Vedika - Sanskrit NLP Toolkit

Vedika is a comprehensive toolkit for Sanskrit text processing, offering deep learning-based tools for sandhi splitting and joining, text normalization, sentence splitting, syllabification, and tokenization.

Features

• Sandhi Processing
  • Split compound Sanskrit words using attention-based neural networks
  • Join Sanskrit words with proper sandhi rules
  • Support for beam search to get multiple suggestions
• Text Processing
  • Syllabification
  • Tokenization
  • Sentence splitting
  • Text normalization

Installation

# Install from PyPI (Soon Coming)
pip install vedika

# Install from source
git clone https://github.com/tanuj437/vedika.git
cd vedika
pip install -e .

Requirements

• Python >= 3.8
• PyTorch >= 1.9.0
• NumPy >= 1.19.0
• Pandas >= 1.3.0
• tqdm >= 4.62.0
• regex >= 2021.8.3

Quick Start

Sandhi Splitting

from vedika import SanskritSplit

# Initialize splitter
splitter = SanskritSplit()

# Split a single word
result = splitter.split("रामायणम्")
print(result['split'])  # Output: राम + अयन + अम्

# Batch processing
words = ["रामायणम्", "गीतागोविन्दम्"]
results = splitter.split_batch(words)
for result in results:
    print(f"{result['input']} → {result['split']}")

Sandhi Joining

from vedika import SandhiJoiner

# Initialize joiner
joiner = SandhiJoiner()

# Join split words
result = joiner.join("राम+अस्ति")
print(result)  # Output: रामास्ति

# Batch processing
texts = ["राम+अस्ति", "गच्छ+अमि"]
results = joiner.join_batch(texts)
print(results)  # ['रामास्ति', 'गच्छामि']

Advanced Usage

Beam Search for Multiple Suggestions

# Get multiple suggestions with beam search
result = splitter.split("रामायणम्", beam_size=3)
print(f"Best split: {result['split']}")
print(f"Confidence: {result['confidence']}")
print("Alternatives:")
for alt in result['alternatives']:
    print(f"- {alt['split']} (confidence: {alt['confidence']})")

Model Information

# Get model details
info = splitter.get_model_info()
print(f"Vocabulary size: {info['vocabulary_size']}")
print(f"Device: {info['device']}")
print(f"Configuration: {info['model_config']}")

Project Structure

vedika/
├── __init__.py
├── normalizer.py
├── sandhi_join.py
├── sandhi_split.py
├── sentence_splitter.py
├── syllabification.py
├── tokenizer.py
└── data/
    ├── cleaned_metres.json
    ├── sandhi_joiner.pth
    └── sandhi_split.pth

Model Architecture

The sandhi processing models use:

• Bidirectional LSTM encoder
• GRU decoder with attention
• Multi-head attention mechanism
• Character-level processing

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Authors

• Tanuj Saxena
• Soumya Sharma

Citation

If you use Vedika in your research, please cite:

@software{vedika2025,
  title={Vedika: A Sanskrit Text Processing Toolkit},
  author={Saxena, Tanuj and Sharma, Soumya},
  year={2025},
  url={https://github.com/tanuj437/vedika}
}

Contact

• Email: tanuj.saxena.rks@gmail.com, soumyasharma1599@gmail.com