Enhanced MarianMT Indonesian-English Translation (Meeting Domain Adaptation)

This model is an enhanced fine-tuned version of Helsinki-NLP/opus-mt-id-en with domain-specific adaptation for meeting and business contexts.

🎯 Model Highlights

• Domain Adaptation: Specialized for meeting and business translation
• Enhanced Dataset: TED2020 + 2000+ meeting-specific sentence pairs
• Improved Performance: Better BLEU scores on meeting contexts
• Robust Training: 80% dataset usage with domain mixing
• Production Ready: Optimized for real-world meeting scenarios

📊 Performance Metrics

🚀 Model Details

• Base Model: Helsinki-NLP/opus-mt-id-en
• Training Dataset: TED2020 (80%) + Meeting Domain (10%)
• Training Strategy: Domain adaptation with enhanced learning
• Specialization: Business meetings, technical discussions, formal conversations
• Training Date: 2025-05-27
• Languages: Indonesian (id) → English (en)
• License: Apache 2.0

🛠️ Usage

from transformers import MarianMTModel, MarianTokenizer

# Load model and tokenizer
model_name = "dhintech/marian-ted2020-id-en-lg"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)

# Translate Indonesian to English
def translate(text):
 inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=128)
 outputs = model.generate(
 **inputs,
 max_length=128,
 num_beams=3,
 early_stopping=True,
 do_sample=False
 )
 return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Example usage
indonesian_text = "Tim marketing akan bertanggung jawab untuk strategi ini."
english_translation = translate(indonesian_text)
print(english_translation)
# Output: "The marketing team will be responsible for this strategy."

📝 Example Translations

Meeting Context Examples

🎯 Intended Use Cases

• Business Meeting Translation: Real-time translation during meetings
• Technical Documentation: Translating technical meeting notes
• Corporate Communication: Formal business correspondence
• Project Management: Translating project updates and reports
• Training Materials: Educational and training content translation

📊 Training Configuration

• Dataset Size: 118,626 sentence pairs
• TED2020 Data: 80% of cleaned dataset
• Meeting Domain Data: 10% specialized meeting content
• Max Sequence Length: 128 tokens
• Training Epochs: 12
• Learning Rate: 1e-05
• Batch Size: 12 (effective)

🔧 Technical Specifications

• Model Architecture: MarianMT (Transformer-based)
• Parameters: ~74M (with selective fine-tuning)
• Max Input/Output Length: 128 tokens
• Inference Time: ~0.14s per sentence
• Memory Requirements:
  • GPU: 3GB VRAM minimum
  • CPU: 4GB RAM minimum

🚨 Limitations

• Domain Specificity: Optimized for formal business/meeting contexts
• Informal Language: May not perform optimally on very casual Indonesian
• Regional Dialects: Trained primarily on standard Indonesian
• Cultural Context: Some cultural nuances may be lost in translation

📚 Citation

@misc{enhanced-marian-id-en-2025,
 title={Enhanced MarianMT Indonesian-English Translation (Meeting Domain Adaptation)},
 author={DhinTech},
 year={2025},
 publisher={Hugging Face},
 journal={Hugging Face Model Hub},
 howpublished={\url{https://huggingface.co/dhintech/marian-id-en-enhanced}},
 note={Enhanced with TED2020 and meeting-specific domain adaptation}
}

🙏 Acknowledgments

• Base Model: Helsinki-NLP team for the original opus-mt-id-en model
• Dataset: TED2020 corpus and custom meeting domain data
• Framework: Hugging Face Transformers team

---

This model is specifically enhanced for Indonesian business meeting translation scenarios with domain adaptation techniques.

Base model

Helsinki-NLP/opus-mt-id-en

(18)