Managing Translation Memory for Consistent Localization

Any effective localization effort starts with managing translation memory. That memory is the tool that all other translations are built on. Without taking advantage of the valuable translation work your organization has already accomplished, you’re going to deal with creeping costs and inconsistent messages across business segments.

Your translation memory (TM) is a cornerstone of any streamlined localization workflow. With it, commonly used phrases and strings are translated easily, leaving the necessary time for linguists to spend on more challenging projects. By looking at translation memory not just as a service but as its own product, you can simplify your efforts and avoid expensive mistakes.

Treating Translation Memory as a Product

When it comes to localization, most managers take the same type of approach to content creation. They end up navigating four basic phases:

‍Phase 1

‍The team sends out files for translation.

Typically, this is a service where the main focus is receiving the translated product and not the translation memory behind it. The person managing this part may not even be aware that translation memory exists.

‍Phase 2

‍Marketing, product, and HR business segments all look to different vendors to further their localization efforts, again with little concern for the underlying translation memory. This creates multiple “sources of truth” which can result in further content segmentation among business units.

‍Phase 3

‍This is often the first time the organization decides to seek out the underlying translation memory created during content localization. Typically, it’s the product team that wants to own it because localization is closest to the product. Unfortunately, this creates further segmentation, as the team doesn’t get the feedback of everyone else who will need access to the same database.

‍Phase 4

‍If teams aren’t aligned, different arms of the organization are talking in different ways. At this point, the company realizes that they must align the content of product teams, documentation, and HR or risk an inconsistent voice.

A lot of the time, problems occur because we don’t understand the nuances involved in our corporate lexicons. In English, our message across departments is consistent and clear.

We expect similar results when we’re undertaking a localization project, but we miss the most important part. The translation isn’t just a service. It’s a product because what you get is a translation memory that you should use to guide all your other content.

‍Translation memory should be an integral focus from phase one on. It isn’t just a tool in the localization project. It is the project.

Centralization as a Cornerstone in Managing Translation Memory

Segmented translation leads to a translation memory that creates inconsistent messages. Taking a centralized approach to translation allows you to build one cohesive language engine that saves you time and allows for process flexibility. Supporting this requires:

• Training: With many different people making changes to language strings, the system can soon get bogged down with competing thoughts. Feedback and consistent updates support an aligned vision and the necessary level of quality. Employees must know who to contact and what to report when they see issues with the translation memory. It’s important that everyone who handles the translation memory is trained in best practices and that select individuals are given permissions and responsibilities to respond to feedback and keep the TM reliably up to date.
• Alignment of teams: Product, marketing, and documentation teams must be prepared to work within the same ecosystem to provide a cohesive user experience through all parts of the sales funnel, from discovery to support. There must be a baseline team consciousness in the translation memory and terminology database.
• Database audits: One of the biggest problems in translation is redundancy—when the same work is done over again. You need to audit your translation memory to fully understand the various strings, sentences, and paragraphs that are repeatable and predictable because you already know how to translate them. Ensuring they’re part of the memory will improve the results of translation assisted by automation so you don’t waste funds on fully manual translation.
• Realistic trust levels: While you don’t want to waste money on unnecessary translation, you do want to ensure that you’ve measured quality in your translation memories. You might have the translation memory tell you that it has found a 100% match, but you can’t necessarily trust the TM without a second glance. Otherwise, you could have a syntax error that will make its way through subsequent translations without ever being corrected. Instead, you should penalize your translation memory when finding errors in syntax and context. Subtract points from the match level to ensure a manual review. As you improve consistency and trust, those levels will go up, and you’ll have greater confidence in your team’s translation memory use.

When the translation memory is wrong, you’ll have to take one of two approaches. You may scrap it or scrub it. Scrapping it means starting from a new database.

Scrubbing is about cleaning it up to ensure more accurate results. Neither is a particularly easy or cost-effective endeavor, which is why it’s so important to get the translation memory right early and to keep it faithfully up to date along the way.

Whether you use a single centralized system or different vendors for all your various tasks, you must make managing translation memory a centralized and consistent part of all of them.

That ensures brand cohesiveness both for the customer and all your employees. It builds a baseline of consistency for your product as you enter new markets around the world, which is a vital component of consumer trust.