Studies and benchmarks. Measuring instead of claiming
Everyone claims the best translation quality. Almost no one publishes their method. We start there: the criteria, the protocol and the pitfalls, before the results.
See the publicationsPublications
An evaluation method first, then the results it will produce on a corpus of Swiss law.
What we rule out
Three rules that explain why you will not find a spectacular ranking here.
No figures without a corpus
A percentage without a corpus, a date and a protocol means nothing. Until the corpus is described, no figure is published.
Named engines, dated versions
The engines compared are named, their versions are dated, and the faulty outputs of our own engine appear alongside the others.
Accuracy before speed
Speed depends on load and network. It says nothing about the accuracy of a statutory reference, the only criterion that counts in a binding document.