South African Bantu Languages Research Articles

Synthesised bigrams using word embeddings for code-switched ASR of four South African language pairs

Code-switching is the phenomenon whereby multilingual speakers spontaneously alternate between more than one language during discourse and is widespread in multilingual societies. Current state-of-the-art automatic speech recognition (ASR) systems are optimised for monolingual speech, but performance degrades severely when presented with multiple languages. We address ASR of speech containing switches between English and four South African Bantu languages. No comparable study on code-switched speech for these languages has been conducted before and consequently no directly applicable benchmarks exist. A new and unique corpus containing 14.3 hours of spontaneous speech extracted from South African soap operas was used to perform our study. The varied nature of the code-switching in this data presents many challenges to ASR. We focus specifically on how the language model can be improved to better model bilingual language switches for English-isiZulu, English-isiXhosa, English-Setswana and English-Sesotho. Code-switching examples in the corpus transcriptions were extremely sparse, with the majority of code-switched bigrams occurring only once. Furthermore, differences in language typology between English and the Bantu languages and among the Bantu languages themselves contribute further challenges. We propose a new method using word embeddings trained on text data that is both out-of-domain and monolingual for the synthesis of artificial bilingual code-switched bigrams to augment the sparse language modelling training data. This technique has the particular advantage of not requiring any additional training data that includes code-switching. We show that the proposed approach is able to synthesise valid code-switched bigrams not seen in the training set. We also show that, by augmenting the training set with these bigrams, we are able to achieve notable reductions for all language pairs in the overall perplexity and particularly substantial reductions in the perplexity calculated across a language switch boundary (between 5 and 31%). We demonstrate that the proposed approach is able to reduce the unseen code-switched bigram types in the test sets by up to 20.5%. Finally, we show that the augmented language models achieve reductions in the word error rate for three of the four language pairs considered. The gains were larger for language pairs with disjunctive orthography than for those with conjunctive orthography. We conclude that the augmentation of language model training data with code-switched bigrams synthesised using word embeddings trained on out-of-domain monolingual text is a viable means of improving the performance of ASR for code-switched speech.

Deverbal nominals derived from intransitive state verbs in isiXhosa: A Generative Lexicon approach

The relationship between deverbative noun classification and the semantic meaning of the derived deverbal nominal has been the focus of many studies in linguistics from very divergent perspectives. This article maintains that the descriptive analysis of deverbal nominals in Bantu languages does not fully interrogate the predicate argument structure of the verbs that host these deverbal nominals. An investigation of how the syntactic properties of verbs from which deverbal nominals are derived is invoked within the paradigm of decompositional lexical semantics so as to explain the argument structure, event structure and qualia properties of deverbal nominals. This is demonstrated in this article, firstly, by the Bantu languages of sub-Saharan Africa. Secondly, this article explores the descriptive nature of deverbal nominals as they manifest themselves either inflectionally or derivationally in the South African Bantu languages, with emphasis on isiXhosa. Thirdly, deverbal nominals drawing on Pustejovsky’s Generative Lexicon paradigm are analysed. Lastly, the findings of the analyses of deverbal nominals derived from intransitive state verbs demonstrate the need for further research on deverbal nominals.

Tswana finite state tokenisation

Tswana, a Bantu language in the Sotho group, is characterised by an agglutinative morphology and a disjunctive orthography, which mainly affects the verb category. In particular, verbal prefixes are usually written disjunctively, while suffixes follow a conjunctive writing style. Therefore, Tswana tokenisation cannot be based solely on whitespace, as is the case in many alphabetic, segmented languages, including the conjunctively written Nguni group of South African Bantu languages. This paper shows how a combination of two finite state tokeniser transducers and a finite state morphological analyser are combined to solve the Tswana (verb) tokenisation problem. The approach has the important advantage of bringing the processing of Tswana, beyond the morphological analysis level, in line with what is appropriate for the Nguni languages. This means that the challenge of the disjunctive orthography is met at the tokenisation/morphological analysis level and does not in principle propagate to subsequent levels of analysis such as POS tagging and shallow parsing, etc. The tokenisation approach is novel and, when implemented and evaluated, yields an F1-score of 95 % with respect to a hand tokenised gold standard.

Idiosyncratic sound systems of the South African Bantu languages: Research and clinical implications for speech-language pathologists and audiologists.

The objective of this article is to create awareness amongst speech-language pathologists and audiologists in South Africa regarding the difference between the sound systems of Germanic languages and the sound systems of South African Bantu languages. A brief overview of the sound systems of two Bantu languages, namely isiZulu and Setswana, is provided. These two languages are representative of the Nguni language group and the Sotho group respectively.Consideration is given to the notion of language-specific symptoms of speech, language and hearing disorders in addition to universal symptoms. The possible impact of speech production, language and hearing disorders on the ability to produce and perceive speech in these languages, and the challenges that this holds for research and clinical practice, are pointed out.

PRS43 - Comprehension Of Copd-Specific Terminology In Translated Patient Reported Outcomes Questionnaires

To analyze subjects’ interpretation of concepts specific to Chronic Obstructive Pulmonary Disorder (COPD) during pilot testing of translated Patient Reported Outcomes (PRO) questionnaires. As observed during pilot testing of translated COPD questionnaires, COPD-specific concepts were observed to cause more comprehension difficulties across languages, compared to observed questionnaires for other diseases. Solutions for these comprehension difficulties in COPD questionnaires will be explored. Pilot testing data for the concepts rescue medication, wheezing, the acronym COPD, and chest tightness within PROs were analyzed. Testing was conducted with five subjects per language for three questionnaires. Subjects were asked to paraphrase the COPD questionnaires in their entirety. Suggestions for improved comprehension of COPD concepts were explored, resulting from pilot testing data. Rescue medication caused difficulty within half of the languages analyzed. To improve comprehension, subjects suggested supplementing rescue medication with “(fast acting and quick relief medication)” for improved comprehension. Wheezing was a problematic concept for Spanish, South African Bantu and Indian languages. The subjects’ proposed solution for these languages was supplementation of wheezing with “(heavy breathing accompanied by a whistle sound).” Use of the COPD acronym proved problematic for 27% of languages. Even in languages where an equivalent acronym exists, comprehension issues persisted. Subjects suggested expansion of the acronym to “Chronic Obstructive Pulmonary Disorder.” Chest tightnesscaused minimal comprehension issues which were not significant enough to merit usable suggestions. Following translation of COPD questionnaires, pilot testing in the target language is essential, as concepts specific to COPD are found to be difficult. Because medical and technical concepts cannot be substituted, supplementation with brief definitions for increased comprehension and expansion of acronyms is recommended when developing source PRO questionnaires specific to COPD.

Assessment of speech in neurological disorders: Development of a Swahili screening test

The objective of this article is to create awareness amongst speech-language pathologists and audiologists in South Africa regarding the difference between the sound systems of Germanic languages and the sound systems of South African Bantu languages. A brief overview of the sound systems of two Bantu languages, namely isiZulu and Setswana, is provided. These two languages are representative of the Nguni language group and the Sotho group respectively.Consideration is given to the notion of language-specific symptoms of speech, language and hearing disorders in addition to universal symptoms. The possible impact of speech production, language and hearing disorders on the ability to produce and perceive speech in these languages, and the challenges that this holds for research and clinical practice, are pointed out.

Management and Internal Standardization of Chemistry Terminology: A Northern Sotho Case Study

One of the many implications of the process of language democratization which started post-1994 in South Africa is the empowerment of the previously marginalized South African Bantu languages to become languages of higher functions, i.e. languages of learning and teaching, and also of scientific discourse. This in turn implies the development, consolidation and especially standardization of terminology for each of these languages, and the compilation of LSP dictionaries. This article describes the terminological processing of a technical source text prior to translation, which formed part of the compilation of a Quadrilingual Explanatory Dictionary of Chemistry . It reports on the model of terminology management that was utilized and explores strategies for the internal standardization of terms in the absence of readily available, standardized chemistry terminology.

The development of a comprehensive machine-readable lexicon for Zulu

A comprehensive machine-readable (MR) lexicon is a strategic resource in the development of language technology. Two essential aspects are the development of a comprehensive data model and its accurate and efficient implementation. This study focuses on the development of MR lexicons for the South African Bantu languages and more specifically Zulu.

Semi-automatic retrieval of definitional information: a northern Sotho case study

Corpus-based terminology is currently gaining ground on the international front. It is therefore important that terminologists working on the South African Bantu languages not only take note of this development, but that they should also follow this trend, even if they do not have the same measure of access to highly sophisticated software. The aim of this article is therefore to establish whether it is possible to retrieve definitional information on key concepts from untagged, running text by making use of affordable and easily accessible software such as WordSmith Tools. In order to answer this question, a case study is done in Northern Sotho, using textual material on linguistics as basis for a special field corpus. Syntactic and lexical patterns serving as textual markers of definitional information are identified and the success rate of the computational retrieval of definitional information is analysed and evaluated. Attention is also paid to the retrieval of specifically conceptual information, which turned out to be a fortunate by-product of semi-automatic retrieval of definitional information. Finally, it is illustrated how definitional information retrieved can be utilised in the writing of a formal terminological definition. Keywords: terminology, south african bantu languages, definitional information, semi-automatic information retrieval, terminological definitions, conceptual relationships, lexical patterns, syntactic patterns, textual markers, keyword-in-context (kwic), wordsmith tools

Translating technical (LSP) texts into the official South African languages: A corpus-based investigation of translators’ strategies

Translating technical texts into the South African Bantu languages, and to a lesser extent into Afrikaans, is an activity fraught with problems, the biggest one being the lack of terminology in these languages. In this article-undertaken within the framework of Corpus-based Translation Studies (CTS), and more specifically within the subfield of Applied CTS-we focus on technical translation in the official South African languages, with specific reference to the manner in which (professional) translators deal with the issue of terminology. To this end, and growing out of an earlier pilot study (Gauton et al. 2003), a multilingual corpus of ten sets of parallel texts in all eleven official South African languages was compiled and studied. Taking our cue from the general perception that the translation of terms by means of loanwords and/or transliterations is a dominant strategy, particularly in the South African Bantu languages, these strategies form the focus of the first phase of our investigation. This is followed by a comparative analysis of the translation strategies used by African language translators in finding suitable translation equivalents for English terms foreign to the African languages, using the term'formation strategies' as identified by Mtintsilana and Morris (1988) and'translation strategies' as explicated by Baker (1992) as a basis. For comparative purposes, 20 sample terms were selected from the parallel corpora. The translation strategies utilised in the translation of these terms were then identified and statistically compared. The study concludes with suggestions to African language translators of technical texts in dealing with the problem of non-equivalence encountered in this type of translation work.

Phonetics of intonation in South African Bantu languages

Much is already known about the prosodic systems of the indigenous South African languages from descriptions and analyses in the existing literature. All of the existing work has been carried out in the field of African studies or formal linguistics. In order to be able to implement the generalisations obtained into computational models in speech processing, the existing sources and results must be made accessible to researchers in these areas. In the opposite direction, the modelling of Bantu intonation in speech processing shows the need for more quantitative data. The present article presents a review of durational and pitch-related tonal aspects relating to intonation in South African Bantu languages. Its aim is, on the one hand, to make the results of Africanist and linguistic studies accessible to research in speech processing, laying a foundation on which work in the speech-processing branch of Human Language Technologies can be based. On the other hand, by pointing out gaps in our knowledge, it wants to draw attention to the fact that more quantitative research is needed in order to advance our knowledge also in the theoretical linguistic field.

On the development of a tagset for Northern Sotho with special reference to the issue of standardisation

Working with corpora in the South African Bantu languages has up till now been limited to the utilisation of raw corpora. Such corpora, however, have limited functionality. Thus the next logical step in any NLP application is the development of software for automatic tagging of electronic texts. The development of a tagset is one of the first steps in corpus annotation. The authors of this article argue that the design of a tagset cannot be isolated from the purpose of the tagset, or from the place of the tagset and its design within the bigger picture of the architecture of corpus annotation. Usage-related aspects therefore feature prominently in the design of the tagset for Northern Sotho. It is explained why this proposed tagset is biased towards human readability, rather than machine readability; this choice of a stochastic tagger is motivated, and the relationship between tokenising, tagging, morphological analysis and parsing is discussed. In order to account at least to some extent for the morphological complexity of Northern Sotho at the tagging level, a multilevel annotation is opted for: the first level comprising obligatory information and the second optional and recommended information. Finally, aspects of standardisation are considered against the background of reuse, of sharing of resources, and of possible adaptation for use by other disjunctively written South African Bantu languages. It is not the aim of this article to evaluate the results of any tagging procedure using the proposed tagset. It only describes the design and motivates the choices made with regard to the tagset design. However, an evaluation is in process and results will be published in the near future (cf. Faaß et al., s.a.).

Cattle and cattle colour terminology in South Africa

A comparison between cattle colour terminology found in the Khoikhoi languages and that found in South African Bantu languages brings noticeable similarities to light, leading to the assumption that both the cattle and the terms used to describe their characteristic features were acquired by the South Eastern Bantu-speakers through contact with Khoikhoi pastoralists. In this article we investigate the possibility that these terminological similarities are the result of mutual relatedness to a single progenitor set of terms, rather than of borrowing that took place in a restricted geographical area. To this end, three hypotheses are proposed, respectively termed the East African link-hypothesis, the Northern Botswana link-hypothesis and the Sudan link-hypothesis. We argue that the latter, pointing to a common Nilotic proto-source, represents the most plausible explanation for the origin of cattle colour terminology shared by amongst others, the South African Bantu and Khoikhoi languages.

Corpus-based linguistic investigation for the South African Bantu languages: a Northern Sotho case study

The advent of electronic corpora has revolutionized linguistic investigation internationally and is starting to make inroads into the study of the South African Bantu languages. In this article, it is argued that the move from traditional language studies to corpus-based investigation would be a natural one for South African Bantuists, since it proffers a familiar methodology in a new guise. To illustrate the value and potential of corpus-based investigation, a case study was done on Northern Sotho adjectives, highlighting certain semantic aspects which have hitherto not been noted by grammarians, and illustrating the historical relationship between adjectives and enumeratives.

Development of reusable resources for Human Language Technologies (HLT) applications: practice and experience

Language resources, by their very nature, serve as a repository of linguistic knowledge. They are therefore essential in the building and improvement of natural language applications. The aim of this paper is to elaborate on the practice and the experience gained in the development, maintenance and management of such resources with specific reference to African languages. The focus is on the methods of collection and the formats concerning word lists, morphological analysis and lemma lists. The resources discussed, are those developed in collaborative research with North-West University's Spelling Checker Project. As a broader perspective, the reusability of such resources is highlighted. Recommendations are also made regarding the way forward nationally in developing a resource centre to facilitate the technological development of South African Bantu languages.

Sotho-Nguni Orthography and Tone-Marking

FOR years now the question of orthography in the South African Bantu languages has kept the authorities busy. It flares up periodically in different parts of the country, and local committees are called from time to time to deal with the conflagration.In the past, unfortunately, these committees have too often concerned themselves with the problems of one language only, and apparently ignored both the effect their decisions would have on literature exchange and the attempts of previous committees to solve similar problems in related languages. Thus we have Zulu and Xhosa, very closely related languages, with considerable orthographical differences, while North Sotho and Tswana now differ from each other and from South Sotho, which (perhaps wisely) has set its face against orthographic change since 1906.Until Dr. Jacob Nhlapo launched his campaign for Southern Bantu linguistic unity, few people had seriously considered the possibility of an orthographic system which would cover both the Nguni and Sotho groups, the feeling being that such unity lay in the realms of wishful thinking in view of the virtual impossibility of achieving unity within the groups.The present article is an attempt to combine the experience of past efforts into a system that would actually be applicable to these two language groups. The acceptance of such a system would naturally involve great sacrifices of tradition and prestige from both parties. This aspect is not to be ignored, but at the same time can have little connection with the scientific elaboration of an alphabet. Consequently the argument “The Sotho (or the Nguni) would never accept such a solution” must be relegated to the sphere of practical politics.

South African Association for the Advancement of Science

THE South,african Journal of Science of June 1946 (volume 42; from the Association, Johannesburg) contains the repot of the Johannesburg meeting in 1945 of the South African Association for the Advancement of Science, including the presidential address of E. C. Chubb, director of the Durban Museum, on “Museums and the Advancement of Science”, the sectional addresses and the papers read or presented to the sections. The president, after indicating the services which the British Museum had rendered to the advancement of science, gave an interesting review of the activities of the museums of South Africa. Dr. H. van Gent's presidential address to Section A dealt with the contribution of variable star research to the progress of astronomy, while Mr. H. Wilson's presidential address to Section B, “Methane: a Neglected National Asset”, discussed the potentialities of methane and urged the serious co-ordinated exploration of the country's resources of natural gas containing methane. Prof. C. J. van der Horst's presidential address to Section D, “Revolution in Evolution”, reviewing the influence of Hugo de Vries' conclusion regarding evolution by sudden mutations, discussed more particularly the nature and origin of the trophoblast or feeding layer surrounding the mammalian embryo, and Prof. C. P. Lestrade, addressing Section E, dealt with “Some Problems of Bantu Language Development”. Prof. Lestrade was concerned mainly with whether all or any of the South African Bantu languages should survive, and if so, with the part they should play in the life of the Bantu-speaking peoples, and secondly, how we could best ensure that the languages deemed worthy of survival could play the part desired, and he referred in conclusion to the lessons to be learned from the language situation in the Philippines. Of equal practical interest is A. J. Limebeer's presidential address to Section F on “The Employment of the Partially Disabled considered as a Social Policy.”

IV. Critique on Mr Hyde Clarke's Theory of the Relation of the Australian to the South African (Bantu) Languages

IV. Critique on Mr Hyde Clarke's Theory of the Relation of the Australian to the South African (Bantu) Languages