Skip to content
problem genome

One of 592 unsolved problems in the Problem Genome, a collection written up for student design and engineering teams.

health · social · digital

a voice mostly in english

South African adults who use communication aids understand several languages, but their speech-generating devices spoke almost only English: among 18 device users from non-English homes, one had a synthetic voice in her home language

Source: Tönsing et al., “Multilingualism and augmentative and alternative…” Open ↗source: primaryLast updated: 2026-11-17

Problem statement

People who cannot rely on speech, for example because of cerebral palsy, brain injury or motor neurone disease, often communicate with augmentative and alternative communication (AAC): picture boards, alphabet boards, or speech-generating devices (tablets, laptops or dedicated machines that speak what the user selects or types). South Africa had 11 official languages when the survey was done, and many people use several every day. A survey published in 2019 (ethics approval 2016) by the University of Pretoria's Centre for AAC and the CSIR's Human Language Technology group asked 27 multilingual adults who use AAC how they communicate. All understood at least two languages well, and their communication partners used an average of 3.6 languages with them, but they expressed themselves in an average of 2.1. The aids were the narrowest channel. Of the 24 participants from non-English homes, 18 had a speech-generating device, and only one had a synthetic voice in her home language (a free, low-quality Afrikaans voice); three others used recorded home-language speech, which does not easily allow new sentences. No participant used a picture board in more than one language. Most wanted more access to their home language and other languages, and the barrier they named most often was that their device did not offer the language. The unsolved problem is an AAC system that lets a multilingual user say new things in the languages their family and community speak, including users who cannot yet read and write those languages.

Why this matters

Only 9.6% of South Africans have English as their first language, the paper notes, yet AAC devices, software and intervention are provided mainly in English. For a person who uses AAC, a system in one language means family, neighbours and congregations who do not speak English cannot be addressed directly, and the participants' most-chosen reason for wanting more languages was simply that more people would understand them. Group cohesion or identity came next, followed by its being the community's language and part of who they are. The authors also report a knock-on effect from earlier work: some caregivers choose English rather than home-language literacy teaching for their children because speech-generating devices with home-language voices are scarce. The South African Constitution grants every citizen the right to use the language of their choice; the authors argue that a one-language device is a further exclusion for people already at risk of powerlessness.

What’s been tried and why it hasn’t worked

(1) Speech-generating devices as supplied. Participants used Windows laptops with text-to-speech, iPads with communication apps, Android tablets and phones, and two dedicated devices. In the paper's table of speech output, 19 of the 23 entries were English text-to-speech only (21 participants used a device, two of them two types of device). More than half of the participants from non-English homes (54%) had no access to their home language through either typing-based or picture-based aided methods. The barriers participants chose, in order, were: no access to the language on their device, no text prediction in the language, no pre-stored words or phrases in the language, and their own spelling skills. (2) Recorded speech. Three participants added recorded home-language speech to English text-to-speech. Recorded messages are fixed, so they do not "easily allow the spontaneous generation of novel utterances". (3) Translating English systems. Picture-symbol systems are built and sold mainly in English. The authors cite service providers' experience and a comparison of English and isiZulu core vocabulary (the most-used words in natural conversation) as showing the "limitations of translation of English systems into other languages", especially into languages with a different structure. In isiZulu, words are built from many small meaningful parts (formatives), and a 2019 study found that a core vocabulary works better when counted by formatives than by whole words. (4) What service providers do. In focus groups with 15 South African AAC service providers (2018), practice ranged from home language only, to English only, to both. Although many saw multilingual access as positive, "this did not always translate into multilingual AAC practices", because of the lack of devices, software and apps in non-English languages and because of providers' own language skills and their beliefs that several languages on one system would be too demanding. (5) Pieces now built. South African-language voices exist: the CSIR's Qfrency text-to-speech covers "all the 11 official South African languages", and the CSIR's page names its commercialisation focus as "the business-to-business sector", such as call centres; the page does not mention AAC or disability. The voices are, however, sold for AAC: a South African assistive-technology supplier, Inclusive Solutions, sells the Qfrency voices for Windows as SAPI 5 voices in all 11 languages (some in one gender only) and names Grid 3 and Communicator 5 among the AAC software that can use them (of Communicator 5: "Support included for isiZulu & isiXhosa!"); its page says voices for Android are "in development" and offers a software kit and API for other platforms. Smartbox's help page confirms that Grid 3 "can support SAPI5 voices". So typed home-language speech is on sale for Windows devices, the kind 10 of the survey's participants used with text-to-speech; it does not yet reach the iPads, Android tablets and phones that 9 participants used, and it serves only users who can spell. Commercial vocabularies stop at Afrikaans (makers' own language lists, opened 2026-11-17): Smartbox lists Afrikaans grid sets for Grid 3, among them Super Core, Voco Chat and Text Talker (page revised 2024), but none in isiZulu, isiXhosa, Sesotho, Setswana or Sepedi, and its voice list (revised 2022) has no voice in any South African language except South African English; Proloquo2Go is offered in English, Spanish, French and Dutch; the TD Snap voice-language lists (on a Tobii Dynavox page that also offers neural voices in over 100 languages without listing them) and Avaz's language list name no South African language other than English, nor does the Ace Centre's language-library entry for the open-source Cboard. Among mainstream voices, Microsoft's Azure speech service lists Afrikaans and isiZulu neural voices and Google Cloud lists Afrikaans only; Acapela's voice list names none, and Apple's iOS feature list names Afrikaans, Xhosa and Zulu only as keyboards. An earlier prototype existed: the CSIR Meraka Institute's GNApp, "An augmentative and alternate communication (AAC) device which generates synthesised (or pre-recorded) speech" from icons, was released as a demonstration in English, Afrikaans and isiZulu (version September 2009); its repository record lists no files and no evaluation. The South African Speech-Language and Hearing Association offers printed medical, preschool and school communication boards revised at Stellenbosch University, each as separate English, Afrikaans and isiXhosa files (posted 2025); no isiZulu or two-language board is listed there. Printed two-language boards with isiZulu do exist for one topic: in 2011 the University of Pretoria's Centre for AAC, with Temple University, published boards for telling someone about crime, abuse or neglect, built from the 56 words named by at least two of four language focus groups (English, Afrikaans, Sepedi and isiZulu) and shown as Picture Communication Symbols, with an alphabet side. The boards were made in English and Sepedi and then translated into Afrikaans and isiZulu; each carries English and one other language, "with one gloss at the top and the other below the symbol". The authors call it "a generic topic-based board", meant for a first disclosure, not a vocabulary for everyday conversation. The same CSIR and Pretoria team reported integrating South African-language text-to-speech into accessibility applications in 2017 (conference paper). A University of Cape Town team built synthetic child voices in South African English, Afrikaans and isiXhosa matched to three children with expressive communication difficulties (2024), and the children, caregivers, teachers and therapists who heard them accepted them and preferred them to British or US English voices (2025); but that paper names as a limitation that "there are limited AAC devices/apps which are currently equipped to load synthetic African voices", and calls for core vocabularies to pair with the voices. A Zulu core vocabulary of 213 formatives (2019) and a validated Sepedi core vocabulary list (2024) have been published, and a team is gathering design requirements for a non-electronic Sepedi picture-symbol board from teachers, therapists, parents, linguists and adults who use AAC (2025). A 2022 scoping review of AAC systems built in languages other than English (22 studies) opens by stating that access to such systems and to multilingual systems "remains limited". No source read reports these pieces combined into a system that multilingual users have taken up.

What would unlock progress

The voices and vocabulary lists now exist separately, and the voices can already be bought for Windows AAC software; the gap is getting them onto the iPad and Android apps many users carry (few AAC apps can load African-language voices, per the 2025 child-voice study), building picture-based vocabularies in the African languages (commercial ones stop at Afrikaans), and a system design that holds more than one language without doubling the user's effort. Open design questions the sources raise: how to organise picture symbols so the same symbol can be voiced or labelled in either language (the 2011 abuse-reporting boards show one labelling answer, a gloss above and below each symbol, on a single topic); how to support switching mid-conversation (code-switching), which multilingual speakers do routinely; how to lay out vocabulary for a language built from formatives rather than whole words; and how to give non-literate users new-sentence power without spelling. The authors call for collaboration between people who use AAC and their families, service providers, linguists and language-technology specialists to build adjustable templates, then studies of how such systems work in daily life. A 2022 study by the same group found that 30 English–Afrikaans bilingual children aged 4–5, without disabilities, taught nine symbol meanings in English partly carried them over to Afrikaans (median correct rose from 0 to 6 of 9), which suggests one symbol set can serve two languages; it has not been tested with children who use AAC.

How we’d know it’s being solved

(a) The share of multilingual AAC users who can produce novel sentences, not only stored phrases, in their home language as well as English (here: 1 of 18 device users from non-English homes had home-language synthetic speech). (b) The number of languages a user expresses in, approaching the number they understand and are addressed in (here 2.1 against 3.6 used by partners). (c) Use of a picture-based system in more than one language (here none). (d) Service providers reporting that a multilingual option exists and that they offer it. (e) Communication partners who do not speak English rating the user as understandable.

Where to start

Facility-free door (make a thing): a non-electronic Sepedi board is already being designed, and translated English boards are what the sources say falls short, so the team's object should test the open layout question instead: a printed, low-tech bilingual communication board or book, isiZulu–English, for everyday conversation rather than one topic, in which the core cells follow the published Zulu core vocabulary rather than a translated English list. Start from the one printed isiZulu–English symbol board found, the Centre for AAC's 2011 abuse-reporting board, which already puts both labels on each symbol (one above, one below) so that the partner reads the gloss in their own language; it is a topic board translated from an English–Sepedi original, and no source on the page reports a general-purpose isiZulu–English core board, or a tested way for the user to switch language on one. In the team's board each symbol cell carries both labels and the user (or partner) can switch language by flipping an overlay, turning a page edge or using a colour code. A board needs no synthetic voice, so it does not wait on the device problem, and pointing at symbols lets a user who cannot spell build new sentences. Check what is sold first: the makers' language lists checked in November 2026 showed picture vocabularies in Afrikaans but none in isiZulu, and the printed South African boards found were topic boards: the speech-therapy association's medical, preschool and school boards, one language each, and the Centre for AAC's two-language abuse-reporting boards. Check again before building; if an isiZulu–English core board has appeared, test it against the team's instead of building from scratch. Build two or three layouts and compare them with campus student volunteers, ideally multilingual ones who read both languages, in timed message-building tasks (how fast a set of target sentences can be pointed out, in each language and mixed). What carries over from such a test: how fast cells are found, how much a language switch costs, and whether labels stay legible. What does not: the motor and access methods of people who use AAC, real conversation with family, and, with volunteers who do not speak isiZulu, any judgement of the isiZulu itself. The only materials needed are the published vocabulary studies (the Zulu and Sepedi lists have open author copies in the University of Pretoria repository, as does the 2011 paper describing the abuse-reporting boards), a printer and lamination. User door: the real users, multilingual South Africans who use AAC and their families, cannot be reached from a campus outside South Africa, but the same design question, one board for two languages, faces bilingual families elsewhere. A campus speech-language-hearing clinic, or a school serving bilingual families, can connect a team with speech-language therapists and with bilingual (for example Spanish–English) families of children who use AAC, to critique the switching method and layout. What carries over is the switching method and layout logic; the isiZulu core vocabulary and South African family settings do not. Therapists and parents can critique the boards; any session with a person who uses AAC, child or adult, is design-the-test only, with the campus's ethics review. A field study in South Africa is designed, not run, and needs a local partner of the kind the sources describe (a university AAC centre with a speech-language therapy service). Relevant skills: communication and graphic design, linguistics, speech-language pathology, human-computer interaction.

Genes

Related problems

Stuck the same way, in other fields.

field: society & learningmore than the buildThe first large trial of LEGO-based therapy for autistic pupils in English mainstream schools found a small social-skills benefit at 20 weeks, non-significant in the main analysis and not sustained at 52; its team's later fidelity analysis judges gaps in the adult's facilitation of social interaction key contributorsbuilt without the people who'd live with itsource: primaryfield: water & energygrants built for big utilitiesFederal resilience grants for water systems are designed for applicants that small utilities cannot be — the cost share alone can exceed a small town's annual budget, and no one tracks who gives upbuilt without the people who'd live with itsource: primaryfield: society & learningexperiments without facesDigital learning platforms can now run cheap randomized experiments on hundreds of thousands of students — but cannot answer "for whom does it work?" Because they may not hold student demographic databuilt without the people who'd live with itsource: secondaryfield: digital & spaceten times the data, same computing budgetThe high-luminosity LHC will produce ten times the collision data, and current computing cannot scale to itbuilt without the people who'd live with itsource: primary

Sources

  • Tönsing KM, van Niekerk K, Schlünz G, Wilken I. "Multilingualism and augmentative and alternative communication in South Africa – Exploring the views of persons with complex communication needs." African Journal of Disability 2019;8:a507, doi:10.4102/ajod.v8i0.507 (publisher AOSIS, South Africa), accessed 2026-11-17. Link ↗
  • Tönsing KM, van Niekerk K, Schlünz GI, Wilken I. "AAC services for multilingual populations: South African service provider perspectives." Journal of communication disorders 2018;73:62-76, doi:10.1016/j.jcomdis.2018.04.002 (author's accepted manuscript read in full from the University of Pretoria repository), accessed 2026-11-17. Link ↗ Link 2 ↗
  • Schlünz GI, Wilken I, Moors C, et al. "Applications in accessibility of text-to-speech synthesis for South African languages." Proceedings of the South African Institute of Computer Scientists and Information Technologists 2017:1-10, doi:10.1145/3129416.3129445 (read in full from CSIR ResearchSpace), accessed 2026-11-17. Link ↗
  • Amery R, Thirumanickam A, Barker R, Lowell A, Theodoros D, Raghavendra P. "Developing Augmentative and Alternative Communication Systems in Languages Other Than English: A Scoping Review." American Journal of Speech-Language Pathology 2022;31(6):2900-2919, doi:10.1044/2022_ajslp-21-00396 (abstract read via the Europe PMC record), accessed 2026-11-17. Link ↗
  • Terblanche C, Schnoor TT, Harty M, Tucker BV. "The development of synthetic child speech in three South African languages." Augmentative and Alternative Communication 2025;41(4):333-344 (published online 2024), doi:10.1080/07434618.2024.2374312 (abstract read via the Europe PMC record), accessed 2026-11-17. Link ↗
  • Terblanche CC, Pascoe M, Harty M. "Do you like my voice? Stakeholder perspectives about the acceptability of synthetic child voices in three South African languages." International Journal of Language & Communication Disorders 2025;60(1):e13152, doi:10.1111/1460-6984.13152, accessed 2026-11-17. Link ↗
  • Inclusive Solutions, "Qfrency South African Voices" (supplier page), accessed 2026-11-17. Link ↗
  • Smartbox, "Which voices are available with Grid 3?" (maker's help page), accessed 2026-11-17. Link ↗
  • Smartbox, "Which grid sets are available in my language?" (maker's help page), accessed 2026-11-17. Link ↗
  • AssistiveWare, "Proloquo2Go - AAC app with symbols" (maker's page), accessed 2026-11-17. Link ↗
  • Tobii Dynavox, "AAC voice options" (maker's page), accessed 2026-11-17. Link ↗
  • Avaz, "In what languages is Avaz available?" (maker's help page), accessed 2026-11-17. Link ↗
  • Ace Centre, "Cboard AAC" (language-library entry), accessed 2026-11-17. Link ↗
  • Microsoft Learn, "Language and voice support for Azure Speech", accessed 2026-11-17. Link ↗
  • Google Cloud, "Supported voices and languages", accessed 2026-11-17. Link ↗
  • Google Cloud, "Gemini-TTS", accessed 2026-11-17. Link ↗
  • Acapela Group, "Repertoire", accessed 2026-11-17. Link ↗
  • Acapela Group, "Available Languages", accessed 2026-11-17. Link ↗
  • Apple, "iOS and iPadOS 27 - Feature Availability", accessed 2026-11-17. Link ↗
  • Meraka Institute, CSIR, "GNApp (VSep2009)" (SADiLaR repository record), accessed 2026-11-17. Link ↗
  • South African Speech Language Hearing Association, "Communication Boards", accessed 2026-11-17. Link ↗
  • CSIR, "Converting text into automated South African languages", accessed 2026-11-17. Link ↗
  • Bornman J, Bryen DN, Kershaw P, Ledwaba G. "Reducing the Risk of Being a Victim of Crime in South Africa: You can tell and be Heard!" Augmentative and Alternative Communication 2011;27(2):117-130, doi:10.3109/07434618.2011.566696 (open author copy read in full), accessed 2026-11-17. Link ↗
  • Bornman J, Bryen DN, Kershaw P, Ledwaba G. "Communication for ALL: English to Zulu Alphabet Adult Board" (printed board, 2011; the University of Pretoria address now returns 404; Internet Archive copy), accessed 2026-11-17. Link ↗
  • Mngomezulu J, Tönsing KM, Dada S, Bokaba NB. "Determining a Zulu core vocabulary for children who use augmentative and alternative communication." Augmentative and Alternative Communication 2019;35(4):274-284, doi:10.1080/07434618.2019.1692902 (open author copy read in full), accessed 2026-11-17. Link ↗
  • Tönsing KM, Mothapo NRB, Morwane RE, Soto G. "Stakeholder validation of a Sepedi core vocabulary list as a resource for augmentative and alternative communication." International Journal of Speech-Language Pathology 2024;26(4):566-577 (published online 2023), doi:10.1080/17549507.2023.2241671 (open author copy read in full), accessed 2026-11-17. Link ↗
  • Tönsing KM, Dada S. "Learning graphic symbols in two languages: Effects of monolingual teaching." International Journal of Speech-Language Pathology 2023;25(2):281-291 (published online 2022), doi:10.1080/17549507.2022.2048889 (open author copy read in full), accessed 2026-11-17. Link ↗
  • Mothapo NR, Tönsing K, Soto G. "Hearing stakeholder voices in the design of a non-electronic sepedi graphic symbol-based AAC resource." Augmentative and Alternative Communication 2025:1-13, doi:10.1080/07434618.2025.2588554 (published version read in full from the University of Pretoria repository), accessed 2026-11-17. Link ↗
Revision notes

This brief gets checked against its sources and contains the following revisions. Future revisions get added here.

Worth knowing

  • The main evidence is one survey of 27 adults who use communication aids, published in 2019.

Full check

Sources read

  • Tönsing and others, survey of multilingual South African adults who use communication aids (2019): full text
  • Tönsing and others, focus groups with service providers (2018): full text (author accepted manuscript)
  • Schlünz and others, South African voices in communication software (conference paper, 2017): full text
  • Mothapo, Tönsing and Soto, Sepedi communication board (2025): full text
  • Mngomezulu and others, Zulu core vocabulary (2019): full text
  • Tönsing and others, Sepedi core words (2024): full text
  • Tönsing and Dada, symbol study with bilingual children (2022): full text
  • Terblanche, Pascoe and Harty, synthetic child voices in three South African languages (2025): full text
  • Bornman and others, crime- and abuse-reporting communication boards (2011): full text (with the archived English and isiZulu board)
  • Terblanche and others, child voices study (2024): abstract only
  • Amery and others, scoping review (2022): abstract only
  • Makers' and suppliers' pages: Inclusive Solutions, CSIR, Smartbox, AssistiveWare, Tobii Dynavox, Avaz, Ace Centre, Microsoft, Google, Acapela, Apple, SADiLaR, SASLHA: full text

What changed

  • Not found in the sources, replaced with what they say: the three Cape Town children whose voices were built were said to use communication aids; both papers describe them as children with expressive communication difficulties.
  • Stated more strongly than the source: recorded speech was said to be unable to build new sentences; the paper says it does not easily allow them; 'most' South Africans using several languages every day is now 'many', as the paper says.
  • Numbers: the count of English-only speech output is now given as 19 of the 23 entries in the survey's table, since 21 people used a device and two of them used two kinds.
  • Title: the full title now says one device user had a synthetic voice in her home language, since three others had recorded home-language speech (display title unchanged).
  • Entry points: the first door now starts from a two-language board with isiZulu that already exists for one topic and from the published Zulu core vocabulary, and says what is still open: a general-purpose isiZulu and English board and a tested way to switch language on it.
  • Other: What's Been Tried now notes that printed two-language boards with isiZulu exist for one topic: the 2011 crime- and abuse-reporting boards, each in English and one other language.

Left open

None noted in this check.

Separate check of specific claims: The count of 11 official languages is now tied to the time of the survey, What would unlock progress now points to how the 2011 two-language boards label each symbol, and citations for the sources the page uses were added to its source list.

Earlier full check: The iPads, Android tablets and phones were 10 devices used by 9 people, a supplier's line about isiZulu and isiXhosa support applies to one program only, a maker's unlisted voices are now mentioned so its language lists are not read as proof, a quotation now matches the paper, and two papers are now marked as read in full.

Separate check of specific claims: What's Been Tried now adds South African voices on sale for Windows communication software and what the main apps and voice services cover; the display title changed from 'A Voice in One Language Only' to 'A Voice Mostly in English', because home-language voices are now sold for some devices.

Separate check of specific claims: The first door keeps a printed two-language board and adds a check of what is sold and campus volunteers; the user door says the real users are out of reach and extends the ethics line to every person who uses a communication aid.

Earlier full check: The survey is now dated by publication, existing South African synthetic voices and a newer review were added, an unsupported claim about one voice maker's software was replaced with its own page's words, and a symbol study was redated and limited to children without disabilities.