hinglish

Convert romanised Hindi (Hinglish) to Devanagari and back, and make URL-safe slugs from Hindi text. Rule-based, offline, zero dependencies.

$ npx hinglishify "kaise ho"
कैसे हो

$ npx hinglishify "mera naam Ujjwal hai aur yeh mera blog post hai"
मेरा नाम Ujjwal है और यह मेरा blog post है

$ npx hinglishify --latin "नमस्ते दुनिया"
namaste duniya

$ npx hinglishify --slug "मेरा Blog पोस्ट #1"
mera-blog-post-1

$ printf 'namaste dost\nzindagi ek safar hai\n' | npx hinglishify
नमस्ते दोस्त
ज़िंदगी एक सफ़र है

Install

npx hinglishify "kaise ho"

npm i hinglishify

Node 18 or newer. ESM only.

Usage

CLI

$ hinglish --help
hinglish - romanised Hindi <-> Devanagari, offline and rule-based

Usage
  hinglish <text>            Hinglish -> Devanagari
  hinglish --latin <text>    Devanagari -> Latin
  hinglish --slug <text>     URL-safe slug
  cat file.txt | hinglish    read from stdin

Flags
  -l, --latin      romanise Devanagari instead of converting to it
  -s, --slug       print a URL-safe slug
      --strict     leave words that are not in the dictionary untouched
  -h, --help       show this help
  -v, --version    print the version

Examples
  $ hinglish "kaise ho"
  $ hinglish --slug "मेरा Blog पोस्ट #1"
  $ echo "नमस्ते" | hinglish --latin

| Flag | Effect | | --- | --- | | -l, --latin | Devanagari to ASCII instead of the other way round | | -s, --slug | print a URL-safe slug | | --strict | leave words that are not in the dictionary exactly as typed | | -h, --help | print help | | -v, --version | print the version |

Text can come from arguments or from stdin. Each input line is converted separately.

API

import { toDevanagari, toLatin, slugify, isHinglish, addWords } from "hinglishify";

toDevanagari("namaste duniya");        // नमस्ते दुनिया
toDevanagari("mera naam Ujjwal hai");  // मेरा नाम Ujjwal है
toLatin("नमस्ते दुनिया");                // namaste duniya
slugify("मेरा Blog पोस्ट #1");           // mera-blog-post-1
isHinglish("kaise ho");                // true
isHinglish("the quick brown fox");     // false

addWords({ jugaadu: "जुगाड़ू" });
toDevanagari("jugaadu banda");         // जुगाड़ू बंदा

Output captured from that exact snippet:

नमस्ते दुनिया
मेरा नाम Ujjwal है
namaste duniya
mera-blog-post-1
true false
जुगाड़ू बंदा

API

toDevanagari(text, options?)

Hinglish to Devanagari. Punctuation, digits, emoji, whitespace and text that is already Devanagari pass through untouched, so running it twice is a no-op.

Words are resolved in this order: tokens containing digits, ALL CAPS tokens and domain-looking tokens are left alone; then the dictionary; then the English stop-list (blog, post, api, www, https, …); then capitalised words, which are treated as proper nouns; then the phonetic rules.

• options.strict (default false) — when true, a word that is not in the dictionary is left exactly as typed instead of being guessed phonetically.

toDevanagari("chai peelo yaar");                 // चाय पीलो यार
toDevanagari("chai peelo yaar", { strict: true }); // चाय peelo यार

toLatin(text)

Devanagari to plain ASCII — no diacritics, because the point is slugs and search. Non-Devanagari characters are left alone.

slugify(text, options?)

toLatin for the Devanagari parts, then lowercase, strip punctuation, collapse to the separator and trim. Never returns an empty string for input that contains Latin letters, digits or Devanagari.

• options.separator (default "-")
• options.maxLength (default: no limit) — truncates at a separator boundary
• options.lowercase (default true)

isDevanagari(text)

true when the text contains Devanagari characters and no Latin letters.

isHinglish(text)

Heuristic. Scores each Latin token against the dictionary, a shape test (aa, kh, bh, -ji, -wala, …) and a common-English veto list, then asks for a simple majority.

transliterate(text, options?)

Wrapper. options.to is "devanagari" (default) or "latin".

dictionary, addWords(words), getDictionary(), resetDictionary()

dictionary is a read-only live view of the 298 built-in words — assigning to it throws. addWords adds or overrides entries (keys are lowercased) and returns how many were applied. getDictionary() returns a detached mutable copy; resetDictionary() drops everything addWords added.

Transliteration rules

Implemented in src/mappings.js (tables) and src/transliterate.js (the engine):

• Inherent vowel and matras. k ka kaa ki kii ku kuu ke kai ko kau kam kan → क का का की की कू कू के कै को कौ कम कन. A vowel at the start of a word, or straight after another vowel, is independent (अ आ इ ई …); inside a word it is the matra (ा ि ी ु ू े ै ो ौ).
• Long word-final vowels. Hinglish hindi, tu, mera mean हिंदी, तू, मेरा, so a final i/u/a is written long.
• Aspirates and retroflex. kh gh ch chh jh th dh ph bh sh, plus T→ट, t→त, D→ड, d→द, N→ण, n→न, Sh→ष. Lowercase-only typing still works (tha → था).
• Conjuncts. kya → क्या, pyaar → प्यार, shraddha → श्रद्धा, kshama → क्षमा, dost → दोस्त, shabd → शब्द.
• Where not to join. A halant is only inserted for a word-final cluster, a word-initial cluster, a geminate, or a listed conjunct pair. That is what keeps karna → करना rather than कर्ना, and kitna, apna, sabse, matlab, bachpan.
• Nasals. n/m before a stop become anusvara: hindi → हिंदी, chand → चंद, sundar → सुंदर, angrezi → अंग्रेज़ी. m only nasalises before a labial, so kamzor stays कमज़ोर. A final -aan becomes candrabindu: haan → हाँ, yahaan → यहाँ.
• Nukta. z → ज़, f → फ़, q → क़, R → ड़, and a trailing dot adds a nukta to the previous consonant (kh.aas → ख़ास). Output is always NFC, so nukta letters have one encoding.
• Dictionary. 298 frequent words override the rules, because exact Hindi spelling is not recoverable from Hinglish spelling.
• Mixed English. ALL CAPS, tokens with digits, domains, the English stop-list and capitalised proper nouns are left as typed.
• Reverse. toLatin collapses matras, maps nukta letters to z/f/q, and applies Hindi schwa deletion right to left: राम → ram, लड़का → ladka, करना → karna, मतलब → matlab, while कमल stays kamal.

Honesty about the limits

Transliteration between Hinglish and Devanagari is lossy and genuinely ambiguous. This library gets common text right and will get uncommon text wrong. Known cases:

• Schwa deletion is a guess. Hindi writes vowels it does not pronounce. karna is करना but dost is दोस्त, and nothing in the Latin spelling says which. The rule set above covers the usual shapes; jaldi → जल्दी and similar irregulars live in the dictionary instead.
• त/ट, द/ड, न/ण, स/श cannot be recovered from lowercase typing. theek is ठीक, not थीक, only because the dictionary says so. Type T, D, N, Sh for the retroflex letters when it matters.
• Vowel length is lost in both directions. sharab → शरब, not शराब. Going back, काम → kam and राम → ram, so kaam does not round-trip byte-identically. 24 of the 30 round-trip words in the test suite come back unchanged; the other 6 lose a doubled vowel. The exact expectation for each word is asserted in test/to-latin.test.js.
• Candrabindu vs anusvara. chand → चंद, although चाँद is the moon. The rule follows the anusvara default; put the word you want in the dictionary.
• Hinglish and English collide. the is in the dictionary as थे, so English the becomes थे. --strict, the stop-list and addWords are the ways out.
• Capitalised words are assumed to be proper nouns. Ujjwal survives, which is the point, but so does Takkar — type it lowercase, or add it to the dictionary.
• माँ romanises as man. Candrabindu becomes n, which is right for यहाँ → yahan and wrong here.

The escape hatch

Anything the rules get wrong is one line away from being right:

import { dictionary, addWords } from "hinglishify";

addWords({ "jugaad": "जुगाड़" });

Keys are lowercased and the dictionary always wins, including under strict: true. dictionary itself is read-only — assigning to it throws — so nothing can corrupt the table by accident.

Why

Hindi content on the web is typed in Latin letters and published in Devanagari, and the tools in between are either online APIs, machine-learning models, or 40 MB of training data. This is a few tables and a scanner: it runs offline, in a build step, with no network and no dependencies, and when it is wrong you can fix it with one dictionary entry. The slug half exists because encodeURIComponent("नमस्ते") is not a URL anybody wants to click.

Support

If this saved you some time, you can support the work here:

• Patreon: https://www.patreon.com/cw/BABU_ISHU
• PayPal: https://www.paypal.com/ncp/payment/SECBQ62TRZZ6Y

License

MIT