A rule-based Sanskrit Computational Linguistics engine that executes full-sentence Padanta Sandhi (पदान्त सन्धि) according to Panini's Ashtadhyayi with step-by-step grammatical derivations (प्रक्रिया).
Live Demo • Features • Installation • Python Usage • Supported Sutras
सन्धीराट् (Sandhify) is an open-source Sanskrit computational engine designed to conjugate space-separated Sanskrit words (Padas) in complete sentences according to classical Paninian grammar (Vyakarana).
Unlike statistical/ML models that produce hallucinations, सन्धीराट् is 100% rule-based and deterministic, providing:
- The exact sandhi-fied text (सन्धियुक्त रूपम्).
- A summary table of all word boundaries analyzed.
- A complete step-by-step grammatical derivation (Prakriya / प्रक्रिया) citing the exact Ashtadhyayi Sutra numbers (e.g., इको यणचि ६.१.७७, झलां जशोऽन्ते ८.२.३९).
- 📜 Strict Paninian Compliance: Implements classical rules for Ach Sandhi (vowels), Hal Sandhi (consonants), and Visarga Sandhi.
- 🔍 Step-by-Step Prakriya (प्रक्रिया): View intermediate phonetic transformations with referenced Sutra numbers at each transition.
- ⚙️ Configurable Vaikalpika (Optional) Sandhis: Toggle optional rules directly from the sidebar:
- लोपः शाकल्यस्य (8.3.19) — द्वावपि vs द्वा अपि
- वा शरि (8.3.36) — तपस्स्वाध्याय vs तपः स्वाध्याय
- यरोऽनुनासिकेऽनुनासिको वा (8.4.45) — एतद् मुरारिः vs एतन्मुरारिः
- शशछोऽटि (8.4.63) — तद् शिवः vs तच्छिवः
- झयो होऽन्यतरस्याम् (8.4.62) — वाग् हरिः vs वाग्घरिः
- वा पदान्तस्य (8.4.59) — सं कल्पः vs सङ्कल्पः
- 🌐 Multi-Script Transliteration: Instant script conversion powered by Aksharamukha (Devanagari, IAST, Telugu, Kannada, Malayalam, Bengali, Grantha, etc.).
- 🌐 Bilingual Interface: Toggle the entire UI seamlessly between संस्कृतम् and English.
-
Clone the repository:
git clone https://github.com/samvadah/sandhi.git cd sandhi -
Create a virtual environment & install dependencies:
python -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate pip install -r requirements.txt
-
Launch the web app:
streamlit run app.py
You can import and use the engine in your own Sanskrit NLP pipelines, bots, or research scripts without launching the web UI:
from core import vaakya_sandhi
# Input Sanskrit sentence (space-separated words)
sentence = "अत्र अपि मुनिः उवाच तत् शिवः सम् कल्पः"
# Optional Paninian settings (all default to True / False as standard)
settings = {
"lopa_shakalyasya": True,
"vaa_shari": True,
"yaro_anunasike": True,
"shashcho_ati": True,
"jhayo_ho": True,
"anusvarasya_yayi": True,
}
# Generate Sandhi, Summary DataFrame, and Step-by-Step Prakriya DataFrame
result_text, summary_df, prakriya_df = vaakya_sandhi(sentence, settings=settings, lang="संस्कृतम्")
print("Sandhi Output:")
print(result_text)
# Output: अत्रापि मुनिरुवाच तच्छिवः सङ्कल्पः
print("\nGrammatical Derivations (Prakriya):")
print(prakriya_df)| Sutra (सूत्रम्) | Number | Rule Type | Description |
|---|---|---|---|
| इको यणचि | 6.1.77 | अच् सन्धि | Ik vowels replaced by Yan before vowels |
| एचोऽयवायावः | 6.1.78 | अच् सन्धि | Ech vowels replaced by ay, av, āy, āv |
| आद्गुणः | 6.1.87 | अच् सन्धि | a/ā + vowel results in Guna |
| वृद्धिरेचि | 6.1.88 | अच् सन्धि | a/ā + Ech results in Vriddhi |
| अकः सवर्णे दीर्घः | 6.1.101 | अच् सन्धि | Homorganic vowels lengthen (Savarna Deergha) |
| एङः पदान्तादति | 6.1.109 | अच् सन्धि | e/o + short a results in Avagraha (ऽ) |
| अतो रोरप्लुतादप्लुते | 6.1.113 | विसर्ग सन्धि | aḥ + a turns into o'ऽ |
| हशि च | 6.1.114 | विसर्ग सन्धि | aḥ + voiced consonant turns into o |
| एतत्तदोः सुलोपः... | 6.1.132 | विसर्ग सन्धि | Dropping of s/ḥ for saḥ / eṣaḥ before consonants |
| छे च | 6.1.73 | हल् सन्धि | Short vowel + ch adds augment c (Tuk) |
| रो रि / ढ्रलोपे... | 8.3.14 / 6.3.111 | विसर्ग सन्धि | Elision of r before r and lengthening of prior vowel |
| ससजुषो रुः | 8.2.66 | विसर्ग सन्धि | Pada-final s transforms into Ru (r) |
| झलां जशोऽन्ते | 8.2.39 | हल् सन्धि | Non-nasal stops turn into voiced unaspirated stops (Jash) |
| मोऽनुस्वारः | 8.3.23 | हल् सन्धि | Pada-final m becomes Anusvara (ṃ) before consonants |
| खरवसानयोर्विसर्जनीयः | 8.3.15 | विसर्ग सन्धि | r becomes Visarga (ḥ) before Khar or pause |
| लोपः शाकल्यस्य | 8.3.19 | अच् सन्धि | Optional elision of y/v before vowels |
| स्तोः श्चुना श्चुः | 8.4.40 | हल् सन्धि | Dental to Palatal assimilation (Schutva) |
| ष्टुना ष्टुः | 8.4.41 | हल् सन्धि | Dental to Retroflex assimilation (Shtutva) |
| यरोऽनुनासिकेऽनुनासिको वा | 8.4.45 | हल् सन्धि | Consonants assimilate to class nasal |
| खरि च | 8.4.55 | हल् सन्धि | De-voicing of stops before voiceless consonants (Chartva) |
| तोर्लि | 8.4.60 | हल् सन्धि | Dental stops assimilate to l (with nasalization for n) |
| झयो होऽन्यतरस्याम् | 8.4.62 | हल् सन्धि | Voiced aspirate substitution for h after stops |
| शशछोऽटि | 8.4.63 | हल् सन्धि | ś becomes ch after stops |
| वा पदान्तस्य | 8.4.59 | हल् सन्धि | Optional conversion of Anusvara to matching nasal |
- Python — Core rule engine implementation
- Streamlit — Fast, interactive frontend web application
- Pandas — Structured derivation tables and data handling
- Akshara — Phonetic parsing & varna-vyavastha representation
- Aksharamukha — Multi-script Indic transliteration engine
Contributions, feature suggestions, and rule additions are welcome!
- Fork the repository.
- Create your feature branch (
git checkout -b feature/NewRule). - Commit your changes (
git commit -m 'Add support for Vartika ...'). - Push to the branch (
git push origin feature/NewRule). - Open a Pull Request.
If you find an edge case or mistake in derivation, please Open an Issue or email samvadah@proton.me.
Distributed under the MIT License. See LICENSE for more information.
Built with ❤️ for Sanskrit Computational Linguistics by Samvadah