A useful voice assistant should keep English measurements and Twi family instructions in one conversation, then answer the recipe as one whole task. Switching languages at the crucial step should preserve meaning, context, and the person’s intent.
The instruction that changes everything
Consider Afia in her kitchen in Kumasi, phone propped against a tin of tomatoes, trying to cook a stew her aunt has explained over a voice note. Rain is tapping the louvre windows. A pot is already warming on the stove.
The note starts in English: “Use two tomatoes, one onion, and a small piece of ginger.”
Then her aunt’s voice changes naturally: “Sɛ wokyew no wie a, fa gu so kakra. Nanso mma no nnyɛ dodo.”
Afia pauses. The amount matters now. Too much of the next ingredient could change the stew from rich to overpowering, and dinner is meant to be ready before her brother gets home.
She could translate the Twi phrase separately. She could repeat the whole note in English. She could type fragments into different tools and try to rebuild the instruction herself. But a family recipe is rarely delivered in tidy, single-language steps. Measurements may arrive in English. The method, warning, and small adjustment that make it taste right may come in Twi.
The task is to hold all of that together.
A recipe is one conversation, even when the language shifts
Code-switching often carries the meaning that a strict language boundary loses. English may be useful for an ingredient name or a measurement. Twi can carry the family method, the caution, the rhythm of what to look for in the pot.
“Fa gu so kakra” is more than a request to add something. In the context of a recipe, it can signal restraint. The next line, “mma no nnyɛ dodo,” makes the warning clear: do not let it become too much.
A helpful response needs to connect the quantities, sequence, and caution. It might say: “After the mixture has cooked, add only a little of that ingredient. Your aunt is warning you not to overdo it, so start small and taste before adding more.”
That response respects the mixed instruction instead of treating the language change as a break in the request.
This is why forcing English-Twi switching into separate turns can cost the conversation more than convenience. A missed phrase can change the action someone takes. In a recipe, that may mean a pot that tastes wrong. In a family voice note, it may mean the person listening has to call back, replay the message, and hope they have understood the part that mattered most.
Speak the way the instruction arrives
Nkomo is built for natural Twi and Ghanaian English conversation by voice or text. You can hold to speak, pause to think, interrupt naturally, and hear the reply. The point is simple: keep talking in the mix of languages you actually use.
Afia can replay the voice note, then ask in the same conversation: “So should I add it now, or wait until the tomatoes reduce?”
She does not need to turn the request into a language exercise first. She can explain that the tomatoes are still watery, switch to Twi for the phrase she wants checked, then ask for the next step in plain Ghanaian English.
A coherent answer also needs room for uncertainty. If a phrase is unclear in the recording, an assistant should say so and ask for the part that needs repeating. Silent failure is especially unhelpful when someone is holding a spoon over a hot pot. Nkomo shows errors rather than swallowing them, so the conversation gives a visible signal when something has gone wrong.
The same need appears outside the kitchen. A parent forwarding a medication instruction, a student checking a voice note about an assignment, or a relative abroad asking how to make a dish they miss may all move between English and Twi without planning it. The conversation has to follow the person, not make the person adapt to the tool.
Keep control of the voice note before you share it
Family recipes can hold more than cooking instructions. A voice note may include a relative’s voice, a name, a joke, or a detail Afia never intended to send beyond the family chat.
Before she asks for help, she should know what will reach the cloud and what will remain on her device. Nkomo makes that choice explicit: never send to the cloud, ask each time, or allow it for the current session. History is controlled on the device, and turning it off purges it immediately. Data export and account deletion are available in one tap.
That control changes Afia’s next move. She chooses the consent setting she wants, asks about the crucial phrase, then turns history off when she is done. The next morning, the pot is empty except for a spoon by the sink, and the recipe is written in her own words: two tomatoes, ginger, cook it down, then fa gu so kakra.
The instruction stayed whole long enough for her to use it.
Comments
No comments yet.