When your parent holds the phone and asks a question in Twi, then naturally adds English mid-sentence, then hears a single, complete reply that understood the whole thing, that moment works because the software stopped forcing a choice she never thought to make. Most voice assistants break here: they demand a language picker before the conversation even starts, which means your parent has to split up how she actually thinks.
In 1991, a team at Apple including Mark Davis faced almost the same problem with text on computers. A document had to declare itself English or Japanese or Arabic before anything was written. If your message needed two scripts, if you wanted Greek letters mixed with English text or English mixed with Hebrew, you were out of luck. The workaround was tedious: save multiple versions of the same document, each in a different encoding, and manually switch between them. The solution wasn't a better language picker. They designed Unicode to let multiple scripts live in the same document at once, so English and Arabic could flow together in one sentence without a mode switch. The system didn't ask the writer to choose first; it let what was written carry the information about which script was which. What made it work was that they abandoned the premise of the forced choice altogether. (The Unicode Standard, maintained continuously since that release, documents the full design and the reasoning behind each decision.)
Your parent code-switches because that's how thought works. A question in Twi lands naturally with an English phrase woven through. A voice assistant that makes her pick a language first is asking her to perform something she doesn't actually do: to decide which language this turn belongs to before she opens her mouth. When she speaks to Nkomo, Twi and English together, the system recognizes it as one continuous thought. The reply comes back in a single, concise response.
Why language pickers interrupt how people actually talk
Code-switching isn't a bug in how people speak. It's how conversation works when you're fluent in two or more languages. Your parent isn't trying to be complicated; she's just thinking, and the English word lands because it's the right word in that moment. A language picker forces a mental interrupt: stop, decide which box this turn goes in, then continue. No actual speaker does that before opening their mouth. They speak, then the language falls out naturally.
Imagine asking your parent a question about the house. She might start in Twi, describing a wall or an appliance, then slip into English for a specific term or brand name that has no good Twi translation. If an app made her choose a language before speaking, she'd have to rehearse the answer first in her head, guess which language would dominate, or mentally rehearse twice. None of those are how humans actually communicate.
This is especially true in Ghana, where English and Twi aren't separate compartments but two strings of the same conversation. You ask a question in Twi, someone answers with English because that's where their thought went, and the meaning flows without hesitation. A system that stops that flow to ask "which language?" doesn't understand how you actually communicate. The Language Selector Nkomo Removed, and the Context You No Longer Lose walks through why that moment of choice, even if it takes just a second, changes everything.
What it means when the system gets it all at once
When your parent speaks naturally to Nkomo, something quiet happens: she doesn't get interrupted. She speaks in whatever mix comes out. Twi, English, both at once, whatever that thought needs. The system listens to the whole thing as one complete idea, not as fragments to be sorted into language bins. The reply comes back as a single spoken answer, concise and complete, in Ghanaian English or Twi or both, depending on what fits.
That's not trivial. Most voice assistants, when they hear code-switching, either ask for a language picker, or they process each language separately and return a confused answer, or they fail silently and show an error. Nkomo does none of those things. It takes what you said, understands it as one continuous turn, and answers in a way that recognizes the mix you used. No picker. No language-by-language processing. Just understanding. Your parent gets a response as natural as the question was.
Clear control over what stays with you and what leaves
Understanding code-switching is only half the story. The other half is trust. Your parent should know exactly what happens to what she says: what stays on the device, what goes to the cloud for processing, and for how long.
Nkomo is explicit about all of this. Your parent can set cloud consent to never send anything, ask her each time, or allow it for this session only. History stays on the device by default, and she can turn it off and purge it immediately. There's no hidden logging, no vague "we analyze speech patterns" buried in a terms page. If something goes wrong, you see the error, not a silent failure. If she wants to export everything or delete her account, both are one tap away.
This matters when you're speaking in your home language, about things close to home. You shouldn't have to guess whether an AI company is keeping what you said or selling access to it. Nkomo shows you exactly what's happening, no assumptions required.
Comments
No comments yet.