Translation API for operating multi-lingual stores

Translation is a hard problem, where the really hard part is solved. You can point your camera at a japanese restaurant menu and get a perfect idea whether you're ordering scrambled eggs or poached eggs.

The unsolved part is personality. George Orwell and Joan Didion both wrote in English and both are excellent but distinct. Just like brands. Nike doesn't communicate the same way as Supreme. Hence, machine translation sounds robotic and not immediately usable in marketing campaigns.

v1 of Wafrow translation going live today is our first crack at addressing this problem. It borrows brand context, tone of voice and grammar from your own landing pages while staying true to branding elements like names, coupon codes, occassions to give translations which do not change that often.

This approach is fundamentally different from our existing, more creatively fine-tuned LLM which is trained on ecommerce best practices that try to generate variants for outperforming the user input on conversions while maintaining ~90% of the intent. This tries to give a fresh answer everytime you ask.

It manifests in beautiful translations of this form:

1. Tone of voice: A call to action text such as Discover more is translated into German as Entdecken Sie mehr on a luxury fashion app that rhymes with Chermes and Entdecke mehr on a mainstream fashion app that rhymes with Orlando. On a spectrum of more to less formal.
2. Brand context: Context is retained during translations that makes running tools on autopilot easier. For instance:

For a fair comparison across tools, Google Translate often gives verbose and overly formal answers not recognizing context. DeepL translations are much more sound and it's a tool I've personally used and recommended to friends. Since Wafrow Translate has access to the existing landing page context and is fine-tuned to only write marketing messages, it outperforms DeepL in a lot of instances. However, Wafrow is optimized only for 6 languages (more incoming soon).

In practice, this is the sort of output you can expect from Wafrow that you cannot from other translation APIs. Notice the usage of #1 and initial case where the first letter of every word is capitalized.

For reference, the Wafrow Creative API response for the same input was: Die Nr. 1 unter den Kochboxen. Same language, just amped up a few decibels for conversions.

The two APIs are available at https://wafrow.com/api/getTranslation and https://wafrow.com/api/getAlternative and you can refer to the API release blog post for documentation.

Please give it a spin and let me know at [email protected] how you feel about this separation.

Disclaimer: HelloFresh is (currently) not a customer of Wafrow, just my past employer, a firm I still hold a lot of stock options in and I've just used them for explanatory purposes.