What is translation memory and why it matters

What is translation memory?

Translation memory is a database recording all your translated text, like words, sentences, and paragraphs. This database saves your translation segments as you work on a project. Then, it automatically shares suggestions the next time you translate a similar piece of text.

In other words, translation memory helps you avoid translating the same thing twice. It saves you time and ensures all your translations stay consistent. Not only can it help you speed up your translation process, but it also ensures consistency across all target languages.

Learn more about it in the video below.

Translation memory vs. machine translation: How do they differ?

Translation memory (TM) vs. machine translation (MT) are fundamentally different technologies solving different problems in your translation workflow. Translation memory helps you reuse your existing translations, while machine translation automatically generates new ones. 

The key difference: Translation memory recalls what humans have already translated, while MT creates new translations using AI. The latter is quicker, but may not be fully accurate.

• Translation memory: It’s a database storing all of your previously translated segments (terms, phrases, and sentences). When you translate new content, your TM database automatically shows the best matches from past translations. It’s essentially a smart recycling system to reuse translations from the past and save human effort.
• Machine translation: It uses AI algorithms to generate new translations from scratch for any language. MT systems like DeepL analyze patterns across millions of translated texts to come up with the most likely translation for any text. 

Both these technologies speed up your translation process. 

In practice, it isn’t a question of translation memory vs. machine translation. Instead, most professional translation workflows combine both to create quality output quickly.

You might use MT for a first draft, then have human translators refine it while building up your TM. Over time, as your TM grows with approved translations, you rely less on MT for repetitive content because the TM provides consistent, pre-approved translations instantly.

Who came up with translation memory? 

Martin Kay, a pioneering linguist and researcher at Xerox PARC, first proposed the idea of translation memory in his paper, “The Proper Place of Men and Machines in Language Translation”.

Believe it or not, the paper was published in 1980, which means that translation memory has been around for more than 40 years. The evolution of it is exciting, so let's take a look.

Early beginnings (1970s-1980s)

Translation memory was born from a very practical need–to reduce repetitive translation work. The concept itself emerged in the late 1970s, but it was really brought to life in the early 1980s.

Let’s explore what was Martin Kay’s stance shortly.

Martin Kay, an expert in computational linguistics, photo from the 1980s

Kay believed that professional translators need to work with machines in order to make their work more human:

“Translation is a fine and exacting art, but there is much about it that is mechanical and routine and, if this were given over to a machine, the productivity of the translator would not only be magnified but his work would become more rewarding, more exciting, more human.”

Although he never used the phrase “translation memory”, he did suggest putting a word in a “local store” and reusing the translation where appropriate. 

Growth and commercialization (1990s)

By the early 1990s, translation memory became more accessible and practical. The first widely adopted commercial TM system, Trados (developed by Jochen Hummel and Iko Knyphausen), launched in Germany.

It transformed translation memory from a theoretical concept into a practical tool professional translators could actually use.

Trados quickly became the industry standard, making translation memory more common among freelance translators, agencies, and global companies.

The digital boom (2000s)

The internet changed everything, and translation memory is not an exception. New web-based tools emerged. Translators and localization teams could now collaborate and share translation memories more easily. For those exploring “what are translation services“, this evolution highlighted how technology could enhance efficiency and consistency. In 2005, SDL acquired Trados, making TM an integral part of broader localization platforms.

Cloud technology opened up even more possibilities. Translators could now access translation memories anytime, anywhere.

Integrations and AI (2010s-today)

In recent years, translation memory started integrating closely with machine translation (MT), AI-driven quality assurance tools, and content management systems (CMS).

Smart TM tools began automatically suggesting similar segments of text, merging human translations with AI suggestions for even faster workflows.

Today, translation memory is smarter than ever. It streamlines translations, helps with consistency, and significantly reduces costs.

How does translation memory work?

Translation memory uses machine learning algorithms and similarity calculations to quickly scan and match new text segments with your saved translations. This is how it’s able to give you instant suggestions. 

It’s a database with segments of already translated text. So, you have the original (source) text and its corresponding translation across different languages.

TM systems work by comparing new text you're translating with segments you’ve already translated and saved in your database. Depending on how closely these new segments match your stored translations, you’ll see one of four common matches.

How do you create a translation memory?

Creating a translation memory is actually pretty simple. Your translation memory software handles most of it automatically.

Here’s how to get started:

• Pick your tool: Start by choosing a CAT tool or a translation management system with built-in translation memory (like Lokalise)
• Set it up: Usually, you just click “create new translation memory”, give it a name, and you're good to go
• Start translating: As you translate content, the software automatically saves your translations segment-by-segment, building your translation memory behind the scenes
• Reuse it: Next time you encounter similar text, your CAT tools suggests your stored translations so you can reuse them

Just translate as you normally would, and your translation memory grows and becomes more useful over time.

What are some of the best practices?

Here are some best practices to use translation memory in your localization workflow:

• Regularly maintain your database: A bloated TM with conflicting suggestions slows translators down. Occasionally review and clean up your translation memory to remove outdated translations and update terminology. This could be discontinued features or changed product names.
• Be consistent: The clearer and more consistent your translations, the better your TM works. Mismatched translations like “Empezar” vs “Comenzar” for “Start” create conflicting suggestions. Use a centralized glossary and style guide, so translators don't make judgment calls on core terms.
• Segment properly: A TM algorithm splits text into segments, typically by sentence. Review these segmentation rules because overly aggressive segmentation can break context, while segments that are too long reduce match rates.
• Backup regularly: Always back up your translation memory regularly, just to make sure you don’t lose valuable work. Set up automatic daily or weekly backups to a separate location.
• Collaborate carefully: If multiple translators share access to a single translation memory database, set clear guidelines and processes to maintain consistency across your team. You can also assign a TM manager who reviews and approves entries before they become part of the shared database.

What are the benefits of translation memory?

Translation memory has multiple benefits. It helps you translate faster, reduce translation rates, and keep your content consistent by reusing previously translated text. More specifically, it:

• Speeds up the translation process by reusing previously translated content
• Ensures consistency across all translations and languages
• Reduces translation costs by minimizing repeated work
• Improves collaboration between translators and teams
• Helps maintain brand voice and terminology across projects
• Increases translation quality by relying on approved, human translations
• Makes large-scale localization projects more manageable and efficient

Now let’s see a real-life example.

How translation memory saved 16% of translation costs for ActiveCampaign

ActiveCampaign had a common translation problem: their process relied heavily on machine translation and manual fixes, with little to no translation testing, which meant they were repeatedly translating the same phrases, wasting time, and spending too much money.

Once they started using Lokalise and its translation memory feature, their translations got a lot easier.

They began saving every sentence or phrase they translated into a database (translation memory) organized by different content types (e.g., marketing materials, product descriptions, support articles), which was crucial for ensuring consistency and quality in marketing translation, where adapting content for new audiences is key.

Because of this, whenever similar text came up again, the previous translation popped up instantly, ensuring consistency and supporting translation quality assurance. They didn’t have to translate the same thing twice.

Results were clear:

• 16% annual savings on translation costs
  A dramatically accelerated workflow, launching localized campaigns in just two days
• Improved consistency and quality across their global content

If you want to explore Lokalise, you can sign up for a free 14-day trial, no credit card required. Bear in mind that Lokalise is more than a translation memory software. It’s an all-in-one platform that helps you translate and localize your content, and successfully collaborate with all stakeholders.Want to continue learning? Explore the Lokalise blog for other useful resources.