Tag
This paper proposes an end-to-end sequence-to-sequence approach for contextual Tamil spelling and grammar correction, using progressively fine-tuned mT5 and mBART models on synthetic data, achieving 69.3% exact-match accuracy on a diagnostic set.