apakose-ezekiel/Yoruba-Oral-Knowledge-Corpus

# Yoruba Oral Knowledge Corpus A structured, primary-source dataset of spoken Yoruba language collected from elder speakers in Lagos, Nigeria. Built for African NLP research, low-resource language model development, and Yoruba linguistic documentation. --- ## Why This Exists Yoruba is spoken by over 40 million people. Existing Yoruba NLP datasets are sparse, web-scraped, and built from written text — which misses the tonal complexity, proverbial depth, and oral register that define how the language actually lives. This corpus is being built from the source: structured interviews with elder speakers who carry vocabulary, expressions, and cultural knowledge that no internet dataset contains. --- ## Methodology Data collection follows a structured oral documentation protocol: - **Speakers** — Elder Yoruba speakers in Lagos, Nigeria. Participants are selected for dialect range, age, and depth of oral knowledge. - **Session format** — Recorded interviews covering open narrative, prompted vocabulary, proverb elicitation, and conversational speech. - **Recording** — Audio captured at minimum 44.1kHz. Video documentation where consent is given. - **Transcription** — Orthographic transcription with tonal marking following Yoruba standard orthography. - **Annotation** — Linguistic annotation using ELAN. Tags include: speaker metadata, topic domain, speech act type, cultural context notes, and confidence rating for tonal marking. - **Ethics** — All participants provide informed verbal and written consent. Speaker anonymisation is applied at the participant's request. No data is published without explicit speaker approval. --- ## Data Structure ``` yoruba-oral-knowledge-corpus/ ├── audio/ # Raw and cleaned audio files (.wav) ├── transcripts/ # Orthographic transcriptions (.txt, .eaf) ├── annotations/ # ELAN annotation files (.eaf) ├── metadata/ # Speaker and session metadata (.csv) ├── dataset-cards/ # Hug …