Context-aware translation drafts. Original English remains available.
Russian celebrates Daria and Adémidé’s family heritage; Igbo celebrates Lékan’s maternal roots.
Translation notes & sourcesSuggest a correction ↗CULTURAL AI / YORÙBÁ / LIVING MEMORY
A name spoken with care. A family story retold. A child finding familiar words. These are the moments that make language and heritage matter to me.
My Yorùbá heritage, AI studies and Tapestries work meet in a shared question: how can technology help us pass knowledge on while keeping its meaning, context and human relationships intact?
I want to explore that question with speakers, educators and cultural custodians—listening closely, making room for different accounts and reviewing what we create together.
I want to build on my thesis work and further research to explore how applied AI, low-resource language systems and immersive media can support cultural learning. My ambition is to develop this with researchers, language speakers, cultural custodians, educators and purpose-led organisations. I want to connect the discipline of management consultancy with cultural intelligence: understanding whose knowledge is represented, how decisions are made and what a useful, respectful experience looks like to the people involved.
The starting question is practical: how can we make heritage more accessible while respecting the people who hold it? I bring business analysis, systems thinking, governance and AI research to that question. Cultural authority, consent and interpretation must be worked through with communities.
Stage: The directions below are proposals for future research and collaboration. They do not represent a deployed language model, operating VR archive, booked travel programme or confirmed institutional partnership.
RESEARCH FOUNDATION / THREE CONNECTED STUDIES
My Social Context of AI and Data Science papers explore technology through the people, institutions and cultural responsibilities around it. The studies use a fictional college, Òrúnmìlà College at the University of Ọmọlúàbi (OCUO). They are academic scenarios and proposals, not reports of an institution or archive I have deployed.
The formative study asks why a technically promising Yorùbá VR classroom could fail in use. Its scenario brings together limited connectivity, insufficient local support, maintenance costs and uneven access.
What I take forward: offline content, a screen-based alternative, trained local owners and a realistic support budget belong in the design from the beginning.
The first summative paper examines low-resource Yorùbá NLP: tone, dialect coverage, accountability, explainability and cultural data sovereignty. It asks who can authorise reuse and whose speech becomes the default.
What I take forward: invite speakers into annotation and evaluation; document permitted uses; make errors and limitations understandable in Yorùbá and English.
The second summative paper explores a digital heritage archive joining oral histories, artefacts and immersive scenes. It considers consent, attribution, benefit-sharing, cultural interpretation and the ongoing cost of stewardship.
What I take forward: community review and correction must continue after publication. A persuasive reconstruction still needs a source, a label and an accountable editor.
| Question from my work | Design response to test | Evidence I would seek |
|---|---|---|
| Can it work without reliable internet? | Downloadable, versioned learning packs; local playback | Offline completion, load times and support needs |
| Whose Yorùbá is represented? | Speaker-reviewed examples across agreed dialect groups | Coverage and error patterns for each group |
| Who may reuse a voice or story? | Specific permissions, attribution and access levels | A reviewed record of permitted uses and restrictions |
| Can viewers distinguish evidence from invention? | Visible source and interpretation labels | Whether learners correctly recognise those distinctions |
| Who benefits and maintains it? | Shared decisions, contributor recognition and funded support | Agreed responsibilities, community feedback and recurring costs |
FROM MY MSc / INTO TAPESTRIES
The connection between AI, data and culture extends through the making of Tapestries. I intend to apply what I learned in programming, data engineering, machine learning and responsible AI to the work around each episode: finding sources, tracing claims, preparing language, checking continuity and learning from production decisions.
A beautiful shot is one output. The larger aim is a production process in which I can explain where a scene came from, whose account it represents, what changed during development and why it was accepted. The table below translates my production augmentation plan into practical experiments.
| Skillset | How I intend to use it | What I would check |
|---|---|---|
| Python, SQL & data engineering | Build a searchable asset register; connect sources, scenes, recordings and versions through stable identifiers and file checksums. | Metadata completeness, retrieval time and safe reuse. Preserve governed originals. |
| NLP, embeddings & relationship graphs | Retrieve relevant passages and map people, places and claims. Keep alternative traditions and source authority visible beside a proposed script. | Whether retrieved passages support the claim. Contradictions stay visible for editorial review. |
| Yorùbá language processing | Check names, tone marks, agreed terminology and subtitles against reviewed references; build a reusable correction memory. | Language-review corrections and subtitle mismatches. Competent speakers approve meaning and pronunciation. |
| Generative AI & experiment design | Translate approved scene specifications into versioned prompt packs and compare controlled production variants. | Generations, time and cost per accepted shot. The producer decides what serves the story. |
| Computer vision & deep learning | Explore pretrained image embeddings and detection models to flag possible drift in characters, costume, props and environments. | Human-confirmed defects and false alarms. A similarity score cannot establish historical or cultural accuracy. |
| Speech, sequencing & subtitle alignment | Prepare transcript drafts, timing information and subtitle checks around approved native or competent-speaker recordings. | Timing errors, language corrections and rework. Keep the approved master recording unchanged. |
| Statistics, evaluation & production analytics | Compare acceptance rates, recurring failures, turnaround time and subscription use; use audience feedback to identify questions for later episodes. | Like-for-like shot comparisons and meaningful audience responses. Cultural purpose remains part of the decision. |
| MLOps, testing & cultural governance | Make indexing, metadata and reporting jobs repeatable, with logs, retries, rights records and explicit review stages. | Reliable reruns, complete provenance and accountable release decisions. No automatic resolution of disputed history. |
My proposed starting point is one Episode 2 sequence: assemble its sources, record parallel accounts, prepare a reviewed script, generate and compare shots, then check language, visual continuity and attribution before assembly. Each accepted or rejected result becomes a labelled production example for improving the next attempt.
Reuse suitable MSc code, define metadata and build a small source-and-claim index. Remove confidential or restricted material from any reusable derivatives.
Use structured scene specifications, prompt versions and a handful of representative shots to establish a baseline for time, cost and quality.
Test continuity and subtitle support; record reviewer corrections, unresolved questions and the reasons for accepting or revising a scene.
The supplied v2.1 plan assigns ChatGPT to research and structured preparation, Fawna to default creation and assembly, and Higgsfield to selected cinematic escalations before returning to the release master. These are intended production roles, not a claim that this website runs or connects those services.
For histories remembered differently by living communities, I intend to retain the source tradition, custodian perspective, alternative account and approved narration status with each sensitive claim. Research tools may help locate a difference; relevant custodians and editorial reviewers decide how it is represented. Automated agreement is not a verdict on cultural authority.
The plan’s first 90 days are a proposed sequence for building capability, not a report of completed work. More complex models and cloud services follow only when a small pilot shows a useful role for them.
Follow the Tapestries project → · Explore my AI & data studies →
IFÁ / LIVING MEMORY / A RESEARCH DIRECTION
My spiritual and philosophical interest in Ifá brings a question into my AI work: how can technology help people preserve experience and explore patterns while keeping interpretation accountable to the people who carry the tradition?
I would begin with a small, permissioned collection of published or consented material. Each item would retain its performer or author, language variety, source, translation choices and access conditions. AI could assist transcription, search and comparison; fluent speakers and knowledgeable practitioners would review the results. Different interpretations should remain visible.
Choose one theme, such as character or the consequences of a decision. Link every excerpt to its source, compare the original wording with reviewed translations, and keep commentary separate from the verse. A scene could then explore the question through clearly identified dramatic dialogue. Generated text would never be labelled an authenticated ẹsẹ̀ Ifá.
Useful evaluation would ask whether a reader can find the source, recover the intended meaning and see where interpreters disagree. Tone accuracy, attribution, consent and the ability to correct or withdraw material matter alongside search relevance. This is a proposed research approach, not a deployed Ifá model or automated divination service.
The ethical thread is ìwà: knowledge and capability call for care, humility and responsibility. Binary-like structure is an interesting comparison; the community’s living knowledge supplies the meaning.
01 / LANGUAGE, PLACE & BELONGING
My academic work considered tonal marks, dialect representation, provenance and permission to reuse voices and texts. Future research could test how these affect learning and representation in real settings. “Low-resource” describes the availability of suitable digital resources for a task; it does not describe the richness or value of the language.
I would begin with a small, permissioned collection and agreed tasks: transcription, translation or narrated learning. Human reviewers would check meaning, tonal marks and context. Results would be reported separately for the varieties represented, with disagreement and missing coverage visible. Sentiment analysis would need its own culturally reviewed labels; an English classifier cannot simply be treated as a Yorùbá model.
The papers discuss dialect-specific approaches and model fact sheets. They do not train or benchmark a released Yorùbá model. Choosing and evaluating a model would be part of the next research stage.
02 / AN IMMERSIVE HERITAGE RESEARCH CONCEPT
My preferred first experiment is a proposed 5–7 minute, offline heritage experience for educators, museums and diaspora learners. The working brief imagines Ilé-Ifẹ̀ in the 12th century; that period and each reconstruction would need review against appropriate historical and archaeological sources before production.
Smooth cheeks, carefully formed lips and small attachment holes give this head its distinctive presence.
Collection and accession number not yet confirmed
Source & context ↗
Introduce place and the limits of the reconstruction.
Connect an object with its source and interpretation.
Explore craft and innovation through a reviewed workshop scene.
Ask how people organised exchange, authority and everyday life.
The first version would favour guided Yorùbá and English narration, optional captions, modest interaction and a small set of reviewed questions. “Ask the City” could initially retrieve pre-written answers with sources and say when the pack contains no answer. Open-ended generation would be a separate research decision.
Test downloadable heritage packs, offline playback and a screen-based alternative alongside a headset version.
Begin with consented human narration. Investigate speech tools and local question-answering only where quality and permission support them.
Separate documented evidence, attributed oral tradition and creative reconstruction. Offer correction and withdrawal routes.
PEOPLE BEFORE THE PLATFORM
A linguist, storyteller, educator and developer may have different definitions of a successful experience. I want those differences to shape the brief, the source selection and the evaluation—not appear only at final sign-off.
A second aspiration is a gentle, adult-guided collection of stories, landscapes, music and language: starting with audio, illustrated stories and shared screen experiences. Yorùbá would sit alongside other family languages, with accurate pronunciation and permissioned recordings.
I would explore slow pacing, optional participation and space for conversation. This is a design ambition, not a validated Montessori intervention or a claim of developmental benefit. Headset-based work would be a separate, age-appropriate research question; the proposed family materials do not depend on infant or toddler headset use.
PROPOSED SYSTEM / A CLEAR ROLE FOR EACH LAYER
Texts, images and consented recordings carry origin, language, permissions and restrictions.
Python-supported checks, draft transcription and candidate tags reduce manual work; speakers and editors review meaning.
A versioned pack brings together narration, captions, scenes and approved answers, on screen or in a future headset prototype.
Correction requests feed a change log and new editions. Restrictions remain attached to the source and its derivatives.
Supervised models could help classify carefully labelled material; unsupervised grouping could help curators explore a collection. Speech tools could draft transcripts. None of these determines ownership, historical truth or permission. I would start with simple, reviewable methods and compare their errors before adding complexity.
A future provenance record would include source, contributor preference, dialect or language, evidence type, permitted uses, review date and a correction contact. Private or restricted material would not enter a public learning pack. Withdrawal procedures would also explain the limits of recalling copies already downloaded.
03 / THREE CONNECTED DIRECTIONS
I see three complementary ways to turn research into useful practice, starting with a narrow, jointly defined problem.
Help purpose-led organisations explore AI readiness, simpler workflows and clearer decisions through discovery, process mapping and small, evaluated experiments.
Explore cultural impact reviews, content and curriculum audits, and systems maps that connect values with service journeys, representation and governance.
Develop the Tapestries practice into carefully researched stories, learning resources, playbooks and possible immersive experiences, with rights and attribution considered from the start.
Further possibilities include community learning labs, a future social enterprise and heritage journey design with established travel operators. These would follow demonstrated usefulness, suitable partners and the capacity to deliver responsibly.
04 / A GOVERNANCE MODEL TO TEST
Before using cultural material, establish its origin, the intended purpose and audience, the relevant rights holders and custodians, and the terms they have agreed. Possessing a recording does not establish permission to publish it or train a model with it.
Keep a provenance record with creator, date, place, language variety, edits, sources, access restrictions and review history. Label reconstruction separately from documentary evidence. A named human reviewer needs time, knowledge and authority to correct or stop an action.
Design for the people who will use the work: low-bandwidth alternatives, readable text, accessible media and appropriate offline access. Agree how benefits, attribution, corrections and withdrawal requests will be handled with the community. These responsibilities require continuing judgement; ticking boxes cannot discharge them.
For the wider framework, see UNESCO’s Recommendation on the Ethics of AI.
LANGUAGE / REPRESENTATION / EVIDENCE
The meeting of AI and culture needs more than impressive outputs. Three focused experiments examine tonal integrity, representation and transcription differences. The research and stewardship principles on this page explain why those technical choices matter.
The course’s Òrúnmìlà College / University of Ọmọlúàbi is a fictional teaching setting. Speculative archive and VR examples are proposals to examine, not claims of completed deployments. The tools use transparent calculations and curated prompts; they do not pretend to be fluent Yorùbá AI systems.
Loading the Cultural AI tools…
Explore the full interactive catalogue · Look up an AI or data term
05 / RESEARCH INTO PRACTICE
The aim would be to learn whether a small experience is understandable, useful and maintainable. I would agree an evaluation plan with partners before recruiting participants, and report limitations alongside positive findings.
| Dimension | Proposed measure | A reason to revise |
|---|---|---|
| Historical understanding | Short questions before and after; ability to identify interpretation | Beautiful scenes mistaken for documented fact |
| Language quality | Speaker review of meaning, tone and dialect-specific errors | One variety performs poorly or disappears from the sample |
| Access and comfort | Offline task completion, captions, screen alternative and participant feedback | The experience excludes users or causes discomfort |
| Trust and agency | Contributor review of credit, permissions and correction routes | People cannot understand or challenge how material is used |
| Operational viability | Support time, device performance and cost of updates | A pilot depends on support the host cannot sustain |
Confirm a learning need, partners, sources, rights and a small script. Test the historical scope.
Gate: approved content briefAssemble narration, captions and a simple scene; test offline delivery with adult reviewers.
Gate: a reviewable prototypeReview language, interpretation and accessibility; record findings and decide whether to revise, extend or stop.
Gate: evidence for the next decisionConversations with my old school friend Dr Tokunbo Tijani in Louisiana opened a related question: could African football analytics support grassroots development while making room for education, community and sporting heritage? The proposed Football Intelligence & Development Platform is one possible practical starting point for a wider Africa Futures Observatory idea.
I welcome conversations with language researchers, educators, heritage custodians, museums and community organisations interested in shaping a small research or learning pilot.
READING THE EVIDENCE
The three studies are the foundation of this page. The diagrams and tables are new explanatory adaptations of their arguments; the papers provide no empirical pilot dataset or model performance benchmark. The map gives geographical context, and generated imagery is labelled as concept art.
The papers discuss regulatory frameworks as part of their academic analysis. This page does not repeat a blanket claim that all cultural or educational AI is high-risk: classification depends on the specific use and applicable rules. See the European Commission’s AI Act Annex III and Article 6. UNESCO’s Recommendation on the Ethics of AI provides a reference for cultural diversity and participation.
The public adaptation also avoids presenting illustrative vocabulary, proposed historical dates or emerging technology claims as independently validated results. Voice replication is not part of the proposed first pilot; consented human narration is the starting point. Distributed storage alone would not establish cultural ownership or make withdrawal effective.
ARA / LANGUAGE & EXPERIENCE
Mark Dingemanse’s The Body in Yorùbá investigates body-part vocabulary, grammar and meaning. Terms can extend into spatial relationships: orí can refer to a head or a top; inú to an interior as well as the belly; etí to an ear or an edge. Context determines the reading.
A dictionary equivalent alone cannot decide whether a term is anatomical, spatial or part of an idiom. An AI system needs examples of usage and evaluation by speakers.
The thesis treats the body as a rich semantic domain. It invites careful comparison between languages, rather than assuming that an English category divides the world identically in Yorùbá.
Dingemanse (2006), Leiden thesis, especially §§2.2–2.5 and Appendix I, p. 69. These examples explain language use; they are not anatomical or medical claims.
Return to names and meanings · Language and the history of schooling · Digital representation and its limits.
Core reading for the Yorùbá historical and cultural context discussed on this page. Follow the specific sources alongside individual claims, interpretations and language examples.
Cultural stewardship also involves character: Ọmọlúàbí offers this project an ethical vocabulary for truthful, respectful work.
BIRKBECK / THE CRITICAL FOUNDATION
My Birkbeck learning connects cultural representation with questions of power, institutions and whose knowledge counts. That critical foundation informs the way I approach AI and the development of Tapestries.
Enable JavaScript to choose optional feedback-based suggestions.
YORÙBÁLAND / DEVELOPMENT
The development hub connects Yorùbá language, historical skills, AI and biotechnology with education and industrial questions. Eighteen opportunity areas distinguish present evidence from proposals and technical readiness.
Explore the development hub → · Open both interactive maps → · Sources and review · A civilization’s next chapter. →
These learning tools expose selected representations and comparisons. They cannot authenticate a tradition, infer ancestry or replace language speakers and cultural custodians.
Number of concepts in this guide, not performance or usage measures.
9 concepts shown
On small screens, scroll the table sideways. “Learning context” explains a related idea; it does not mean that method runs on this page. Deep learning is the standard term for “deep ML”.
| Technical term | Technical description / definition | In plain English | How this page uses it |
|---|---|---|---|
| Unicode normalisationAlgorithm | Transforming equivalent character sequences into a consistent Unicode representation.Primary reference ↗ | Help differently encoded letters match without throwing away meaningful accents. | Method guide Tone and underdot preservation matter when representing Yorùbá text; spelling differences may carry meaning. |
| Natural language processing (NLP)AI Capability | Computational methods for representing, analysing or generating human language.Primary reference ↗ | Help a computer work with words, while remembering that context matters. | Learning context Language technology needs context and evaluation by speakers, especially for underrepresented varieties. |
| Word error rate (WER)Evaluation & XAI | Substitutions, deletions and insertions divided by the reference word count.Primary reference ↗ | Compare a transcript against a checked reference; the number alone cannot judge cultural meaning. | Learning context Where transcription comparisons appear, word-level errors are only one quality signal; tone and meaning need review. |
| Rule-based logicAlgorithm | Explicit authored conditions map inputs to outputs without fitting parameters from training examples.Primary reference ↗ | A transparent set of if-this-then-that instructions. | Running here The retained learning activities use authored examples and deterministic choices, not divination, identity prediction or generated historical evidence. |
| Visual encodingVisualisation | Mapping values or categories to marks, position, length, colour and labels.Primary reference ↗ | Make the comparison visible, with a scale and a clear label. | Published analysis Diagrams encode the stated categories or relationships; visual resemblance does not establish historical or scientific equivalence. |
| Coverage, bias and representationResponsible AI | Assess how sampling, labels, missing groups and design choices affect error and applicability.Primary reference ↗ | Ask whose experience the data includes—and whose it misses. | Responsible use Missing language varieties and permissions can make a seemingly polished dataset unrepresentative. |
| Data minimisationResponsible AI | Limit collected, disclosed and retained information to what is needed for the stated purpose.Primary reference ↗ | Use the smallest appropriate amount of information. | Responsible use Cultural access, family privacy and agreed uses matter alongside technical quality. |
| Human oversight and consentResponsible AI | People retain appropriate review authority, permission controls and correction routes for data use and outputs.Primary reference ↗ | Keep someone able to question, correct or stop the result. | Responsible use Sources and knowledgeable people retain interpretive authority; corrections are welcome. |
| Deep learningAI Capability | Machine learning using neural networks with multiple representation-learning layers.Primary reference ↗ | A layered model learns useful representations; more layers do not guarantee a better answer. | Learning context A proposed image or language capability is distinct from an implemented, evaluated model on this page. |
No matching concepts. Choose another category or clear the search.
Olukotun Studios Ltd · Registered in England and Wales · No. 17537910
Registered office: 5 Brayford Square, London, E1 0SG, England.