The Twi Word sɛ That Stops the Waveform, and What Silence Costs Speakers

A voice assistant should continue when a Twi word arrives mid-sentence, preserve the meaning of the whole request, and answer in the same natural mix of Twi and Ghanaian English. Silence at that moment turns an ordinary conversation into a test the speaker did not ask to take.

“Can you remind me sɛ me meeting no starts at 3?” The English words arrive cleanly. Then sɛ lands in the middle, doing the small but essential work of joining the thought. The assistant’s confidence drops. The moving waveform stops. Nothing comes back.

That second is familiar because code-switching is ordinary speech for many people in Ghana and across the diaspora. A conversation can move from English to Twi and back because that is the clearest way to say the thing. The speaker has not changed the subject. They have not made a mistake. They have simply spoken naturally.

A mixed sentence needs one continuous listener

A voice assistant has to hear the sentence before it can help with it. Splitting a request into “English mode” and “Twi mode” asks the speaker to do technical work in the middle of a thought: stop, choose a language, repeat the request, and hope the important detail survives.

That is particularly frustrating with voice. You may be walking through Kumasi, cooking, driving, or calling family overseas. You hold to speak because typing is inconvenient. A stalled assistant leaves you with the task still open and no clear explanation of whether it heard you, misunderstood you, or failed somewhere else.

The problem goes beyond pronunciation. Words such as sɛ, yɛ, or phrases that carry context can shape the request around them. Remove them or treat them as noise, and the assistant can lose the thread. A reminder becomes incomplete. A question becomes unclear. A personal message has to be spoken again.

Nkomo is built for natural Twi and Ghanaian English conversation, typed or spoken. You can speak in the mix you actually use, hold to talk, interrupt naturally, and hear a concise reply. When something goes wrong, it shows the error instead of quietly swallowing it. That matters because an honest error gives you a next move. Silence gives you none.

For a closer look at what changes when an assistant breaks the thread, read What Happens When Your Voice Assistant Loses the Thread Between Twi and English?.

Apollo 13 had the wrong connection at the worst time

In 1970, the Apollo 13 crew faced a carbon-dioxide problem in space after an explosion forced the mission to change course. The lunar module carried square lithium hydroxide canisters. The command module had round ones. The equipment both sides had was useful, but the parts did not fit together.

At NASA’s Manned Spacecraft Center in Houston, engineers worked on an adapter using the limited materials available to the crew. The outcome was uncertain while the crew’s carbon-dioxide levels were rising. The improvised solution was sent up as instructions, and the crew built it. Apollo 13 returned safely to Earth.

NASA’s Apollo 13 history documents the episode, including the incompatible square and round canisters and the ground team’s effort to make the available materials work. The story is memorable because the failure was not a lack of intelligence or effort. It was a mismatch at the connection point, precisely when the system needed to keep working.

A Twi word inside an English request creates a much smaller version of that test. The words belong to one thought, but an assistant that expects only one language at a time may treat the connection as a dead end. The speaker pays for that mismatch by repeating themselves, simplifying how they speak, or giving up on voice altogether.

The better response is clear, useful, and under your control

A capable response begins by staying with the request. If you ask, “Please draft a message to my landlord, na mention sɛ I’ll pay on Friday,” the assistant should understand that the language mix carries both the task and the tone. It should respond to the message request, rather than forcing you to translate your own voice into a format it can accept.

Privacy deserves the same clarity. A natural conversation can include family details, health concerns, work matters, or a voice note you would rather keep close. Nkomo makes the cloud choice visible: never, ask each time, or this session. Its history stays on your device under your control, and turning history off purges it immediately. You can also export your data or delete your account in one tap.

Those controls do not make every conversation risk-free. They make the choice explicit before you speak, which is the standard a private companion should meet. Ghana AI privacy controls: Why Adwoa Needs a Visible Choice Before She Speaks explores why that visible choice matters.

Keep speaking in the language of the thought

Professor Isaac Wiafe has urged investment in indigenous AI platforms that understand Ghanaian dialects, warning that dependence on foreign platforms can weaken local ownership of knowledge. The point reaches beyond a single assistant. When everyday speech repeatedly fails at the same boundary, people learn to edit themselves for the technology.

Apollo 13’s team did not ask the crew to breathe differently because the canister shapes did not match. They worked on the connection. Voice technology should meet the same obligation. The next time Twi arrives midway through an English sentence, the conversation should continue, and the speaker should remain in control of where it goes.