A translation memory, or TM, is a centralized database of translated content stored in segments. Think of a sentence as being a segment in writing. However, for a TM, it may not actually be a full sentence, but rather a grouping of sequential words, which allows the translator to have some context to provide an appropriate translation of the segment.
Translation Memory Categories
There are three categories for TM matches:
Repetition is a full segment match within a source file for words in a duplicate segment.
100% match is a match against an existing translation memory for the client and not within the current source file being reviewed. This is leveraging translation previously performed from a prior project.
Fuzzy match is a match of a segment that is very similar to a 100% match against an existing memory, but not exact. It may be just an additional space, a change of a word or two, change in punctuation or some other very minor difference. Usually, a percentage is seen on quotes for the segment variation such as 75% to 99% or some mix within this range. But it is not as simple as a percentage. An algorithm will dictate if a ‘fuzzy’ is close enough to be a useful match. Matches below 75% are not expected to be useful and they would most likely begin to take as much time or more time for a translator to address rather than if translating from scratch. These lower matching fuzzy segments will be counted as new words and treated as such.
New words are when a segment has no match within the source file or against an existing client translation memory and are classified as new words and quoted at a full translation rate.
How does TM impact translation costs?
Across the localization industry, repetition and 100% matches are charged at 30% of a new word rate for a language. Fuzzy matches are charged at 70% of a new word rate (with some companies offering a variation of this if they offer broken out ranges between 75% and 99%). But again, the percentage concept is misleading and may not be a true percentage of the difference within a segment.
This is consistent with how vendors pay translators for their review of these matches. A fuzzy match will take a translator more time than a full match (repetition or 100% match) to review so the cost is higher. A fuzzy match may be a simple, quick edit or could require a full segment to be rewritten. But on average, it takes less time for a translator to address than it would for a new segment.
Translation memories are useful for both the client and translation teams as they help create consistent and cost-effective translations.