Translation memory management is one of those terms that is regularly thrown around by language industry professionals. It often appears on web pages in a list of benefits that your language service provider can offer you. We’re guilty of it ourselves…!

One of the themes of this blog is to decode some of the jargon and to help translation buyers make an informed decision about who they purchase translation services from. We published an Ultimate Guide to Translation with just this aim in mind. It alluded (very) briefly to the idea of reference management; a concept that will hopefully be fully explained today.

So. Let’s go back to basics and define a few of the key concepts involved in translation memory management:

What is translation memory?

Translation memory refers to a software database containing source language content and the corresponding target language translation. These existing translations can be leveraged for new translation projects in order to speed up the translation process and reduce costs.

How is translation memory stored?

Before you can look at translation memory management, it’s important to know exactly what it is that you might be managing! In the case of Transit NXT, the STAR Group proprietary tool, translation memory is stored as language pairs that can be opened and amended using Transit NXT. Other translation memory tools store translation memory in a database file, often in XLIFF or XML format.

I’m not going to lie – the above paragraphs still contain a fair amount of jargon, so to break down translation memory management even further, I would suggest the following definition.

Translation memory management ensures that any existing translations are of the highest quality possible so that you can gain the most amount of benefit from them.

There. Much better.

Formats for translation memory databases

Although I stated above that translation memory is usually stored as XML or XLIFF, this is not always true. It’s true that translation memory software uses these formats, but for companies or individuals working outside of a tool (Yes, they do still exist!), Excel spreadsheets or CSV files are also workable formats. In this case, translation memory management is therefore about manipulating text stored in columns and rows.

What does translation memory management involve?

As a term, translation memory management covers a few different processes to do with storing and updating translations.

For me, the most important consideration for translation memory management would be to look at the first part: storage.

Storing translation memory

Are you storing your translation memory in a format that can be easily leveraged? If you are still working with XLS or CSV files, these can become unwieldy very quickly and you might not be able to enjoy the benefits of fuzzy matches.

Are you storing your translation memory in a format that can be easily navigated? For example, in the case of Transit NXT language pairs, are you using a folder structure that has a logical hierarchy?

Managing translation memory on a large scale

Here at STAR UK HQ, we have working relationships with customers that span nearly the entire lifetime of the company (over twenty years at time of writing). As you can imagine, we’ve done many millions of words for them, and translation memory management is important because of the sheer volume of reference material available to our team.

We need to ensure that each translation can make use of every scrap of material that we have for that customer, but at the same time, we cannot send several gigabytes of data to our team for each project.

Should we organise translation memory by document types such as manuals, press releases and contracts? Should we organise chronologically? Should we organise by text types such as technical, marketing, legal and financial?

There’s no correct answer to that question. For us, translation memory management is about ensuring maximum leverage of existing material, so we organise by language, then chronologically.

For some customers, we further distinguish between text types, but a customer’s press release may still contain technical terminology so making the technical manual translations available as reference will be helpful for terminology.

So it’s just about a sensible folder structure?

Well, no, not really. Translation memory management is also about ensuring that your reference material is the best possible quality.

What does that mean in practise?

dictionary definition

Terminology preferences

Many of our clients have strict terminology preferences. Sometimes these take the form of approved terminology lists that are sent before translation begins. However, sometimes preferences only come to light when signalled by a customer reviewer.

In these cases, it is important that any disallowed terms or preferred terms are updated throughout the existing reference material so that these are not used for any future translations.

A project manager responsible for translation memory management for that customer will comb through the reference material and will update the translations for every occurrence of the term.

Updating translations based on corrections

Translation memory management also involves correcting translations in the case of errors. Although a thankfully rare occurrence, I wouldn’t be doing my chosen topic justice if I omitted this one.

If an inaccurate translation is suggested as a fuzzy match, it is possible that the error will be included in the new translation and will propagate through the reference material. At this point, it is far harder to resolve as the error may appear in so many locations.

Customer corrections can also sometimes relate to preferential changes (a far more frequent occurrence). We understand this one well – your brand needs to be the same across languages and as the customer, you know it best.

In this case, we need to update the reference material so that we continue to learn what the customer likes and so that they don’t need to make the same correction twice.


Remove duplication

The final task that we class as translation memory management is to remove duplicate, or variant, translations from the database.

The principle of translation memory is that you only work on what is new. So, in other words, if a translation exists, you can use it in your text without needing to start again. However, sometimes the practicalities of the industry get in the way.

For certain customers, workflows and internal deadline pressures mean that certain translation projects need to run concurrently. Where there is any overlap between projects, it is possible that a duplicate translation will be created.

Translation memory management therefore involves finding these translation variants and choosing one translation to use for all future projects. This could sound like a needle-in-a-haystack task, though Transit NXT has a handy variant checker for just these occasions.

Who should be working on translation memory management?

Really, it’s THE question, isn’t it? All of the tasks listed above are important for ensuring that translations are high quality, but potentially they can fall through the gap of where responsibility lies.

Although the customer is best placed to make preferential and terminological changes to their material, they often do not have access to TM tools and the reference material.

In my humble opinion, translation memory management is therefore within the remit of your translation supplier.

Your translation supplier will often have tools at their disposal to simplify some of these tasks, as well as quality assurance checkers to make sure that every instance of a change has been made.