What is a Translation Memory?

A translation memory (TM) is a database of previously translated content. Whenever text is translated and confirmed as correct, the original source text and translated target text are stored in your company’s translation memory.

Translation Memory 2@3x

What is a Translation Memory?

A translation memory (TM) is a database that maintains segments of all previously translated content. Any time a piece of source text is translated, its individual sentences or paragraphs, known as segments, are stored in your company’s translation memory alongside their target equivalents. As new segments are translated in a project, they are automatically saved into the translation memory. As a result, translation memories continue to grow over time with every new translated segment.

Translation memories aren’t static, monolithic databases. Companies will create separate TMs for each language pair, content type, use case, and more. Since each of those TMs is associated with its own specific project, the TMs grow and become more useful for future projects as translation work is completed.

According to a recent CSA Research study, most translators that rely on modern tools use TMs in their day-to-day workflows, with only 12% saying that they never use them.

How does it work?

Translation memories start off in one of two ways - from existing resources or empty. Empty TMs build a database of translated content by continually taking in each newly translated segment, word by word and sentence by sentence, over time.

As each new segment is added, the original source text and its translation are paired together and stored for future reference. From that point on, any time that specific segment appears in new content with which the TM is associated, the TM will automatically replace it with the saved translation. This saves time for linguists, enabling them to spend their efforts on new, previously unseen segments, as well as segments that are particularly challenging for machine translation systems to understand - instead of phrases that have been previously translated.

When it comes to translating new content using existing segments within a TM, there are a few key measurements to think about. While there are other other statistics to measure match quality, these are the most important to keep an eye out for:

Context Match: A context match, meaning that the preceding and following words are also 100% matches.

Exact Match: The matched segment is identical to a segment already in the TM.

Fuzzy Match: The matched segment is 75-99% similar to a segment already in the TM.

TMs are especially powerful when it comes to content that references commonly-repeated words or phrases. If you’re translating product resources, for instance, parts, colors, and sizes may be repeated often. Instead of constantly needing to translate the same content over and over, your TM will jump in and automatically replace the previously translated text for you. The more you translate in a specific language, the less manual translation required overall.

Why leverage a Translation Memory?

There are a few important benefits to using a TM on a regular basis. Over time, the efficiency gains, content consistency, added contextual clues, cost savings, and more help make TMs a valuable resource that every company should use.

Consistent Content: Translators no longer need to worry about translating the same segments over and over, meaning your content is more likely to be consistent throughout because the previously translated segments will be used often. If you have a certain phrase that needs to be translated a certain way, your TM will provide that consistency.

Context Enhancements: Since TMs store more than just single words, any context that comes with certain phrases or sentences can be saved in your database. The result? Added context automatically appears without the need for manual work or translation.

Cost Savings: By enabling translators to spend less time on manual translation, you can localize more of your content into more languages without increasing your budget. Companies who build and maintain large TMs can see an enormous return on their localization spend, as linguists spend all their time on previously untranslated segments.

Long Term Database: Translation memories can be created from previously used source materials or TMs, so your database can be taken with you wherever you go. The advantage of technology like a TM is that it can save time and money for the long haul.

Overall Efficiency: In general, using tools like this help translators become more efficient, as they no longer need to undertake the same repetitive tasks over and over. And since TMs can be edited, segments can be updated or removed for improved efficiency.

When do I use a Translation Memory?

The simple answer is always.

No matter the type of content that you’re looking to translate, your TMs will help you maintain consistency throughout your content. Translators will be able to access the appropriate TMs through a computer-aided translation (CAT) tool, so it’s important to use TMs and always keep them updated.

One way to measure the effectiveness and usage of your translation memory is to look at TM Leverage - the statistic that shows the percentage of a given document that was translated using the translation memory. When you first start using a translation memory, its leverage is low. But as more segments are added to the TM over time, the leverage of that TM increases.

Since Lilt’s adaptive machine translation workflow allows translators to provide immediate feedback, your TMs will automatically update to reflect the most accurate translations for your brand.

Things to consider

First and foremost, a translation memory is NOT machine translation. Though the two have similar names, they function differently in the localization workflow.

To summarize, machine translation systems use neural networks to translate a source text into a target language based on predictions made by a neural machine translation engine. Translation memories, on the other hand, automatically translate specific segments of text based transaltions that have been previously completed and confirmed by linguists.

Translation memories are also distinct from termbases and style guides. While the latter generally contain syntax rules and terms across multiple languages, TMs hold actual translated content. All three will, in fact, help translators efficiently translate and maintain consistency, though in different ways.

With Lilt, users can upload translation memories and termbases. As translators work through content within Lilt, the system will take the changes and edits based on translator review and update the databases automatically.