EPIC
TAUS EPIC uses intelligent Quality Estimation to filter your translations, ensuring your team only invests valuable human expertise where it is needed.
Quality Estimation (QE) is a technology that predicts the quality of a machine translation. While traditional quality checks are slow, expensive, and difficult to scale, especially where humans are involved, QE provides an instant, objective quality signal.
With massive volumes of content to translate, most companies have no way to distinguish “good” translations from “misleading” ones. QE acts as a quality filter, analyzing your Machine Translation (MT) output to determine how likely it is to be “good enough” for your specific audience.
Our EPIC application evaluates translations in real-time by taking your source text and the machine-translated output to return an objective quality score — 0 to 1, on every segment.
Based on that score you define your own quality gate. Segments above your threshold go straight to production, while risky segments are routed as per your workflow.
There is no single “magic number” for quality. A score that is perfect for a user-generated support forum might be a disaster for a legal contract. At TAUS, we help you define your own quality gates based on your specific industry, content type, and risk tolerance.
The Baseline Model
If you are using a generic QE model, you’ll get a reliable “general purpose” score. It is excellent for high-volume, low-risk content where you just need to know if the translation is readable and safe.
For brand-critical content — like marketing or technical documentation — a custom model is the difference between “readable” and “on-brand”. Because it is trained on your specific terminology and style, the resulting QE score is far more trusted. It understands your unique definition of good.
PROTIP
If you are using a generic QE model, you’ll get a reliable “general purpose” score. It is excellent for high-volume, low-risk content where you just need to know if the translation is readable and safe.
BELOW 0.60
Re-translate
Or re-evaluate the MT engine.
SCORE 0.60–0.84
Human-in-the-loop
Standard post-editing.
SCORE 0.85+
Publish-ready
Automated workflow.
QE isn’t just a technical metric; it is a business tool that allows you to control quality at scale without exploding your operational costs.
Catch misleading or wrong translations before they go live.
Review low-quality segments only.
Stop waiting for manual checks; get instant feedback on the performance of your MT engines.
QE is designed to be an engine-agnostic “neutral judge”. Whether you use Google, DeepL, or Amazon, EPIC sits within your translation management workflows to ensure only the right content receives human attention.
Start scoring your MT quality today.
(reclaim your localization budget in under 5 minutes)