Febe de Wet

Documents containing mathematical content remain largely inaccessible to blind and visually impaired readers because they are predominantly published as untagged PDFs, which do not include the semantic data necessary for effective accessibility. Equations in such documents consist of text interlaced with lines and other graphical elements and cannot be interpreted using a screen reader. We present a browsing approach for print-disabled readers specifically aimed at such mathematical content. This approach draws on the navigational mechanisms often used to explore the virtual worlds of text adventure games with audio-visual sensory substitution for graphical content. The relative spatial placement of the elements of an equation are represented as a virtual world so the reader can navigate between elements. Text elements are announced conventionally using synthesised speech, while graphical elements, such as roots and fraction lines, are rendered using a modification of the vOICe algorithm. The virtual world allows the reader to interactively discover the spatial structure of the equation, while the rendition of graphical elements as sound allows the shape and identity of elements that cannot be synthesised as speech to be discovered and recognised. The browsing approach was evaluated by 11 blind and 14 sighted participants in a user trial that included identifying twelve equations extracted from PDF documents. Overall, equations were identified completely correctly in 78% of cases (74% and 83%, respectively, for blind and sighted subjects). If partial correctness is considered, then the performance is substantially higher. Feedback from the blind subjects indicated that the technique allows spatial information and graphical detail to be discovered. We conclude that the integration of a spatial model represented as a virtual world in conjunction with audio-visual sensory substitution for non-textual elements can be an effective way for blind and visually impaired readers to read currently inaccessible mathematical content in PDF documents.

Recent research on deep-learning architectures has resulted in substantial improvements in automatic speech recognition accuracy. The leaps of progress made in well-resourced languages can be attributed to the fact that these architectures are able to effectively represent spoken language in all its diversity and complexity. However, developing advanced models of a language without appropriate corpora of speech and text data remains a challenge. For many under-resourced languages, including those spoken in South Africa, such resources simply do not exist. The aim of the work reported on in this paper is to address this situation by investigating the possibility to create diverse speech resources from unannotated broadcast data. The paper describes how existing speech and text resources were used to develop a semi-automatic data harvesting procedure for two genres of broadcast data, namely news bulletins and radio dramas. It was found that adapting acoustic models with less than 10 hours of manually annotated data from the same domain significantly reduced transcription error rates for speaking styles and acoustic conditions that are not represented in any of the existing speech corpora. Results also indicated that much more automatically transcribed adaptation data is required to achieve similar results.

Appropriate training data is a prerequisite for the development of natural language processing (NLP) techniques. Vast amounts of language data are typically required to develop NLP tools that perform at state-of-the-art level. Such abundant resources are currently only available in a few languages. The remaining languages have to find alternative ways to become ``NLP-enabled''. The aim of the study reported on here is to make more language data available to support NLP development in the official languages of South Africa. In this paper we present the idea of generating text data by means of translation. We also propose the use of topic modelling to identify text in a highly resourced source language that will yield meaningful translations in under-resourced target languages. More specifically, the paper describes how topic modelling was used to identify English Wikipedia articles that should be suitable for translation into South Africa's 10 other official languages.

Despite many attempts to address the situation, South Africa's official languages remain under-resourced in terms of the text and speech data required to implement state-of-the-art language technology. To ensure that no language is left behind, resource development should remain a priority until a strong digital presence has been established for all indigenous languages. This paper provides an overview of previous projects that were specifically aimed at speech resource development and introduces an ongoing initiative to launch South Africa's languages on the Mozilla Common Voice platform.

Background: Language sample analysis (LSA) is invaluable to describe and understand child language use and development for clinical purposes and research. Digital tools supporting LSA are available, but many of the LSA steps have not been automated. Nevertheless, programs that include automatic speech recognition (ASR), the first step of LSA, have already reached mainstream applicability. Summary: To better understand the complexity, challenges, and future needs of automatic LSA from a technological perspective, including the tasks of transcribing, annotating, and analysing natural child language samples, this article takes on a multidisciplinary view. Requirements of a fully automated LSA process are characterized, features of existing LSA software tools compared, and prior work from the disciplines of information science and computational linguistics reviewed. Key Messages: Existing tools vary in their extent of automation provided across the process of LSA. Advances in machine learning for speech recognition and processing have potential to facilitate LSA, but the specifics of child speech and language as well as the lack of child data complicate software design. A transdisciplinary approach is recommended as feasible to support future software development for LSA.

This paper presents recent progress in the acoustic modelling of under-resourced code-switched (CS) speech in multiple South African languages. We consider two approaches. The first constructs separate bilingual acoustic models corresponding to language pairs (English-isiZulu, English-isiXhosa, English-Setswana and English-Sesotho). The second constructs a single unified five-lingual acoustic model representing all the languages (English, isiZulu, isiXhosa, Setswana and Sesotho). For these two approaches we consider the effectiveness of semi-supervised training to increase the size of the very sparse acoustic training sets. Using approximately 11 hours of untranscribed speech, we show that both approaches benefit from semi-supervised training. The bilingual TDNN-F acoustic models also benefit from the addition of CNN layers (CNN-TDNN-F), while the five-lingual system does not show any significant improvement. Furthermore, because English is common to all language pairs in our data, it dominates when training a unified language model, leading to improved English ASR performance at the expense of the other languages. Nevertheless, the five-lingual model offers flexibility because it can process more than two languages simultaneously, and is therefore an attractive option as an automatic transcription system in a semi-supervised training pipeline.

The South African Journal of Communication Disorders (SAJCD) publishes reports and papers concerned with research, and critically evaluative theoretical, philosophical and conceptual issues dealing with aspects of human communication and its disorders, dysphagia, service provision, training and policy.

Being a multilingual country rich in culture and diversity, South Africa has residents who use a variety of the 11 official languages for daily communication. Multilingual communication occurs especially in hospitals and clinics. People visiting healthcare facilities often struggle to locate a health professional that shares the same language or even a language which is similar to their own and thus intelligible to them. The research which is reported on here aimed to assist the University of Pretoria in its endeavours to address this situation by teaching students an additional language. This was done by providing three custom-designed, mobile-assisted Sepedi language learning applications to students from the Faculty of Health Sciences enrolled for the Sepedi language module, Sepedi for Beginners. The applications were specifically designed to assist students with the acquisition of vocabulary related to healthcare. The students used the applications as supplementary tools and gave feedback thro...