Building a Frequency Dictionary of Setswana from Radio Transcripts

A frequency dictionary shows which words occur most often in a language sample and how speakers use them in context. For Setswana, radio transcripts offer a valuable record of spoken language: interviews, news bulletins, phone-ins, public announcements, cultural programmes and everyday conversation. The resulting resource can support learners, translators, teachers, broadcasters and researchers interested in the structure and development of Setswana.

Radio is especially useful because it captures language as it is produced for real audiences. A carefully designed corpus can reveal common vocabulary, regional variation, borrowed terms, recurring phrases and differences between formal broadcasting and spontaneous speech. For readers in Australia, the project also provides a practical model for documenting African languages in multilingual settings such as Sydney, Melbourne and Brisbane.

Defining The Corpus

The first task is to decide what the corpus should represent. A useful collection might include national news, rural affairs, sports, religious programmes, youth discussions, call-in segments and recorded advertisements. Each programme should be accompanied by metadata: date, station, programme type, speaker role, location, subject, language variety and level of formality.

Sampling needs to be balanced rather than simply large. If the corpus contains many political broadcasts but few conversations, its word frequencies will reflect public affairs vocabulary more than everyday Setswana. A stratified sample could divide recordings by genre, duration and broadcast period. Repeated episodes should be limited unless the aim is to study formulaic language or the circulation of particular announcements.

The Australian broadcasting environment offers useful points of comparison. ABC and SBS programmes use different editorial styles, while community radio often gives more space to local speech and multilingual audiences. A Setswana project can apply the same principle by treating radio genres as distinct linguistic environments rather than blending every transcript into one undifferentiated dataset.

Transcribing Spoken Setswana

Automatic speech recognition can speed up transcription, yet it should not be treated as an impartial solution. It may struggle with Setswana pronunciation, code-switching, overlapping speech, names, place names and background noise. Human review is essential, particularly where a single misheard vowel or consonant can alter a word’s identity.

A transcription policy should specify how to represent hesitations, repetitions, interrupted words, laughter and unclear speech. Researchers must also decide whether to retain discourse markers and grammatical particles. Removing these small items would produce a cleaner-looking text, but it would distort the frequency profile of spoken language. The objective is a consistent representation, not an artificial version of polished written Setswana.

Transcribers should distinguish between the spoken form and the editorial form used for dictionary searches. A broadcaster may pronounce a borrowed English term in a Setswana sentence, while a presenter may shift briefly into English or another Bantu language. Such instances should be tagged rather than silently corrected. This is particularly relevant for audiences accustomed to linguistic movement between Australian English, heritage languages and Indigenous languages in everyday communication.

Normalising Words And Lemmas

Frequency counts depend on the unit being counted. A token is each occurrence in the transcript, while a type is a distinct written form. A dictionary usually groups related forms under a lemma, but Setswana morphology makes this process demanding. Prefixes, agreement markers, derivational extensions and inflectional patterns can create many forms that share a lexical base.

A practical workflow begins with tokenisation, spelling checks and part-of-speech tagging. The researcher can then create lemma mappings, recording the evidence for each decision. Ambiguous forms should remain visible in the data, with notes explaining whether they function as nouns, verbs, particles or names in a particular context. A frequency dictionary should never conceal uncertainty behind a deceptively precise total.

Etymology can help explain relationships, though it should not replace distributional evidence. An entry about Setlhopha in Bantu languages illustrates how a word connected with “group” or “team” can carry historical and comparative information beyond its immediate translation. In the dictionary, such information would complement frequency, collocation and usage examples.

Reading Frequencies In Context

A word’s rank is only the beginning of its description. High-frequency grammatical items may appear in almost every programme, while a lower-frequency noun may be central to a particular domain such as agriculture, health or sport. Each entry should therefore include concordance lines showing the word in its surrounding context.

Collocations are equally important. A word may regularly occur beside particular verbs, adjectives or particles, revealing meanings that a bilingual gloss cannot fully capture. Comparing collocations across news, interviews and phone-ins can show whether an expression belongs to formal public language, conversational speech or a specialised register.

Some expressions also carry pragmatic meanings. Apologies, admissions and public corrections may use conventional wording whose force depends on the situation. A discussion of mea culpa in apologetic language provides a useful comparative reminder: dictionary entries should explain how an expression works socially, not merely list an equivalent in another language.

Managing Contact And Translation

Radio language often displays contact between Setswana, English, Afrikaans and other languages. Borrowed vocabulary may be fully integrated, used only in particular professional settings or inserted for stylistic effect. A corpus should record the source language where this can be established, while avoiding the assumption that every unfamiliar form is a loanword.

Translation decisions also affect the final dictionary. Literal glosses can obscure idioms, cultural references and grammatical distinctions. Work on translating tanka poetry demonstrates a broader principle relevant here: translation must preserve function and form as far as the target context permits. A Setswana frequency dictionary may therefore need explanatory notes, usage labels and example translations rather than one-word equivalents.

The Australian market has a practical interest in this work. Universities, libraries, schools, migration services and language technology companies increasingly need resources for languages used by African communities. A carefully annotated Setswana dataset could support subtitling, interpreter training and educational materials for learners in Perth, Adelaide or regional centres, while remaining grounded in the language communities represented by the recordings.

Making The Resource Reproducible

A dictionary built from broadcasts should preserve a clear chain of evidence. Researchers need to record the source files, permissions, transcription versions, annotation standards, software, exclusion rules and date of access. Version control allows later users to understand why a frequency count changed after a spelling correction or a revised lemma decision.

Privacy and consent require particular care in call-in programmes. A public broadcast is not automatically free of ethical obligations. Personal names, telephone numbers, medical information and vulnerable speakers may need to be anonymised or withheld. Access conditions can distinguish between material suitable for public search, restricted research and internal verification.

The project can be aligned with wider linguistic documentation priorities through the author’s research interests, especially work involving lexicography, corpus linguistics, etymology and language preservation. These fields meet in the frequency dictionary: corpus evidence supplies the examples, lexicography shapes the entries, and historical inquiry explains selected forms without confusing history with present-day usage.

Designing The Published Dictionary

A useful entry might include the lemma, pronunciation, part of speech, frequency per million words, rank, inflected forms, common collocations, semantic domain, register label and several authentic examples. A digital edition could allow users to filter results by programme type, year, region or speaker role. A printed edition would need careful selection so that the most useful information remains accessible.

Frequency should be normalised when comparing sections of unequal length. Reporting raw counts alongside occurrences per million words prevents a long news series from appearing more linguistically important simply because it contributes more text. Measures such as dispersion can also show whether a word is evenly distributed or concentrated in a small number of programmes.

Quality assurance should involve Setswana speakers, teachers, broadcasters and independent reviewers. Their feedback can identify unnatural glosses, missing senses and misleading translations. The resource could eventually serve both specialist and public users, with a technical dataset for researchers and a readable learner’s version for classrooms and community organisations. Professional language work connected with translation and language services can help bridge these audiences without reducing the dictionary to a commercial product.

A successful project would make spoken Setswana more visible in a form that can be searched, taught and compared. Its value would lie in the relationship between numbers and voices: frequency ranks show patterns, while transcripts preserve the contexts in which speakers make meaning. The next practical step is to assemble a pilot corpus of ten hours of permission-cleared broadcasts, transcribe it consistently and publish a reviewed list of its first one thousand lemmas.