Choosing a language before speaking adds a burden that bilingual speakers never agreed to carry. A useful voice assistant should follow the conversation as it moves between Twi and Ghanaian English, instead of making the speaker reorganize a natural thought for the software.
Auntie has already pressed record. The greeting comes in English, the family matter arrives in Twi, and the date, place name, or instruction slips back into English because that is how the sentence formed in her head. She did not stop to label each part. She spoke.
Then the assistant asks her to choose a language.
Now the voice note has become a settings problem.
When the system makes the human build the adapter
In April 1970, the crew of Apollo 13 faced a dangerous rise in carbon dioxide while sheltering in the lunar module. The spacecraft had lithium hydroxide canisters that could help remove the gas, but there was a physical mismatch: square canisters from the command module had to work with round openings in the lunar module’s system.
NASA engineer Ed Smylie and a team on the ground had to devise an adapter using materials available aboard the spacecraft. NASA documents the episode in its history of Apollo 13, including the improvised device built from items such as plastic bags, cardboard, tape, and a hose.
The crew could not change the equipment they had. The system had to accommodate the mismatch, quickly and clearly, while the outcome remained uncertain.
Everyday bilingual conversation carries the same design lesson at far lower stakes. When a person naturally mixes Twi and English, requiring one language per turn pushes the adaptation work onto the speaker. The person must pause, translate, restart, or flatten the thought into whichever language the interface currently accepts.
That friction belongs in the product, not in Auntie’s mouth.
A language selector interrupts more than speech
Code-switching can carry meaning, emphasis, familiarity, and relationship. English may hold the appointment time. Twi may carry the warning, affection, frustration, or part that must land properly. A sentence can move between both without feeling divided to the person saying it.
A language selector treats that sentence as two separate technical inputs. The speaker experiences one thought.
That difference matters in ordinary moments. Someone in Kumasi may be sending instructions to a relative in Accra. Someone in the diaspora may be drafting a reply to a parent while walking home. They should not have to decide whether the next sentence is “English enough” or “Twi enough” before saying it.
The interruption also changes what people say. A speaker may replace the precise Twi phrase with a weaker English approximation because switching modes feels tedious. They may shorten the message, repeat themselves, or abandon voice altogether.
As another look at bilingual thought explains, the selector does more than add a tap. It asks the user to perform language classification before communication.
Let the conversation move the way people move
Nkomo accepts typed or spoken conversation in natural Twi, Ghanaian English, or the mix people actually use. In voice mode, you hold to speak, hear the reply, and interrupt naturally when you need to correct or add something.
The benefit is simple: you can begin with the thought, not the setting.
Privacy choices remain explicit. You can choose never to send something to the cloud, ask each time, or allow it for the current session. History stays on the device under your control. Turn history off and it is purged immediately. Data export and account deletion each take one tap.
Those controls matter because natural conversation can become personal quickly. A harmless question can turn into family news, health concerns, work details, or a message you are still deciding whether to send. The interface should tell you what may reach the cloud and let you choose, without vague assurances.
It should also speak plainly when something goes wrong. Nkomo shows errors instead of swallowing them. If a reply fails, you should know. Silence should never leave you wondering whether the problem was your accent, your Twi, the English word in the middle, or the connection.
Design around the speaker’s reality
The Apollo 13 adapter is remembered because engineers refused to pretend the mismatch belonged to the astronauts. Ed Smylie’s team worked with the equipment already aboard and found a way to make incompatible parts function together.
Voice products face a related choice. They can demand that people reshape speech around rigid categories, or they can accept that real conversation crosses those categories before anyone reaches for a language menu.
For a bilingual speaker, the practical test is easy. Start speaking the way you would to someone who knows you. Change language when the meaning calls for it. Add the English noun that comes first. Return to Twi without announcing the switch.
Auntie has already pressed record. The assistant’s job begins there.
Comments
No comments yet.