University language planning â a Stellenbosch perspective on policy and AI

This story has significance for readers across Africa and beyond.

This year marks the 30th anniversary of the South African Constitution. Section 6 establishes 11 official languages (South African Sign Language became the 12th in 2023) and states that these âmust enjoy parity of esteem and must be treated equitablyâ. At South African universities, this constitutional provision underpins various language policies and language planning initiatives. Stellenbosch University, the oldest historically Afrikaans university, is currently reviewing its language policy. This article has two objectives: to summarise and provide a public comment on the first draft of the new policy; and to situate the issues raised within the wider context of university language planning in South Africa. This includes a critical reflection on the emerging role of generative AI in language planning.

The wider language planning context

To situate the subsequent analysis of the draft policy, I begin with a brief discussion of the wider context and history of university language planning. A more detailed discussion is available here. Language planning emerged as a subfield of American applied linguistics in the aftermath of World War 2. The field was premised on an awkward distinction between âcorpusâ and âstatus planningâ, which built on Saussureâs foundational linguistic distinction between internal (structural) and external (functional) changes in âa languageâ. Corpus planning is the dominant orientation, concerned with the formal and technical development of a language (e.g. grammar, lexicon). Status planning emerged as a residual or correlate of corpus planning, i.e. an inert and politically neutral process of adding functions to extant languages. Status planning effectively airbrushes the history of socioeconomic contestation and political conflict (nationalism⦠colonialism) out of language standardisation and institutionalisation.

This intellectual tradition entered South African universities during the democratic transition, inspired by the multilingual settlement in the 1993/4 interim Constitution and the 1996 final Constitution. The interim Constitution established 11 official languages, along with a language-centric impetus for status planning and âequal useâ. The 1996 Constitution retains this commitment, but qualifies it in terms of âequity; practicability; and the need to redress the results of past racially discriminatory laws and practicesâ. More recently, numerous constitutional and other court judgments have reflected this founding tension between âpureâ language status claims and the need to address the historical intersection of language, race and socioeconomic status. These court cases invariably involved Afrikaans interest groups challenging language policy changes at historically Afrikaans universities (HAUs).

Language planning became a formal institutional process at South African universities in the years after 2002, in a process that was subservient to the fundamental restructuring (via mergers) of the higher educational landscape. Around this time, Stellenbosch University and other HAUs shifted to various forms of institutional bilingualism. These universities were central to the institutionalisation of language planning, for two reasons. First, they provided a focused rationale or status goal for language planning: the defence of Afrikaans within a transforming university sector. Second, given that Afrikaans is the only South African language with a local history of standardisation and development at university level, they provided the corpus planning expertise. At South African universities we see a clear trend: institutional language planners tend to be language professionals, typically equipped with the technical skills associated with corpus linguistics or applied language studies more generally. These are now supplemented by a new category of expertise: computational linguistics and the machine learning competence that underpins the currently dominant form of âartificial intelligenceâ â large language models.

The new draft language policy

Stellenbosch University recently released âreview draft 1â of its new language policy, which will be implemented in 2027. The initial sections of the document commit the university to âadvancing multilingualism as both a resource and a strategic priorityâ and asserts that âinstitutional support for language use will be context-dependentâ. In a section titled âThe multilingual contextâ, the draft policy distinguishes âinstitutional multilingualism (i.e. establishing multilingual institutional spaces where a diverse group of people coexist)â and âindividual multilingualismâ. The latter is framed in terms of âplurilingual repertoires â the ability to use multiple languages and dialects in complementary ways â which can be harnessed and expanded to support communicationâ. The draft policy notes that âwithin this broader commitment to multilingualism, SU focuses its institutional language practices on Afrikaans, English and isiXhosa (in alphabetical order), as languages predominantly used in the Western Cape and widely represented in the University communityâ.

The problem with this initial framing is that both institutional multilingualism and individual multilingualism are framed in general, and hence vague, terms. Moreover, the commitment to multilingualism as a resource is abstract, effectively masking the actual economic and social costs associated with specific languages used and developed in specific university contexts. This becomes clear as one moves from the initial discussion of context and principles, to the more concrete âpolicy provisionsâ in section 7. Here context and resources become more specific: âAfrikaans and English function as the primary languages of instruction in our classes (depending on institutional resources and capacity), while isiXhosa is increasingly supported where practicable.â This section builds on the current (2021) policy and retains the tacit commitment to English-Afrikaans institutional bilingualism. As with previous policies, the commitment to âincreasing the use of isiXhosaâ is a feint, effectively obscuring the continued commitment of resources to the defence of Afrikaans and English-Afrikaans institutional bilingualism.

The undergraduate policy provisions in section 7 retain the essential structure of the language specifications contained in two previous language policies (2016 and 2021). This is the same formula that was developed, rapidly, in response to the 2015 Open Stellenbosch and Fees Must Fall student protests. The core provisions are rendered more âflexibleâ by relinquishing the labels that had previously (pre-2016) been used and contested. I reinsert the labels (in bold), to show what I mean:

- â7.1.3 For undergraduate modules where it is reasonably practicable and pedagogically sound to have more than one class groupâ¦â (parallel medium);

- â7.1.4 For undergraduate modules where both Afrikaans and English are used in the same class groupâ¦â (dual medium); and

- â7.1.5 In the following instances, undergraduate lectures will be offered in one language onlyâ¦â (English or Afrikaans single medium.

The old labels have been dropped, I suspect, because they are untenable. Parallel medium instruction is not feasible, as the universityâs own 2021 language policy review showed. The old dual medium option (then known as the âT-optionâ) was effectively discontinued in 2016 â in the face of student protests. Since then, the distinction between âdualâ and âsingleâ medium has blurred, for two reasons. First, the quasi-dual medium provision (7.1.4) once again establishes English as the principal undergraduate medium, by asserting that âall information is conveyed at least in English, and summaries or emphasis on content are also provided in Afrikaansâ. Second, both the dual (7.1.4) and single medium (7.1.5) provisions are effectively conceptualised as types of digital parallel medium instruction.

At undergraduate level, English-Afrikaans institutional bilingualism is thus retained â more on paper than in practice â by two techno-linguistic fixes. The first is simultaneous interpreting in the classroom, which has been used for more than 10 years. This dubious pedagogical practice is once again presented as a solution to the decline of English-Afrikaans bilingual competence among students and â to a lesser extent â staff. The university does not release actual use statistics, but my sense is that interpreting is used very unevenly across faculties and modules.

The second is new: âICT-enhanced and AI-assisted learning strategies. These may include educational audio and/or video recordings, interactive digital platforms and adaptive learning tools, which may be offered in English and/or Afrikaans and, where feasible, in isiXhosa and/or SASL.â This somewhat speculative nature of this strategy is evident in the call for âconsidered, research-led use of appropriate technologies to support multilingual engagement and accessâ. The draft policy also qualifies this option as follows:

The university recognises that AI-mediated translation and interpretation do not uniformly capture the nuances of all languages, particularly African languages and SASL, and therefore promotes their considered, research-led use.

I agree with this statement, but it is discordant with the generally techno-optimistic orientation of the draft policy. âAI-mediated translation and interpretationâ are clearly envisaged, and the policy once again imposes general technological solutions with little regard for changing language trends (discussed below) and context/discipline-specific academic needs.

Website publishing is one noteworthy example of a context in which the rigid enforcement of institutional bilingualism has very deleterious consequences. The universityâs new Drupal-based website compels the translation of all web-material at all levels into Afrikaans. The training material states that this must be done by human translators. This would be exorbitantly expensive, in both money and time, so it is not happening. Web content must therefore be machine translated, but this process is slow and problematic. If this practice remains in place, the general effect will be the continuous delay and neglect of website updates.

The changing context of language use at Stellenbosch University

Over the past decade, Stellenbosch University has shifted rapidly to English, not simply as a lingua franca, but also as an implicit institutional commitment. This is reflected most obviously in the fact that Afrikaans competence is no longer an entrance requirement, for student enrolment or academic appointments. This institutional commitment is also manifested in the universityâs transformation objectives, its research status (and research status ambitions) and its commitment to internationalisation.

The current draft ignores this shift, and the fact that growing diversity at the level of âplurilingual repertoiresâ has reinforced the status of English as the dominant academic medium. The following graph presents the three-decade home language trend at undergraduate level.

The âotherâ category masks the growing diversity of first language speakers. This year, the other category includes 57 minority home languages. The university also records English and Afrikaans as âtuition languagesâ. Here the shift is also evident within the Afrikaans student community (undergraduate and postgraduate): this year 78% of students who indicated Afrikaans as their first language chose English as their tuition language.

The draft policy needs to acknowledge these trends and the de facto status of English as the principal academic language at Stellenbosch. In the age of AI this status is a double-edged sword, as students (and staff) resort increasingly to ChatGPT and other large language models, which are trained predominantly on North American English texts. However, abstract commitments to âmultilingualismâ will not resolve this dilemma. The policy needs to reflect the actual costs and trade-offs associated with the development and use of specific languages in specific academic contexts. My concrete recommendation: relinquish the centralised bilingual apparatus, and channel more resources into context-specific language support â and academic writing in particular.

The issues that I have highlighted in the new draft SU language policy reflect a more general trend in South African university-based language planning. This can be summarised in terms of the tension between the formal and technical processes associated with discrete language development, and the more messy and politically charged processes that manifest as unequal language status. The toolbox of the institutional language planner is now, seemingly, turbo-charged by the advent of generative AI and a host of new techno-optimistic LLM-as-tool discourses. But these run up against fundamentally non-technical asymmetries: the deep repertoire and institutional legitimacy of English; the declining academic status of Afrikaans; and the collective aspirational status of the official African languages. DM

Reporting originally appeared via Daily Maverick. Read the full source for additional context.