What is "predicted quality"?
ModelFront takes an original segment and translated segment, and returns a score that ranges from 0% to 100%.
0% quality is bad, and 100% quality is good.
In most scenarios, ModelFront predicts post-editing probability.
The probability that a segment will get used as-is vs. post-edited if your professional human translators look at it.
How does ModelFront make predictions?
The prediction is based on your post-editing data.
ModelFront trains an ML (machine learning) model on your post-editing data (original segment, machine-translated segment, final post-edited segment).
It's basically learning the human decision to edit or not.
What languages are supported?
ModelFront is built to be language-agnostic. It's been used for more than 100 languages, from Afrikaans to Chinese to Zulu.
It learns from your data, so if you're translating that language pair, ModelFront can support it.
One model can support all your language pairs. Language pairs can be to English or from English, or between two other languages.
See docs.modelfront.com/#languages for more technical details.
If you can predict quality, can you just give us better translations?
That's harder, but not impossible.
We started with quality prediction because it provides more savings to more companies now.
Ask us about APE (Automatic Post-Editing).