Nigeria, Sept. 25 -- A former colleague walked into my office two months ago, and our conversation was barely 20 minutes on before he jokingly accused me of not including his state of origin's popular languages (Ibibio and Eket) among the languages our data collection outfit, Goloka, recently collected and open-sourced for Meta to use in their ongoing multilingual language model projects.

A similar conversation ensued in June at a meeting where I had noted that we would add Fulfulde to the first five languages we planned on collecting. A mentor who was moderating the conversation jokingly remarked that I wasn't politically correct by picking Fulfulde over Kanuri, which I know was his own language, and I would be paying a fine for that. ...