What is Fuzzy Matching?

Fuzzy matching is how translation memories handle almost-repeats. When a new source segment closely resembles a previously translated one, differing by a word, a number, or punctuation, the TM returns the old translation with a similarity score, and the translator edits it instead of starting from scratch.

Match scores drive both workflow and pricing. Exact matches may be applied automatically; high-similarity fuzzy matches are fast edits; low-similarity matches are barely worth showing. Translation vendors price accordingly, with fuzzy bands charged at a fraction of the new-word rate, which is why weighted word counts exist.

The term predates TM tooling in one important place: gettext marks an entry fuzzy when the source string changed and the old translation was carried over as a guess. A fuzzy entry in a PO file is explicitly untrusted, excluded from compiled output by default until a human confirms it. Both uses share the same idea: a probably-right translation that still needs eyes.