A speaker should not have to choose one language and stay inside it for an assistant’s convenience. Natural conversation in Ghana often moves between Twi and English within the same thought, so a useful assistant should follow that switch without asking the speaker to start again.
At 6:47 on a wet evening in Kumasi, Auntie Adwoa is standing beside a provisions counter, holding her phone close while a taxi waits outside. She begins a voice note in Twi: “Mepa wo kyɛw, sɛ wokɔ hɔ a…” Then the detail she needs arrives in English: “Ask whether the parcel can still go tonight.” She returns to Twi without pausing.
Her nephew needs the message before he leaves. If the assistant drops the English sentence, mistranslates the instruction or treats the switch as a new request, he may set off without the parcel. The shop is preparing to close. Adwoa has no time to reconstruct a sentence she already said clearly.
This scene is an invented composite, but the language pattern is ordinary. Auntie never signed a contract promising to speak Twi from beginning to end.
Code-switching carries meaning
A language switch can do real work inside a sentence. English may supply the workplace term, the school phrase or the exact wording someone used. Twi may carry the relationship, urgency, humour or softness around it.
Consider Adwoa’s message. “Ask whether the parcel can still go tonight” is not stray English waiting to be translated. It is the instruction at the centre of a Twi conversation. Separating the languages can break the thought even when every individual word is recognised.
The switch can also shape tone. A person might explain a family matter in Twi, repeat an official phrase in English, then return to Twi to say what they actually think. Flatten that movement and the response may sound technically related yet socially distant.
That is why code-switching support involves more than placing two language options in a menu. The assistant has to treat the whole turn as one piece of communication. A May 2026 overview of Twi voice tools describes natural-language understanding as a major unresolved part of the stack, alongside demand for assistants that understand ordinary Twi speech rather than stopping at translation or narrow commands.
The useful question is simple: can I speak the way I already speak?
One thought should remain one thought
Nkomo supports typed and spoken conversation in natural Twi and Ghanaian English, including the mix people use in daily life. In voice mode, you hold to speak, hear the reply and interrupt naturally when you need to correct or add something.
For Adwoa, the turn arrives when she keeps the message intact. She says it once, in her own rhythm, and hears a concise reply that follows the parcel instruction across both languages. With the taxi still waiting, she can check the wording and send the message to her nephew before he leaves.
She does not have to translate herself first. She does not have to decide which half of the sentence counts as the “real” language. She can focus on the job she was trying to finish.
The same principle matters in quieter moments too. A student may rehearse an explanation before calling home. Someone in the diaspora may draft a message that sounds too formal in English alone and too imprecise when forced entirely into Twi. [Kofi’s private rehearsal](\/blog\/twi-and-english-voice-notes-how-private-rehearsal-helped-kofi-sound-like-himself-11ad8c69\/) shows how language choice can affect whether a voice note still sounds like the person sending it.
Privacy belongs inside the conversation
Voice carries more than words. It can reveal who is nearby, which name makes you hesitate and what you are worried about. Language freedom loses much of its value when the speaker cannot tell where that recording or text may go.
Nkomo makes the cloud choice explicit: never, ask each time or allow it for this session. “Never” means the content does not reach the cloud, which may limit what Nkomo can do with that request. The app should make that boundary visible rather than create the impression of offline capability.
Conversation history is kept on the device under your control. Turn history off and it is purged immediately. You can also export your data or delete your account with one tap. If something fails, Nkomo shows the error instead of swallowing it and leaving you to wonder whether the message worked.
Before recording a sensitive family matter, choose the cloud setting that matches the conversation. [The three cloud choices](\/blog\/the-three-cloud-choices-that-decide-where-your-mother-s-voice-note-goes-d49bd3a2\/) explain the practical difference.
Let the speaker finish naturally
Adwoa listens once more beside the counter. The Twi opening is still there. The English instruction has not been treated as an interruption. Her final Twi sentence still lands with the warmth she intended.
Then she sends the note.
That small moment offers a useful test for any bilingual assistant. Speak one real thought in the languages you normally use. Change direction halfway through. Add a correction. Include the phrase your family would actually say. Then check whether the assistant follows the meaning, shows you when something fails and gives you a clear choice before personal content reaches the cloud.
Your conversation already has a natural shape. The technology should learn to meet you there.
Comments
No comments yet.