Browsing by Subject "isiXhosa"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Open Access
    Are we there yet? on a journey towards more contextually relevant resources in speech-language therapy and audiology
    (AOSIS Publishing, 2013) Pascoe, Michelle; Rogers, Christine; Norman, Vivienne
    Audiologists and speech-language therapists working in developing contexts like South Africa have the opportunity to address a range of needs through their research. One of these needs is the development of assessments and therapy materials that are appropriate for their clients’particular language needs and cultural background. This editorial paper aims to introduce original research in speech-language therapy and audiology, which has been carried out in South Africa and other developing contexts and is presented in this volume of the journal. In addition we suggest that while the focus of much research is on the destination or end product that is developed, there is a need to share the methodologies that are used to reach that goal so that more research can be carried out by a wider pool of students, researchers and clinicians. We describe some of the methods that we have used in our research – often in small scale projects with budgetary constraints, which would be feasible for clinicians in their routine clinical contexts. Our hope is that others can build on these approaches, critique and share their own strategies for the ultimate advancement of the professions in southern Africa.
  • No Thumbnail Available
    Item
    Open Access
    Linguistic and sociolinguistic aspects of the omission of sounds and syllables in isiXhosa speech: an explanatory guide for 2nd language learners of the language
    (2025) Mahlumba, Esethu; Dowling, Teresa
    The shortening of words, sometimes even to just one syllable, variously referred to as clipping, reduction and truncation, is a recognized phenomenon in slang but also in authentic speech. Most of us know that the English farewell “Goodbye!” often reduced to just “Bye!” is a clipped form of “God be with you” and that “ad” is a shortened form of “advertisement”. In fact, a body of scholarship on historical and contemporary clippings in English and many of the world's languages has, and continues to be, developed. The reduction of sounds in the lexicons of African languages, particularly Xhosa, is, however, a neglected field – a lacuna that I noticed when struggling to explain the phenomenon to English and Afrikaans speakers who could not align the written Xhosa word with its spoken form. In this thesis, therefore, I foreground the struggles of 2nd language learners of Xhosa at the University of Cape Town in comprehending spoken Xhosa and explain that being a tutor of the language at this level led me to conduct the research. I felt the need to establish a data-base of the most commonly truncated forms in the language and the phonological processes at play so that entry level students of the language (as a 2nd language) as well as more advanced learners would have access to commonly reduced words as well as explanations of the most widely occurring phonological processes involved in the omission of sounds. To ensure that the research kept its focus on actual student needs, a short survey was conducted with 2nd language students of Xhosa at the university to ascertain the extent to which they see the development of listening skills as desirable in a communication course. In addition, a pilot listening test was conducted with two groups of students – the one group receiving an intervention providing information on reduced forms in Xhosa, the other group taking the test without the intervention. Scores are tabulated and discussed and recommendations made based on the outcomes of both the survey and the tests.
  • No Thumbnail Available
    Item
    Open Access
    Subword segmental neural language generation for Nguni languages
    (2025) Meyer, Francois Rolihlahla; Buys, Jan
    Deep learning models for text generation are now able to produce fluent and coherent text in many conversational settings. However, such models require large training datasets and are primarily designed for a limited number of high-resource languages. These advances are not directly applicable to low-resource languages with distinctive linguistic characteristics. In this thesis we develop text generation models for the Nguni languages of South Africa -- isiXhosa, isiZulu, isiNdebele, and Siswati. The Nguni languages are agglutinative and conjunctively written, so words are formed by stringing together morphemes. We design neural models that suit the morphological complexity of the Nguni languages by explicitly modelling the segmentation of words into subword units. We propose subword segmental modelling, a neural architecture and training algorithm that learns subword segmentation during training. The standard approach to subword modelling is to apply data-driven algorithms such as byte-pair encoding (BPE) during preprocessing. Subword segmental modelling represents a departure from this paradigm: instead of casting subword segmentation as a preprocessing step, we incorporate it into end-to-end learning to allow the model to discover the optimal subword units for a particular language and task. Explicitly modelling the complex subword structure of Nguni languages serves as an inductive bias for more efficient training on the typically limited training data. In this thesis we present subword segmental models for three natural language generation tasks. Our first model is for autoregressive language modelling. We propose the subword segmental language model (SSLM), a decoder-only model that learns subword segmentation to optimise its language modelling objective. SSLM achieves lower (better) perplexity-based intrinsic evaluation scores than tokenisation-based language models, on average across the four Nguni languages. We also evaluate SSLM as an unsupervised morphological segmenter, showing that its learned subwords are closer to morphemes than standard subword tokens. Since SSLM is our first instantiation of subword segmental modelling, we present a detailed analysis of the architectural components and hyperparameters we found to be influential during development. Our second model extends subword segmental modelling to neural machine translation (NMT). We propose subword segmental machine translation (SSMT), an encoder-decoder model that learns target language subword segmentation to optimise its sequence-to-sequence translation objective. To generate translations with SSMT, we propose dynamic decoding, a decoding algorithm for generating text with subword segmental architectures. SSMT outperforms tokenisation-based NMT on Nguni languages, achieving large gains in the extremely low-resource setting of English to Siswati translation. As for SSLM, we show that SSMT learns subword boundaries more aligned with morpheme boundaries than tokenisation-based subwords. SSMT also exhibits greater morphological compositional generalisation, the ability to generalise to novel combinations of known morphemes. We extend SSMT to multilingual translation, where it learns a single target-side subword segmentation scheme to optimise performance across multiple translation directions. We compare multilingual SSMT to multilingual tokenisation-based NMT. Multilingual SSMT does induce cross-lingual transfer, but to a lesser extent that multilingual tokenisation. In cross-lingual finetuning experiments, SSMT improves transfer between unrelated languages. Our experiments confirm that decisions around subword segmentation greatly affect cross-lingual performance. We also show that differences in orthographic word boundary alignment between languages can impede cross-lingual transfer. Our third and final model combines subword segmental modelling with a copy mechanism, for the task of data-to-text generation. We propose the subword segmental pointer generator (SSPG), which jointly learns to segment words and copy subwords to optimise data-to-text generation. We also propose unmixed decoding, a text generation algorithm for copy-equipped subword segmental models. On isiXhosa data-to-text, SSPG outperforms tokenisation-based architectures trained from scratch. Besides reference-based evaluation, we develop an extractive evaluation framework to measure how faithfully models capture the expected data content of generations. This shows that SSPG more effectively combines entity copying and morphological composition. Across all three tasks, and for all four Nguni languages, subword segmental modelling consistently equals or outperforms equivalent tokenisation-based models. Its performance gains are greatest for extremely low-resource languages and tasks. Through linguistically informed evaluations, we show that subword segmental modelling successfully acquires particular aspects of Nguni-language morphology. Its subword units resemble morphemes more closely than subword tokens and it effectively applies morphological composition. Subword segmental modelling proves effective for the Nguni languages, offering a promising new approach to text generation for low-resource, morphologically complex languages.
  • No Thumbnail Available
    Item
    Open Access
    Towards the development and validation of an isiXhosa tool for the assessment of apraxia speech in adults: A descriptive study
    (2024) Allie, Nasreen; Singh, Shajila; Pascoe, Michelle
    South Africa is a culturally and linguistically diverse country. Serving isiXhosa speakers is a challenge for many Speech-Language Pathologists as there is a lack of appropriate assessment tools for this population. Current methods of adapting existing English tools are not appropriate as this does not allow the isiXhosa linguistic features to be assessed. Changing the method of scoring renders the tool invalid and unreliable. This project consisted of 3 sequential studies based in an exploratory quantitative framework. Each study has its own methodology and sub-aims with the overall aim to develop and determine the validity and reliability of isiXhosa speech stimuli for the assessment of Apraxia of Speech (AOS) in adults. Study 1 aimed to describe and generate criteria and corresponding speech stimuli for an assessment of AOS in isiXhosa. IsiXhosa has distinguishing linguistic features – such as additional phonemes, alternate places and manners of articulation (e.g. ejectives), as well as phonetic features unique to the language (e.g. prenasalised consonants and tonal contrasts). Inclusion of these features were considered to be important in the creation of the tool. Study 2 aimed to determine the face, content and construct validity of the generated criteria and speech stimuli. The speech stimuli which consisted of words, phrases and sentences were found to be culturally appropriate and to have face, content and construct validity as judged by a group of isiXhosa speakers and a Delphi panel. The speech stimuli met the criteria for an assessment of AOS as set out in Study 1. Study 3 assessed the theoretical constructs outlined in Study 1 and Study 2 determined whether the revised criteria, which generated the newly devised speech stimuli, was valid and reliable in diagnosing AOS. The speech stimuli were based on criteria for the assessment of AOS and considered the isiXhosa linguistic features. There was high inter-rater reliability (79.2 – 98.4%) for determining the presence of features of AOS. Many of the error patterns displayed were similar to that documented in the literature for AOS, such as a higher number of errors were present on less frequently used speech stimuli and stimuli with an increasing number of syllables. It was it was hypothesized that the differences in the language such as clicks and tonal contrasts may add to complexity. Clicks presented with a higher number of errors whereas tonal contrasts had fewer errors suggesting tone was less affected by AOS. Results further suggested that the first consonant of the stimulus rather than the initial phoneme in words add to complexity in isiXhosa. Further research in this area and refinement of the speech stimuli are required to create a comprehensive tool for assessment of AOS in adults.
  • Thumbnail Image
    Item
    Open Access
    uPeter Tshobisa Mtuze neenoveli zakhe
    (2015) Ndlela, Lulama Judith; Nyamende, Abner
    This thesis is about the author, Peter Tshobiso Mtuze, pertaining to his life and works. Mtuze is one of the prominent writers of isiXhosa. He is a highly endowed writer who has produced books in all literature genres: drama, novel, short stories, poetry and essays. Some of his books were and are prescribed to be taught as part of the syllabus in schools and universities. He has not only produced literature books but also written books that serve as a guide to the teaching and learning of Xhosa grammar. Furthermore, he has produced a very rare book, especially amongst the Xhosa authors, an autobiography. He has written two autobiographical works. Mtuze is one of the authors who have played a major role in the upliftment of isiXhosa. The thesis looks intently at how Mtuze has influenced the public through his literary works. This has been done by a full critical analysis of his novels, to portray the main themes embodied in these works. Chapter one is the introduction. Chapter two is the author's historical background. Chapter 3 analyses the novels that tackle moral issues (UDingezweni and Umsinga). Chapter 4 explores the social novels (Indlel' ecand' intlango, Alitshoni lingaphumi and Uyese namahlandinyuka obomi) and chapter 5 evaluates his historical novel (Iingada zibuyile endle). Mtuze is influenced by different factors in his writings. These issues have been explored in the last chapter of the thesis.