Departing from traditional linguistic models, advances in deep learning have resulted in a new type of predictive (autoregressive) deep language models (DLMs). Using a self-supervised next-word prediction task, these models generate appropriate linguistic responses in a given context. In the current study, nine participants listened to a 30-min podcast while their brain responses were recorded using electrocorticography (ECoG). We provide empirical evidence that the human brain and autoregressive DLMs share three fundamental computational principles as they process the same natural narrative: (1) both are engaged in continuous next-word prediction before word onset; (2) both match their pre-onset predictions to the incoming word to calculate post-onset surprise; (3) both rely on contextual embeddings to represent words in natural contexts. Together, our findings suggest that autoregressive DLMs provide a new and biologically feasible computational framework for studying the neural basis of language.
Departing from rule-based linguistic models, advances in deep learning resulted in a new type of autoregressive deep language models (DLMs). These models are trained using a self-supervised next word prediction task. We provide empirical evidence for the connection between autoregressive DLMs and the human language faculty using spoken narrative and electrocorticographic recordings. Behaviorally, we demonstrate that humans have a remarkable capacity for word prediction in natural contexts, and that, given a sufficient context window, DLMs can attain human-level prediction performance. Leveraging on DLM embeddings we demonstrate that the brain constantly and spontaneously predicts the identity of the next word in natural speech, hundreds of milliseconds before they are perceived. Finally, we demonstrate that contextual embeddings derived from autoregressive DLMs capture neural representations of the unique, context-specific meaning of words in the narrative. Our findings suggest that DLMs provides a novel biologically feasible computational framework for studying the neural basis of language.
BackgroundThe Eastern Cape province of South Africa has one of the highest burdens of HIV in the world. Emergency Departments (EDs) can serve as optimal clinical sites for the identification of new HIV infections and entry into care. We sought to determine the current burden of HIV disease among ED patients in the Eastern Cape.MethodsWe conducted a prospective cross-sectional observational study in the EDs of three Hospitals in the Eastern Cape province of South Africa from June 2017 to July 2018. All adult, non-critical patients presenting to the ED were systematically approached and offered a Point-Of-Care (POC) HIV test in accordance with South African guidelines. All HIV-positive individuals had their blood tested for the presence of antiretroviral therapy (ART) and the presence of viral suppression (≤ 1000 copies/ml). HIV incidence was estimated using a multi-assay algorithm, validated for a subtype C epidemic.FindingsOf the 2901 patients for whom HIV status was determined (either known HIV-positive or underwent POC HIV testing), 811 (28.0%) were HIV positive, of which 234 (28.9%) were newly diagnosed. HIV prevalence was higher in Mthatha [34% (388/1134) at Mthatha Regional Hospital and 28% (142/512) at Nelson Mandela Academic Hospital], compared to Port Elizabeth [22% (281/1255) at Livingstone Hospital]. HIV incidence was estimated at 4.5/100 person-years (95% CI: 2.4, 6.50) for women and 1.5 (CI 0.5, 2.5) for men. Of all HIV positive individuals tested for ART (585), 54% (316/585) tested positive for the presence of ARTs, and for all HIV positive participants with viral load data (609), 49% (299/609) were found to be virally suppressed.InterpretationOur study not only observed a high prevalence and incidence of HIV among ED patients but also highlights significant attrition along the HIV care cascade for HIV positive individuals. Furthermore, despite developing an optimal testing environment, we were only able to enrol a small sub-set of the ED population. Given the high HIV prevalence and high attrition in the ED population, HIV services in the ED should also develop strategies that can accommodate large testing volumes and ART initiation.
The field of implementation research (IR) is growing. However, there are no recognised IR core competencies in low/middle-income countries (LMICs), nor consistent curriculum across IR training programs globally. The goal of this effort is to develop a framework of IR core competencies for training programs in LMICs. The framework was developed using a mixed-methods approach consisting of two online surveys with IR training coordinators (n = 16) and academics (n = 89) affiliated with seven LMIC institutions, and a modified-Delphi process to evaluate the domains, competencies and proficiency levels included in the framework. The final framework comprised of 11 domains, 59 competencies and 52 sub-competencies, and emphasised competencies for modifying contexts, strengthening health systems, addressing ethical concerns, engaging stakeholders and communication especially for LMIC settings, in addition to competencies on IR theories, methods and designs. The framework highlights the interconnectedness of domains and competencies for IR and practice, and training in IR following the outlined competencies is not a linear process but circular and iterative, and starting points for training may vary widely by the project, institution and challenge being addressed. The framework established the need for a theory-based approach to identifying proficiency levels for IR competencies (ie, to determine proficiency levels for IR based on generalisable educational theories for competency-based education), and the relevance of various IR competencies for LMICs compared with high-income settings. This framework is useful for identifying and evaluating competencies and trainings, and providing direction and support for professional development in IR.
Background South Africa faces the highest burden of HIV infection globally. The National Strategic Plan on HIV recommends provider-initiated HIV counselling and testing (HCT) in all healthcare facilities. However, HIV continues to overwhelm the healthcare system. Emergency department (ED)-based HCT could address unmet testing needs. Objectives This study examines the reasons for accepting or declining HCT in South African EDs to inform the development of HCT implementation strategies. Method We conducted a prospective observational study in two rural EDs, from June to September 2017. Patients presenting to the ED were systematically approached and offered a point-of-care test in accordance with national guidelines. Patients demographics, presenting compaint, medical history and reasons for accepting/declining testing, were recorded. A pooled analysis is presented. Results Across sites, 2074 adult, non-critical patients in the ED were approached; 1880 were enrolled in the study. Of those enrolled, 19.7% had a previously known positive diagnosis, and 80.3% were unaware of their HIV status. Of those unaware, 90% patients accepted and 10% declined testing. The primary reasons for declining testing were ‘does not want to know status’ (37.6%), ‘in too much pain’ (34%) and ‘does not believe they are at risk’ (19.9%). Conclusions Despite national guidelines, a high proportion of individuals remain undiagnosed, of which a majority are young men. Our study demonstrated high patient acceptance of ED-based HCT. There is a need for investment and innovation regarding effective pain management and confidential service delivery to address patient barriers. Findings support a routine, non-targeted HCT strategy in EDs.
This open-access article is distributed under Creative Commons licence CC-BY-NC 4.0.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.