The AILQA program will evaluate translation quality using AI.
The AILQA Beta program is presented by Welocalize. The goal of the development is to enhance translation assessment through AI. The AILQA Beta program is patented by Welocalize, as reported by https://xrust.ru/. It is designed to use AI to detect and classify translation errors. An improved approach to evaluating linguistic quality and supplier performance has been proposed. The value of AILQA lies in the fundamental flaw of traditional LQA, which relies on sampling a portion of translated segments assessed by a linguist to detect errors. AILQA is used to automate the assessment of all translated segments and pre-label errors to direct the linguist’s attention. The result is the maximization of the linguist’s time value and a much more efficient detection of errors across the entire body of translated content. The development identifies translation errors through a two-step process: determining the presence of errors in a segment and subsequently classifying the type and severity according to the standard industry DQF-MQM framework. The system combines a general error detection model (LLM) with a terminology-specific LLM and employs a hierarchical prompt strategy with a chain of thought. For performance evaluation, AILQA uses human-annotated data from its own quality assessments at Welocalize, feedback loops from translators-LQAers, and publicly available DQF-MQM datasets. Initial results show that AILQA is an excellent complement to human review.
https://xrust.ru/news/310966-programma-ailqa-ocenit-kachestva-perevoda-s-ispolzovaniem-ii.html
