sandhi

Denormalized Sanskrit Sandhi (based on Pāṇini's sūtras) implemented in pure Python.

🚀 Run in Google Colab

Use the link below to open and run the notebook directly in Google Colab (no installation required):

https://colab.research.google.com/github/shantanuo/sandhi/blob/main/sanskrit_sandhi.ipynb

Run all the cells in the notebook, then test your word at the end of the script. For example:

sandhi_builder('पितृ उपदेश')

returns:

{'पित्रुपदेश'}

---

If you do not want to use Python, look for the last character of the first word and the first character of the second word in the index file (sandhi_code_out.txt), similar to looking up a telephone directory.

For example, if you are looking for the Sandhi of यज् + न, search for ज् न in the index file:

grep 'ज् न' sandhi_code_out.txt

Output:

ज् न ज्ञ 2.1.1 श्चुत्व
ज् न ञ्न 2.7.1 अनुनासिक

You get ज्ञ, and hence the Sandhi word becomes:

य + ज्ञ = यज्ञ

This is a poor man's Sandhi builder. For a richer experience, visit:

https://sanskrit.uohyd.ac.in/scl/#

---

If more than one level of Sandhi is required, enter the output again.

For example, when you type:

सत् जन

you may expect:

सज्जन

but the software will first return:

सद् जन

Enter सद् जन again to obtain the expected result:

सज्जन

---

1) Underscore (_) in Sandhi Output

If an underscore (_) appears in the output, it indicates that the words cannot be joined.

Example:

Input

मुनी इमो

Output

मुनी_इमो

This indicates that no Sandhi is possible because of प्रकृतिभाव सन्धि (ईदूदेद्द्विवचनं प्रगृह्यम्).

Even if मुनीमो is also returned, it should be ignored.

2) Asterisk (*) in Sandhi Output

If a word is followed by an asterisk (*), it indicates the preferred form.

Example:

Input

महा ऋषी

Output

महर्षी* महार्षी

Here महर्षी* is the preferred form.

---

A) Telegram Bot

If you use Telegram, add the Sanskrit One bot to your contacts:

https://t.me/sanskritonebot

B) Android App

If you use an Android phone, install the Marathi Spell Check app. It includes a Sanskrit Sandhi tab.

https://play.google.com/store/apps/details?id=com.myapp.marathispellcheckandsanskritsandhi

You get three tools in one:

• Sandhi Builder
• Sandhi Splitter
• Spell Checker

---

Achieved 93.5% accuracy on the Bhagavad Gītā benchmark corpus.

Compare it with other solutions mentioned in this paper:

http://www.lrec-conf.org/proceedings/lrec2018/pdf/755.pdf