Language access solutions in healthcare have focused almost exclusively on the provision of verbal medical interpretation, despite federal and state laws that mandate translation of written information for patients withlimited English proficiency (LEP). In recent years, machine translation (MT) has made significant strides, butwhen it comes to mission-critical technical materials such as healthcare information, the accuracy rate ofmachine-only translations plummets. Thus, experts recommend MT as a starting point for translatinghealth-related material then supplementing with human quality assurance editing. However, coordinatingmachine translation with bilingual human editors who have technical medical knowledge is a challenge,especially for less commonly supported languages. Translation vendors currently pass on the associated costsof human assistance to healthcare institutions. The Canopy Translate project will address these deficits byimplementing a novel, human-assisted machine translation (HAMT) process. The envisioned workflowmanagement platform will leverage MT engines to expedite the initial rendering of source documents into atarget language, then invite bilingual healthcare professionals around the world to apply human editing to themachine-generated translation. The bilingual contributors, who will gain complimentary access to our MedicalEnglish eLearning courses as an incentive for their participation, will complete the editing task through gamifiedlearning exercises. For example, a nurse in the Philippines has native fluency in Tagalog and advancedgeneral English but desires to improve his medical English. He can edit a machine-generated Tagalogtranslation one sentence at a time in the form of a gamified activity. Other contributors will edit the same text foradditional quality assurance to form the final, polished version in Tagalog. The system will then organize thefinal translated content into a reusable document library. In Phase I, we will test the feasibility of this hybridHAMT approach for medical content. Upon meeting feasibility benchmarks, we will advance to Phase II, duringwhich we will create a minimum viable product, encompassing several novel natural language processing(NLP) algorithms, and evaluate the translation output according to a set of quality benchmarks. If successful,this project will significantly improve the availability, speed, and cost-effectiveness of producing multilingualhealth content, with potential to reduce health disparities in LEP populations.
Public Health Relevance Statement: PROJECT NARRATIVE Despite a number of federal and state laws that mandate translation of written information for patients with limited English proficiency, this aspect of language equity has been largely neglected. The current process for translation seldom supports marginalized languages and the power of machine translation for medical documents has not been fully harnessed due to lack of infrastructure for human oversight, creating significant cost and time burdens. This innovation will leverage international medical English students to complete gamified learning exercises as they edit machine translations in order to improve the availability, speed, and cost-effectiveness of creating translated medical content.
Project Terms:
