Web Development

Enhancing Translation Efficiency and Quality: How we are utilizing DeepL

Being a Swedish agency, localization is near and dear to us. Not only because we have to suffer bad attempts at it into Swedish, but because we know how important it is to engineer for multilingual support.

Over the years, we have done projects that called for easier localization, such as a company website in two languages (in our case often English and Swedish). Other projects have required much more: A online magazine in 15 languages. A cross-European organization with a localized website. An internal application for a global company with 7 languages.

The challenge of localization

Technically, adding languages to an application is trivial. There are indeed considerations and things to think about ahead of time. Once you have done so, technology isn’t stopping you.

The first thing I tell our clients when they want to support multiple languages is the required workflow and the investment that goes with it.

Do you think you aren’t moving fast enough now? You are definitely not moving faster when everything needs to be localized. For many, supporting multiple languages ends up being a pain.

The added process might be feasible with a website but less so when you are developing a SaaS application. You will quickly end up either:

  1. Shipping before you translate leads to a fragmented user experience for users not using the default language.
  2. Not shipping until translations are done leads to longer lead times.

You would be surprised at how many changes we make to an application that involves changing a text string or adding a new one. When you go multilingual, this will require extra steps.

Even with tight integration with a great translation agency, you’ll be forced to move slower and take the additional investment.

Machine translation is getting good

Enter machine translation. Historically it hasn’t been viable. The quality has been lacking; more creative copy in an industry context just hasn’t worked well.

Over the past few years, this has changed so much that we have tightly integrated machine translation into our solutions.

For websites and our Company Cloud platform on WordPress, our BM Concierge service integrates a one-click translation directly into the editor.

For applications and tools, our autotranslate package for Laravel makes it easy for us always to ship an up-to-date localized version.

While favorite consumer tools such as Google Translate (or Microsoft’s Bing Translate) aren’t bad, I’ve found DeepL to be better. It does a great job of localizing any content into great language. It also has a solid API for integrating.

Having tried it at everything from easy interface copy to in-depth articles with industry jargon, I’m surprised at how well it manages. Most mistakes it makes are because of ambiguous phrases or context sensitivity, where it needs to be fed more details.

A successful AI-assisted workflow

The key to adopting an AI-assisted machine translation workflow is to look at it in two stages:

  1. Bulk translation
  2. Proofing and editing

Let the AI machine translation service do the bulk work, and do the required light editing afterward. Even better, combine the proofing with something like Grammarly and have assisted editing.

Our workflows differ slightly depending on where the copy ends up.

Some of our clients opt not to proofread and edit, relying solely on the machine translation quality. This strategy can be useful for applications or SaaS platforms where moving quickly is important and proofing or editing lots of languages is worse than an occasional less desirable translation.

Most, however, opt for the combination. This is even more important with marketing copy on websites, presentations, and more.

This strategy can be even more powerful when dealing with languages that you and your teams master internally. For example, work that we do in English and Swedish.

We could easily manage the translations ourselves. By running the text through DeepL and doing light editing and proofing, we free ourselves from spending time translating and can add more value instead.

PS: This post was written in English but is also available in Swedish on our Swedish website using the strategies from this article.