South African Tuberculosis Vaccine Initiative
UCT Computer Science Research Document ArchiveNot a member yet
1270 research outputs found
Sort by
CoSMo: A multilingual modular language for Content Selection Modelling
Full text linkRepresenting snippets of information abstractly is a task that needs to be performed for various purposes, such as database view specification and the first stage in the natural language generation pipeline for generative AI from structured input, i.e., the content selection stage to determine what needs to be verbalised. For the Abstract Wikipedia project, requirements analysis revealed that such an abstract representation requires multilingual modelling, content selection covering declarative content and functions, and both classes and instances. There is no modelling language that meets either of the three features, let alone a combination. Following a rigorous language design process inclusive of broad stakeholder consultation, we created CoSMo, a novel Content Selection Modeling language that meets these and other requirements so that it may be useful both in Abstract Wikipedia as well as other contexts. We describe the design process, rationale and choices, the specification, and preliminary evaluation of the language
Neural Machine Translation between Low-Resource Languages with Synthetic Pivoting
Full text linkTraining neural models for translating between low-resource languages is challenging due to the scarcity of direct parallel data between such languages. Pivot-based neural machine translation (NMT) systems overcome data scarcity by including a high-resource pivot language in the process of translating between low-resource languages. We propose synthetic pivoting, a novel approach to pivot-based translation in which the pivot sentences are generated synthetically from both the source and target languages. Synthetic pivot sentences are generated through sequence-level knowledge distillation, with the aim of changing the structure of pivot sentences to be closer to that of the source or target languages, thereby reducing pivot translation complexity. We incorporate synthetic pivoting into two paradigms for pivoting: cascading and direct translation using synthetic source and target sentences. We find that the performance of pivot-based systems highly depends on the quality of the NMT model used for sentence regeneration. Furthermore, training back-translation models on these sentences can make the models more robust to input-side noise. The results show that synthetic data generation improves pivot-based systems translating between low-resource Southern African languages by up to 5.6 BLEU points after fine-tuning
Choose-Your-Own-Adventure (CYOA): An empathy digital training tool for healthcare workers in maternity settings
Full text linkThe study aimed to explore the potential of adapting the Secret History (SH) workshop method to an independent digital learning tool using the Choose-Your-Own-Adventure (CYOA) framework to enhance empathy skills and support continuous healthcare worker education among healthcare workers in maternity settings. In the context of SH training, healthcare professionals are encouraged to participate in roleplay activities while simulating scenarios involving patients, observing their own reactions and responses in each role. This enables them to gain insights into both the patients' backgrounds and the assumed roles of healthcare workers. While this training has enhanced the empathic abilities of healthcare workers, there are challenges in expanding the reach of this intervention.
The implementation of the SH workshop presents cost implications in terms of the logistics required to implement in-person workshops. This study makes a meaningful contribution to the field of healthcare training and has important implications for the development and implementation of digital storytelling technologies in healthcare. The research provides a valuable resource for healthcare workers looking to improve their empathy skills in the healthcare industry, particularly in maternity settings. The CYOA tool developed in this study consists of a mobile application that presents users with a series of interactive narratives that simulate real-life scenarios in maternity settings. Our mobile application (SHiMA) has the potential to either enhance or introduce SH concepts on a larger scale for healthcare workers.
The study employed a mixed-methods approach, incorporating two interviews, four co-design workshops, and two focus group discussions. The participants included expert informants associated with the Perinatal Mental Health Project (PMHP), Bhabhisana Baby Project (BBP), and midwives from Al-Nisa Maternity Home. The narratives have been collaboratively developed with this group of participants utilizing a CYOA framework to gradually reveal the characters’ stories similar to the original SH workshop method. The data collected from these methods was analysed using thematic analysis to identify recurring themes and patterns on the adaptation of the SH workshop method to a digital training tool. The study identified key design considerations for the development of the CYOA tool, including the need for engaging and interactive narratives, the significance of customizing the tool to meet the distinct requirements of healthcare workers, and the need for ongoing evaluation and feedback
Evolving Folding Bodies and Brains in Origami Robots
Full text linkEvolutionary robotics has produced a vast array of adaptive design paradigms applicable to
body-brain (controller-morphology) adaptation. However, within the purview of adaptive body-brain evolutionary robotic architectures, folding (origami) robotics has received relatively little research attention. An open problem in evolutionary robotics, and more
broadly embodied evolution, is how to automatically design robots that are general problem-solvers across various task environments. Proposals include AutoFacs: self-designing methods for producing novel robot (body-brain) designs for given environments, evaluated as problem-solvers in such environments and then re-configured (with adapted body-brain designs) for the next generation of robots
Surface realisation architecture for low-resourced African languages
No full textThere has been growing interest in building surface realisation systems to support the automatic generation of text in African languages. Such tools focus on converting abstract representations of meaning to a text. Since African languages are low-resourced, economical use of resources and general maintainability are key considerations. However, there is no existing surface realiser architecture that possesses most of the maintainability characteristics (e.g., modularity, reusability, and analysability) that will lead to maintainable software that can be used for the languages. Moreover, there is no consensus surface realisation architecture created for other languages that can be adapted for the languages in question. In this work, we solve this by creating a novel surface realiser architecture suitable for low-resourced African languages that abides by the features of maintainable software. Its design comes after a granular analysis, classification, and comparison of the architectures used by 77 existing NLG systems. We compare our architecture to existing architectures and show that it supports the most features of a maintainable software product
Automating the Generation of Competency Questions for Ontologies with AgOCQs
Full text linkCompetency Questions (CQs) are natural language questions drawn from a chosen subject domain and are intended for use in ontology engineering processes. Authoring good quality and answerable CQs has been shown to be difficult and time-consuming, due to, among others, manual authoring, relevance, answerability, and re-usability. As a result, few ontologies are accompanied by few CQs and their uptake among ontology developers remains low. We aim to address the challenges with manual CQ authoring through automating CQ generation. This novel process, called AgOCQs, leverages a combination of Natural Language Processing (NLP) techniques, corpus and transfer learning methods, and an existing controlled natural language for CQs. AgOCQs was applied to CQ generation from a corpus of Covid-19 research articles, and a selection of the generated questions was evaluated in a survey. 70% of the CQs were judged as being grammatically correct by at least 70% of the participants. For 12 of the 20 evaluated CQs, the ontology expert participants deemed the CQs to be answerable by an ontology at a range of 50-85across the CQs, with the remainder uncertain. This same group of ontology experts found the CQs to be relevant between 70-93across the 12 CQs. Finally, 73of the users group and 69of the ontology experts judged all the CQs to provide clear domain coverage. These findings are promising for the automation of CQs authoring, which should reduce development time for ontology developers
A Computational Method to Support Chemical Product Design Based on Multi-objective Optimisation and Graph Transformers
Full text linkChemical product design refers to the practice of developing
novel chemical products given properties to be optimised and
constraints to be satisfied. Strategies for chemical product
design can be based on multi-objective constrained optimisation
in a large search space of compounds whose properties
are uncertain and partially known. Advances in machine
learning, multi-objective optimisation, formal representation
of chemical compounds and identified correlations between
molecular structures and relevant properties, have fostered
increased interest in computer-based techniques to identify
candidate compounds for innovation in chemical products.
In this paper we empirically explore a combination of stateof-
the-art machine learning and evolutionary multi-objective
optimisation methods to support chemical product design. In
order to ground our arguments as concrete examples, we consider
the design of domestic detergents, and explore how automating
computational design can be controlled via specification
of hyper-parameters, so as to generate solutions (detergents)
with desired features. Our results contribute to the
methodological problem of automating chemical product design,
and more broadly functional molecular design
Policy-based Reinforcement Learning for Generalisation in Interactive Text-based Environments
Full text linkText-based environments enable RL agents to learn to converse and perform interactive tasks through natural language. However, previous RL approaches applied to text-based environments show poor performance when evaluated on unseen games. This paper investigates the improvement of generalisation performance through the simple switch from a value-based
update method to a policy-based one, within text-based environments. We show that by replacing commonly used value-based methods with REINFORCE with baseline, a far more
general agent is produced. The policy-based agent is evaluated on Coin Collector and Question Answering with interactive text (QAit), two text-based environments designed to test zero-shot performance. We see substantial improvements on a variety of zero-shot evaluation experiments, including tripling accuracy on various QAit benchmark configurations. The results indicate that policy-based RL has significantly better generalisation capabilities than value-based methods within such text-based environments, suggesting that RL agents could be applied to more complex natural language environments
Digital Health Technologies for Maternal and Child Health in Africa and Other Low- and Middle-Income Countries: Cross-disciplinary Scoping Review With Stakeholder Consultation
Full text linkBackground: Maternal and child health (MCH) is a global health concern, especially impacting low- and middle-income countries (LMIC). Digital health technologies are creating opportunities to address the social determinants of MCH by facilitating access to information and providing other forms of support throughout the maternity journey. Previous reviews in different disciplines have synthesized digital health intervention outcomes in LMIC. However, contributions in this space are scattered across publications in different disciplines and lack coherence in what digital MCH means across fields. Objective: This cross-disciplinary scoping review synthesized the existing published literature in 3 major disciplines on the use of digital health interventions for MCH in LMIC, with a particular focus on sub-Saharan Africa. Methods: We conducted a scoping review using the 6-stage framework by Arksey and O'Malley across 3 disciplines, including public health, social sciences applied to health, and human-computer interaction research in health care. We searched the following databases: Scopus, PubMed, Google Scholar, ACM Digital Library, IEEE Xplore, Web of Science, and PLOS. A stakeholder consultation was undertaken to inform and validate the review. Results: During the search, 284 peer-reviewed articles were identified. After removing 41 duplicates, 141 articles met our inclusion criteria: 34 from social sciences applied to health, 58 from public health, and 49 from human-computer interaction research in health care. These articles were then tagged (labeled) by 3 researchers using a custom data extraction framework to obtain the findings. First, the scope of digital MCH was found to target health education (eg, breastfeeding and child nutrition), care and follow-up of health service use (to support community health workers), maternal mental health, and nutritional and health outcomes. These interventions included mobile apps, SMS text messaging, voice messaging, web-based applications, social media, movies and videos, and wearable or sensor-based devices. Second, we highlight key challenges: little attention has been given to understanding the lived experiences of the communities; key role players (eg, fathers, grandparents, and other family members) are often excluded; and many studies are designed considering nuclear families that do not represent the family structures of the local cultures. Conclusions: Digital MCH has shown steady growth in Africa and other LMIC settings. Unfortunately, the role of the community was negligible, as these interventions often do not include communities early and inclusively enough in the design process. We highlight key opportunities and sociotechnical challenges for digital MCH in LMIC, such as more affordable mobile data; better access to smartphones and wearable technologies; and the rise of custom-developed, culturally appropriate apps that are more suited to low-literacy users. We also focus on barriers such as an overreliance on text-based communications and the difficulty of MCH research and design to inform and translate into policy
Reflections on Digital Maternal and Child Health Support for Mothers and Community Health Workers in Rural Areas of Limpopo Province, South Africa
Full text linkIntroduction: Digital health support using mobile and digital technologies, such as MomConnect and WhatsApp, is providing opportunities to improve maternal and child healthcare in low- and middle-income countries. Yet, the perspective of health service providers, pregnant women, and mothers as recipients of digital health support is under-researched in rural areas. Material and Methods: An exploratory-descriptive qualitative research approach was adopted to reflect on the experiences of mothers, community leaders, and community health workers on mobile health opportunities in the context of maternal and child health in rural areas. Purposive sampling was used to select 18 participants who participated in the two focus groups and individual semi-structured interviews for data collection about digital maternal and child health support. The thematic open coding method of data analysis assisted authors in making sense of the given reflections of mothers, community leaders, and healthcare workers about digital health support. Results: Participants commented on different existing digital support apps and their importance for maternal and child health. For example, MoMConnect, Pregnancy+, WhatsApp, and non-digital resources were perceived as useful ways of communication that assist in improving maternal and child health. However, participants reported several challenges related to the use of digital platforms, which affect following the health instructions given to pregnant women and mothers. Conclusions: Participants expressed the significant role of digital support apps in maternal and child health, which is impacted by various challenges. Addressing the lack of digital resources could improve access to health instructions for pregnant women and mothers