Preprint on arXiv

This organization hosts the fine-tuned models and datasets developed as part of the LMSpell research, along with datasets collected from previous research and other authors. If you use any of these datasets, please make sure to properly credit and cite the respective authors. The source code, training toolkit, experiments, and documentation are available on GitHub.

• 💻 View the LMSpell GitHub Repository

LMSpell aims to make neural spell correction more accessible for low-resource languages by providing reusable models, datasets, and tools for further research and development.