Docs

Technical overview of the ModelFront API

To get started with ModelFront, create an account access to the console and get an API key. Then you're ready to run evaluations in the console and hit our REST API directly.

For your convenience, both the console and the API have an option to use the translation from major translation APIs like Google, Microsoft, ModernMT and DeepL.

If you have an urgent issue or question, you can contact support right away.

Evaluation

API

Running evaluations in the console is an easy way to get risk predictions for a dataset.

You can just upload a file to to filter training data, evaluate and compare engines or just play with the API with no coding.

Hitting the ModelFront API directly gets you instant segment-level risk predictions.

You can integrate ModelFront into your application or machine learning pipeline.

Read our Evaluation documentation
Read our API documentation

Error types

ModelFront is built to catch many kinds of errors, both the common ones and the rare but painful ones, including:

Named entities

Errors translating names of persons, places, organisation and products

Negation

Revered meaning due to missing or inserted negation

Offensive

Offensive or ambiguous translations

Untranslated

Untranslated words or phrases

Noise

Garbage input or output characters, words or phrases

More

Many other types of translation errors, for example translating from the wrong language

Languages

ModelFront is built to support any language pair.

We leverage additional high-scale open data and our own curated data in both our default generic and custom models.

The benefits of open data are greatest for those languages with a sizable open corpus, including but not limited to Albanian, Arabic, Armenian, Azerbaijani, Basque, Belarusian, Bengali, Bosnian, Bulgarian, Burmese, Catalan, Cebuano, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, Galician, Georgian, German, Greek, Gujarati, Haitian, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Irish, Italian, Japanese, Javanese, Kazakh, Kyrgyz, Korean, Latvian, Lithuanian, Macedonian, Malay, Marathi, Mongolian, Nepali, Norwegian, Persian, Polish, Portuguese, Punjabi, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish, Swahili, Swedish, Tagalog, Tajik, Tamil, Telugu, Thai, Turkish, Ukrainian, Urdu, Uzbek, Vietnamese, Welsh and Yoruba.

Customizing

ModelFront returns a risk score, not a label, so you can sort or set your own quality threshold to control the balance of quality, scale and speed for each content type. We recommend using different thresholds for different types of content.

You can use our generic base model or get your own custom models trained on your own monolingual data, parallel data, post-edited data or labelled data.

To get a custom model or early access to the training API, please contact us.