A Caption Is Two Products With One Name
Two people press the same captions button and need opposite things: a full record of every sound, or a parallel translation that stays in step with the speech. Platforms only recently admitted the difference.
Two people press the same button and get different things. The first cannot hear the audio, so the captions have to stand in for sound itself — the words, who is speaking, the door that slams off screen, the music that suddenly goes tense. The second can hear every syllable and understands none of it, so the captions have to stand in for meaning — a version in another language that keeps pace with what is being said. The button says captions on both screens. It is not the same product.
One word is doing two jobs here, and the jobs are close enough that every platform ships them under a single label while the underlying files were always two different things.
Two people, two opposite jobs
For someone who cannot hear, a caption is a replacement for a channel. Everything that arrived by ear has to arrive by eye: dialogue, but also speaker identity when the face is off screen, and the sounds a hearing viewer would not think to mention — a ring in the distance, an engine starting, a song. Leave those out and the person is not watching the same scene. They are watching a thinner one.
For someone who can hear and cannot follow, a caption is a bridge, not a replacement. The ears already deliver the door slam and the music. What is missing is meaning, so the useful text is a translation that stays in step with the speech. This reader does not want [tense music building] interrupting the sentence they are trying to map word to word.
One product is measured by how much it includes. The other by how well it lines up. They cannot both be optimised at once.
The platform files already keep them apart
The split is not a theory. It is in the delivery rules.
Netflix’s own style guide for subtitles for the deaf and hard of hearing opens by banning the normal habit of subtitling. “Include as much of the original content as possible. Do not simplify or water down the original dialogue,” it says, and a few lines later, “Do not paraphrase dialogue in SDH, keep transcriptions as close to the audio as possible.” Slang, dialect, word order — all of it must be written the way it was spoken. On top of the dialogue the file has to carry speaker labels (-[Kimmy] Are you coming?) and the sounds a hearing viewer gets for free, in square brackets: [door slams], [phone buzzing], [dramatic music swells].
That is the opposite of a translated subtitle, whose job is to be condensed, recast and matched to the reading speed of someone hearing the original. A translation is allowed to cut. An SDH file is told not to. Two files, two sets of rules, one word on the screen.
Netflix only said so out loud in 2025
For years Netflix shipped the difference without naming it. If you wanted subtitles in the original language — say you are learning English and want the English words on screen — your only option was SDH/CC, which carried the audio cues and speaker names as well as the dialogue. There was no “just the words” track.
In April 2025 that changed. Netflix announced an original-language subtitle option showing only the spoken dialogue, and spelled out the problem it solved in one sentence: “Until now, if you wanted subtitles in the original language, you’d turn on Subtitles for the Deaf and Hard of Hearing (SDH/CC), which includes not only dialogue but also audio cues like [phone buzzing] or [dramatic music swells], as well as speaker names.” The language picker now offers two English options: one with the cues, one without.
Read that as an admission rather than an announcement. The two audiences had been sharing one track, and it took the platform more than a decade of streaming to split them. The split is that recent.
YouTube puts both behind one button
YouTube does the opposite: it keeps the two jobs merged and hides the merge behind a single CC button. But the merge is uneven. YouTube’s automatic captions transcribe speech and nothing else — no sound cues, no speaker labels. That makes them a partial deaf caption: a viewer who cannot hear gets the words and misses every sound a hearing viewer would use to place them.
The full version has to be built by hand. A creator who wants real captions edits the automatic draft and adds the missing layer themselves — the [music], the speaker names, the off-screen noise. Which is another way of saying the machine does the easy half and leaves the accessibility half to a person.
So on YouTube the word captions can mean three different things at once: a machine transcript, a hand-completed deaf caption, or a translation into another language. All three appear under the same icon.
The two jobs pull in opposite directions
Now the tension is visible, it is worth naming, because it explains why one tool rarely serves both readers well.
| Who | Deaf caption | Translation |
|---|---|---|
| Reader | Cannot hear | Cannot follow |
| Goal | Complete | Aligned |
| Sound cues | Required | No |
| Speakers | Labeled | No |
| Condensing | Forbidden | Allowed |
Every row is a choice that points one way. A deaf caption adds; a translation aligns. The same screen cannot easily be the fullest record of the sound and the clearest parallel of the meaning, because the first wants more on the line and the second wants the two lines to match.
Which half is ours
We are on the translation side, and we say so rather than letting the word captions absorb it.
Tablingo reads the tab you are already playing and puts the original line above its translation. That is a bridge for the second reader — the one who can hear and wants to follow. We do not detect the door slamming, we do not label the speaker, and we do not produce a full accessibility transcript. Those are the jobs of a deaf-caption track, and the platforms already make one; pointing someone who needs it back to the platform’s own CC track is the honest answer, not a dismissal.
Two audiences, two files, two tools. The only thing they share is the word printed on the button.
If you want to try ours, it’s called Tablingo — a browser extension that reads the tab you are already in and puts the original line above the translation. What a caption tool actually reads is here, why a podcast has a transcript but nowhere to put a caption is here, and the language decision a caption makes before its first word is here.