If you work in localization, you’ve probably felt it: the pressure to ship more languages, more content, in less time, without sacrificing quality or your sanity.
In a recent episode of The Agile Localization Podcast by Crowdin, host Stefan Huyghe sat down with Jaap van der Meer, Founder and CEO of TAUS, to talk about how TAUS’s new system, EPIC, is changing that equation. Together with its integration into Crowdin, EPIC promises up to 80% efficiency gains in MT-heavy workflows, while turning quality from a vague cost center into hard business intelligence.
Listen to the new episode on:
From Data Think Tank to Quality Estimation Pioneer
Most people know TAUS as a think tank and a massive data hub. Since 2008, they’ve been collecting translation memories from across the industry to help others build better MT engines.
At one point, TAUS had a choice: use that data to build their own MT system and compete with their customers, or find another way to unlock value. The turning point came when Uber asked TAUS a different question: “Can you build a quality estimation (QE) model that tells us which MT output is safe to show to users?”
That project became the seed of EPIC. Instead of producing translations, TAUS focused on judging them, building models that score MT or LLM output segment-by-segment and tell you whether it’s good enough or needs human help.
Fast-forward a few years, and now Gartner lists TAUS as one of only two pure-play quality estimation providers globally.
Why QE Is More Than Just ‘Better MT’
A common question arises: If QE models can spot errors, why not simply train the MT systems to stop making them? According to Jaap, MT and QE serve entirely different functions.
- MT models generate translation output.
- QE models evaluate that output with the sole purpose of detecting errors.
Jaap compares it to traditional workflows, where translators and reviewers have distinct responsibilities. QE is the machine equivalent of a reviewer, an independent layer that checks the work of MT or LLM engines. And because QE sits above the MT layer, it works across any engine, providing a neutral quality standard.
How EPIC delivers 80% efficiency gains
EPIC’s value becomes clear when you look at how it processes content:
1. Quality Estimation (QE)
EPIC scores every segment generated by MT or an LLM. On average, about 50% of these segments meet the required quality threshold and can be auto-approved and locked in Crowdin. That means half your content can bypass human post-editing entirely, your first 50% efficiency gain.
2. Automatic Post-Editing (APE)
The remaining 50% is automatically sent to a Large Language Model with specialized prompts designed to improve the translation. After this automated refinement, another 30% of the total content reaches the quality threshold.
Together, these steps result in roughly 80% of content requiring no human post-editing. And because EPIC and Crowdin connect through a single API call, these enhancements fit into the existing workflows. Try TAUS QE & APE integration in Crowdin:
Teaching EPIC Your Brand Voice
EPIC supports 100+ languages with generic models, but quality is rarely one-size-fits-all. Organizations often have specific requirements: terminology, tone, regulatory conventions, or regional variants.
To address this, TAUS fine-tunes custom QUE models using:
- Branded translation memories
- Glossaries
- Style guides
- Domain-specific data
Their NLP team even generates negative examples to help the model recognize incorrect translations. This level of tuning is crucial for linguistic nuances (think Canadian French vs. European French) or specialized industries like legal or pharmaceutical content.
This creates a new opportunity for specialized LSPs and linguists: instead of merely editing text, they can help train, validate, and maintain AI-driven quality systems. Jaap calls this the rise of AI service integrators within the localization ecosystem.
What This Means for Localization Teams and Linguists
For localization leaders:
- 50% immediate savings through auto-approval.
- 30% additional savings via automatic post-editing.
- Continuous quality visibility.
- Faster feedback loops and smarter routing.
For linguists:
- More meaningful work focused on nuance and expertise.
- New roles in data validation, model tuning, and AI-integrated workflows.
- Opportunities to become specialists in a domain where quality still demands deep human insights.
- Content volume will continue to explode. The key is deciding which content needs a human touch and which can be safely automated.
Jaap’s Background
Jaap van der Meer is the Founder and CEO of TAUS, a leading language technology company specializing in translation quality estimation and automatic post-editing solutions. With over two decades of experience in the localization industry since founding TAUS in 2005, he has established the company as one of only two pure-play quality estimation providers globally, as recognized by Gartner’s 2024 market analysis.
Listen to the new episode on:
Localize your product with Crowdin
Yuliia Makarenko
Yuliia Makarenko is a marketing specialist with over a decade of experience, and she’s all about creating content that readers will love. She’s a pro at using her skills in SEO, research, and data analysis to write useful content. When she’s not diving into content creation, you can find her reading a good thriller, practicing some yoga, or simply enjoying playtime with her little one.