Nobody sends me films. They send me screen recordings. A forty-minute walkthrough of a CRM, a training clip for warehouse software, an onboarding video for a payroll app — and the voiceover is English, the viewer is Turkmen, and the software itself has never been localized into Turkmen and probably never will be.
That's the job most of the time. Not a drama with jokes to rescue. A guy clicking through menus, saying "now click Submit, then go to Settings." And on the screen, in plain pixels, the button says Submit. In English. It will always say Submit, because nobody paid to translate the interface.
So what goes in my subtitle?
The button isn't mine to translate
If I write the Turkmen word for "submit" in the subtitle, the viewer reads it, looks at the screen, and sees a button that doesn't match. They hunt for a word that isn't there. I've just made the video harder to follow than if I'd left it in English.
My rule: whatever the viewer has to find and click stays exactly as it appears on screen. Submit stays Submit. Settings stays Settings. I'm not translating the UI; I'm translating the instructions around it. The verb, the sequence, the warning — that's mine. The label is a landmark, and you don't rename landmarks on a map.
But leaving a raw English word stranded in a Turkmen line reads badly too. So the first time a term shows up, I gloss it: the Turkmen explanation, then the on-screen label in parentheses or quotes. After that, I trust the viewer and just use the label. One glossing pass, then get out of the way.
The hard part isn't the decision. It's being consistent across forty minutes and eleven subtitle files when the same button appears in clip 3 and clip 29 and you spotted them two weeks apart. This is where a termbase earns its keep — not for "synergy" and "stakeholder," but for a flat list of every clickable string and exactly how I handled it the first time.
The clock doesn't care about your gloss
Here's the squeeze. Turkmen runs long. English says "Click Submit" — two beats. The honest Turkmen, with the agglutinated endings and a gloss for a first-time term, wants to be three times the length. And the voiceover talent didn't slow down to let me fit it.
So the gloss costs me characters I don't have. A line that explains and names the button blows past any sane reading speed, the viewer's eyes are still on line one when line two is already gone, and the whole point — matching screen to text — collapses because they never finished reading.
My fix is to move the gloss off the busy moment. If I know "Submit" is coming at 04:12, I introduce it earlier, during a calm stretch where the narrator is just talking, not clicking. By the time the button appears, the viewer already knows the word and I can use the bare label. Spotting isn't just where the audio sits. It's deciding when the viewer has spare attention to spend. On a dense demo, attention is the budget, not words.
This is why I quote these by the minute, never the word. A technical SRT with thirty UI terms, each needing a placement decision and a consistency check, is slower per minute than a talking-head interview of twice the length. The word count lies. The screen is the work.
What the PM should send and almost never does
The English SRT. The video, not a link that expires in three days. And — the one I beg for — the UI string list, if the software exists in any localized language.
Because sometimes it does. The app isn't in Turkmen, but it's in Russian, and half my audience uses the Russian build. Now Submit isn't Submit on their screen; it's Отправить. If I knew that, I'd gloss against what they're actually looking at. If I don't know it, I guess, and I guess wrong, and the video confuses the exact people it was made for.
Nobody thinks to send this because the person who ordered the subtitles isn't the person who knows which build ships to Ashgabat. The PM has a video and a deadline. The product team has the string table and no idea subtitles are happening. I sit between them asking a question neither expected: what language is the interface the viewer will see? Half the time asking it is the most useful thing I do on the whole job.
Subtitling a demo isn't creative localization in the tagline sense. There's no wordplay to reinvent. But deciding what to translate and what to leave frozen on the glass — that's a localization judgment every single line, and it's the part a machine fed only the audio track will cheerfully get wrong.
Translate the instruction. Leave the button alone. Tell the viewer once what it means, then let them click.