I consider myself privileged, since I can fluently produce mistakes in several languages. I do this diligently, because I believe content matters more than form. Too many mistakes, however, get in the way of reading. Mistakes need no apology, and I’m grateful that many people generously overlook mine.

How does one improve a text? One can point to different aspects — theological or didactic ones, or linguistic ones. Wording is one thing; grammatical correctness is a genuine challenge in itself.

Today, that last point is also a matter of technology. There are now many good software solutions that support writers. In other words, my weaknesses can partly be caught by technology and specialized systems. That’s already been happening for years. Every text goes through an automatic correction pass. But that isn’t error-free either, as some people are happy to point that out to me.

Besides general language shortcomings, there are other reasons why language sometimes doesn’t land. Not everyone connects with theological terms, and beyond that, there’s the “language of Canaan” — the distinct linguistic culture found in each community. Capturing that language, using it, and putting it to work as a means of communication doesn’t always succeed.

So the particular challenges lie not only in general correction, but also in the needs of the target audience. I want to be understood, and I would like to speak and write in a way that as many people as possible can follow.

Moreover, further challenges appear — for instance, from people stepping out of an evangelical culture. They reject certain wording because they’re directly tied to a particular Christian subculture they’ve already left behind.

Conflicts with language are therefore, in a sense, built in on several levels.

A particular weak point of this website lies in the translations. These aren’t done by hand, but automatically. That requires two components: a system for displaying multiple languages, and a translation method. Both have to build on the current platform and be integrated into it.

Here’s how it works: the website runs on WordPress. Multilingual support is currently handled by WPML, a suite of plugins for WordPress. WPML allows automated translation using several solutions. There are good but expensive options, as well as cheaper alternatives. I use a mid-range solution built on the DeepL.com platform. DeepL produces excellent translations, but they don’t come through properly within WPML.

The problem with the DeepL integration is that WPML only supports translation line by line, which means lines get pulled out of their own context before being translated. Much like with some Bible translations, that results in flawed translations. Ideally, texts would always be evaluated in context, taking into account the type of text and the vocabulary typical of it. It’s really a fair comparison to biblical translation and the challenges that come with it.

Ideally, one would look at the text in context and only then translate it. There are far more ways to do that today than just a few years ago. Right now I’m testing translation with the help of AI. The advantage here is that I can feed specific terms into a glossary in a targeted way, so translations improve. Larger passages, or even entire texts, can also be improved this way. That results in noticeably better translations.

Nothing here is a given. Still, reaching for better results is exciting, everyday work. Does that automatically apply to all texts already published? No, of course not. There are many older texts, some written more than 10 years ago, that were corrected with whatever tools were available at the time and translated later. One would have to regularly comb through every text to improve it. Every now and then, I republish an old text after an appropriate correction.

Speech recognition and automated translation for videos are similarly demanding. Here too there are many options today, mostly built on AI. Often these features are bundled into expensive subscriptions. I use a different approach: MacWhisper is a solution for automatic speech recognition on Mac. It’s connected to DeepL and allows for a fairly good translation. Newer models keep delivering better results. Still, every text in every language has to be checked manually.

Subtitles are essential for people with limited hearing, and often for speakers of other languages too. YouTube’s own data, moreover, suggests that many people like to follow videos with subtitles on.

Subtitles are uploaded as separate text files in a special format alongside the video.

I’ve learned that absolute quality isn’t attainable. Many people seem to know better than I do myself, both theologically and linguistically. From a pragmatic standpoint, that doesn’t concern me much, because moving forward is more promising than alleged perfectionism. It’s also worth remembering that every choice comes with its own set of challenges.

My motto is: Not perfect, but steadily better.