Amsterdam, 9 November, 2023 – TAUS, a leading provider of data-services and human-powered language data for AI solutions, has gone officially in production with their DeMT™ Estimate API. Work on this quality estimation feature started in October 2022. After a period of scoping, building the infrastructure, training the models and thorough testing with our early adaptors, TAUS is proud to announce that the Estimate API is now out of beta and into production.
Quality estimation plays a pivotal role in ensuring the accuracy and reliability of machine-translated content. The model thoroughly assesses and compares the source and target segments in order to quantify the quality of the output. Through this, organizations are able to gauge the reliability of the MT output, reducing the dependence on human post-editing and decreasing their time-to-market as well as their costs. Quality estimation involves advanced neural modeling techniques that analyze various aspects of a translation, such as fluency, adequacy, and adherence to domain-specific terminology.
“The official production launch of our Quality Estimate API marks a significant milestone in streamlining modern translation workflows, making them more efficient and cost-effective for organizations seeking to bridge language barriers and reach global audiences”, says Amir Kamran, Solution Architect at TAUS. “With our API, businesses can integrate state-of-the-art quality estimation into their applications, gaining real-time insights into their quality, and ultimately, delivering high-quality multilingual experiences to their customers.”
The Estimate API is actively processing millions of characters each day, serving clients from both enterprise and LSP domains. The out-of-the-box generic models, pre-trained in over a hundred languages, are already well-suited for a number of use cases. However, the team is dedicated to ongoing enhancements through extensive model training, fine-tuning and exploring new QE frameworks, particularly for languages with limited representation. Looking forward, TAUS is also developing improved metrics to obtain not just semantic, but also grammatical and linguistic insights. Furthermore, TAUS offers model customization to ensure that the quality scores are well aligned with the distinctive content, style and quality expectations of each client.
“At Milengo, we initiated collaboration with TAUS to evaluate their MTQE technology. Following a comprehensive series of test projects, we were notably impressed by the seamless integration of MTQE technology into our existing workflows and computer-assisted translation (CAT) tools. Consequently, we have made the strategic decision to procure a bundled package, ensuring its continued utilization across our project spectrum”, says Sarita Vásquez, Machine Translation Product Owner at Milengo. “Our intention is to harness this cutting-edge technology to systematically appraise MT quality, thereby enhancing our capability to identify and address MT-related challenges and enhance productivity.”
TAUS的解决方案架构师Amir Kamran表示：“我们的Quality Estimate API正式推出，标志着简化现代翻译工作流程的一个重要里程碑，使其对于寻求跨越语言障碍和接触全球受众的组织来说更加高效和经济。”“通过我们的API，企业可以将最先进的质量评估集成到他们的应用程序中，实时了解他们的质量，最终为他们的客户提供高质量的多语言体验。”
“在Milengo，我们开始与TAUS合作，评估他们的MTQE技术。在一系列全面的测试项目之后，MTQE技术与我们现有工作流程和计算机辅助翻译（CAT）工具的无缝集成给我们留下了深刻的印象。因此，我们做出了购买捆绑包的战略决策，以确保它在我们的项目范围内持续使用”，Milengo机器翻译产品负责人Sarita V á squez说。“我们的目的是利用这一尖端技术系统地评估机器翻译的质量，从而提高我们识别和应对机器翻译相关挑战的能力，并提高生产率。”