savinugunarathna/singlish-to-sinhala-mt5-small

Singlish to Sinhala Translation Model (mT5-Small)

This model translates Singlish (romanized Sinhala mixed with English) to Sinhala script. Built on google/mt5-small.

Model Description

• Base Model: google/mt5-small
• Task: Translation (Singlish → Sinhala)
• Languages: Singlish (romanized Sinhala) → Sinhala (සිංහල)
• Training Date: 2026-01-16
• Architecture: Multilingual T5 (subword tokenization)

Training Details

• Dataset Size: ~490,000 translation pairs
• Data Source: Phonetic transcriptions + ad-hoc Singlish variants from Swa-bhasha Resource Hub
• Hardware: Tesla P100 GPU
• Framework: Hugging Face Transformers

Usage

Using Transformers Pipeline

from transformers import pipeline

translator = pipeline("translation", model="savinugunarathna/singlish-to-sinhala-mt5-small")
result = translator("translate Singlish to Sinhala: oyage nama mokakda")
print(result[0]["translation_text"])
# Output: ඔයාගේ නම මොකක්ද

Manual Loading

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("savinugunarathna/singlish-to-sinhala-mt5-small")
model = AutoModelForSeq2SeqLM.from_pretrained("savinugunarathna/singlish-to-sinhala-mt5-small")

input_text = "translate Singlish to Sinhala: mama pasal yanawa"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_length=80, num_beams=5)
translation = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(translation)
# Output: මම පාසල යනවා

Batch Translation

texts = [
    "translate Singlish to Sinhala: kohomada",
    "translate Singlish to Sinhala: mama hodata innawa",
    "translate Singlish to Sinhala: api yamu"
]

inputs = tokenizer(texts, return_tensors="pt", padding=True)
outputs = model.generate(**inputs, max_length=80, num_beams=5)

for i, output in enumerate(outputs):
    print(f"{texts[i].split(': ')[1]} → {tokenizer.decode(output, skip_special_tokens=True)}")

Example Translations

Singlish Input	Sinhala Output
oyage nama mokakda	ඔයාගේ නම මොකක්ද
api koheda yanne	අපි කොහෙද යන්නේ
kohomada	කොහොමද
mama hodata innawa	මම හොඳට ඉන්නවා

Model Capabilities

✅ Handles phonetic romanization (standard Latin script)
✅ Understands informal Singlish (conversational variations)
✅ Subword tokenization (efficient processing with mT5)
✅ Prefix-based translation (requires "translate Singlish to Sinhala:" prefix)

Limitations

• Performance may vary with non-standard Singlish spellings
• Best suited for conversational Singlish text
• Requires the prefix "translate Singlish to Sinhala:" for optimal results
• May struggle with very informal or heavily code-mixed text

Citations

If you use this model, please cite:

@misc{singlish-sinhala-mt5-20260116,
  author = {savinugunarathna},
  title = {Singlish to Sinhala Translation Model (mT5-Small)},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/savinugunarathna/singlish-to-sinhala-mt5-small}}
}

Data Source Citation

This model uses data from the Swa-bhasha Resource Hub:

@article{sumanathilaka2025swa,
  title={Swa-bhasha Resource Hub: Romanized Sinhala to Sinhala Transliteration Systems and Data Resources},
  author={Sumanathilaka, Deshan and Perera, Sameera and Dharmasiri, Sachithya and Athukorala, Maneesha and Herath, Anuja Dilrukshi and Dias, Rukshan and Gamage, Pasindu and Weerasinghe, Ruvan and Priyadarshana, YHPP},
  journal={arXiv preprint arXiv:2507.09245},
  year={2025}
}

License

Apache 2.0

Acknowledgments

• Base model: google/mt5-small
• Training data: Swa-bhasha Resource Hub (Sumanathilaka et al., 2025)
• Training framework: Hugging Face Transformers
• Compute: Tesla P100 GPU

Model Card Contact

For questions or issues, please open an issue in the model repository.