Welocalize is introducing the AILQA Beta program, a collaborative initiative aimed at refining AI-driven translation quality assessment. Welocalize’s patent-pending AILQA technology is designed to leverage AI to detect and classify translation errors, offering a dramatically improved approach to evaluating linguistic quality and vendor performance.

The value of AILQA lies in a fundamental shortcoming of traditional LQA, which relies on sampling a fraction of translated segments, evaluated by a linguist, to detect errors. AILQA is used to automate the evaluation of all translated segments and to pre-label errors to direct the linguist’s attention. The result is maximizing the value of a linguist’s time and far more effectively detecting errors across an entire corpus of translated content.

"AI-driven quality assessment is the next module of our OPAL-Enable suite, which currently includes AI-driven post editing and quality estimation, both in production with clients,” said Konstantinos Karageorgos, AI/ML Engineering Lead, Welocalize. “Our goal with AILQA is to create a system that both detects errors and enhances translation quality assurance processes. This beta program allows us to work closely with customers to refine performance.”

The AILQA technology identifies translation errors through a two-step process: detecting whether a segment contains errors and then classifying the type and severity of those errors according to the industry-standard DQF-MQM framework. The system combines a general error detection large language model (LLM) with a terminology-specific LLM, and uses a hierarchical chain-of-thought prompting strategy.

To evaluate performance, the AILQA technology leverages human-annotated data from proprietary Welocalize quality assessments, translator-LQAer feedback loops, and publicly available DQF-MQM datasets. Initial results show that the AILQA technology is an excellent complement to human review.

Welocalize is partnering with a few customers to participate in the AILQA Beta program to help refine and enhance the system. Early participants will have the opportunity to contribute to AI-powered translation quality assessment and gain early insights into future developments.

"By collaborating with customers in this beta phase, we can ensure AILQA evolves into a practical, real-world solution for translation quality assessment," said Hanne Gregg, Quality Operations Leader at Welocalize. “Our aim is to strike the right balance between automation and human expertise, reducing the burden on linguistic reviewers while maintaining high-quality multilingual content.”