In the dynamic world of localization, ensuring consistent and high-quality translation has always been a top priority. At Translated, we’re not just aiming for “good enough” – we’re actively redefining what “good” means by developing robust systems and metrics that provide transparency and drive continuous improvement. For those familiar with the industry, you’ll know that traditional metrics often fall short. That’s why we’ve pioneered a more practical and human-centric approach, focusing on how quality directly impacts efficiency and, ultimately, flawless communication.
How quality gets monitored at Translated
Our commitment to quality is woven into every step of our process, from selecting the right linguist to delivering the final product. Here’s how we ensure that our translations consistently meet and exceed expectations:
T-Rank™: The brains behind translator selection
Before a single word is translated, we leverage our proprietary AI system, T-Rank™, to select the absolute best human translator for your specific project. Think of T-Rank™ as an intelligent matching engine. It analyzes over 30 factors for each incoming project, including:
- Content of the document: What’s the subject matter? Legal, medical, marketing?
- Translator’s expertise: Does the translator have a proven track record in that specific domain?
- Quality and timeliness of previous jobs: How well have they performed on similar projects in the past?
- Daily turnaround: Can they meet the deadlines?
- Availability: Are they ready to take on the work?
T-Rank™ creates a ranked list of translators, with the top-ranked individual being statistically the best fit for that particular job. It’s a meritocratic system that ensures the right expert is always on the job, and it constantly learns from feedback from project managers and clients to make even better choices in the future.
Projects powered by Matecat: Our collaborative environment
Once a translator is selected, they work within Matecat, our advanced Computer-Assisted Translation (CAT) tool. Matecat is designed to enhance human-AI collaboration and streamline the translation process. Here, we utilize Time to Edit (TTE), which quantifies the actual time a professional translator needs to refine a machine-translated text to a human-quality standard. By focusing on TTE, we shift the focus from abstract scores to a tangible metric that directly impacts project efficiency and cost, measuring the real-world effort required to achieve excellence.
Matecat tracks the time a translator spends on each segment, providing real-time insights into the post-editing effort (PEE). This data is crucial for:
- Measuring TTE: As mentioned, TTE is calculated as the total time a translator spends post-editing a segment divided by the number of words.
- Continuous improvement: The system collects feedback on suggested translations (whether from machine translation or translation memories) and uses this data to refine our AI models, specifically our advanced Language AI, Lara.
- Internal Quality Assurance (QA): Matecat has built-in QA functionalities. It automatically checks for common errors like punctuation, numerals, links, and spelling. It also identifies “translation conflicts” where the same source segment might have different target translations within the project, allowing translators to address inconsistencies. Our internal QA processes, often involving a second native translator (reviser), provide detailed feedback through Matecat’s Quality Report, which includes:
- A summary of job data, including final quality score, average TTE, and reviewed words.
- A detailed breakdown of all issues found and their associated “error points.”
- Segment-level details, showing the history of edits and the origin of suggestions.
This feedback loop ensures that any issues are identified and addressed promptly, leading to higher quality outputs and ongoing improvement for both human linguists and our AI.
Error per thousand (EPT): A granular look at linguistic quality
Beyond TTE, we also utilize a more granular metric: Errors per Thousand (EPT) words. This metric provides a detailed assessment of linguistic accuracy and fluency. During the quality review process, errors are categorized (e.g., mistranslation, terminology, grammar, style) and assigned severity levels (minor, serious, critical).
- How it works: A review is conducted, and for every 1,000 words of translated text, the number of identified errors is calculated.
- Setting the bar: While machine-translated content might initially have an EPT of around 50 errors per 1,000 words, our rigorous human post-editing process significantly reduces this. After review by our top translators, the average EPT drops to around 10. An additional review by a second professional further refines this, bringing the EPT down to an average of 5.
This metric allows us to precisely quantify the impact of human expertise and track our progress in delivering increasingly error-free translations.
The human-AI symbiosis: Lara and TranslationOS
Our translations are powered by Lara, Translated’s advanced Language AI. Unlike generic AI models, Lara is purpose-built for translation, leveraging full-document context to produce superior initial drafts that significantly reduce TTE. This symbiotic relationship between human expertise and cutting-edge AI is central to our quality promise.
All of these processes, from T-Rank™ selecting the best translator to the granular EPT analysis, are managed within TranslationOS, our AI-first platform. TranslationOS provides clients with transparent visibility into quality and performance, demonstrating our commitment to a human-in-the-loop collaboration model that consistently delivers superior results.
Tailored quality for every need
We understand that “quality” isn’t a one-size-fits-all concept. Different industries and purposes demand different levels of precision and style. That’s why our approach is inherently flexible. Our AI solutions adapt to specific needs, from the stringent accuracy required in medical translations to the creative nuance needed for marketing campaigns. This “fit-for-purpose” model, managed seamlessly through TranslationOS, ensures that every client receives the right level of quality for their strategic goals and budget.
The future of translation quality
At Translated, we’re not just reacting to industry standards; we’re actively shaping them. Our continuous research into TTE trends reveals a clear path of progress toward the “singularity in translation” – the point where machine translation reaches human parity. By focusing on practical, data-driven metrics like TTE and EPT, and by empowering our expert linguists with advanced AI, we are establishing new benchmarks that will define the future of global communication.
We invite you to explore how Translated’s unique approach to quality, and the power of human-AI collaboration, can transform your localization workflows and help you achieve truly flawless communication, every time.