The pattern is documented. When Ibom NLP created the first dataset for Efik, Ibibio, Anaang, and Oro, it became an IJCNLP 2025 paper precisely because nobody had done it before. The paper's leverage came from the zero-baseline gap, not from algorithmic novelty.

The Izzi cluster is in the same position today. Millions of speakers. No ASR, no TTS, no benchmark. The combination is rare: high speaker count, total absence of resources, and a research community that actively rewards first-mover datasets. GPT-4o scores roughly 59 out of 100 on African-language benchmarks. Whisper's error rate drops by 76% when fine-tuned on 1,838 hours of NaijaVoices speech. For a language with zero current data, any hours collected are publishable. A 125-hour first corpus, like Kallaama in Senegal, is enough.

The documentation funders know this. ELDP pays up to 10,000 euros per grant for a first-ever language documentation. Lacuna Fund's agriculture and health rounds list under-resourced African languages as a standing priority. The verified gap in elder speech data (only 2.4% of new African speech data comes from speakers over 50) makes an oral-history-style Ebonyi collection fundable on two separate grounds at once.