A voice assistant should follow the meaning of a thought when it moves between office English and Twi. People should not have to translate their concern into one approved language before they can ask for help.
At 6:18 p.m., Kwame was standing beside the trotro stop at Circle with his laptop bag pressed against his side and his phone in his hand. This is an illustrative composite, but the moment will feel familiar to anyone whose workday follows them home.
He had started the question at his desk: “Can you help me write a short message to my manager explaining I need to leave early tomorrow?”
Then his sister called.
Their mother had mentioned a hospital appointment, and Kwame’s careful office wording suddenly felt too distant from the real issue. He held the voice button again and continued in Twi: “Mepɛ sɛ metumi ka sɛ me maame yare no nti, ɛhia sɛ me kɔka ne ho. But I don’t want it to sound like I’m making excuses.”
The trotro was already filling. If he sent the wrong message, he could sound careless at work. If he waited until he got home, his manager might see it too late. The concern had changed language because the concern had become personal.
A real thought rarely stays in one language
At work, English can carry the first layer of a question: the meeting, the deadline, the message that needs to sound professional. Then the detail that gives the question its weight arrives in Twi, because that is where family, worry, humour, correction, and urgency may come most naturally.
That shift is not a mistake to clean up. It is part of the meaning.
Forcing a person to pause and translate before continuing creates a small but important break. They may leave out the detail that matters, soften it until it sounds unlike them, or stop asking altogether. A conversational assistant built for Twi and Ghanaian English should stay with the thought as it develops, typed or spoken.
Natural code-switching also makes practical tasks easier. A person may ask for a work message in English, explain the family context in Twi, then ask for a final version that sounds respectful without becoming stiff. The language mix carries tone as well as information.
That is why the important instruction can arrive late and in Twi. What Happens When the Important Instruction Arrives in Twi? explores the risk of treating that later sentence as an aside when it is actually the condition everything depends on.
Voice matters when both hands are busy
Kwame did not need to arrange his concern into a neat paragraph while he was getting home. He needed to say it while keeping an eye on his bag, the queue, and the people moving around him.
Nkomo lets people hold to speak, hear a reply, and interrupt naturally when a better detail comes to mind. That matters in the middle of ordinary life. A spoken question is often unfinished on the first attempt. Someone remembers the key point halfway through. Someone wants to correct a word. Someone hears a reply and says, “No, make it softer,” before the thought disappears.
The goal is not to make every conversation sound formal. The goal is to let the person keep speaking in the mix of languages they already use.
The broader research direction supports taking that mix seriously. KasaSpeech contains 54,855 manually transcribed recordings from speakers across Ghana, including natural English and Twi code-switching in everyday scenarios. Everyday speech has its own rhythm. Technology should meet it with attention rather than make it perform for a form.
The personal part deserves a clear cloud choice
The moment Kwame moved from a work message to his mother’s appointment, the question became more sensitive. Before he spoke further, he should be able to decide what reaches the cloud.
Nkomo makes that choice explicit: never, ask each time, or this session. It also gives people control over on-device history. Turn history off, and it is purged immediately. Data export and account deletion are available in one tap.
These controls do not ask people to become privacy experts. They give a plain answer to a plain concern: “Where will this conversation go?”
That answer matters before the first personal sentence, not after it. Voice Assistant Consent: Why the Cloud Choice Must Come Before the First Sentence looks more closely at why timing changes the meaning of consent.
The message that reached his manager
By the time Kwame got home, he had a short message that still sounded like him: direct enough for work, respectful about his mother, and free of details he did not want to share. He read it once at the kitchen table before sending it.
The next time a question begins in English and the real concern surfaces in Twi, let the language change. Say the part that matters. Then choose, before you continue, what you want kept on the device and what you are comfortable sending to the cloud.
Comments
No comments yet.