Off first, then English to find out what you missed, then off again. Those three states in that order, inside one scene, is the whole answer, and the state to be careful about is the middle one because it is the most comfortable.
English subtitles print the words as written. Spoken English does not contain those words: sounds are dropped, endings vanish and whole syllables reduce to nothing, so a subtitle quietly repairs everything you needed to notice. You read the repaired version, you understand it, and the thing that defeated your ear is now invisible.
Eight rules for using them without depending on them
- English subtitles, never your own language. Translated subtitles turn the exercise into reading your own language while sound plays.
- subtitles hide the reductions. They print going to where the actor said gonna, so what you read is not what was said.
- use them to check, not to follow. Watch a scene without, then again with, and the gap between the two is the lesson.
- subtitles first, then without is the right order for material well above your level, and the reverse for anything near it.
- turn them off for the second watch, which is where the listening actually improves.
- a scene twice beats two scenes once, because the second pass is the one where the ear does the work.
- captions differ from subtitles. Captions transcribe sound including tone and noise; subtitles assume you can hear.
- pause and repeat aloud. Saying the line at the speed it was said is the fastest way to hear it next time.
Three states, and what each of them actually trains
Subtitles in your own language train reading in your own language. The soundtrack becomes atmosphere, and a whole series can go past with the ears contributing nothing at all. That is a legitimate way to enjoy a programme and it is not practice of any kind.
English subtitles train reading in English, which is worth something: vocabulary, spelling and reading speed all improve. What they do not train is the ear, and they feel exactly like they do, which is what makes this the most expensive setting on the menu.
Nothing trains listening. It is uncomfortable, you will follow less, and it is the only one of the three that touches the skill in question.
They are not even a transcript
Worth knowing before you rely on the middle state: subtitles are edited for reading speed. A line delivered fast gets shortened so a viewer can read it in the time available, so words are cut, clauses are merged, and repetitions and fillers disappear. What is on screen is a summary that fits.
The closed caption or SDH track is closer to what was said, because it exists for viewers who are not hearing the audio at all, and it therefore keeps more of the fillers, the false starts and the repetitions. Selecting it rather than the ordinary subtitle track is a small menu choice that changes what you are checking yourself against.
It replaced the single burned-in track on a DVD, where one translation was welded to the picture and there was nothing else to choose. Its limitation is that SDH is written for a different purpose, so it adds descriptions of sounds and music that are not dialogue, and it is still edited rather than verbatim.
Where it stops: leaving them on
The stall is English subtitles permanently on, and it is the most common listening mistake there is. A learner switches them on because a scene was hard, keeps them on because the next scene is easier with them, and three years later has watched several hundred hours of English with their eyes.
Every signal available says it is working. Comprehension is high, vocabulary grows, and the material gets steadily more ambitious. The learner has genuinely improved, at reading, and their listening is roughly where it was when they turned the subtitles on, which only becomes visible when there is no screen: a phone call, a meeting, somebody talking in a corridor.
The exit is to accept a much lower comprehension for a few weeks and to use the middle state as a check rather than as a companion. Watch a scene with nothing, then replay the same scene with English subtitles, and read what you missed. That costs twice the time per scene and it is the only version of this that is practice.
What it looks like when it goes wrong
The recognisable failure is discovering, on the replay, that almost every line you missed contained only words you already knew. That is the ordinary result and it is the finding: the gap was not vocabulary, it was that the words did not sound the way they look.
The second failure is reading ahead. Subtitles appear slightly before or with the line, so the eyes reach the end of the sentence before the voice does, and you have understood the line without hearing any of it. There is no way to stop this while they are on, which is the argument for the first state rather than a technique for improving the second.
The deletions that a subtitle hides are catalogued in why spoken English runs words together, and how to hear it anyway, and choosing material where the first state is survivable at all is choosing what to watch, and what to do while watching. Both sit under real speed, real accents, no subtitles.
What follows is a question about material with no subtitle track available in any language, which removes the decision entirely and is the reason a lot of learners never go near it: listening with no pictures to lean on.