The Twi Condition That Could Turn “Maybe” Into a Promise

An assistant can produce a polished but useless answer when it handles the English part of a request and drops the Twi condition that carries the real instruction. In mixed Twi and Ghanaian English, the final phrase can change the promise, tone, or action the speaker needs.

In 1999, NASA lost the Mars Climate Orbiter as it approached Mars. The investigation found that one team used imperial pound-seconds while another expected metric newton-seconds. The spacecraft had travelled through space correctly enough to reach Mars, yet the missing conversion at a critical boundary meant it entered the atmosphere on the wrong path and was lost.

Edward Weiler, then NASA’s Associate Administrator for Space Science, described the failure as an inability to recognise and correct a simple error. The Mars Climate Orbiter Mishap Investigation Board documented the cause plainly: two parts of the same mission were working with different meanings for the same kind of data.

The condition often arrives at the end

A real request can begin in English because that is where the practical details sit: “Help me reply to Auntie about Sunday.” Then the part that decides whether the reply is safe may come in Twi: “ma nka sɛ mɛtumi aba.”

That condition means the assistant must not write as though the speaker has promised to come. A fluent English reply that says, “I’ll definitely be there on Sunday,” sounds helpful. It also creates the exact commitment the speaker asked to avoid.

This is why code-switching is more than a language preference. It carries instructions. The switch can mark a boundary, soften a refusal, signal respect, or add the one condition that makes the answer usable.

A system that treats the Twi phrase as optional decoration can return something polished enough to send and wrong enough to cause trouble.

A smooth answer can hide a missing instruction

The Mars Climate Orbiter did not fail because every part of the mission was useless. The problem sat at the handoff. One measurement was accepted in a form the next system could not correctly interpret.

Mixed-language requests have their own handoffs. A person may ask for a voice note in English, add a family-sensitive instruction in Twi, then expect one answer that holds both together. If an assistant hears only the English request, it may sound confident while silently losing the constraint.

That is especially risky in messages about money, health, travel, family plans, or work. “Tell him I have seen it, but don’t make it sound like I agree” is a different job from “tell him I have seen it.” The extra clause is the job.

The same issue appears in voice. People do not pause to choose a language before correcting themselves. They speak while cooking, walking, holding a child, or passing a phone across the table. A correction in Twi can arrive halfway through the sentence. It still needs to remain attached to the answer.

The Phone Passed Across the Table, and the Question That Had to Be Answered explores why that moment matters when the phone becomes the shared place where a question gets resolved.

Keep the whole request in view

Nkomo is built for natural Twi and Ghanaian English conversation by voice or text, including the ordinary mixing people use when one language alone does not say the whole thing.

That does not mean every answer should be trusted without reading it. Important messages deserve a quick check before sending. The useful standard is simpler: the assistant should preserve the condition you gave it, rather than quietly replacing it with the most likely English-only version of your request.

When speaking, you can interrupt naturally and correct the instruction as you go. When typing, you can write the request in the mix that feels normal to you. There is no need to force one language per turn just to make the tool understand the shape of the task.

Check for the promise, refusal, or limit

Before you send an AI-written message, look for the part most likely to be lost:

• Did it preserve what you refused to promise?
• Did it keep the respectful tone you asked for?
• Did it include the deadline, amount, person, or next step you added in Twi?
• Did it turn a careful “maybe” into a definite “yes”?

NASA’s lesson was not that units are complicated. It was that a small mismatch can survive long enough to become expensive. In a family message, the cost may be an avoidable argument, a false expectation, or a promise you never meant to make.

Say the whole request as you would say it. Then read the answer for the condition that matters most.