FAQ Help Center

Category: Glossary Translation Memory

Translation Memory (TM) 101 [2024 Guide]

Translation Memory (TM) is powerful technology integral to the productivity of those who work as professional translators or bilingual employees of organizations that need translation support.

It is a productivity enhancer often included in premium translation software, most commonly translation management software

Pairaphrase CTA Banner

In this post you’ll learn what is translation memory, benefits, how translation memory works and definitions of related buzzwords such as “fuzzy matches,” “alignment,” and “context matches.”

What is Translation Memory?

A translation memory is a database that stores previously translated segments of text, sentences or paragraphs. Translation memories can be bilingual or multilingual depending on the specific translation memory software used.

It enhances the productivity of translators by storing their translations for re-use, so that they never need to translate the same word or phrase twice.


The origin of Translation Memory can be traced back to 1978. The concept was mentioned in a paper authored by Martin Kay of the Xerox Palo Alto Research Center, “The Proper Place of Men and Machines in Language Translation.”

In this paper, he advocated for enhancements of “cooperative man–machine systems” so that the translation industry could truly benefit from the original goal of machine translation.

Since the development of Translation Memory, the translation industry has come to rely on this technology for developing human-quality level translations in the least amount of time possible, which results in significant cost savings.

How Does Translation Memory Work?

We’re about to explain how Translation Memory works, and this gets a bit technical. However, your experience with this technology doesn’t have to be (depending on your choice of translation memory software). If need be, feel free to skip ahead to the section named “A Simplified Explanation.” matching

Step 1: Import or Create your Translation Memories

In order to create a Translation Memory you’ll need to use a CAT tool (Computer Assisted Translation). Depending on your choice of Translation Memory software, you may need to manually create a Translation Memory while other tools will create one automatically on-the-fly for you.

There are several ways to create a Translation Memory. You can create brand-new Translation Memories in a CAT tool, or, if your organization is already using Translation Memory, you should be able to import those as TMX files. TMX stands for Translation Memory eXchange, a translation industry standard developed so that users can export and import Translation Memories into any software that supports TM.

Creating new Translation Memories can be easy or difficult depending on your choice in CAT tool.

Step 2: Analyze New Files for 100% and Fuzzy Matches

One of the features included with your CAT tool is the ability to pre-translate and “analyze” a file against a Translation Memory.

What this means is that the software will identify what has already been previously translated and stored in the Translation Memory for reuse with your new file(s). Reusing previous translations does two things in the best CAT tools. They will train the machine translation to produce better initial translation quality and repetitive text will be stored in a Translation Memory for future reuse.

The combination of continuously learning machine translation and translation memory produces the best quality translations at the lowest possible cost and in the shortest amount of time.

translation memory benefits

Step 3: Translate or Edit Your Translation Using the Translation Memory Software

In order to save your translations in a Translation Memory, you’ll need to use Translation Memory software to translate and edit your files. You can’t do this in Excel or MS-Word. Depending on your choice of software you’ll either get a complex user interface that will require some learning on your part or a simple-to-use interface that requires little effort for you to learn how to use.

The best Translation Memory software will support a wide variety of file formats. Everything from InDesign to XML files. This is where it can add them the most value and save you a lot of time and money. 

That’s it! With three simple steps, you can become a more effective and efficient translator. Producing translations in a fraction of the time, at a fraction of the cost and producing more consistent translations. Translation Memory is the oldest translation technology in use today at every commercial translation company. It’s a proven cost and time saver.

Using Machine Translation and Translation Memory

Today, more and more human translators will start their translation process with machine translation. Pure machine translation will give the user a bare-bones first-draft translation of a file without any quality improvements.

In order to improve the quality of the translation, the translator will make manual edits to the machine-translated file but the machine translation remains static and quality improvements are delivering solely by the machine translation software.

The best Translation Memory software uses AI, Artificial Intelligence, to “train” machine translation software. Training machine translation is done by applying machine learning to teach the machine translation software the proper translations of your text. In short, your edits and content help teach the software the words and phrases you want to use in your translations.

how translation memory works

Another way a Translation Memory application improves your translation is the software actually aids the user in creating higher-quality translations from machine-translated content by storing translations in a bilingual repository called a Translation Memory.

When a user uploads a file for translation to software that supports translation memory, the text file is separated into segments. When a translator changes the translation, these segments are then stored in a Translation Memory.

Segments, 100% Matches & Fuzzy Matches Explained

Using Translation Memory software provides you with a statistical analysis of the file(s) you want to translate. This analysis is used to estimate the amount of time and cost that a file(s) will require to translate. The more 100% matches and Fuzzy matches the less time you’ll spend translating and editing.

Your file is broken down into segments and then compared against translated segments that are already stored in the Translation Memory (a bilingual repository) within a source language < > target language XML file.

It will search for several kinds of matches, also known as 100% matches (Context Matches), and imperfect matches (Fuzzy Matches). 100% and Context matches mean that the text is identical and you won’t need to do anything; the Translation Memory software will provide the translation for you. A Fuzzy Match is a segment that is statistically almost identical to a previous segment that has already been translated and stored.

how does translation memory work

At the push of a button, the software retrieves the previous translations from the database and places them in the translated file for the user. The more files a user translates and the more segments a user edits, the larger the database grows and the more valuable it becomes.

Over time, the user will translate less and less of the same words and phrases that his or her company often uses. The application will just automatically populate them for the user when he or she runs their files and text through its engine.

A Simplified Explanation

It’s a bilingual database for storing translations that are manually created or improved from their machine-translated origins. Storing them allows a user to draw on them later, so that the next time they need to produce a translation, they don’t need to manually produce the same translation.

Understanding Translation Memory By Analogy

Translation Memory is Like a Retirement Account

An easier way to understand Translation Memory is by comparing it to a retirement account.
The key to successful retirement accounts is starting early and making steady contributions. The more you contribute to it early on, the more quickly your account balance will grow. Your earnings continue to compound over time, meanwhile, you continue to add to your account to grow your retirement fund.

translation memory explained

If Translation Memory is the retirement account, editing or producing a translation segment is the equivalent of depositing money into a retirement account. The more effort a user puts in upfront, the faster they will reap the benefits of recycled translations, saving a significant amount of time and money.

The point of a retirement fund is to sustain you during your non-working years. Eventually, you no longer will want to (or perhaps physically be able to) put the time and effort into working.
Similarly, Translation Memory will help a user quickly put in less work for translations so they can focus on other tasks.

The great thing is that unlike a retirement account, there is no annual limit to the amount of contributions one can make to it (except for when a software plan has a word limit). Furthermore, the user doesn’t need to wait until a specific time to start making withdrawals.

They can start reaping the benefits of Translation Memory right away.

translation memory

Amplifying its Power 

Dynamic Machine Learning

With Translation Memory, the user doesn’t need to manually withdraw previously translated words and phrases each time they want to use one. That would take more time than it’s worth.

Instead, its power is often put into motion by a technology called Dynamic Machine Learning. If you haven’t yet heard of machine learning, you will hear a lot of buzz about it in the coming years as artificial intelligence permeates multiple facets of human life.

Dynamic Machine Learning is AI technology that has been incorporated into the most effective Translation Memory applications. It’s responsible for the auto-filling of previously translated words and phrases across a file or batch of files.

translation memory software

When a user makes an edit and it’s stored in Translation Memory, that edit will simultaneously and automatically be applied to any repetition of the segment they edited that exists within the document. If that file is part of a batch of files uploaded together, the edit will be applied within that file and across all the files within that batch. This is the power of Dynamic Machine Learning.

In effect, the user doesn’t need to manually edit the same word or phrase repetitively. This results in significant cost savings due to the reduction of required time and effort.

The more Translation Memories a user creates over time, the better the translation quality becomes and the less input is needed from the user. Eventually, little involvement will be required from the user, as they’ll have adequately trained their machine translation engine.

Translation Memory API

A Translation Memory Application Programming Interface (API) allows developers to integrate Translation Memory with their applications.

One example of a use case for adopting an API is to power a content management system with real-time translations. So, as a user updates their translations, it can update their website content in real-time.

There are many use cases for a translation API, but it depends on the user’s specific needs.

Get Easy-To-Use Translation Memory Software

Need to save time and money on translation? Try Pairaphrase, the web-based translation management system for enterprises. It includes Translation Memory and uses Dynamic Machine Learning to save organizations 50% of the time it normally takes to produce high-quality translations.

Schedule a demo or share this article with a colleague.

Recommended Posts