Predicting Risks of Machine Translations of Public Health Resources by Developing Interpretable Machine Learning Classifiers

Xie, Wenxiu; Ji, Meng; Huang, Riliu; Hao, Tianyong; Chow, Chee W.

doi:10.3390/ijerph18168789

Cited by 4 publications

(24 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“… 16–19 Numerous studies have documented errors in machine translation of public health and medical information that reduce accuracy, understandability, and usability of the information. 20–22 Consistent with this research, multiple lay press articles have documented serious errors in machine-translated COVID-19 vaccine information on PHD websites. 17 , 23 , 24 Anecdotal evidence suggests that the lack of linguistically and culturally appropriate information about COVID-19 vaccines hampers access for many non-English-communicating Americans, which likely exacerbates health disparities they experience.…”

Section: Introductionsupporting

confidence: 64%

Analysis of Language Translations of State Governments' Coronavirus Disease 2019 Vaccine Websites

et al. 2022

View full text Add to dashboard Cite

Introduction: During the coronavirus disease 2019 (COVID-19) vaccination campaign, non-English-communicating individuals have faced inequities in access to resources for vaccine education and uptake. We characterized the language translation status of states' COVID-19 vaccine websites to inform discussion on the sufficiency of translated information and strategies for expanding the availability of multilingual vaccine information. Methods: We identified the primary COVID-19 vaccine website for all 50 states, the District of Columbia, and the federal government (“jurisdictions”) and determined the languages into which information about obtaining the vaccine (access) and vaccine safety and efficacy had been translated, as of October 2021. We compared these findings with data from the American Community Survey to determine how many individuals had these online resources available in their primary language. Results: Only 56% of jurisdictions provided professionally translated information about COVID-19 vaccine safety and efficacy, and only 50% provided professionally translated information about how to register for or obtain the COVID-19 vaccine, in at least one language. Consequently, ∼26 million Americans may not have accurate vaccine safety and efficacy information available, and ∼29 million Americans may not have vaccine access information available, from their jurisdiction in their primary language. Furthermore, translated information often was limited in scope and/or number of languages provided. Conclusion: Translation of COVID-19 vaccine information on state government websites currently is insufficient to meet the needs of non-English-communicating populations. This analysis can inform discussions about resource needs and operational considerations for adequate provision of multilingual, critical health information.

show abstract

Section: Introductionsupporting

confidence: 64%

Analysis of Language Translations of State Governments' Coronavirus Disease 2019 Vaccine Websites

et al. 2022

View full text Add to dashboard Cite

show abstract

“…The records were published between 2009 and 2023 as either conference papers (12/46, 26%) [ 48 , 53 , 55 , 60 , 61 , 63 , 65 , 71 , 78 , 80 , 82 , 84 ] or articles in traditional journals (34/46, 74%) [ 29 , 42 - 47 , 49 - 52 , 54 , 56 - 59 , 62 , 64 , 66 - 70 , 72 - 77 , 79 , 81 , 83 , 85 , 86 ] ( Multimedia Appendix 4 ).…”

Section: Resultsmentioning

confidence: 99%

“…Four types of information transmitters (ie, the end users of MT) could be identified: PH departments and research institutions (21/46, 46%) [ 29 , 49 , 50 , 52 - 54 , 56 , 59 , 60 , 62 , 63 , 65 , 70 , 73 - 78 , 80 , 86 ]; clinical and hospital staff (15/46, 33%) [ 44 - 47 , 51 , 58 , 61 , 66 , 68 , 69 , 71 , 72 , 79 , 81 , 85 ]; international and national health organizations, such as the WHO, the US Centers for Disease Control and Prevention, and the UK National Health Service (8/46, 17%) [ 42 , 43 , 55 , 57 , 64 , 82 - 84 ]; and developers of web-based health information platforms (eg, Cochrane) or social media outlets (eg, Facebook; 2/46, 4%) [ 48 , 67 ].…”

Section: Resultsmentioning

confidence: 99%

“…The types of PH materials translated with MT fell into 6 broad categories: official guidelines and educational resources (11/46, 24%) [ 42 , 43 , 53 , 62 , 67 , 70 , 75 , 79 , 80 , 82 , 83 ], simplified medical information and lexica (11/46, 24%) [ 48 , 54 , 55 , 61 , 63 , 68 , 69 , 71 , 81 , 85 , 86 ], PH promotional material (10/46, 22%) [ 29 , 49 , 50 , 59 , 60 , 64 , 65 , 76 - 78 ], instruction handouts (6/46, 13%) [ 51 , 52 , 57 , 58 , 66 , 72 ], academic research (6/46, 13%) [ 44 - 47 , 73 , 84 ], and survey instruments (2/46, 4%) [ 56 , 74 ].…”

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

The Use of Machine Translation for Outreach and Health Communication in Epidemiology and Public Health: Scoping Review

Herrera-Espejel,

Rach

2023

JMIR Public Health Surveill

View full text Add to dashboard Cite

Background Culturally and linguistically diverse groups are often underrepresented in population-based research and surveillance efforts, leading to biased study results and limited generalizability. These groups, often termed “hard-to-reach,” commonly encounter language barriers in the public health (PH) outreach material and information campaigns, reducing their involvement with the information. As a result, these groups are challenged by 2 effects: the medical and health knowledge is less tailored to their needs, and at the same time, it is less accessible for to them. Modern machine translation (MT) tools might offer a cost-effective solution to PH material language accessibility problems. Objective This scoping review aims to systematically investigate current use cases of MT specific to the fields of PH and epidemiology, with a particular interest in its use for population-based recruitment methods. Methods PubMed, PubMed Central, Scopus, ACM Digital Library, and IEEE Xplore were searched to identify articles reporting on the use of MT in PH and epidemiological research for this PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews)–compliant scoping review. Information on communication scenarios, study designs and the principal findings of each article were mapped according to a settings approach, the World Health Organization monitoring and evaluation framework and the service readiness level framework, respectively. Results Of the 7186 articles identified, 46 (0.64%) were included in this review, with the earliest study dating from 2009. Most of the studies (17/46, 37%) discussed the application of MT to existing PH materials, limited to one-way communication between PH officials and addressed audiences. No specific article investigated the use of MT for recruiting linguistically diverse participants to population-based studies. Regarding study designs, nearly three-quarters (34/46, 74%) of the articles provided technical assessments of MT from 1 language (mainly English) to a few others (eg, Spanish, Chinese, or French). Only a few (12/46, 26%) explored end-user attitudes (mainly of PH employees), whereas none examined the legal or ethical implications of using MT. The experiments primarily involved PH experts with language proficiencies. Overall, more than half (38/70, 54% statements) of the summarizing results presented mixed and inconclusive views on the technical readiness of MT for PH information. Conclusions Using MT in epidemiology and PH can enhance outreach to linguistically diverse populations. The translation quality of current commercial MT solutions (eg, Google Translate and DeepL Translator) is sufficient if postediting is a mandatory step in the translation workflow. Postediting of legally or ethically sensitive material requires staff with adequate content knowledge in addition to sufficient language skills. Unsupervised MT is generally not recommended. Research on whether machine-translated texts are received differently by addressees is lacking, as well as research on MT in communication scenarios that warrant a response from the addressees.

show abstract

“…Naïve Bayes is not a single algorithm, but a family of algorithms that share a common principle, i.e., features being used to classify are assumed to be independent of each other. It predicts the membership probabilities for each class [12], and the class with the highest chance is considered the most likely [21]. Before deep learning, the Naive Bayes classifier was a commonly used classification algorithm.…”

Section: Gaussian Naïve Bayesmentioning

confidence: 99%

Predicting HIV Status among Men Who Have Sex with Men in Bulawayo & Harare, Zimbabwe Using Bio-Behavioural Data, Recurrent Neural Networks, and Machine Learning Techniques

et al. 2022

View full text Add to dashboard Cite

HIV and AIDS continue to be major public health concerns globally. Despite significant progress in addressing their impact on the general population and achieving epidemic control, there is a need to improve HIV testing, particularly among men who have sex with men (MSM). This study applied deep and machine learning algorithms such as recurrent neural networks (RNNs), the bagging classifier, gradient boosting classifier, support vector machines, and Naïve Bayes classifier to predict HIV status among MSM using the dataset from the Zimbabwe Ministry of Health and Child Care. RNNs performed better than the bagging classifier, gradient boosting classifier, support vector machines, and Gaussian Naïve Bayes classifier in predicting HIV status. RNNs recorded a high prediction accuracy of 0.98 as compared to the Gaussian Naïve Bayes classifier (0.84), bagging classifier (0.91), support vector machine (0.91), and gradient boosting classifier (0.91). In addition, RNNs achieved a high precision of 0.98 for predicting both HIV-positive and -negative cases, a recall of 1.00 for HIV-negative cases and 0.94 for HIV-positive cases, and an F1-score of 0.99 for HIV-negative cases and 0.96 for positive cases. HIV status prediction models can significantly improve early HIV screening and assist healthcare professionals in effectively providing healthcare services to the MSM community. The results show that integrating HIV status prediction models into clinical software systems can complement indicator condition-guided HIV testing strategies and identify individuals that may require healthcare services, particularly for hard-to-reach vulnerable populations like MSM. Future studies are necessary to optimize machine learning models further to integrate them into primary care. The significance of this manuscript is that it presents results from a study population where very little information is available in Zimbabwe due to the criminalization of MSM activities in the country. For this reason, MSM tends to be a hidden sector of the population, frequently harassed and arrested. In almost all communities in Zimbabwe, MSM issues have remained taboo, and stigma exists in all sectors of society.

show abstract

Predicting Risks of Machine Translations of Public Health Resources by Developing Interpretable Machine Learning Classifiers

Cited by 4 publications

References 17 publications

Analysis of Language Translations of State Governments' Coronavirus Disease 2019 Vaccine Websites

Analysis of Language Translations of State Governments' Coronavirus Disease 2019 Vaccine Websites

The Use of Machine Translation for Outreach and Health Communication in Epidemiology and Public Health: Scoping Review

Predicting HIV Status among Men Who Have Sex with Men in Bulawayo & Harare, Zimbabwe Using Bio-Behavioural Data, Recurrent Neural Networks, and Machine Learning Techniques

Contact Info

Product

Resources

About