@lacspace/nepali-typing

Type Nepali in Latin letters and get Devanagari, with a list of candidates per word like a phonetic input method. It's built for search boxes and comment fields. Pure JS with no dependencies, so it is safe for React Native.

import { suggest, toDevanagari, createTyper, buildLexicon } from "@lacspace/nepali-typing";

suggest("sarkar");        // ["सरकार", "सर्कर", "सरकारी", …]
suggest("kathmandu");     // ["काठमाडौं", …]
suggest("netaharulai");   // ["नेताहरूलाई", …]  (typed case endings)
toDevanagari("NEPSE aaja 20 ankale badhyo");  // "NEPSE आज २० अंकले बढ्यो"
toDevanagari("mero desh nepal ho.");          // "मेरो देश नेपाल हो।"

How it works

1. Lexicon: a list of Nepali words, matched through a loose key that forgives how people actually romanise. All of these fold into one key:

   • long and short vowels: aa/a, ee/i, oo/u;
   • consonant variants: sh/s, w/v/b, ch/chh, ph/f;
   • the final or medial a that Nepali doesn't pronounce;
   • nasals.

   Candidates are ranked by a weighted edit distance plus word frequency. Words that start with what was typed are offered too (autocomplete).

2. Case endings: typed endings are split off the word. These include le, lai, ko, ka, ki, ma, bata, sanga, dekhi, samma and haru…, so "sarkarko" gives सरकारको.

3. English loanwords: these are written the Nepali way: facebook → फेसबुक, mobile → मोबाइल, budget → बजेट.

4. Phonetic engine: handles words the lexicon doesn't know. ITRANS-style capitals give the retroflex letters: T Th D Dh N Sh → ट ठ ड ढ ण ष (DhuNgaa → ढुण्गा). Its spellings always appear as extra candidates.

The bundled lexicon is small: about 750 entries: frequent words, the 77 districts and news vocabulary. Feed it your own Nepali text for real coverage:

const typer = createTyper({ words: buildLexicon(myPublishedNepaliStories) });
typer.suggest("balen");             // ["बालेन", …] once your stories use it
typer.learn("pani", "पानी");         // the user picked पानी: it ranks first next time
save(typer.exportLearned());         // persist (AsyncStorage / localStorage) and pass back as { learned }

API

• suggest(word, { limit = 5, complete = true }) and toDevanagari(text, { digits = true, danda = true, keep }): these use the shared default typer.
  • ALL-CAPS words (NEPSE) stay in Latin; keep adds brand names.
  • digits turns 0–9 into ०–९.
  • danda turns a sentence-ending "." into "।".
• createTyper({ words, noBase, learned }): returns { suggest, convert, learn, addWords, exportLearned, size }.
• buildLexicon(texts): returns { word: count }, counting the Devanagari words in your text.
• latinKey() / devKey() / phonetic(): the building blocks.

Licence

Lacspace Free Licence v1.0: free for personal and commercial use.