mukh-xadna-hunna

Appearance

Introduction ​

no-nepali-profanity is a small, dependency-free profanity filter for Python. It detects and censors profanity in English, Romanized Nepali and Devanagari Nepali, plus the Hindi slang common in Nepal.

py
from no_nepali_profanity import censor, check, contains_profanity, find_profanity

contains_profanity("Great teacher!")    # False
contains_profanity("मुजीको क्लास")       # True
find_profanity("f.u.c.k this sh1t")     # ["fuck", "shit"]
censor("you muji")                      # "you ****"
check("you muji").censor()              # "you ****"

Why another profanity filter? ​

General-purpose profanity filters are built for English. On a Nepali site they fail in two ways:

  • They miss Nepali. People write Nepali in Devanagari (मुजी) and in Latin letters (muji, mujiko), often in the same sentence as English. An English word list catches neither.
  • They block real names. Substring matching flags names like Shitij or Randip. On a site where people sign up with their real names, that is a worse failure than a missed swear.

This package is built for that situation: it matches whole words, understands Nepali postpositions, and checks its word lists against common Nepali names.

Features ​

  • Three scripts in one pass. English, Romanized Nepali and Devanagari, including mixed text.
  • Handles common dodges. Leetspeak (sh1t, @ss), ! for i (sh!t), * for a hidden letter (f*ck), stretched letters (fuuuuck) and spelled-out letters (f.u.c.k, f u c k).
  • Understands Nepali grammar. Postpositions and plurals like -ko, -lai and -haru (mujiko, मुजीहरू).
  • Handles Devanagari spelling variants. Nukta, chandrabindu vs anusvara, and zero-width joiners.
  • Censors in place. Masks exactly what the user typed, even F.U.C.K or Sh1t, and leaves the rest alone.
  • Catches phrases. Word pairs that are only offensive together, like sasto manche.
  • Configurable. Turn each language on or off, and pick one of three strictness levels.
  • The same in every language. JavaScript, Python, Go, PHP and Dart share the word lists and matching rules, and are tested against the same inputs, so a comment gets the same result on every part of your stack.
  • Zero dependencies. Pure Python, standard library only, with type hints (py.typed).

What it isn't ​

  • It doesn't understand meaning. Insults without a listed word, sarcasm and context are out of scope. Treat it as a first-pass filter, and send anything that matters to a human moderator.

Next steps ​