AI to check and fix translations
ModelFront provides AI to accelerate translation and keep human quality.
Under the hood, two types of private custom large language models (LLMs) work together.
- Quality prediction, to check translations
- Automatic post-editing, to fix translations
These two models are integrated together seamlessly, but transparently.
Quality prediction
The quality prediction model checks if a translation needs a fix or not. Verification, not generation.
The input is a sentence and its translation, for example a machine translation. The output is a boolean flag (✓ or ✗).
| Input | Output |
|---|---|
Hello, world!¡Hola, mundo! | ✓ |
One two three¡Catastrófico! | ✗ |
Automatic post-editing
The automatic post-editing model fixes a translation. It generates edits.
The input is a sentence and its translation. The output is a new translation.
| Input | Output |
|---|---|
One two three¡Catastrófico! | Uno dos tres |
The system then checks the new translation with the quality prediction model.
FAQ
Which languages does ModelFront support?
ModelFront supports hundreds of languages and locales. ModelFront was fundamentally built to more than 100 languages from day one.
Does ModelFront provide generic models?
No, ModelFront provides private custom models only, that keep human quality. ModelFront does not provide generic models.
Which foundation model does ModelFront use?
ModelFront is foundation model-agnostic. We're constantly pulling in better foundation models into our own infrastructure, and modifying, training, deploying and testing them.
Does ModelFront use ChatGPT?
No, ModelFront does not send data to external, third-party APIs to models like GPT. ModelFront models are deployed within ModelFront infrastructure only.