Heewon Lee-Laminack

Dr. Heewon Lee-Laminack


Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
25 March 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
Iris FeinbergIris Feinberg
Iris Feinberg
HLHeewon Lee-Laminack
Heewon Lee-Laminack
ETElizabeth L. Tighe
Elizabeth L. Tighe
IOIfedola M Owoeye
Ifedola M Owoeye
Background/Objectives: AI voice interpretation applications are increasingly used in clinical settings to address language access challenges, yet evidence comparing their performance to qualified in-person medical interpreters in authentic clinical encounters remains limited. This study compares the linguistic and clinical accuracy of AI-based voice interpretation and certified in-person medical interpretation using recorded real-world clinical encounters. Methods: Outpatient clinical encounters involving patients with limited English proficiency were audio-recorded. Fourteen physician speech segments (mean length 78.5 words) representing common diagnostic, treatment, and counseling content were extracted and translated into seven languages using both an AI voice interpretation application and certified in-person medical interpreters. Bilingual reviewers and a clinician evaluated translations for accuracy, completeness, and clinical fidelity. Qualitative analyses examined error patterns and contextual loss; quantitative comparisons assessed error rate differences across languages and interpretation conditions. Results: AI voice app translations exhibited significantly higher linguistic errors (χ2[1] = 19.78, p < .001) and clinical accuracy errors (χ2[1] = 45.07, p < .001) than qualified medical interpreter translations. Clinical error rates were 33.3% for AI-generated versus 4.8% for interpreter-generated translations. Error rates were also higher for less commonly spoken languages compared to commonly spoken languages when using the AI voice app (42.9% vs. 14.3%). Conclusions: Qualified in-person interpreters remain essential for safe, accurate clinical communication. Hybrid models integrating professional interpretation with appropriately deployed AI technology may offer a balanced approach to expanding language access while maintaining communication safety and equity.

11 August 2026 in Healthcare
Verified
Journal Article
Open access
PDF file
Iris FeinbergIris Feinberg
Iris Feinberg
HLHeewon Lee-Laminack
Heewon Lee-Laminack
ETElizabeth L. Tighe
Elizabeth L. Tighe
IOIfedola M Owoeye
Ifedola M Owoeye
Background/Objectives: AI voice interpretation applications like Google Translate in speech/voice mode are increasingly used in clinical settings to address verbal language access challenges, yet evidence comparing their performance to qualified in-person medical interpreters in authentic clinical encounters remains limited. The aim of this study is to compare the linguistic and clinical accuracy of AI-based Google Translate in speech/voice mode and qualified in-person medical interpretation using spoken real-world speech segments from clinical encounters. Methods: Outpatient clinical encounters involving patients with limited English proficiency were audio-recorded. Fourteen physician speech segments (mean length 78.5 words) representing common diagnostic, treatment, and counseling content were extracted, spoken in English into Google Translate speech/voice mode, and translated into six languages. Audio recordings of the same sentence segments were provided to qualified in-person medical interpreters. All non-English translations were back-translated into English by professional interpreters. Researchers and a clinician evaluated the back-translated English speech segments for linguistic accuracy, completeness, and clinical significance. Qualitative analyses examined error patterns and contextual loss; quantitative comparisons assessed error rate differences across languages and interpretation conditions. Results: Google Translate in speech/voice mode exhibited significantly higher linguistic errors (χ2[1] = 19.78, p < 0.001) and clinical accuracy errors (χ2[1] = 45.07, p < 0.001) than qualified medical interpreter translations. Clinically significant error rates were 33.3% for Google Translate in speech/voice mode versus 4.8% for interpreter-generated translations. Error rates were also higher for less commonly spoken languages compared to commonly spoken languages when using Google Translate in speech/voice mode (42.9% vs. 14.3%). Conclusions: Qualified in-person interpreters provided translated speech segments that had fewer errors that were either linguistically or clinically significant, and remain essential for safe, accurate clinical communication.

11 October 2026 in Healthcare
Verified
Journal Article
Open access
PDF file
GKGal Kaldes
Gal Kaldes
SMSarah E. Miller
Sarah E. Miller
HLHeewon Lee-Laminack
Heewon Lee-Laminack
ETElizabeth L. Tighe
Elizabeth L. Tighe
Background/Objectives: Digital health literacy refers to the ability to find, understand, and use health information from digital sources. Disparities in these skills contribute to health inequalities and poorer outcomes. This study examines how adults with varying literacy skills navigate a digitally presented health-related literacy task. We identify engagement profiles and examine relations to demographics, daily skills-use, self-reported health, and literacy performance. Methods: We analyzed public-use data from the 2022–2023 U.S. Program for the International Assessment of Adult Competencies (PIAAC) Cycle 2 assessment. The sample included 3722 adults who completed a digitally presented medicine-label literacy item. Process indicators included total item time, number of actions, visits, latency before the first action, and interaction speed. Latent profile analysis was used to identify engagement patterns, which were then descriptively examined in relation to demographics, home skills-use, and literacy proficiency. Results: Latent profile analysis of total time, actions, visits, hesitation and speed revealed five engagement patterns: Early Exit (8 s, 0 actions, 0% correct; minimal interaction); Hesitant-Minimal (36 s, 2 actions, 30% correct; long pauses with minimal interaction); Rapid Exploratory (54 s, 44 actions, 51%; fast, high-volume clicking); Direct/Routine (57 s, 6 actions, 58%; few actions at a moderate pace); and Deliberate Exploratory (80 s, 20 actions, 62%; longer task time with higher action counts). The identified profiles showed descriptive variation in age, education, self-rated health, skills-use, and literacy level. Conclusions: Process data revealed distinct engagement patterns on a digitally presented health-related literacy task that were not captured by accuracy alone. These findings highlight the value of examining how adults interact with digitally presented health-related information, while additional tasks are needed to determine whether similar patterns generalize to broader digital health contexts.