Transifex, the world’s leading AI-Powered Translation Management System, today announced the public beta of the Transifex Quality Index (TQI), the industry’s first translation quality scoring index designed to increase the availability of valuable information by global audiences faster and affordably. TQI is integrated with Transifex AI, an artificial intelligence-powered, all-in-one language technology platform for managing enterprise translation and localization announced separately today, and will also be available to measure the quality of other outputs soon.
TQI is essential for removing the cost and speed barriers associated with human intervention required for localization and translation, making it possible for organizations to offer their global customers the same experience as their local ones, increasing revenue, customer satisfaction, brand loyalty, and more.
“Today, organizations have to make a choice which content they can afford to localize and make available to their global customers,” said Chris Menier, Chief Executive Officer, Transifex. “They build up a rich library of content that benefits their native customer base, but when they enter new markets, making the same content available is often cost prohibitive. TQI solves that problem by identifying AI-localized or transcreated content that can go straight to production, meaning that localization budgets can go 10x further.”
TQI leverages machine learning to provide quality estimation and monitoring, which preserves human intelligence at the core with provided metadata and model training but removes the need for human-in-the-loop. TQI assesses translation quality in real-time, allowing for efficient reviews and automated workflows. Not only can TQI assess which content can be published without human intervention, it can also identify the kind of human intervention needed to make processes as efficient as possible. Automating the quality assurance process with AI can significantly reduce turnaround times and costs associated with traditional, manual methods while maintaining high accuracy.
Initial tests have shown that on average 62% of strings can go right to production because they pass TQI quality checks that look at everything from grammar and style to brand voice, and more. 38% are handed over for human review, where TQI identifies the issues it found to make it easy to assign the right reviewer, be it a linguist, editor, creative writer, etc. Out of the 38%, for example, 5 strings might completely fail the quality evaluation and need to be rewritten, 8 might have stylistic issues and should be looked at by an editor, 3 might fail some localization features and need review by someone who specializes in German locale conventions.
Transifex AI’s emphasis on localization use cases, combined with technology advancements such as TQI that significantly reduce costly human interaction, is now providing enterprise customers with personalized content creation and human quality translations at massive scale, effortlessly, at a fraction of the cost, and orders of magnitude faster than today’s conventional processes.
TQI goes far beyond current conventional quality estimation by incorporating quality assurance as well as measurement of adherence to the unique linguistic style of the source content. It leverages machine learning to provide quality estimation, which preserves human intelligence at the core but removes the need for human-in-the-loop.
TQI reduces perceived dependance of large language models (LLMs) on human-in-the-loop processes for delivery of human quality translation and localization by minimizing the cost and time of human review by indicating where humans are needed and where they are not.
TQI is proven to enable Transifex customers to move more than 60% of their AI-generated translations directly into production without the need for human review. This translates to a 10x+ reduction in translation costs primarily associated with previous human involvement, unlocking vast amounts of content that was previously not localized due to cost constraints.
“What makes Transifex AI unique is its ability to generate human quality content that maintains our customers’ brand voice and linguistic style,” Chris Menier, continues. “It is a language hub in the center of their global content strategy supporting all of their personas ranging from product to marketing to development with workloads and content being continuously localized. TQI is at the center of Transifex’s value proposition and clearly demonstrates the measured impact and value that Transifex AI is bringing to their business as they execute on their globalization strategy.”
The Transifex Quality Index (TQI) consists of 3 key component scores:
Translation Complexity: we use machine learning to score results from multiple LLMs against each other in order to quantify the complexity of a translation and select the best one.
Quality Assurance: we score the quality of a translation related to grammar, syntax and language specific rules.
Quality Estimation: we score QE using AI with a reference-free model similar to MQM. We further weight the component scores based on individual organizations’ feedback, which ensures quality scoring is unique to their needs and content utility.
Transifex AI then auto-corrects translations until a maximum TQI score is achieved.
“As large enterprises continue to expand globally, generative AI is fueling exponential increases in content requiring translation,” said Mohamed Alaa Saayed, Senior Program Director & Fellow, Head of Digital Content Services, Frost & Sullivan. “We have been closely monitoring the development of Transifex’s AI-powered translation quality index that is enabling a new breed of highly scalable translation management solutions at a much lower cost.”
Transifex AI is now consistently outperforming major Machine Translation (MT) providers in terms of translation quality as measured by industry-recognized quality estimation tools like COMET and BERT. By applying AI beyond the typical approach by translation management vendors of inserting a generic ChatGPT prompt into the user interface, Transifex reduces dependance on but maximizes the power of large language models (LLMs), by leveraging the full depth of customers’ translation activity and learnings across all translated content, glossaries, brand guidelines, workflows, and more, to deliver a level of translation quality previously unattainable.