Evaluating the feasibility and utility of machine translation for patient education materials written in plain language to increase accessibility for populations with limited english proficiency.

Background: Populations with limited English proficiency (LEP) face comprehension barriers with health information as navigating healthcare systems involves encountering health information that is written at high reading grade levels, utilizes complex medical jargon, and unfamiliar or abstract terms and concepts. Despite the serious consequences of language discordance there is limited funding available for the translation of patient education material in the public healthcare setting. In response to the imperative need to provide equal access to patient education materials to all patients, regardless of English language proficiency, some have raised the feasibility of leveraging machine translation software. This study investigates the feasibility and utility of using machine translation (Google Translate) to translate patient education materials written in plain language.

Methods: A sample of 5 patient education pamphlets were selected for inclusion based on their high usage and importance. These were assessed for their readability and translated by both human translators and using Google Translate into Spanish, Portuguese, Punjabi, Simplified Chinese, and Vietnamese. Medical translators conducted blinded appraisal of both sets of translations on four domains.

Results: Spanish and Vietnamese language pamphlets achieved the highest overall scores. There were significant differences between human and machine translation in favour of the former for all of the languages, although machine translation scored above 3/5 in 90 % of the domains tested. There was no correlation between readability scores and translation scores.

Discussion: Google Translate performs well in multiple translation domains despite its continued inferiority relative to professional human translation. The high scoring of machine translated pamphlets, particularly in the most crucial domain of severity however, points to its potential adoption in a limited capacity in healthcare settings, with processes in place, like pre-screening for high-risk content that may pose a threat to patient well-being.