The first Dialectal Arabic Code Switching - DACS corpus from broadcast speech. Annotated at the token-level, considering both the linguistic and the acoustic cues. This dataset is a potential benchmark for DCS in spontaneous speech.

Fine-tuning NLLB-200 for Hinglish & Spanglish → English translation using the LinCE benchmark — with training, BLEU/ChrF/COMET evaluation, batch inference, and ONNX export.

EACL 2021 paper (SJ_AJ@DravidianLangTech-EACL2021: Task-Adaptive Pre-Training of Multilingual BERT models for Offensive Language Identification)

Predict whether individual words are ENG or MRI.

📖 A repository of all sociolinguistics data analyses and research papers written during my studies at NUS University Scholars Programme