Digital Humanities and AI in the Social Sciences

Situated practice: digitisation agreements as living documents, ontologies open to revision
Sovereignty: design grounded in access and governance; ethics revisited throughout a project
Shared problems: consent, standardisation, ill-fitting metadata, hidden labour, language bias, maintenance, all more visible in African contexts
Reading as data: 217 works annotated, mapped into 145 concepts and 1,219 links

Journal of the Digital Humanities Association of Southern Africa (JDHASA)

Articles drawn from the Stellenbosch papers

Slide 6

Datafication

Rendering social life, objects and records into machine-readable data

van Dijck 2014: Datafication as a paradigm shift in knowledge production, and the belief in data it fosters
Mejias & Couldry 2019: The quantification of social life, whose appropriation they call data colonialism
Drucker 2011: Data are capta: taken, not given
Gitelman (ed.) 2013: “Raw data” is an oxymoron

Question for today: what is decided, and by whom, when a source becomes data?

Slide 7

One object, two forms

A memorial in Hasselt, Belgium, and its record in the Global Heshmat code

Hassan Heshmat's memorial wall in Hasselt, Belgium, photographed in 1970 — Hasselt memorial, 1970 (historical photo, Global Heshmat)

// artworks/018-Heilig Kruiskerk.ts
const artwork: Artwork = {
  id: 18,
  slug: "heilig-kruiskerk",
  name: 'Heilig Kruiskerk',
  lat: 50.92058026384286,
  lng: 5.325863864013154,
  country: 'Belgium',
  city: 'Hasselt',
  status: 'located',
  trails: ['east-west', 'faith-remembrance'],
  address: 'Kruisherenlaan 29, 3500 Hasselt',
  desc: 'In memory of 23 Belgian missionaries murdered
    in Buta (Congo) in 1965, Hassan Heshmat was …',
  images: [{src: "Mahnmal_Hasselt_Kopie.jpg",
    caption: "Mural next to the Heilig Kruiskerk"}, … 7 more],
  links: [{label: 'KADOC',
    url: 'https://kadocheritage.be/items/show/11473'}]
};

Slide 8

Global Heshmat

Following Hassan Heshmat's works around the world

Slide 9

Global Heshmat

Egyptian sculptor Hassan Heshmat (1920–2006) · a project by Sonja Hegasy

heshmat.zmo.de Open ↗

Open interactive resource: Global Heshmat: the map of Hassan Heshmat's works

29works located

13still to be found

160photographs

31people profiled

46 entries, each with its own page: story, photographs, sources, a citable address

Missing works keep their entries, with a public call to report them.

Slide 10

What the mapping showed, and what it took

31 profiles from Sonja Hegasy's research, and the decisions behind the site

What the mapping showed An unexpected cluster: Heshmat's ties to the Armenian community in Cairo.

The Global Heshmat page for Shant (Chant) Avetisyan

One file per work

Each work and each person is a typed data file: adding one is a one-file edit

Statuses

Located, to be found, relocated: a work's history becomes a field and a line on the map

Names and scripts

Search ignores Arabic vowel marks and accents, but each name still takes one chosen spelling

After the project

Who maintains the record, and who in Egypt can correct it?

Slide 11

Datafication is where the work is

Most of the time goes into turning sources into data; analysis comes after

Dataficationfind · scan · transcribe · correct · describe · link

Analysisvisualise · count · compare

AI speeds up the first part by an order of magnitude

Text extraction

Manuscripts, typescripts and print, in several scripts

Transcription

Interviews, lectures and sermons, with speakers told apart

Entities

People, places and organisations, ready to link

Its own problems: errors that read well, bias, cost, and extraction (whose data trains the models, and who profits?)

Slide 12

Kalamu

A desktop app I am developing: the same work, without writing code

Read scanned and handwritten documents
Transcribe interviews, in the cloud or offline
Analyse texts with fields you define: people, places, themes
Check every result against its source; corrections never overwrite the AI's text
Export with a record of model, prompt and settings

github.com/fmadore/Kalamu
Free, open source
Windows and macOS

Slide 13

Albani Zaria's sermons

From DVD to word counts: Hausa sermons in the Islam West Africa Collection (IWAC)

Slide 14

Who was Albani Zaria?

Muhammad Auwal Adam Albani Zaria (1960–2014)

Born: Zaria, northern Nigeria, 27 September 1960
Educated: Barewa College, Zaria
Role: Salafi scholar and Qur'anic exegete; taught at Daarul Hadeethis Salafiyyah, Zaria
Languages: Taught in Hausa, quoting Arabic
Died: Assassinated in Zaria, 1 February 2014

The collection includes the lessons he gave on the day of his death.

islam.zmo.de Open ↗

Open interactive resource: The Islam West Africa Collection home page

Slide 15

The collection

45 recordings, c. 1999–2014 · 42 transcribed so far: 145 h, 1,089,269 words

Recordings by date and length: each dated recording placed by date and running time

Running time (hours) of the 15 dated recordings; the other 30 carry no date. Catalogued for IWAC by Vincent Favier and Aleksei Akseshin; 3 recordings not yet transcribed.

Slide 16

The pipeline

Five stages turn the recordings into figures and passages

How the Albani Zaria sermons become figures: five stages from the recordings to figures and passages

Slide 17

Every step is a decision

A word list is a classification scheme

Spelling

ɓ/b, ɗ/d, ƙ/k, ƴ/y, q/k, double letters and apostrophes count as one word

Common words

76 published Hausa stop words (Bashir, Bichi & Adamu 2024), plus this speaker's fillers ainihin, wato

Formulae

sallallahu alaihi wa sallam, in sha Allah: counted as whole phrases, outside the word lists

Loanwords

addini, shari'a, ilimi: Arabic words that count as Hausa

Meaning

sarki is not counted where it means God (1,184 of 2,094); ƙasa, “country” or “to fail”, stays out of totals

Speakers

Voices are told apart only within each 20-minute segment; the 35 dialogue segments can be excluded

Slide 18

The places named

Each circle a place, its area proportional to how often the sermons name it

A map of the places the sermons name, each circle sized by how often it is named

Slide 19

The people named

Seven groups of people, the most named first in each

The people the sermons name, grouped by who they are, each with its count and the recordings it appears in

Most named, all groups

Al-Bukhari 502
Usman dan Fodio 397
Ja'afar Mahmud Adam 279
Abubakar Gumi 252
Isa (Jesus) 188
Abu Hurayra 176

Al-Bukhari in 36 of 42 recordings; Usman dan Fodio in 25.

Slide 20

Youth: who is being talked about?

Occurrences of the youth word list across the 42 analysed transcripts

The youth word list and its counts, by term

Yara (children) is the largest youth term: much of the theme concerns children and families.

Makaranta (school, education theme) occurs 996 times, in 40 of 42 recordings.

Almajirai, Qur'anic pupils: 164 times in 30 recordings.

Slide 21

Politics: the vocabulary of government

Occurrences of the political word list across the 42 analysed transcripts

The political word list and its counts, by term

sarki (emir, king): 2,094 occurrences, but 1,184 left out because they refer to God, as in Allah Maɗaukakin Sarki

Three in four uses of demokuraɗiyya (201 of 263) come from a single lecture, Demokuradiyyah (9 h 33).

Slide 22

Where youth meets politics

A meeting: a youth term and a political term within ten words, same speaker

Which terms meet: how often each youth term meets each political term

Most frequent pairs: yara × sarki (13), yara × jihadi (12), 'ya'ya × sarki (11), almajirai × sarki (11)

89 segments contain at least one meeting; the exhibit ranks them and links each to its 20 minutes of audio.

The count points to the 20 minutes worth hearing; the analysis comes from listening to them.

Slide 23

The sermons as a score

Each row a recording, each cell a 20-minute segment: click one to listen to it

Slide 24

Read the passages

On the exhibit page, every count leads back to the words it comes from

1Pick a term

or search the lines: 1,814 for the youth list

2Read it in context

the term highlighted in its sentence

3Go to the recording

each line links to its 20-minute segment

4Public text only

drawn from transcripts already public on IWAC

Slide 25

Discussion

Methodological, ethical and epistemological stakes of AI and digital humanities tools

Slide 26

The Global South and data

Digitisation and AI can reproduce older asymmetries

Who digitises

Mostly Western-funded: “digital imperialism” (Breckenridge), a “digital saviour complex” (Shringarpure)

What gets digitised

Selection deepens archival silences; keyword search sidelines what is not online

Who has access

The digital divide: unequal internet access remains a major obstacle (Barringer & Wallace 2014)

Data colonialism

Life turned into raw material for profit, as land once was (Couldry & Mejias 2019, 2024)

Data extractivism

African texts, voices and annotation labour feed models whose value flows north

Data flaring

The reverse: African-language data never collected or kept (Adebara 2025)

One answer, in IWAC: FAIR and CARE principles; catalogue where custodians prefer; credit rights holders.

Slide 27

The dialectics of datafication

Which datafication helps, which damages, and which endangers?

The counter-position

Some projects try to make interlocutors impossible to datafy, so that they do not end up in prison: dispersing data points to prevent mapping.

Datafication varies by region (authoritarian or not), by discipline and by goal.

Historians, already surrounded by texts, tend to see DH as a treasure
Anthropologists are often hostile: everyday experience is not data
Fieldwork is already extraction; AI changes speed and scale
What would consensual, qualitative practice look like?

Slide 28

Questions for the discussion

1Method

Who checks the word lists, and in which language?

What does a 20-minute segment cut through?

Can error rates be measured for Hausa–Arabic speech?

2Ethics

Whose voice becomes data, and who decides?

Should such recordings go through commercial AI services?

When should material stay out of the data altogether?

3Epistemology

Is a count evidence, or only a pointer?

What does building one's own categories teach that automatic labelling skips?

Fieldwork, archives, AI: is it all extraction?

Slide 29

Two more projects to explore

ZMO Visualisations

ZMO as data: 1,985 publications by 671 authors, 1994–2026

zmo-berlin.github.io/visualisations Open ↗

Open interactive resource: ZMO Visualisations: the word explorer

zmo-berlin.github.io/visualisations

Genocide at the Security Council

How is the word used in 167,600 UN Security Council speeches, 1946–2024?

fmadore.github.io/genocide-at-the-security-council Open ↗

Open interactive resource: Genocide at the Security Council: the dashboard

QR code to the Genocide at the Security Council dashboard

fmadore.github.io/genocide-at-the-security-council