Voice assistants built for English speakers can't handle code-switching: when you switch between Twi and English mid-thought without stopping to select a language, they freeze or misunderstand. The hidden cost is constant management of the tool itself, which forces you to simplify your words and sound like someone smaller than you are.
When Ama calls her mum in Kumasi from her flat in London, she doesn't plan what language to use for each thought. Mid-sentence, the right word arrives in Twi, she says it, and keeps going. If the English phrase feels clearer for the next idea, she switches without thinking about it. This is how she thinks. This is how most bilingual Ghanaians in the diaspora actually speak. But when she tried to use a voice assistant last month, it wanted one language per session. Every time she mixed them (dropped an English phrase into Twi, or a Twi word into English) the assistant froze, backtracked, or misheard what she'd said. The call that should have taken ten minutes stretched to twenty. Her mum kept asking if the line was breaking up, and by the end, Ama was drained from managing the tool instead of just talking.
This is the untold cost of language selectors: they force you to manage a tool that doesn't understand how real bilingual conversation works.
What most voice tools assume
Most voice assistants are designed with one language per session. You pick English or you pick Twi before you start, and the system listens for that language alone. If you start speaking both in one breath, if you say a thing in English and then the thought completes in Twi, the system gets confused about what language it is listening for. Transcription breaks. The AI hears the English words but not the Ghanaian context they're swimming in. The meaning falls apart. The app forces you to stop, rewind, select the correct language, and start the phrase over.
It's a small thing, technically. But it's constant. Every call becomes a negotiation between what you want to say and what the tool allows you to say. You end up simplifying your thoughts into single-language chunks. You leave out the exact words that would make your meaning clear. You sound like someone else, someone flatter, someone edited.
For people who live between languages, this isn't friction. It's exhaustion. You start editing yourself before you even speak, picking simpler words, leaving out nuance, sounding like someone smaller than you are.
When you don't have to pick first
Ama tried Nkomo on a Tuesday, calling her mum again. This time she didn't open a menu to choose a language. She just held the phone and spoke: English for the logistics (the traffic on King's Road, the time she'd call back), Twi for what she wanted her mum to hear between the lines (the tiredness, the "I miss you", the parts that didn't translate without losing shape). The app caught both. It didn't freeze when she switched. It didn't ask her to repeat or clarify which language she was using. When she asked a question mixing both, something about whether her mum had seen the photo she'd sent and what did she think. It understood. It got the actual question, not just the English skeleton. The conversation ran at the speed of thought.
That Tuesday call lasted exactly as long as it needed to. No backtracking. No extra explanation. No managing the interface. Just her, her mum, and the words that fit.
What this means for people in the middle
Bilingual life in Ghana and the diaspora isn't a switch between two languages. It's a fluid mix. You code-switch because code-switching is efficient: you pick the language whose word is clearest, whose idiom hits better, whose register fits the moment. You do this without thinking about it, the way you would shift between formal and casual English with a stranger versus a friend. Most apps assume this mixed speech is a bug to fix. But code-switching is how billions of people think and talk.
When you're not stopping to select a language, something deeper shifts. You're not translating your thoughts before you speak them. You're not filtering yourself into one linguistic register. You're just there, in your actual voice, the one that lives in both languages at once. That's where you think fastest. That's where you understand the most. That's where the conversation can be about what actually matters.
For Ama, it meant a call that felt normal. For her mum, it meant not asking twice if the connection was clear. For both of them, it meant the full conversation, unedited by the tool, in the language mix that is their actual shared language.
Comments
No comments yet.