The words were all in my deck. They just sounded completely different coming out of an actual person.
There's this thing that used to happen to me constantly when I first started having real Cantonese conversations, and if you've ever tried to use a language you mostly studied from an app or a deck of cards, I'd put money on it having happened to you too. Someone says a sentence, and there's one word in it that just doesn't land. Nothing comes up. The conversation moves on, because conversations do that, and then a minute or two later, while they're already onto something else entirely, the word quietly shows up in my head with its meaning attached, like it had been out getting coffee. And what made it maddening rather than just embarrassing is that I'd reviewed that word. Sometimes that same morning. It was in my deck, it was marked as known, my app was very pleased with my progress on it, and none of that mattered at conversation speed.
The TV thing was the same deal from a different angle. I'd watch something in Cantonese with the subtitles on and follow basically all of it, and feel great about myself, and then the same show without subtitles was noise. Not completely, you catch islands here and there, but mostly noise. And it's the same vocabulary. That's the part I kept getting stuck on. Nothing about the language changed between those two experiences except which sense it was coming in through.
Here's how I ended up thinking about it. A text flashcard is a reading exercise. Obvious, right, but follow the thread: the prompt is written, you look at it in silence, at your own pace, and you can sit there staring at it for ten seconds if you need to and nobody's moving on without you. So the link you're strengthening, every single review, is written form to meaning. Which is a real skill, and it's why my reading got properly decent.
But speech never shows you that form. A word in conversation is a burst of sound squashed up against the words either side of it, in somebody else's voice, at whatever speed they happen to talk, and it's gone, and the next one's already arriving while you're still processing. Recognising a word under those conditions is a separate skill from reading it. It feels like it should transfer, because it's "the same word", but it transfers way less than you'd expect.
The annoying part is, I knew this. French had already taught me the shape of it, reading comprehension showed up early and conversation took years longer, so I walked into Cantonese knowing exactly where the trap was. I even picked spoken Cantonese as the goal from day one, over reading and writing, because spoken was the only version I actually needed. And then I spent my study time looking at text anyway, because that's what every tool put in front of me, and watched the deck get bigger while conversations sounded exactly as fast as before.
Subtitles will tell you, and it takes one evening. Put on a show in your target language, with subtitles in that language too, not in English, and watch it normally for a few minutes. You'll feel like you're following the audio. Then cover the bottom of the screen with something and find out how much of that was actually your ears. For me the honest answer was: not much. Lines I would have read without even noticing I was reading meant nothing at all as pure sound. The words were in there, all of them, just filed under an appearance that speech never takes.
The fix isn't clever. If the thing you want to get good at is understanding speech, then the prompt has to arrive as speech. That's the whole insight. A recording plays, in a real voice, and your job is to get the meaning before it finishes, because that's the deadline a real conversation gives you whether you like it or not. And all the spaced repetition machinery underneath can stay exactly as it is, by the way. The scheduling works, reviewing a word right before you'd forget it works, I've got no argument with any of that. I was just running an extremely well-scheduled programme of reading practice and expecting listening to fall out of it somehow.
That's why I built Colloquial the way it is: the cards are voice recordings, and a review session is literally a phone call, your phone rings, you answer, and it walks you through your deck out loud while you answer back, screen off the whole time. There's a card at the bottom of this page if you want a look. But honestly, even if you never touch my app, you can steal the principle with whatever you're using now: record yourself saying the words, hide the text, and make your ears do the recognising, because they're the ones that'll be on duty in the actual conversation.
Colloquial is a flashcard app that calls you and walks you through your deck out loud.
See how it works →