ModelFront instantly predicts if a translation is correct or not.
en Tesla loses head of product
es Tesla pierde cabeza de producto
Predicted risk
94%
en Wild week on Wall Street
es Semana salvaje en Wall Street
Predicted risk
7%
94%
7%
Machine translation is great when it works - it scales.
But human language is difficult for machines. Machine translation fails often and unpredictably. The results can be nonsensical, misleading, offensive or even dangerous.
Now you can safely auto-approve raw machine translation.
ModelFront risk prediction technology is the key to hybrid translation, quality evaluation and much more.
As billions of users join the internet, platforms expand to support hundreds of languages.
ModelFront loves the chaos. The approach supports all of the top ten thousand language pairs.
ModelFront supports product titles, descriptions and reviews, emails, technical documents, marketing material, subtitles or UI strings.
The production-strength system is built to work for all verticals — from fashion, gaming, social and travel to hardware, automotive and defense.
It works with any translation system, like those from Google, Microsoft, Yandex (Яндекс), Baidu (百度), IBM, DeepL and ModernMT.
And it even works with custom-trained or pre-trained models like AutoML, Fairseq and Marian.
There are many ways for you to apply the linguistic knowledge, innovative research and scalable infrastructure at the core of ModelFront.
Better training data is key to better machine translation.
ModelFront is an easy and robust way to filter parallel corpora for training better custom machine translation.
Are you estimating post-editing effort, comparing engines or training or building your own?
ModelFront correlates with human evaluation and beats BLEU on accuracy and convenience - no reference translations required.
Automated evaluation is a game-changer for everyone from linguists to machine translation researchers.
Do you need to translate at quality and scale?
ModelFront catches critical errors in machine translation to let you balance machine scale and human quality.
Hybrid translation can be orders of magnitude faster than traditional post-editing.
Checking final human translations is almost as much work as translation itself.
ModelFront makes it super easy to sort by risk for quick and targeted final validation.
ModelFront integrations support custom models and adaptive systems like Google AutoML, Microsoft Custom Translator and ModernMT.
Gartner, the world's leading research and advisory company, published its 2020 Market Guide for AI-Enabled Translation Services.
"Advances in artificial intelligence (AI) are enabling a broad range of new use cases, enabling translations that were previously not possible due to cost or delays"
It lists key players in machine translation technology, among them Systran, Google, Microsoft, Amazon, Unbabel and ModelFront.
"Many enterprises are not aware of the emerging AI-enabled translation services and their capabilities, and as a result are not achieving the savings and efficiencies these might offer"
Gartner features ModelFront as the first and only provider of translation risk prediction.
"Value to enterprises: The solution can be used to automate translation post-editing workflows, and aggregate decision-making and offline tasks."
The TAUS Massively Multilingual Conference & Expo 2022 will be held in the San Francisco Bay Area in November.
The panel on cutting-edge machine translation technology brings together leaders from ModelFront, Amazon, Systran, Google and Apple.
John DeNero, the CTO of Lilt, will grill them on massively multilingual approaches, customization marketplaces, risk prediction and more.
ModelFront is easy to try and easy to integrate.
In the ModelFront console you can run on evaluation on a file with just a few clicks. You can get line-level risk productions on your own parallel data or use one of the integrated machine translation APIs.
Create an account or explore open evaluations
With the ModelFront API you can get instant line-level translation risk predictions programmatically and integrate them into your application.
Require a full-service language service provider, translation management system or computer-assisted translation platform?
ModelFront partners make innovative technology accessible right inside their trusted platforms for translation automation.
Talk to GlobalDoc or translate5 or read more about how to access ModelFront
ModelFront's production-strength stack is built on a decade of quality estimation research and the latest advances in deep learning.
Translation risk prediction is the key to balancing human quality and machine scale.
“Quality Estimation (QE) is an important component in making Machine Translation (MT) useful in real-world applications”
“Significantly lowering the barrier to start experimenting with QE technology”
“The feedback from both our clients and our translators is consistently positive.”
ModelFront is built on decades of open research and open-source technology.
ModelFront supports open research on machine translation and quality estimation.
I-Q-Tel Labs partnered with ModelFront to provide top-quality data for the WMT 2020 Quality Estimation Shared Task.
The labeled dataset designed by Nina Lopatina, now VP of Data Science at I-Q-Tel, featured user-generated content - a challenge for machine translation, quality estimation and human linguists.
10000 sentences were translated from Russian to English by a Fairseq model and labelled by ModelFront human linguists according to Facebook AI Research FLORES guidelines.
Better training data is key to better machine translation. The young research lab YerevaNN used ModelFront and took first place in WMT 2020 Russian-English biomedical translation task.
ModelFront openly filters open datasets for the community, from the TAUS Corona Crisis Corpus to the latest conference competition datasets. Open evaluations are free and easy to browse, filter and download.
For WMT 2021, ModelFront has added hundreds of datasets with tens of millions of translations, supporting better machine translation for languages from Afrikaans to Zulu.
Just upload a file to compare and eval engines, estimate post-editing effort, filter your training data and more.
Free credit for 1.1M characters is included with every new account. No credit card is required.
Hit the REST API or just upload files in the console
ModelFront predicts translation fails instantly, so your application can react in real time.
Train on your own data and set your own risk thresholds