Profanity¶

pygarble.screening.ProfanityDetector finds English profanity at the token level, so words that merely contain a profane substring stay clean. Every finding has kind profanity; the confidence tells strong words from mild ones and plain spellings from obfuscated ones.

from pygarble.screening import ProfanityDetector

detector = ProfanityDetector()
assert detector.detect("Scunthorpe assassin classic bass") == ()

(mild,) = detector.detect("what a damn mess")
assert (mild.kind, mild.confidence, mild.reason) == (
    "profanity",
    0.7,
    "mild",
)
assert [f.reason for f in detector.detect("f u c k this")] == ["spaced"]
assert ProfanityDetector(allowlist=["damn"]).detect("damn") == ()

Tiers and obfuscation¶

The mild tier also holds words with innocent senses, such as cum, prick, cock, tits and pissed. Embedded matching applies only to compounds of “fuck”. Multi-word phrases are matched as phrases. Digit-only tokens are ignored; possessives and trailing ! are handled. allowlist exempts words, and tiers selects strong, mild or both.

The detector covers English only and does not detect hate speech beyond its word list.

Use Scanner(profanity_tiers=["strong"], profanity_allowlist=[...]) from pygarble.screening to configure these rules in a combined screening pass. The detector, normalization and word list live together under pygarble/screening/profanity/. Existing pygarble.profanity imports remain supported. No optional package or model is needed for profanity.

Word list attribution¶

The word list is seeded from the LDNOOBW English list (List of Dirty, Naughty, Obscene, and Otherwise Bad Words) published by Shutterstock under CC-BY-4.0, then filtered and extended by the pygarble maintainers. The attribution text ships as pygarble.screening.profanity.wordlist.ATTRIBUTION (also available at the old pygarble.profanity.wordlist path).