One of the best use cases for LLMs is translation1, but as someone who only speaks English I have no idea if the response is bad quality.
What I’m trying now is:
- Ask AI to translate x to x language
- Copy response, and in a new (preferably private) thread ask AI to translate to English
- Compare original English to the new English response
Each translation “pass” gets further away from the original ala playing Telephone so the actual translation (one pass) should be better than the new English response (two passes).2
An example:
ORIGINAL
The house stood on a slight rise just on the edge of the village. It stood on its own and looked over a broad spread of West Country farmland. Not a remarkable house by any means — it was about thirty years old, squattish, squarish, made of brick, and had four windows set in the front of a size and proportion which more or less exactly failed to please the eye.
TRANSLATION
La maison se dressait sur une légère éminence, juste à la lisière du village. Elle se tenait seule et dominait une vaste étendue de terres agricoles du West Country. Ce n’était une maison remarquable en aucune façon — elle avait une trentaine d’années, était plutôt trapue, plutôt carrée, construite en briques, et possédait quatre fenêtres en façade dont la taille et les proportions échouaient plus ou moins exactement à plaire à l’œil.
TRANSLATION #2
The house stood on a slight rise, right at the edge of the village. It stood alone and overlooked a wide stretch of farmland in the West Country. It was not a remarkable house in any way — it was about thirty years old, rather squat, rather square, built of brick, and had four windows across the front whose size and proportions failed, more or less exactly, to please the eye.
The differences make it clear that there’s still a lot of progress to be made!
Footnotes
-
The modern day Babel fish ↩
-
These are all assumptions. ↩