The Square Canister Apollo 13 Had, and What It Almost Cost the Crew

A voice assistant should adapt when a speaker moves between English and Twi, because code-switching often carries the part of the question that matters most. People should be able to speak naturally instead of translating their concern into one approved language first.

In Houston in 1970, Apollo 13 had a problem that could not be solved by insisting every part stay in its assigned place. The lunar module had to support three astronauts after the command module was shut down, but its carbon-dioxide system was built for two. The command module carried square lithium-hydroxide canisters; the lunar module used round openings. NASA engineers had to devise an adapter from materials already aboard, while the crew’s safe return remained uncertain. NASA’s Apollo 13 history records the improvised solution and the mission that followed.

The useful part of that story is not the famous phrase about fitting a square peg into a round hole. It is the decision behind the fix. The people in Houston did not ask the canister to become round. They made a way for the system to work with what was actually there.

The careful English opening

A person may begin a question in measured English because they have learned that digital tools expect it.

“I want to ask about a message I received from my landlord…”

Then the detail arrives, the one that makes the question urgent or difficult to explain: “but sɛ meka sɛ…” Or a concern about an auntie, a transfer, a clinic visit, a family disagreement, or words that sounded polite but carried a warning.

That switch is not a mistake. It can be the speaker reaching for the language that holds the right meaning, tone, relationship, or emotional weight. English may establish the facts. Twi may carry the concern underneath them.

Forcing the speaker back into English at that moment asks them to do extra work precisely when they are trying to think clearly. It also risks flattening the question. A literal translation can preserve words while losing whether something sounded urgent, respectful, doubtful, irritated, or affectionate.

Natural language has room for both

Ghanaian English and Twi share space in ordinary conversation. A voice assistant built for that reality should treat a mixed-language turn as one thought.

Nkomo lets you type or speak in natural Twi and Ghanaian English, including the mix people use in conversation. In hands-free voice mode, you hold to speak, can interrupt naturally, and hear a reply. The goal is simple: say the question as it comes, then decide what to do with the answer.

This matters especially when a question changes shape mid-sentence. Someone may start with, “Can you help me write a reply?” and continue in Twi because the relationship and intended tone are easier to express there. The useful assistant follows the speaker’s lead. It does not make the speaker prove they belong in one language before it can help.

That is also why a good Twi voice assistant needs more than language labels. The practical evaluation is whether it can handle speech, replies, code-switching, and privacy controls in the situations people actually bring to it. This evaluation checklist lays out what to look for.

Privacy belongs in the same conversation

A sensitive question can require another kind of adaptation: control over where the question goes.

Nkomo makes cloud consent explicit. You can choose 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 choices matter when the question begins casually and becomes personal halfway through. You should not have to decide between speaking naturally and understanding what happens to the conversation afterward.

Clear controls also make it easier to ask directly. There is no password to remember because sign-in uses a magic link. And if something goes wrong, Nkomo shows the error rather than leaving you to wonder whether your Twi, your connection, or the app failed. That last point deserves more attention than it gets. A silent reply can make people blame their language, even when the problem lies elsewhere.

Build for the thought that arrives late

The hardest part of a question is often not its first sentence. It is the clause that follows after a pause, when the speaker stops managing their language for the machine and starts explaining what they mean.

Apollo 13’s engineers worked with the equipment the crew actually had. A conversational AI should work with the language a person actually speaks. When English leads into Twi, the answer is not to send the speaker back to the beginning. Let the conversation continue.