AI & automation
Do you speak Wolof?
AI is advancing quickly. But does it understand the languages we actually use every day?
Imagine an app designed to make your life easier.
You can ask it something in English.
French works too.
But when you speak Wolof, it hesitates, misunderstands or answers something completely unrelated.
Yet Wolof may be the language you use most naturally in everyday conversation.
Technology can be impressive and still feel far away from how people actually communicate.
In Senegal, that matters.
The French interface and the Wolof conversation
In an ordinary day, people may move between languages without thinking.
A written message in French.
A phone call in Wolof.
A sentence mixing both.
Sometimes expressions that make immediate sense in Dakar but are difficult to translate literally.
For an AI system, this is not straightforward.
Recognising speech is one task.
Understanding meaning is another.
Responding appropriately in a cultural context is another again.
The data problem
Language models learn from text and, for speech, recordings paired with transcripts.
When a language has less suitable training data, performance can become uneven.
It isn't only about the number of sentences.
Diversity matters: speakers, accents, settings, vocabulary and language mixing.
An assistant trained on formal speech may struggle with fast everyday conversation or code-switching between Wolof and French.
That's why a few successful demonstrations shouldn't be treated as proof of broad reliability.
Research and community work
Initiatives such as Masakhane support African-language NLP research.
Mozilla Common Voice provides a community-based approach to collecting speech data across languages.
Research published in 2026 also explores how different models and data quantities affect speech recognition in lower-resource African-language settings.
These initiatives don't mean every assistant already speaks perfect Wolof.
They show the serious scientific and community work behind useful systems.
What could this help with?
I can imagine practical applications.
Someone receiving public-service information through speech.
A merchant recording an order by voice.
Customer support understanding a question without forcing someone into an unfamiliar language.
Learning tools that help students move between languages.
The goal isn't to put AI everywhere.
It's to remove barriers where the technology can genuinely help.
Who decides what is correct?
A language is more than a word list.
Speakers, linguists, teachers and people familiar with real usage need to be part of the process.
Data contributors deserve respect too: informed consent, clear usage conditions, safeguards and appropriate rights.
Errors must be tested, not hidden.
An awkward translation in casual conversation is annoying.
A wrong medical or administrative answer can be much more serious.
My developer question
When I see a new AI interface, I'm curious about what it can do.
But I also ask: who can speak to it naturally, and who has to make an extra effort to be understood?
That difference matters.
Accessible technology shouldn't require everyone to abandon the way they speak.
I'd love to see more digital products that understand our languages and everyday communication habits.
AI will have made real progress when it isn't only impressive in a demo.
But useful in an ordinary conversation.
