Real-time QE scores per segment via REST API — MQM, BLEU and TER compatible. Human post-editing lands where it counts: significant QA savings on large volumes.
QE scores that give your post-editors their time back
Real-time quality estimation per segment via REST API: high-risk sentences are flagged for human post-editing on the spot, safe segments pass through untouched. MQM, BLEU and TER compatible.
QE is not a replacement for human quality control — it is a smart way to deploy that
control where it counts. By routing low-quality segments straight to post-editing and
processing high-quality segments faster, you raise the efficiency of your entire
translation process. Our clients achieve significant QA cost savings — without
compromising publication quality.
Language reach
Quality estimation in 225+ languages
From core language combinations to low-resource markets — every language pair gets its own QE scoring profile.
You deliver source segments and machine translation output via API, file or our platform. Integration with mainstream TMS solutions (Trados, memoQ, XTM, Phrase) is supported as standard.
02
Quality estimation
Our QE model analyses every translated segment and calculates a quality score based on metric (MQM/BLEU/TER) and linguistic parameters — without any reference translation.
03
Risk-based ranking
Segments are ranked by quality score. High-risk segments that require human post-editing are identified upfront with an explicit risk indicator.
04
Targeted post-editing
Only the lowest-scoring segments are routed to a native post-editor. That cuts QA cost significantly with no loss of publication quality.
05
Reporting and insights
You receive a complete quality report: per-segment scores, aggregate statistics, deviation patterns per engine and recommendations for further workflow optimisation.
Data-driven translation workflows
QE turns guesswork into focused expertise.
Without quality estimation, your post-editors review every single MT segment — or guess where the errors hide. QE assigns a hard score per segment, so human time lands on the segments that genuinely deserve it. No shift in quality, but a meaningful shift in cost — and a faster path to publication.
Predictive quality scores so human expertise lands exactly where it should — without any reference translation.
Real-time quality scores
Through our REST API you receive real-time QE scores per segment — directly usable in automated translation workflows and TMS integrations.
Significant QA savings
By post-editing only the low-scoring segments by hand, you cut post-editing and QA cost significantly — without sacrificing publication quality.
MQM, BLEU and TER
We support the industry-standard quality metrics MQM, BLEU and TER for full compatibility with your existing QA processes and reporting.
225+ languages via API
QE available for 225+ language combinations via REST API — direct integration into your translation pipeline, TMS or automated content workflow.
Quality assurance
Quality estimation built on industry standards
From MQM to REST API — the foundation of a translation workflow that decides on data.
MQM · BLEU · TERIndustry-standard metrics
Real-time REST APIDirect in your TMS
Significant QA savingsTargeted post-editing
225+ language combinationsFull coverage
Human-in-the-loopExperts on risk segments
GDPR-alignedDatacenter configurable on request
From practice
Concrete QE projects
From e-commerce pipelines and SaaS localisation to large-scale technical documentation.
01E-commerce · Marketplace
Case Study
E-commerce MT pipeline — 2M words/month
A marketplace processes 2M words of MT per month into 14 languages. With QE, high-scoring segments are published directly; only low-scoring segments go to a post-editor. Significant cost reduction.
2Mwords/month
14languages
lowercost
02SaaS · Localisation
Case Study
Software localisation — 85k strings on a continuous cycle
A SaaS provider with 85k UI strings in 20 languages. QE detects strings with high UI-break risk; only those go to human QA. Release cycle cut in half.
85kstrings
20languages
×2release
03Industry · Tech docs
Case Study
Tech documentation — 18 languages
An industrial manufacturer keeps technical documentation translated into 18 languages on a rolling basis. QE decides per segment which MT output is publication-ready and which must go back to a specialist. Lead time down 60%.
18languages
3kpages/yr
fasterlead time
Applications
For which workflow?
8workflow types
QE shines on high-volume MT output — from e-commerce to software localisation and technical documentation.
MTPE workflow optimisation
Technical documentation at scale
E-commerce product catalogues
Software localisation strings
Legal and financial documents
Pharmaceutical content
Website content at platform scale
Automated translation pipelines
Trusted by government, legal institutions & global enterprises
HPMinistry of JusticeDSMSiemensASMLAmazonINGCalvin KleinRocheShellEuropean Court of JusticeBoschBMWPhilipsAudi
HPMinistry of JusticeDSMSiemensASMLAmazonINGCalvin KleinRocheShellEuropean Court of JusticeBoschBMWPhilipsAudi
What is the difference between quality estimation and quality evaluation?
Quality estimation (QE) rates the quality of a translation without a reference translation — purely on the basis of source and target text. Quality evaluation (QA) compares a translation against a human reference translation. QE is therefore better suited to automated workflows where assessment has to happen quickly, without an additional reference.
Which quality metrics do you support?
We support the standard industry metrics: MQM (Multidimensional Quality Metrics) for detailed error analysis, BLEU for automated n-gram match scores and TER (Translation Edit Rate) as a measure of edit distance. On request we also integrate sector-specific metrics.
How do I integrate quality estimation into my workflow?
Through a REST API that integrates seamlessly with mainstream translation management systems such as Trados, Phrase, memoQ and XTM. We guide you through the technical integration and connect QE to your existing QA processes. Custom integration with your own CMS or content pipeline is equally possible.
For what translation volume does QE make sense?
QE delivers most value from roughly 50,000 words per month upwards, where automated routing produces a measurable ROI. We are happy to advise — without obligation — whether QE is a sound investment in your situation, including a pilot test on a sample batch to measure the actual savings.
Does QE work for new or rare language combinations?
Yes, although accuracy varies by combination. For high-frequency language pairs (EN-DE, EN-FR and so on) QE scores are highly reliable. For rare combinations we can train a language-pair-specific QE model on your data — that raises accuracy significantly.
Is QE a replacement for human review?
No. QE is a prioritisation layer — it decides which segments go to human review. Our native post-editors remain responsible for publication quality on the segments QE has flagged. QE reduces the volume that humans see, not the quality bar.
How does your pricing model for QE work?
Rates per 1,000 segments, with tiered discounts from 100,000 segments per month. QE-only (scoring and reporting) is more cost-effective than QE plus integrated MTPE. Custom model training for rare languages is quoted separately. Pilot tests run at an introductory rate so you can validate the business case.
01What is the difference between quality estimation and quality evaluation?
Quality estimation (QE) rates the quality of a translation without a reference translation — purely on the basis of source and target text. Quality evaluation (QA) compares a translation against a human reference translation. QE is therefore better suited to automated workflows where assessment has to happen quickly, without an additional reference.
02Which quality metrics do you support?
We support the standard industry metrics: MQM (Multidimensional Quality Metrics) for detailed error analysis, BLEU for automated n-gram match scores and TER (Translation Edit Rate) as a measure of edit distance. On request we also integrate sector-specific metrics.
03How do I integrate quality estimation into my workflow?
Through a REST API that integrates seamlessly with mainstream translation management systems such as Trados, Phrase, memoQ and XTM. We guide you through the technical integration and connect QE to your existing QA processes. Custom integration with your own CMS or content pipeline is equally possible.
04For what translation volume does QE make sense?
QE delivers most value from roughly 50,000 words per month upwards, where automated routing produces a measurable ROI. We are happy to advise — without obligation — whether QE is a sound investment in your situation, including a pilot test on a sample batch to measure the actual savings.
05Does QE work for new or rare language combinations?
Yes, although accuracy varies by combination. For high-frequency language pairs (EN-DE, EN-FR and so on) QE scores are highly reliable. For rare combinations we can train a language-pair-specific QE model on your data — that raises accuracy significantly.
06Is QE a replacement for human review?
No. QE is a prioritisation layer — it decides which segments go to human review. Our native post-editors remain responsible for publication quality on the segments QE has flagged. QE reduces the volume that humans see, not the quality bar.
07How does your pricing model for QE work?
Rates per 1,000 segments, with tiered discounts from 100,000 segments per month. QE-only (scoring and reporting) is more cost-effective than QE plus integrated MTPE. Custom model training for rare languages is quoted separately. Pilot tests run at an introductory rate so you can validate the business case.
Social proof
Client testimonials
What clients say about working with Ecrivus — from marketplace MT pipelines to SaaS localisation.
“
★★★★★
Certified translations for our international cases are delivered quickly and carefully. Our project manager knows our account inside out.
01 / 03
Need AI quality estimation?
No-obligation — response within one hour on business days