Search Results (7)

In the analysis of isiXhosa literary texts, the role of stylistic sentences in enhancing the meanings and reinforcement of themes and their impact in foregrounding the textual features has been largely ignored and under researched. This study is intended to explore the efficacy of stylistic sentences in the isiXhosa creative work Ukhozi Olumaphiko. In Ukhozi Olumaphiko, the author artfully employs periodic, cumulative, and balanced stylistic sentences for the realization of different purposes in the story. In this study, content analysis has been used as a qualitative and quantitative research technique, as it allowed for a detailed examination of the novel Ukhozi Olumaphiko. Stylistic sentences were identified, interpreted, and coded, using integer coding for classification. Employing literary stylistics as a theoretical approach, the stylistic sentences were analysed according to their literary impact and effect. The findings indicate that the author utilises periodic sentences predominantly in the beginning stages of the story, a spread of cumulative, balanced, and periodic sentences in the middle stages, and periodic and cumulative sentences more in the end stages of the novel. The stylistics mentioned enhance the themes, textual meanings, and narrative features of Ukhozi Olumaphiko text and are useful in weaving suspense in a way that captures the reader’s attention and evokes emotions. Full article

The stigma and invisibility surrounding mental health often lead to alienation and reinforce societal misconceptions. This paper examines how the concept of inxeba elingaphakathi (the invisible wound) in isiXhosa encapsulates the emotional and psychological pain frequently overlooked in mental health discussions, particularly among Black South African women. Employing an Indigenous storytelling methodology, the study explores how social community caregiving can illuminate these hidden wounds and challenge prevailing stigma. By analyzing the societal factors shaping mental health perceptions, this research advocates for a culturally grounded approach to healing and belonging. Full article

In this study, we present improvements in N-best rescoring of code-switched speech achieved by n-gram augmentation as well as optimised pretraining of long short-term memory (LSTM) language models with larger corpora of out-of-domain monolingual text. Our investigation specifically considers the impact of the way in which multiple monolingual datasets are interleaved prior to being presented as input to a language model. In addition, we consider the application of large pretrained transformer-based architectures, and present the first investigation employing these models in English-Bantu code-switched speech recognition. Our experimental evaluation is performed on an under-resourced corpus of code-switched speech comprising four bilingual code-switched sub-corpora, each containing a Bantu language (isiZulu, isiXhosa, Sesotho, or Setswana) and English. We find in our experiments that, by combining n-gram augmentation with the optimised pretraining strategy, speech recognition errors are reduced for each individual bilingual pair by 3.51% absolute on average over the four corpora. Importantly, we find that even speech recognition at language boundaries improves by 1.14% even though the additional data is monolingual. Utilising the augmented n-grams for lattice generation, we then contrast these improvements with those achieved after fine-tuning pretrained transformer-based models such as distilled GPT-2 and M-BERT. We find that, even though these language models have not been trained on any of our target languages, they can improve speech recognition performance even in zero-shot settings. After fine-tuning on in-domain data, these large architectures offer further improvements, achieving a 4.45% absolute decrease in overall speech recognition errors and a 3.52% improvement over language boundaries. Finally, a combination of the optimised LSTM and fine-tuned BERT models achieves a further gain of 0.47% absolute on average for three of the four language pairs compared to M-BERT. We conclude that the careful optimisation of the pretraining strategy used for neural network language models can offer worthwhile improvements in speech recognition accuracy even at language switches, and that much larger state-of-the-art architectures such as GPT-2 and M-BERT promise even further gains. Full article

The creation of linguistic resources is crucial to the continued growth of research and development efforts in the field of natural language processing, especially for resource-scarce languages. In this paper, we describe the curation and annotation of corpora and the development of multiple linguistic technologies for four official South African languages, namely isiNdebele, Siswati, isiXhosa, and isiZulu. Development efforts included sourcing parallel data for these languages and annotating each on token, orthographic, morphological, and morphosyntactic levels. These sets were in turn used to create and evaluate three core technologies, viz. a lemmatizer, part-of-speech tagger, morphological analyzer for each of the languages. We report on the quality of these technologies which improve on previously developed rule-based technologies as part of a similar initiative in 2013. These resources are made publicly accessible through a local resource agency with the intention of fostering further development of both resources and technologies that may benefit the NLP industry in South Africa. Full article

A well-nourished and healthy population is a central tenet of sustainable development. In South Africa, cultural beliefs and food taboos followed by some pregnant women influence their food consumption, which impacts the health of mothers and children during pregnancy and immediately afterwards. We documented food taboos and beliefs amongst pregnant isiXhosa women from five communities in the Kat River Valley, South Africa. A mixed-methods approach was used, which was comprised of questionnaire interviews with 224 women and nine focus group discussions with 94 participants. Overall, 37% of the women reported one or more food practices shaped by local cultural taboos or beliefs. The most commonly avoided foods were meat products, fish, potatoes, fruits, beans, eggs, butternut and pumpkin, which are rich in essential micronutrients, protein and carbohydrates. Most foods were avoided for reasons associated with pregnancy outcome, labour and to avoid an undesirable body form for the baby. Some pregnant women consumed herbal decoctions for strengthening pregnancy, facilitating labour and overall health of both themselves and the foetus. Most learnt of the taboos and practices from their own mother or grandmother, but there was also knowledge transmission in social groups. Some pregnant women in the study may be considered nutritionally vulnerable due to the likelihood of decreased intake of nutrient-rich foods resulting from cultural beliefs and food taboos against some nutritious foods. Encouraging such women to adopt a healthy diet with more protein-rich foods, vegetables and fruits would significantly improve maternal nutrition and children’s nutrition. Adhering to culturally appropriate nutrition education may be an important care practice for many pregnant women in the Kat River Valley. Full article

The English-Xhosa Dictionary for Nurses (EXDN) is a bilingual, unidirectional printed dictionary in the public domain, with English and isiXhosa as the language pair. By extending the digitisation efforts of EXDN from a human-readable digital object to a machine-readable state, using Resource Description Framework (RDF) as the data model, semantically interoperable structured data can be created, thus enabling EXDN’s data to be reused, aggregated and integrated with other language resources, where it can serve as a potential aid in the development of future language resources for isiXhosa, an under-resourced language in South Africa. The methodological guidelines for the construction of a Linguistic Linked Data framework (LLDF) for a lexicographic resource, as applied to EXDN, are described, where an LLDF can be defined as a framework: (1) which describes data in RDF, (2) using a model designed for the representation of linguistic information, (3) which adheres to Linked Data principles, and (4) which supports versioning, allowing for change. The result is a bidirectional lexicographic resource, previously bounded and static, now unbounded and evolving, with the ability to extend to multilingualism. Full article

Rural areas of the Eastern Cape (EC) Province, South Africa have a high incidence of squamous cell oesophageal cancer (OC) and exposure to mycotoxin fumonisin has been associated with increased OC risk. However, to assess exposure to fumonisin in Xhosas—having maize as a staple food—it is necessary to determine the amount of maize consumed per day. A maize-specific food frequency questionnaire (M-FFQ) has recently been developed. This study developed a food photograph (FP) series to improve portion size estimation of maize dishes. Two sets of photographs were developed to be used alongside the validated M-FFQ. The photographs were designed to assist quantification of intakes (portion size photographs) and to facilitate estimation of maize amounts in various combined dishes (ratio photographs) using data from 24 h recalls (n = 159), dishing-up sessions (n = 35), focus group discussions (FGD) (n = 56) and published literature. Five villages in two rural isiXhosa-speaking areas of the EC Province, known to have a high incidence of OC, were randomly selected. Women between the ages of 18–55 years were recruited by snowball sampling and invited to participate. The FP series comprised three portion size photographs (S, M, L) of 21 maize dishes and three ratio photographs of nine combined maize-based dishes. A culturally specific FP series was designed to improve portion size estimation when reporting dietary intake using a newly developed M-FFQ. Full article