The feasibility of using artificial intelligence-powered translation software in lieu of human interpreters for low-risk, Arabic-speaking pregnant people in an Australian antenatal clinic

Background: Language barriers in maternity care can compromise patient safety and experience, particularly for culturally and linguistically diverse (CALD) populations. Mobile transcription and translation technologies have emerged as potential solutions for low-risk clinical interactions; however, evidence on their real-world usability in antenatal settings remains limited. This study aimed to evaluate the feasibility and usability of a real-time transcription and translation software in facilitating communication between clinicians and Arabic-speaking pregnant women requiring interpreter services.

Methods: A feasibility study was conducted in the antenatal clinic at a teaching hospital. Ten Arabic-speaking women attending low-risk antenatal appointments were recruited. Each participant completed two consultations: the first using standard interpreter services and the second using the software with interpreter oversight. Pre- and post- study surveys captured participant experience, while interpreters assessed translation accuracy using a structured tool adapted from New South Wales (NSW) Multicultural Health Communication Service guidelines. Data were analysed descriptively.

Results: Half of the participants reported being satisfied/very satisfied with software-mediated communication, and 60% experienced no difficulty understanding translations. However, interpreter assessments indicated poor translation accuracy, with mean clarity and meaning scores of 3.3 and 3.2 (scale 1–10), and factual errors in 60% of cases. Connectivity emerged as a critical factor, with network disruptions causing delays and impacting usability.

Conclusions: Findings demonstrate the feasibility of the software in real-world, low-risk antenatal consultations and its ability to support two-way communication. However, concerns regarding translation accuracy, reliability, and connectivity currently limit its clinical utility, necessitating ongoing interpreter supervision and substantial refinement before broader implementation. Notably, the discrepancy between participant satisfaction and interpreter-identified errors, together with reliance on interpreter assessments, introduces the potential for observer, expectation, and vested interest biases, and therefore warrants cautious interpretation of the findings.