How we made our most read help articles multilingual

We want to support as many people as possible in their language, not only in our administrator dashboard and player interface but also in our Help Center. We aim to have most of the support be self-service. That is only possible if we translate the help articles into our target audience’s languages. But how can you structurally translate everything without adding more work for the current staff? We decided to ask Anouk Perquin, a lean consultant who focuses on localization processes, to run a Quick Scan and help us out. “The way translation processes are usually viewed often leads to unnecessary complexity in processes and automation.”

Quick Scan results

Anouk ran a Quick Scan that focused on these four things:

• The content we need to create and translate and in what order
• The quality that we need to achieve the goal we set for our content
• The speeds that we need to publish all language’s versions
• Our content’s value

We wanted non-English speaking clients to read the translated help articles more frequently. We also wanted to reduce the number of support tickets from non-English speaking countries. Ultimately, we expected that it would lead to a better user experience, which would result in more conversions. The Quick Scan revealed that we considered speed and scalability to be most important for this new process. Anouk explains, “Easy LMS releases new features pretty quickly. Sometimes, they do multiple feature releases a day. These releases have a huge impact on the support content. It needs to be updated quickly and thus translated quickly. It was also very important that we implemented this new process without waste.”

First experiment

Continuous improvement is the standard at our office

Continuous improvement is the standard at our office. We search for the smallest step that gives immediate results, so we can learn from those steps and iterate on that. In this case, that was using Machine Translation. At first, we had a strong preference for human translators over Machine Translation. We worried that machine translation would incorrectly translate referenced user-interface elements. There was also a general belief that Machine Translation would be insufficient for the text.

Anouk tells, “With my experiment, I wanted to prove differently. We translated 25 articles in our top five languages with machines and uploaded them by hand in the Help Center. We collected feedback from the users and looked at what bottlenecks appeared.”

Identified bottlenecks

During the experiment with actual readers and translators, we identified a few bottlenecks that harm the effectiveness of the help articles. We are going to resolve them along the way.

Quality of the source

“The first was the quality of the actual content. It appeared that several sentences were incomprehensible when translated. When I examined these, it appeared the English source sentences were not well-formed and not so clear to the English reader either,” explains Anouk. Based on those findings, we are trying to improve our content creation process. The first step was expanding our writing rules for the source content. We adhere to a style guide better than before. Also, a native English speaker does a final check to catch inconsistencies.

We set up an experiment with automatic screenshots, and the first results are promising

Screenshots

Our help articles contain a lot of screenshots to support the content. According to Anouk, “A support article would be much more helpful if the screenshots were in the correct language. Therefore, I advised Easy LMS to invest in automatic screenshots and dynamic linking, so that they would be updated for all languages automatically, including English.” This bottleneck was a trigger for us. We set up an experiment with automatic screenshots, and the first results are promising. We managed to update a help article with the needed screenshots without manual work. Also, we expanded the experiment to our Features section. The screenshots we use there are also multilingual, but also revealed that the automatic screenshot tool is not stable enough. So, that is our main focus right now.

Software references

We make quite a lot of references to the software in the help articles. We name specific user-interface elements in the tool, but they are subject to change. How to handle this? Anouk suggests, “This is a hard one. If you look into the root cause, you find that the real bottleneck is the fact that the same content is stored in different places. If we were to look at it from a single source perspective, these references should always be connected to the place they originate. So that any change, in whatever language, including the source language, will be updated as soon as they are changed in the software application.” Anouk advised us to do more research into applying the single-source approach. We are currently looking into that.

Making content is too late in the process

Content creation wasn’t part of our development process. It always happened afterwards and took up to two weeks to publish supporting articles and other information. We discovered that it was one of the major obstacles to providing self-service support. Now, our goal is to make the help articles, including screenshots, available as soon as we release new functionalities. We’re closing the gap, but the process still needs refinement.

The result

The experiment was successful

The goal of making our Help Center available in our five main languages was to make our support more self-service. The translated articles should eventually reduce the number of support tickets and improve user experience. Did this experiment find this task to be feasible? Anouk suggests, “Yes, the experiment was successful. We only translated the top five most read articles into the main five languages, and even then, Easy LMS received fewer questions about these topics.”

What’s next? This experiment only included a few articles. Ultimately, our goal is to have our whole Help Center translated to all twelve supported languages. But that isn’t done with just a mouse click (we wish 😅). In the next period, we are going to invest in an API that makes sure the translated content from the Machine Translation Engine ends up in our database automatically, so we don’t have to do this by hand, like in the experiment. Of course, this API should work together with our screenshot tool and our software reference solution. Long story short: it’s a complex and big puzzle we need to solve, but we are very willing to complete it.

Thanks to this experiment, we gained a whole different perspective on the translation process. We took a step back and looked at the whole picture. It will save us time and money. It will also lead to happier support consultants, clients and easier expansion to international markets. This article is based on the client case Anouk wrote on Medium.

Learn more about our multilingual LMS.