Perceptions of the Feasibility of AI-Based Machine Translation for Linguistic Validation

OBJECTIVES: The availability and application of AI-based neural and statistical machine translation (hereafter MT) across business and translation domains has increased rapidly over the past decade with advancing technology. However, many studies conclude that MT may presently have drawbacks or reduced application within the medical translation field, where higher standards exist to safeguard patients and clinical research.

Linguistic validation (LV) describes the process of translating and culturally adapting clinical outcome assessments (COAs), through iterative review by life science stakeholders including COA professionals, linguists, patients, and clinicians. No systematic review of these stakeholders' perceptions of MT and LVs intersection exists; such data can reveal development gaps and opportunities, and inform best-practices guidelines, for deploying emerging MT technologies in sensitive healthcare settings.

METHODS: We surveyed LV professionals and linguists specializing in COA translation and cultural adaptation (N=198). Respondents provided ratings (0: "very favorable" to 100: "very unfavorable") of their impressions of MT, its applicability to LV, and open-ended responses about benefits, problems, and modifications to LV processes that MT use would entail.

RESULTS: Respondents had favorable opinions of general MT (avg: 40.6), and slightly negative opinions of its appropriateness for LV (avg: 55.3). Notably, favorability ratings of general MT vs use for LV differed significantly (paired t-test: p<.01). Open-ended feedback showed 70% of respondents mentioning time and cost benefits of MT for LV, while problems included diminished linguistic or cultural nuance (45%), reduced medical domain accuracy (24%), and no human oversight (28%), which 58% of respondents considered a necessary addition for QA purposes. 43% of respondents wanted to see MT used for LV, while 57% were unsure (38%) or opposed (19%).

CONCLUSIONS: Altogether, these results show hesitancy regarding MT adoption for LV work, and suggest increased need for regulatory guidance and industry-wide best-practices discussions. Additional surveys will gather feedback from regulatory stakeholders, clinicians, and patients.