Expectations for Translation Memory Reuse

One of the biggest benefits of centralizing translation management is the reuse that is gained through translation memory. If you’ve never used translation memory before, it means that you no longer waste money translating the same sentences over, and over, again. If you have used translation memory, but in a decentralized way–maybe with separate translation memories in the hands of multiple language service providers–it means that you will increase the amount of translation that can be leveraged, across content types and regardless of the translators and language service providers who are providing your translations.

90% Translation Memory Reuse

I’ve seen statements about translation memory reuse reaching as high as 90%, but I would like to caution anyone from expecting reuse to be that high. It depends on the content that is being translated, how it is written, and how unique or repetitive the content is, if pattern-matching has been recognized as a need and implemented, and on how correct the source content is in the first place. Punctuation matters: a comma here, or a colon there, makes a difference.

“Reuse” is any content that is either totally identical to, or very similar to, content that has previously been translated and stored in a translation memory. It means that the translator has little, if any, need to change or correct the translation in the context of a new translation job.

For example, if the sentence “Switch on the power” appears in both a user manual and a Quick Start guide for 10 different products, or if the heading, “Free shipping!” appears multiple times on a website, it’s reuse. The translator does not need to translate it again, each and every time it appears.

Some content provides opportunity for very high reuse rates, due to the way in which it is written and structured. Technical documentation, for example, will yield high reuse, as it is typically written as step-by-step instructions, with short sentences, clear statements, and factual rather than emotional content. A product catalog will also see very high reuse rates, as the statements are simple descriptions of functionality and design.

Reuse can be enhanced further using pattern matching for content that changes, but which changes according to defined rules. For example, the sentence “Our new product XB123F-G” would become “Our new product YC456G-H”, and the only thing that has changed is the product number. If the translation technology can identify the pattern of the product number, and ignore it, then the text “Our new product” is matched from translation memory, and does not need to be retranslated. This is great news for companies who are translating content for many products with many releases and updates.

Emotional, Free-form, and Descriptive Content

However, the same levels of reuse should not be expected for content that is more emotional, free-form, and descriptive. Marketing and advertising content is often a play on words: sentences are long and convoluted, or short and non-grammatical. This is the intent of the content: it makes the reader take notice, or it resonates with the customer’s desires and needs. It’s attention-grabbing rather than instructional. This type of content is less repetitive, and therefore less likely to provide very high translation reuse rates. But–and this is a big “but”–marketing and advertising translations absolutely benefit from other aspects of centralized translation management, such as consistency of terminology and style.

I know of at least one instance where a company’s product was returned by customers, because the functionality description was translated differently in the product’s marketing and on the product’s packaging. Neither translation was incorrect: it was simply a case of two translators choosing a different way to express the same concept, correctly, in their language. This would have been avoided, had the two content types shared the same glossary and leveraged from the same translation memories. That company no longer risks this happening: they have centralized all translation management technology and process.

Centralized Translation Management

I know of many instances of a company choosing to translate website content several times, because several teams were repurposing the same content for different media, for web use and for print. The company no longer wastes money in this way: they have centralized all translation management processes. This was highly descriptive marketing content, which can result in very different translations if done by different people.

I also know of a situation where very unique content, for a marketing newsletter, was being translated through a centralized translation management process, using company-branded terminology, and leveraging central translation memories. The translation memory reuse was really low, 5% at best, and yet the process provided enormous benefits. The news articles were always on-brand, and due to the automated workflow, translators were able to turnaround an extremely high volume of marketing translation and transcreation in a very short amount of time.

Depending on the amount of translation being purchased, a very small increase in the percentage of translation memory reuse can result in an annual saving of thousands, or tens of thousands, of dollars. For a large company, an extra 1% or 2% translation reuse, created through centralizing translation memories across product lines, can provide an enormous ROI.

Three Decision Drivers of Translation

We used to talk about translation decisions being based on three drivers, and choosing any one of three:

  • cost
  • quality
  • time

This model is no longer true. We have to deliver all three. This is how centralized translation technology helps achieve the goal of “all three”.

Type of content Cost Quality Time
Marketing Low to medium translation reuse. Consistent voice, style and brand. Consistent terminology. High-speed workflow.
Technical and user documentation Maximum translation reuse. Building large translation memory for potential training of MT (machine translation) service. Possible integration of MT for post-editing by professional translators. Consistent voice, style and brand. Correct structure of product names and descriptions in translation. High-speed workflow.
Support content Maximum translation reuse. Self-solve reduces cost of local call centres. Consistent terminology across product lifecycle. High-speed workflow.
Outbound social media Minimal translation reuse. Consistent voice, style and brand. Consistent terminology. High-speed workflow.

Setting Expectations

Translation memory reuse builds over time. If a translation memory has never been used before, you begin with zero reuse. A good translation management system will enable reuse to start accumulating immediately: it will identify any identical sentences in the same translation project (batch, document, job), and will make sure it’s not translated multiple times.

The first year of using a centralized approach to translation memory management across all types of content–marketing, product, technical–reuse rates achieving 30-40% are probable by the last months of the year. Two or three years later, it might be as high as 50%-60%. If the content is mostly technical, and if the content is authored using a restricted vocabulary or “Controlled English”, reuse will be higher still.

If all the content is user-generated and/or very emotional, descriptive and flowery marketing content, the reuse rates will be lower, but the benefit of a streamlined, automated process will be providing other, undeniable benefits.