Listening to the Voices of Setswana Elders Through Corpus Linguistics

Setswana, one of the dominant languages of Botswana and parts of South Africa, has been documented in dictionaries and grammars for over a century. Yet much of what its oldest speakers know still lives only in conversation: proverbs offered while pounding sorghum, prayers murmured before a journey, technical descriptions of cattle breeds exchanged in the kraal. A spoken corpus built from these moments can recover a register that no printed reference work has captured.

The urgency behind such a project is demographic. In towns like Gaborone and Mahalapye, the median age of fluent speakers of certain older Setswana forms is rising quickly. When those voices fall silent, whole clusters of vocabulary — directions for thatch roofs, the cadence of a courtship song, the names for rain-bearing clouds — risk vanishing with them. A corpus cannot recreate a living community, but it can preserve enough for descendants and scholars to study on their own terms.

This work sits naturally alongside the lexicography and translation already explored on this site, where Setswana vocabulary, etymology, and language contact receive careful treatment. Researchers at the Australian National University in Canberra have shown comparable dedication to recording Aboriginal elders, and that parallel offers a useful frame for what follows.

A spoken corpus is also a technical artefact. Its value depends on rigorous metadata, ethical consent, and an analytical workflow that allows future users to query the data flexibly. The discussion below walks through that workflow — from the first field recording to the comparative conclusions that connect Setswana to other language-preservation efforts worldwide.

Building a Spoken Corpus of Older Generations

A Setswana elder-speech corpus usually begins with interviews, narrative sessions, and group conversations recorded in homes, churches, and cattle posts. Because the goal is natural rather than elicited speech, the fieldworker often spends days with a community before pressing record. The conversations that follow are usually unstructured: elders tell stories about droughts, weddings, and migrations, drawing on a vocabulary that younger relatives rarely hear any more.

Ethical protocols matter as much as microphone quality. Participants must understand that their words will be transcribed, anonymised where needed, and shared with future researchers. In Botswana this is usually handled through community elders acting as gatekeepers, a familiar pattern of distributed responsibility that parallels how Bogosi chieftainship still operates today.

Once the recordings are secure, the team creates detailed metadata for each session: speaker pseudonyms, ages, regions, topics, and recording conditions. Without that scaffolding a corpus is just a pile of sound files. With it, the same material becomes a searchable record of how Setswana was actually used in a particular time and place.

Transcription Struggles and Annotation Choices

Setswana is a tone language with a complex nominal class system, and those features complicate the move from audio to text. Tone, in particular, does not survive standard orthographic transcription. Some teams mark tonal variation with diacritics; others leave the text in conventional spelling and capture prosody in a separate annotation tier using tools like ELAN or Praat.

Idiomatic phrases add another difficulty. An elder may say "go jela mogopolo" — literally "eat the mind" — to describe deep reflection. A literal transcription that misses the idiom becomes useless to anyone searching the conceptual cluster around thoughtfulness. Annotators therefore mark phrase type, register, and cultural domain alongside the orthographic line.

Code-switching is frequent too. Many elders move between Setswana, English, and occasionally Afrikaans within a single story. Deciding how to tag such switches — and whether to keep each language in its own line — affects every subsequent count and concordance. There is no universally correct answer, only a documented one that the corpus team must explain in its accompanying metadata.

Words for Time, Words for Weather, Words for Kinship

Once a corpus is searchable, certain lexical families stand out. Terms for time are a recurring case in point. The Setswana day is partitioned into named segments that map light, activity, and ritual rather than clock hours, as the survey of Setswana words for time from moso morning to bosigo night shows in detail. In an elder corpus these markers appear constantly, anchoring narratives to particular moments and revealing how earlier generations organised their lives around rhythms the modern working week has largely erased.

Vocabulary for weather is similarly dense. Older speakers distinguish a dozen kinds of rain, several cloud formations, and nuanced wind directions tied to specific agricultural decisions. Corpus queries on rain terms quickly show that certain expressions survive only in the speech of those born before the 1960s. A concordance line that reads as flat text hides an entire weather knowledge system; the corpus surfaces it.

Kinship terms form a third cluster of special interest. Setswana classifies relatives with a precision that English cannot match, and elders use those terms as anchors for memory: a story set in the 1950s often begins by locating the speaker relative to someone else. Building lexical networks around these terms reveals a cognitive map in which relationships, not dates, organise recollection. Such patterns matter for dictionary-makers deciding which senses merit separate entries.

Cultural Frames Hidden in Narrative Structure

Beyond individual words, corpora reveal how elders structure stories, when they pause, and which topics trigger laughter or silence. Recurring opening phrases, fixed transitions, and conventional closings can be catalogued across dozens of recordings. These patterns are themselves cultural data, showing how authority, deference, and shared memory are encoded in the act of speaking.

Several stories collected in such corpora also touch on chieftainship, land allocation, and customary courts — matters at the heart of Botswanan public life. A transcription that turns suddenly to a particular lineage or ward can sound opaque to a modern reader; situating such passages in their historical context is part of the corpus team's interpretive labour.

There is also a question of power: whose voice ends up in the corpus, and whose does not. Fieldworkers who record in only one village may capture a regional accent as if it were the whole language. Researchers who interview only men will miss the lexicon of women's gathering, childbirth, and ritual. Reflective corpus builders therefore publish demographic profiles of their speakers and openly acknowledge the gaps. A corpus is honest about its own silences; treating it as a complete record is the first mistake a user can make.

Comparing Methods Across Continents

Setswana elder-speech research runs alongside similar efforts in Australia, where linguists at the Australian National University in Canberra and the University of Melbourne have built corpora of Aboriginal languages such as Yawuru, Warlpiri, and Kaurna. In Sydney, dictionary makers behind Macquarie have wrestled with similar questions of how to record and represent the speech of older generations. Both continents face parallel choices: how to record, how to transcribe tone or vowel quality, how to share data with descendant communities, and how to weight the voices of elders who may be the final fluent speakers of a particular variety. The methodological literature from Australian projects is a useful resource for anyone designing a Setswana corpus, and the reverse holds as well.

Translators of spoken material in other media face versions of the same problem. Subtitlers, for instance, must compress and reshape an utterance so it fits a viewer who reads rather than hears; the compromises they make, as in anime subtitle adaptation, mirror the choices a corpus annotator makes when marking an idiomatic phrase. None of these fields has a final solution, but the conversation between them is productive.

Each tradition also reminds the others that the goal is not to freeze a language but to keep it usable for the people who still speak it. A corpus is a foundation, not a destination. Used well, it can support new dictionaries, classroom materials, and translation tools for years to come, all grounded in the actual rhythm of speech rather than the abstractions of grammarians.

A spoken corpus is, in the end, an act of listening made permanent. It records the words elderly Setswana speakers choose, the way they arrange them, the rhythm with which they pace a story, and the audience they imagine as they speak. Any future researcher who opens the files will inherit a careful piece of cultural memory — and, with it, the responsibility to handle those voices with the same respect that was shown when the recording began.