Despite the many proposals to solve the neural machine translation (NMT) problem of low-resource languages, it continues to be difficult. The issue becomes even more complicated when few resources cover only a single domain. In this paper, we discuss the applicability of a source-side monolingual dataset of low-resource languages to improve the NMT system for such languages. In our experiments, we used Wolaytta–English translation as a low-resource language. We discuss the use of self-learning and fine-tuning approaches to improve the NMT system for Wolaytta–English translation using both authentic and synthetic datasets. The self-learning approach showed +2.7 and +2.4 BLEU score improvements for Wolaytta–English and English–Wolaytta translations, respectively, over the best-performing baseline model. Further fine-tuning the best-performing self-learning model showed +1.2 and +0.6 BLEU score improvements for Wolaytta–English and English–Wolaytta translations, respectively. We reflect on our contributions and plan for the future of this difficult field of study.
Machine learning has seen enormous growth in the last decade, with healthcare being a prime application for advanced diagnostics and improved patient care. The application of machine learning for healthcare is particularly pertinent in Africa, where many countries are resource-scarce. However, it is unclear how much research on this topic is arising from African institutes themselves, which is a crucial aspect for applications of machine learning to unique contexts and challenges on the continent. Here, we conduct a bibliometric study of African contributions to research publications related to machine learning for healthcare, as indexed in Scopus, between 1993 and 2022. We identified 3,772 research outputs, with most of these published since 2020. North African countries currently lead the way with 64.5% of publications for the reported period, yet Sub-Saharan Africa is rapidly increasing its output. We found that international support in the form of funding and collaborations is correlated with research output generally for the continent, with local support garnering less attention. Understanding African research contributions to machine learning for healthcare is a crucial first step in surveying the broader academic landscape, forming stronger research communities, and providing advanced and contextually aware biomedical access to Africa.
Recently, using deep neural networks for machine translation (MT) tasks has received great attention. In order for these networks to learn abstract representations of the input and store them as continuous vectors, they need a lot of data. However, very few research studies have been conducted on low-resource languages like Amharic. The progress of an Amharic-English machine translation task in both directions is affected by the lack of clean, easy-to-find, and up-to-date parallel language corpora. This paper presents the first relatively large-scale Amharic-English parallel corpora (above 1.1 million) for machine translation tasks. We ran experiments with recurrent neural networks (RNN) and Transformer in various hyper-parameter settings to investigate the usability of our dataset. Additionally, we explore the effects of Amharic homophone character normalization on machine translation. We have released the dataset in both unnormalized and normalized forms. Our dataset is available in train, test, and validation split files.