A voice assistant should show you when it fails to understand an English-to-Twi switch, fails to complete the request, or cannot proceed. Nkomo makes errors visible, so you are not left guessing whether it heard you, failed, or sent anything to the cloud.
When the reply never comes
In 1999, NASA lost contact with the Mars Climate Orbiter as it approached Mars. The spacecraft had entered a critical moment of its mission, and the team did not yet know why communication had stopped. NASA’s Mars Climate Orbiter Mishap Investigation Board later found a mismatch between metric units used by NASA and pound-seconds used in a Lockheed Martin software component. The navigation data did not line up, and the spacecraft was lost.
The lesson was painfully practical: when two parts of a system interpret an input differently, silence hides the real problem. A clear signal gives people a chance to correct it.
A mixed-language voice request can create a much smaller version of that uncertainty. You say, “Please remind me to call Auntie, na ka kyerɛ me sɛn na mɛka no,” then the screen pauses. Did the assistant catch the English part and miss the Twi? Did your connection fail? Did it stop listening before you finished? Did it try to process the request elsewhere?
Those are different problems. They deserve different, visible answers.
Code-switching needs a clear recovery path
Ghanaian speech moves naturally between English and Twi. The switch can happen halfway through a sentence, especially when the right phrase arrives first in one language. Being asked to separate your speech into neat language turns changes the way you speak to fit the tool.
Nkomo is built for typed and spoken Twi and Ghanaian English conversation, including the mix people actually use. Still, no voice system gets every utterance right. A name may be unclear. A pause may cut a thought short. Background sound may get in the way.
When that happens, the useful response is not a blank screen or a confident reply to the wrong thing. Nkomo shows errors instead of swallowing them. You can see that something went wrong, rather than spending the next minute repeating yourself because you cannot tell what happened.
That visibility matters most when the request carries a consequence. A payment name, a message for family, a short instruction before you leave the house. In Kwaku’s missed payment-name moment, the detail that needs correction is the whole point. Clear feedback gives you a place to start correcting it.
Knowing what reached the cloud
An error message answers one part of the question. Privacy controls answer another.
Nkomo gives you explicit cloud-consent choices: never, ask each time, or this session. You choose before cloud use rather than trying to infer it from a missing reply. That is especially useful when you are speaking a mixed-language thought aloud and decide, midway through, that it should stay on your device.
Your conversation history is also under your control. Turn it off and Nkomo purges it immediately. You can export your data in one tap or delete your account in one tap.
These controls do not promise that every request will succeed. They make the state of the request clearer: what you chose to allow, what stayed in history, and when the assistant could not complete an action.
A small error can save a larger misunderstanding
The Mars Climate Orbiter failure was not caused by a lack of data. It came from data that was interpreted differently without the mismatch being caught in time. The board chaired by Arthur G. Stephenson documented how a visible check between systems could have exposed the problem before the mission reached Mars.
Your voice request does not need a dramatic analogy to deserve the same principle. If an assistant misunderstands “call Auntie” after a Twi switch, show the failure. If it cannot continue, say so. If cloud processing needs consent, let the person speaking decide.
The next time a voice assistant goes quiet after you code-switch, pause before repeating the whole sentence. Check what the app is telling you, correct the part that needs correction, and choose the privacy setting that fits that conversation.
Comments
No comments yet.