E.g., 10/30/2020
E.g., 10/30/2020

What is Translation Memory?

Translation memory (TM) reuses past translations to improve translation productivity. It works by splitting text into short segments and storing their translations in a database. Whenever a segment is repeated, the translator can reuse a fitting previous translation, or sometimes paste a translation of a similar segment with minor edits. Compared to machine translation, translation memory relies on actual human translations, which are most often quite accurate and contain the right terminology.

Cost Saving and Consistency

Legal contracts, software interfaces, technical documentation, medical labels, and product catalogues have highly repetitive texts. When translating these types of content, translation memory can cut costs and increase speed by 15-90%. The average saving is around 36%, according to one study by Memsource.

In a team, each member can reuse translations from other members, resulting in higher savings, and improved translation consistency.

More than 90% of translation companies use TM and CAT-tools, a 2016 survey supported by GALA found out. Because of cost-saving, translation memory is the mainstay technology in the industry. It is commonly built in professional translation editing programs, or CAT-tools.

Examples of Translation Memory Providers
SDL Trados Across Translation Editor
Kilgray memoQ XTM Editor
Memsource Editor and Web Editor Memsource personal edition
Wordfast Anywhere and Wordfast Pro Transit NXT
Fluency Translation Suite Wordbee Translator
Smartling Multitrans
Ontram Atril Déjà vu X
CafeTran Espresso MateCat
Google Translator Toolkit Omega-T

Things to Consider

  • Deployment model: Cloud, server, or desktop. Cloud translation memory tools work in the browser and don’t require installation. They store data online on their servers. Cloud is easy to start with, but may not be suitable for some companies with high data confidentiality requirements. Another option is to use desktop tools that are installed on the translators’ PCs, with server components providing exchange of data between several translators working in a group.
  • TM business value. A large translation memory database of good quality can significantly cut costs on a new project. Agencies with their own TM can offer better prices to new clients to win new business, or achieve higher margins. Furthermore, TM databases help train machine translation engines, making them a valuable asset. However, client contracts often limit trading and reuse of their translation memories.
  • TM discount schemes. Large-scale buyers most commonly require their suppliers to provide discounts based on translation memory. Discounts are based on similarity of the segment translated to a match in the translation memory. 100% matches are sometimes called ICE (in-context exact) matches, and anything below 100% is called a fuzzy match.



Example discount


Repeated segment in the same document



A context match: 100% match preceded and followed by other 100% matches



A segment is identical to a segment in the TM



99-95% similarity



85-94% similarity



75-84% similarity



50-74% similarity

Full price


0-50% similarity

Full price

Who Owns the Translation Memory?

Because translation memory contains actual translation, it is treated as intellectual property by law in most countries. In theory, the translator owns the TM by default. In practice, most contracts transfer the IP to the client. Modern tools store TM on central servers / Cloud, which further limits the translator’s ability to download the database and reuse it for another client.

.TMX File format. CAT-tools can export translation memory databases in .TMX (Translation Memory eXchange) formats, allowing limited interoperability between mainstream software suites.

.XLIFF File format is a standardized XML-based format for bilingual documents that most CAT-tools can work with. SDL and other vendors modify XLIFF slightly, limiting interoperability.

Read More Translation Technology Descriptions