Translation memory is an important tool in the modern translator’s toolkit, and one that is currently the focus of a great deal of discussion in the translation and localization community. Simply put, translation memory is a type of shared database that stores translations and continually updates itself as its users work. Translation memory cannot replace the work of a skilled and experienced translator, but in the right hands it can significantly improve the quality and efficiency of localization.
Building a Better Brain
Translation memory is a database of “segments,” which are phrases, sentences, or even entire paragraphs that have previously been translated. Translators often handle tasks, such as product documentation, marketing material, and user guides, which intentionally contain similar or identical language elements that recur within or between documents. SDL, makers of the popular Trados Studio software, note that “recurring phrases and statements…can make up 40% or more” of texts to be localized. These elements are stored in pairs (called translation units) within the language memory, and new translation units are accumulated automatically as translators work.
TM is not MT
Unlike machine translation, translation memory relies on human input. Machine translation prioritizes speed over quality, translating whole documents quickly but with middling accuracy. TM software, conversely, does not automatically translate entire documents. Instead, it searches documents for elements that are full or partial matches for segments in its translation memory database and then inserts those translated segments into the document, for review by a human translator. These matches can be accepted or rejected, and any changes are added back into the translation memory for use in future projects. Another advantage of TM is that because its translation units consist of language pairs that have been added based on user input, there is effectively no limit on the number of languages and language pairs that can be stored.
To Forget is Human
Translation memory software provides many compelling benefits to its users. It speeds up translation by making suggestions based on previous work, reducing costs and increasing efficiency. It also helps to compensate for some of the weaknesses of human translators – in particular, inexperience and inconsistency. In many cases, while translators are language experts, they may be unfamiliar with the jargon or terms used in specific domains, or they may forget how they translated them in the past. Because translation memory is shared among its users, however, translators who are less experienced in a particular domain can rely on the software to handle the subject-specific elements instead of having to become experts themselves. Another problem with flesh-and-blood translators is inconsistency. Not only can individuals forget how they translated specific or unusual passages between documents, but teams can also have consistency issues when different translators localize similar passages in different ways. TM software helps to make translations consistent among multiple translators because a translation memory never forgets. And because it can make “fuzzy” matches that rely on human oversight, it remains flexible in a way that pure machine translation never could.
Ultimately, translation tools are only as good as the people that use them. TM software cannot replace a talented localization team, but it can help them work faster, smarter, and better. It is an exciting development in the field, and one that every language service provider should take full advantage of.