A natural Twi and English conversation can change language inside one sentence, with no cue to wait for. A useful voice companion has to follow that turn as it happens, because asking people to choose a language first changes how they naturally speak.
At 6:42 p.m., Esi is outside a pharmacy in Kumasi with a crumpled paper list in one hand and her phone in the other. Her aunt has called twice. The last message said, “Please ask whether they have it, na me pɛ sɛ meyɛ sure before I send somebody.”
Esi holds the phone to speak. She starts in English, then shifts as the thought becomes more precise: “Can you help me ask if they have this medicine, but me pɛ sɛ wɔnka sɛ prescription no ho hia anaa?”
The answer matters before the shop closes. If she gets the wording wrong, her aunt may send someone across town for nothing, and the medicine may still be unavailable by morning.
That is the moment a language switch stops being a linguistic curiosity. It becomes part of the request.
The switch carries the meaning
People often switch between Twi and English because one phrase reaches the point faster. English might carry the name of a document, a medicine, a job title, or a detail picked up at school or work. Twi can carry urgency, respect, doubt, humour, family context, or the exact shape of a request.
“Me pɛ sɛ meyɛ sure” does more than add another language to Esi’s sentence. It explains why she is asking. She wants certainty before someone makes a trip. Remove that turn, or treat it as an error, and the response can become technically related but practically useless.
A companion that expects a clean handover misses the rhythm of how people actually talk. There is rarely a pause to announce, “I am switching to Twi now.” The sentence moves because the speaker’s thought moves.
This is especially clear in voice conversations. A person speaking while standing in a queue, walking home, or replying to a family voice note does not want to mentally translate each idea into one approved language. They want to say the thing as it comes.
Google has identified scarce, high-quality speech data as a central barrier to useful African-language voice technology. That is part of why the difference between recognising words and following a living, mixed-language request matters. The hard part is holding the thread when the language changes midway through it.
Waiting for a cue puts the assistant behind
Imagine an assistant that hears Esi’s English opening, settles into English, then hesitates when Twi arrives. It might ask her to repeat herself, flatten the meaning into a generic translation, or answer only the first half of the question.
Each response adds work at the wrong moment.
The speaker now has to repair a sentence that was already clear to another person in the same conversation. She may try again in simpler English, leave out the nuance, or stop using voice altogether. The problem becomes less about the original request and more about managing the assistant.
That cost can be small on an ordinary day. It can also land badly when the details matter. A family message, a landlord question, a health concern, or a payment issue can depend on what the speaker meant by one switched phrase. What Happens When an AI Treats English-Twi Switching as an Error? explores what gets lost when mixed speech is treated as something to correct.
The goal is not for an assistant to perform bilingualism for show. The goal is a conversation where the person does not have to interrupt their own thinking to keep the system comfortable.
Follow the request, then let the speaker stay in control
Nkomo is built for natural Twi and Ghanaian English conversation, typed or spoken. You can hold to speak, interrupt naturally, and hear a reply without arranging every thought into one language per turn.
For Esi, that means she can ask her full question in the mix that came naturally. The conversation can stay focused on what she needs to find out, rather than on proving which language she is using.
Language is only one part of the trust involved in speaking freely. Voice can include family details, money worries, health questions, and things someone would never put in a public message. That is why Nkomo makes cloud consent visible: choose never, ask each time, or this session. History stays on the device under your control, and turning it off purges it immediately. You can export your data or delete your account in one tap.
Those controls do not make the conversation less natural. They make it easier to decide when a natural conversation belongs on your phone and when you are comfortable sending it to the cloud. Ghana AI privacy controls: Why Adwoa Needs a Visible Choice Before She Speaks looks at why that choice should come before a voice message leaves the device.
The reply should keep pace with the person
A few minutes later, Esi has a clearer question ready for the pharmacist and a message for her aunt that does not need three rewrites. She is still outside the shop, still holding the same paper list. The pressure has eased because the conversation followed her sentence where it went.
That is the standard worth holding. Speak in the mix you use with your people. Keep the meaning intact. See clearly what happens to your words after you speak.
Comments
No comments yet.