Expert Insights into Translation Memory

Certainly, one value of the Translation Memory is already apparent; a running log of all translations is a useful tool to manage your content in one place. But there is more - continue reading for everything you need to know about translation memory.

Yext
Kyle Robb

Kyle Robb

Customer Success ManagerSmartling

What is the Translation Memory and How is it Used?

Think of your Translation Memory as a library of all of your translated content. This library may contain:

  • Multiple sections: perhaps a dedicated Translation Memory for marketing content and one for legal content
  • Multiple locales: such as English-to-French translations, Spanish-to-English translations, even Chinese-to-Japanese translations
  • Multiple sources: imported TM and built TM from your ongoing translations

Certainly, one value of the Translation Memory is already apparent; a running log of all translations is a useful tool to manage your content in one place.

However, the true value of maintaining a robust, accurate Translation Memory comes from behind-the-scenes action. As you authorize new translation jobs, Smartling’s leverage tooling is constantly reading, re-reading, and reviewing the library of translations you’ve built in your TM.

Here is a use case:

You're publishing some localized marketing material for an upcoming company event, and you have translated for the same event last year. Even if there are changes in your content this year, last year’s translations can still be utilized to reduce costs and the load of effort of translation and speed up delivery. Not to mention you’ll see more consistent translated material providing a smooth and natural experience for your global audience!

How does this all happen? That’s one of the best parts – your translation memory will be utilized behind the scenes in every single translation request in Smartling.

In your Leverage Configuration, you can dictate the read-from Translation Memories. These are the libraries that are referenced for each and every job you authorize. You have the option to prioritize the TM order, cross-reference similar locales (such as European and Canadian French), leverage other projects’ TMs, and customize your SmartMatch behavior. This configuration can be updated and improved at any time, giving you very flexible and direct control over how your previous translations are utilized for ongoing translation work.

Translation Memory Maintenance

Automated Translation Memory leverage and hands-off SmartMatching are powerfully efficient tools, but remember, these tools are only as good as the material we feed into them. To avoid negatively impacting your ongoing translations, it's important to ensure the translations in your Translation Memory are accurate, high-quality, and up-to-date.

One of the most common reasons a Translation Memory falls out of sync with your published content is the external or offline review. Let's say you shared your translations in a Word document with your regional teams, and they made edits directly in the Word file. Now, your TM content does not match your translations that have been put into use in web, print, app, marketing content, and more.

The long-term effect? Your regional teams may have to continue making the same edits over and over again. And for automated content processes, if you’re using a Smartling Connector or our API, these outdated translations from the TM may get automatically published before your teams can revise them.

Here is an example: Your vendor has translated a subtitle string from a blog post: “How Smartling’s Translation Memory Impacts your Bottom Line.” They’ve translated it as “Comment la mémoire de traduction peut-elle avoir un impact sur votre résultat financier,” which is technically accurate French, but it sounds a bit wordy.

And as it turns out, the French content manager prefers to translate “bottom line” as “bénéfice net.” If this is revised directly in the translated content rather than in Smartling, you may see a different translation on the live site such as: “Impact de la mémoire de traduction de Smartling sur votre budget.”

Now, bear with me, as this is a multi-step process. At this point, we have:

  • Source string: How Smartling’s Translation Memory Impacts your Bottom Line
  • Translation in your Smartling TM: Comment la mémoire de traduction peut-elle avoir un impact sur votre résultat financier
  • Translation in your published content (not logged in the TM): Impact de la mémoire de traduction de Smartling sur votre bénéfice net

This mismatch could corrupt the translation quality of future content, even if the exact source string never appears again. This is just one example, but at a grand scale, this disconnect can create an inconsistent in-language experience for your customers, require more effort for your internal reviewer team, and increase translation costs down the road. So keep your TM up to date today to improve your translations tomorrow!

How else does the Translation Memory impact translation quality?

The Translation Memory serves as a valuable reference tool for linguists, even when there is no perfect match available for a source string.

A linguist may not be sure how to handle a colloquialism like “your bottom line,” so they may search for previous examples in the TM.

With this search, they would see the translation for the previously-translated string, and they are likely to use this as a guide. If we are working with an entirely new source string: “Your Bottom Line Depends on Proper Leverage”

You may get translations like this: “Le résultat financier dépend de l'optimisation de votre configuration"

Instead of what you may prefer, based on the translation from earlier (“Impact de la mémoire de traduction de Smartling sur votre bénéfice net”): “Votre bénéfice net dépend de l'optimisation de votre configuration”

The practice of searching for previous translation examples, which aren’t exact matches but still provide context for how a word/term should be translated, is called “concordance search.” This is a great way for linguists to ensure consistency across your translated material, so it’s important that they are referencing accurate translations!

When to Penalize a Translation Memory

In the Leverage Configuration page, you’re able to dictate which Translation Memories are referenced for your Smartling projects.

These TMs may come from a couple of different places: most likely, they are either active TMs (where other projects’ ongoing translations are being saved) or imported TMs from offline or pre-Smartling translation efforts.

There is also a setting on the Leverage Configuration page, which allows you to penalize individual translation memories.

This doesn’t impact SmartMatches, so it is important to know how SmartMatch settings work. What the penalty will do is reduce the “fuzzy match” percentage for strings with an apparent TM match in the penalized TM.

Take a look at this example:

Let’s say you translated “A translation memory is a saved record of previously translated content” in 2018. The translation is a few years old, and your company’s termbase may have changed by then. Perhaps you have a new, preferred translation of “translation memory.”

Today, you’re authorizing a string that is quite similar to the old one but just a bit different: “A translation memory is a saved record of all previously translated content.”

Notice that there’s one new word in the string, so this is not an exact SmartMatch. Rather, this string is now a 94% “fuzzy match.” This is helpful because typically, matches like these will cost less to translate, and the linguist can still reference and utilize the TM entry!

The risk here is that the linguist is referencing this match, understanding that it is in the higher range of match percentages. Should they adjust the terminology or just add “all” into their translation? Very often, they will use the previous translation as a template and make only minor adjustments needed to make the new translation accurate.

Suppose you’ve lost faith in your 2018 Translation Memory (perhaps it was before a company rebrand or it was imported from an old vendor who didn’t produce quality work). In that case, you can add a percentage penalty to this TM while still making use of it for fuzzy matches.

Adding a 10% penalty to this TM will reduce the 94% match to an 84% match, thereby instructing the linguist to adjust their confidence in the prior translation accordingly.

Want More Tips?

Do you have further questions or want to continue the discussion about Translation Memory? Contact your CSM today or visit our help center to find more information on any topic related to your Smartling account!