A women-health application with over 250 million user base
Open up application archive with information on female health to readers around the world
The archive consists of 6,000 articles that needed to be localized into ten potential languages; manual localization was out of scope. The client decided to opt for machine translation; however, dealing with healthcare information and applying machine translation is an extremely delicate task. There is the risk of a misunderstanding due to the automatic translation and the risk of the reader taking wrong actions with potential health impacts based on an incorrect translation.
To mitigate the risk, a company hired CustomMT for a research project. The task consists of two parts – to apply correct healthcare terminology to the machine translation models and to measure the risk of incorrect translation. Additionally, for better user protection, the company designed the ability to switch back to the English original and added a disclaimer indicating that the translation is automatic and may contain incorrect statements.
The CustomMT team organized two rounds of expert review for machine translation, one for linguistic accuracy and the second for medical accuracy. Experts evaluated more than 40 candidate machine translation models applied to 4 tasks each and compiled risk scores to select the best models and also to support the decision to adopt machine translation with data. The client used the research results to move the project to the final stage.