sk-community/romanized-bengali-transliterator-60M

Model Card for Romanized Bengali Transliterator

Model Summary

This model is a Marian-based Seq2Seq transliterator trained on the Romanized Bengali dataset (975k pairs).
It maps Bengali written in the Roman alphabet (Banglish) into native Bengali script.

• Architecture: MarianMT (Seq2Seq Transformer)
• Parameters: ~60M
• Training Data: romanized_bengali dataset (rule-based transliteration from Bengali Wikipedia)
• Languages: Bengali (Romanized ↔ Bangla script)

Intended Use

• Transliteration of Banglish into Bengali for:
  • Social media analytics
  • Search and information retrieval
  • Chatbots and assistants
  • Hate speech detection tasks

Example Usage

from transformers import MarianMTModel, MarianTokenizer

model_name = "sk-community/romanized_bengali_transliterator"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)

input_text = "tumi kemon acho"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
# Output: "তুমি কেমন আছো"

Performance

• BLEU (Bengali): 77.82
• CER: 0.16
• WER: 0.25

Citation

@article{gharami2025indotranslit,
  title={Modeling Romanized Hindi and Bengali: Dataset Creation and Multilingual LLM Integration},
  author={Kanchon Gharami and Quazi Sarwar Muhtaseem and Deepti Gupta and Lavanya Elluri and Shafika Showkat Moni},
  year={2025}
}