So far, we have covered 50 different languages in our projects. We are able to deliver well over 100 languages in high quality. We are happy to receive a list of the desired language pairs and give a concrete statement on the subject of the project.

Through our API architecture, we have ensured that your models can be connected quickly and reliably in all relevant systems. Connections implemented so far range from content management systems, e-shops, product information systems to office software such as Word and Excel.

Of course, with our separate web frontend, you always have the opportunity to use the functionalities and your models across systems.

To train the language models, we extract sentence pairs in the source and target languages.

We obtain these pairs of sentences from the existing publications on the company’s websites, from publications, databases, translation memory systems, glossaries and much more.

We are able to process a wide range of formats and data and make them usable for training.

Most companies have enough language data from the past for training.

We can collect this language data and make it usable.

Through our processes, we are usually able to win 10,000 to 300,000 sentence pairs per language.

In the training projects, we check at the beginning for which languages enough training data are available and communicate the status early, so that transparency is available from the start.

The quality of translations and their improvement can be calculated and expressed mathematically. Of course, we deliver that.

In our experience, the acceptance of specialist departments, countries and translators is more relevant.

That’s why we’ve created an evaluation environment where your company’s employees can test and evaluate the trained models.

Compared to the previous manual translations and also compared to other untrained generic translation services. The qualitative evaluation by your colleagues is then the basis for deciding which trained models are ready for translation.

Translation memory systems are designed to help translators and language managers deliver consistent translation quality. Furthermore, TMS are intended to reduce translation costs, because strings that have already been translated are not resubmitted for translation.   Instead, when there is a sufficient character‑level match, the previously translated string is returned.

This should help reduce translation costs and keep translation quality constant. And this approach makes sense when many manual and changing translators work with the knowledge.

wonk.ai language models combine the previous translation knowledge (like a TMS) with the feedback of the proof readers. This means that a translation memory system is not needed for many applications.

Translation pricing is based on the number of trained language models and the total number of words translated. On average, wonk.translate is 80% cheaper than previous translation solutions.

Machine translation is always useful when many translations in many languages are needed in a short period of time.

This is the case, for example, when global content campaigns are to be rolled out or when changing CMS and relaunching websites. Whenever a lot of new content is created or edited and is to be translated in a short time, machine translation is particularly useful.

For context:
A typical rollout for a website relaunch—assuming 400 content pages per language and 10 target languages—amounts to about 2,000,000 words (2M). That’s a substantial volume where getting support makes sense.

Absolutely. With wonk.ai you can comfortably and quickly translate PDF files. You can translate a variety of file formats either via the API or directly in the web frontend using drag‑and‑drop. .pdf format and many more.

Automatic translation also works for classic Office files (Powerpoint, Excel, Word) as well as for specific formats, e.g. InDesign prepress.