What actually runs out
It is not vocabulary and it is not grammar; both are still there afterwards, as you discover when you replay the conversation in your head on the way home and produce the sentence effortlessly. What runs out is the working attention that unautomated skills demand.
In a first language almost everything is automatic, so conversation costs little and you can sustain it for hours while doing something else. In a second language, most of it is still conscious. Every particle you check, every word you search for and every sound you monitor is drawing on the same pool, and the pool is small.
This reframes the problem usefully. You do not extend stamina by trying harder for longer. You extend it by making individual pieces automatic, so that each minute of conversation costs less.
The warning signs before the cliff
The collapse feels sudden but it is not. There are three or four minutes of clear warning, and learning to notice them lets you intervene while intervention is still cheap.
The most reliable sign is regression: an error you had fixed weeks ago comes back. Monitoring is the first thing to be dropped when attention runs short, so old habits resurface before anything else visibly degrades.
- Errors you had already fixed start reappearing.
- Your sentences get shorter and stop using connectors.
- You start missing the beginning of what the other person says.
- You answer with a set phrase where you would normally elaborate.
- You catch yourself translating rather than composing.
Any two of these together means you have three or four minutes. Use them to steer, rather than discovering the cliff by falling off it.
Buying back attention mid-conversation
Once you notice the signs, the goal is to reduce cost rather than to push harder. Shifting to a topic you know well, asking a question that lets the other person talk for a while, or taking a short break all restore enough capacity to continue.
Asking a question is the cheapest of these. Listening costs less than producing, and a well-placed question buys you a minute of recovery without ending the conversation.
| Japanese | Romaji | Meaning |
|---|---|---|
| ちょっと休憩してもいいですか。 | Chotto kyūkei shite mo ii desu ka. | May we take a short break? |
| そのへんの話、もっと聞きたいです。 | Sono hen no hanashi, motto kikitai desu. | I would like to hear more about that. |
| すみません、少し疲れてきました。 | Sumimasen, sukoshi tsukarete kimashita. | Sorry, I am getting a little tired. |
| ゆっくりお願いできますか。 | Yukkuri onegai dekimasu ka. | Could you speak slowly, please? |
| 水を飲んでもいいですか。 | Mizu o nonde mo ii desu ka. | May I have some water? |
| もう少しだけ、頑張ります。 | Mō sukoshi dake, ganbarimasu. | I will keep going a little longer. |
Lowering the cost per minute
The long-term fix is automation. Every piece of the conversation that stops requiring conscious attention gives that attention back for something else, and the effect compounds because the freed attention makes the next piece easier to automate.
The pieces worth automating first are the ones that recur in every conversation, regardless of topic. Your introduction, the ten questions you are always asked, agreeing and disagreeing, and the phrases for repair are used constantly, so making them free has an outsized effect.
| Piece | Cost when conscious | How to automate |
|---|---|---|
| Your introduction | High, every time | Twenty spoken repetitions |
| Common answers | High | Rehearse the top ten questions |
| Repair phrases | Medium | Use them deliberately for a week |
| Sounds | Medium, constant | Pronunciation lab drills |
| Connectors | Medium | One pattern per week, spoken |
| Topic vocabulary | High if unfamiliar | Flashcards before the event |
This is why the pronunciation lab matters beyond accent. A sound you have to think about is a sound taxing the same pool your grammar needs.
Training the window rather than surviving it
Stamina responds to sessions slightly longer than comfortable, repeated often, rather than to occasional marathons. If you fade at twenty minutes, practise at twenty-five. Two hours once a month builds much less than twenty-five minutes three times a week.
The other half is recovery. Two shorter conversations with a gap between them are far more productive than one long one that spends its second half in collapse, because the collapsed half teaches you almost nothing.
- Find your current fade point by timing one honest conversation.
- Practise at that number plus five minutes, three times a week.
- Stop at the target even if it is going well; consistency beats a record.
- Add five minutes only after two comfortable weeks.
- Track the fade point monthly, not daily; it moves slowly.
Using controlled sessions to measure it
The problem with measuring stamina in real conversations is that everything varies: the topic, the accent, how tired you already were, how much the other person carried you. A conversation that felt worse may simply have been harder.
Practising with an AI Sensei in Unihongo's immersive 3D classroom removes most of that variance, because you set the length and the level and speak aloud under consistent conditions. Unlike a text chatbot, which lets you rest between messages for as long as you like, a spoken session keeps the load continuous, which is what stamina training requires.
The session report is what turns it into a measurement: if the corrections cluster in the last third, the window is the limit rather than the material, and the answer is a shorter session repeated more often rather than a harder one.
Expect the window to grow in steps rather than smoothly. Several flat weeks followed by a jump is the normal shape.