Zendesk on Zendesk: How we translate KCS articles

Zendesk on Zendesk is a discussion about a specific topic and how the Zendesk Support team uses Zendesk products. Each session is hosted by a member of our Support team.

In this Zendesk on Zendesk, Mackenzie a Self Service Specialist explains how our KCS (Knowledge-Centered Service) content is localized (translated) and how we use Zendesk tools and integrations to optimize the localization experience.

See all of the Zendesk on Zendesk series discussions.

How we do it 

At Zendesk, we utilize a mixture of human and machine translation, as well as an integration that is maintained by various teams built to automate the translation process.

Our workflow process is that the top 20% of English content with high page views is translated by both machine translation and humans. The bottom 80% of English content that has lower views is translated by machine only. Here we are using a Knowledge-Centered Service methodology technique called the 80/20 rule, this rule of reuse states that 80% of the issues are resolved by 20% of the knowledge base. Because the top 20% of content has the most resolution it takes priority of human translation over machine translation.

Translation processing is done by trained machine translation engines and human translation editing. The machine translation engines are trained using previously translated and reviewed content. Because of this when visiting an article in the help center, the article will provide the following disclaimer: 

“Translation - disclaimer: This article has been translated by machine translation software to enable a basic understanding of its content. Reasonable efforts have been made to provide a correct translation, but Zendesk does not guarantee the accuracy of the translation. If you have any questions about the accuracy of the information contained in the translated article, consult the English version of the article, which represents the official version.”

Why we do it 

Zendesk provides localization of the user interface automatically allowing us to host and share translated articles. Because Guide does not provide automatic translation of the articles, we have created workflows to best provide localized support. 

There are many reasons why we do things the way we do. Speed of the integration, our Advice and troubleshooting sections of the help center, contains an estimated 1167 English articles that are continually growing. Using machine translation speeds up the translation and translation update process. This also means that we can minimize the use of a human translation that has its additional costs. 

The process  

Note: Our workflow is specific to our use case and involves custom integrations. For more information on process and translation services in the Zendesk Marketplace, see the article: Creating and managing translated content for our Zendesk Guide knowledge base.  

Each week at a specified day and time our automation pulls new and recently edited content from our Advice & troubleshooting sections of our help center for translation. 

The top 20% of English content will be translated over a 7 day period. During this time the content is first passed through a Translation Management System, where the content is translated using machine translation post-editing. MTPE in short is the process of using AI to automatically translate your content from the source (one language) to the target (another language) without the assistance of human input. Once the machine translation is completed the content will get a second edit for correctness by human translators. Our integration then pulls for newly created content every 2 hours, then publishes the newly translated articles to our knowledge base. 

The bottom 80% of English content is translated by machine translation only. Once translated our integration will process the updated content and post it to our knowledge base.

Although a custom integration is used for automated translation processes, there are instances where article translations may need updates due to translation quality. For that reason, we advise our Customer Advocates to flag the article. For more information on flagging, see the article: Flagging articles in Knowledge. 

Translation and content standard

Our content standard defines how articles should be written while in the tone of a unified voice. No matter the language we need our content to translate correctly. At Zendesk we have guides within our content standard to ensure seamless translation for articles and other content. See examples of our content standard guides for localization below:    

• Avoid using metaphors and analogies, they are complicated concepts for machine translation.
• Do not use unessential phrases and extra words.
• Avoid strings of nouns, noun clusters don't work well with machine translations.
• Do not use an exclamation mark or ellipsis as they transmit emotion, and machine translation doesn't understand emotions.

To read more on how Zendesk Support uses Zendesk products, see all of the Zendesk on Zendesk series discussions.