The translation page had two dropdowns: source and target language. After using it a few times, I noticed the model could often recognize the source even when I selected the wrong one. Choosing it afresh each time began to feel unnecessary.
I asked for automatic detection as the first source option. The software could handle that step. But the target I had already chosen needed to stay selected, rather than reverting to English when detection ran.
The swap button needed the same attention. When I swapped the sides, I expected the text to move too. Changing the labels while leaving the contents in place wasn't the action I meant.
These details are easy to miss in a demonstration. Translating several pieces of text in a row shows which actions have to be repeated. Next time I ask for automation, I'll describe the choices I want to keep as well.
Working on something similar?
Bring your setup, your goal and the detail you are stuck on.
Discuss the project
