Merging short dialog lines in SubtitleEdit

[Night457]

This is just a quick example, using the subtitles shared here:
https://www.first-loves.com/forums/view ... 13#p110513

(I decided I did not want to clutter a movie thread with boring tech talk, like I usually do.)

I looked at the numbers in SubtitleEdit and I can see that there were 12 lines that were under one second in length. Fortunately they were short bits of dialog -- at least in English. Only half of them could be lengthened without overlapping other dialog lines, and the ones that could be "fixed" were STILL under a second! And with a minimum gap between subtitles of 24 milliseconds, it only lengthened the "fixable" ones by 1 millisecond! I think the best way to deal with that is to go to a "Too Short" line of dialog and merge it with the one before it or after it, whichever makes the most sense. That way you can get two short lines of dialog combined, on screen together for long enough to read them both. I am not going to actually bother because in this case I don't think it will keep me from understanding it even if I miss a few lines. I showed a sample Merge below, for when someone judges it necessary to do the merges. Carefully read the details shown in the screenshots.

Screenshot 2024-02-07 102346.png

Screenshot 2024-02-07 092503.png

Screenshot 2024-02-07 093945.png

Screenshot 2024-02-07 094208.png

[endurro]

Using "Tools->Fix Common errors" gives good results.
I use this function most often. In my opinion, it is useful. Of course, once it shows its suggestions for correction, it is worth looking at these suggestions and perhaps reject some manually, because AI is not always right.

The above feature for "Nie bede cie kochać" proposes 146 corrections. For the most part, these are cosmetic corrections that do not change much.

As for the remarks I made to the movie "I won't love you", I didn't mean very short lines like "Anka". Such lines can easily be read in less than 1 sec.
I was more concerned with long lines that change quickly in succession and where there is not much way to improve the times.
In such cases, sometimes combining lines and sometimes breaking them helps.

[ghost]

Interesting!

I didn't know this feature in SubtitleEdit. Thanks.

[Night457]

Using "Tools->Fix Common errors" gives good results.
I use this function most often. In my opinion, it is useful. Of course, once it shows its suggestions for correction, it is worth looking at these suggestions and perhaps reject some manually, because AI is not always right.

You said it alright!

I also use "Fix common errors" and I carefully monitor it. It is exasperating when I don't pay attention and it makes fixes that are clearly wrong! Glad you mentioned it, because fixing subtitle time length is one of the options. ("Fix short display times" and "Fix long display times".)

I was more concerned with long lines that change quickly in succession and where there is not much way to improve the times.
In such cases, sometimes combining lines and sometimes breaking them helps.

Oops, I just focused on overall time and not really long text!

In SubtitleEdit / Options/Settings/General, it has adjustable settings for "Optimal chars/sec" and "Max chars/sec". (See first screenshot below.) These are the settings that have to do with how quickly a person can read a long line of subtitle text. If a subtitle has too much text to read in the time allowed, it gets highlighted in the subtitle window. That makes it easier to find those lines to fix them by Merge or Split. See the 2nd, 3rd, 4th screenshots for an example. Of course the person fixing it will have to play it back in the Video window to see if their fix is an improvement! I will sometimes start the subtitle a little earlier than they actually start speaking in order to give more time onscreen to read it.

And I must confess that if I spend many many hours TRANSLATING a subtitle, I am usually too exhausted to focus on how long the subtitle is onscreen. I have also read the same text over and over again, so I do not notice that it may be too long to read quickly, because *I* have it practically memorized!

Screenshot 2024-02-07 140652.png

Screenshot 2024-02-07 141123.png

Screenshot 2024-02-07 142006.png

Screenshot 2024-02-07 142108.png

[endurro]

In SubtitleEdit / Options/Settings/General, it has adjustable settings for "Optimal chars/sec" and "Max chars/sec". (See first screenshot below.) These are the settings that have to do with how quickly a person can read a long line of subtitle text. If a subtitle has too much text to read in the time allowed, it gets highlighted in the subtitle window. That makes it easier to find those lines to fix them by Merge or Split. See the 2nd, 3rd, 4th screenshots for an example. Of course the person fixing it will have to play it back in the Video window to see if their fix is an improvement! I will sometimes start the subtitle a little earlier than they actually start speaking in order to give more time onscreen to read it.

And I must confess that if I spend many many hours TRANSLATING a subtitle, I am usually too exhausted to focus on how long the subtitle is onscreen. I have also read the same text over and over again, so I do not notice that it may be too long to read quickly, because *I* have it practically memorized!

You are absolutely right. I know this very well from my own experience.
I've noticed that professional subtitlers get around the problem with too much "density" of text by simplifying the content. In such cases, there is less literal translation and more intuitive translation so as to convey the meaning of the sentence. This approach to translation is not easy.

[David32441]

Thanks for all the screenshots!
I recall that about 16-20chars per second is a typical average persons reading speed. But clever that the software can join 2 lines together. I presume it automatically updates the line number when you've selected it to do this?

[Night457]

endurro -- I have attempted condensation of text when the sentences get ridiculously long, but I prefer to retain as much of the original as possible.

The professional (which means PAID, not necessarily superior) subtitlers are prone to simplifying TOO much, and they show their contempt for their audience by assuming that they know NOTHING about any thing or any place out of their own narrow culture. When the intended audience for a film is children, it makes sense to keep the sentences as short and simple as possible -- or to dub the film instead. They won't enjoy it if they can't keep up ... and I have more patience with children than with adults. (I'm working on that.)

What I find funny is when they simplify for its own sake and not for time. I watch an old martial arts movie where a character speaks two sentences over the course of 10 seconds and the subtitle says "Go now!"

David -- I really struggle to understand complicated software directions when it is solely text. Once I can SEE it, then it is so much easier. But I struggle even more when it is only pictures (or video) and no words!

They have the default "Optimal" setting at 15 char/sec, I assume because a person reading the subtitles is ALSO watching the movie. That surely slows down reading. And yes, it readjusts line numbers whether you Split or Merge.

[pillowbaker]

Thanks Night and everyone for your ideas. I am keen on learning how to make subtitles easier to read for everyone. So many great points are brought up in this thread, and I can hardly contain myself to see that many of the techniques and ideas listed out here are also ones that I figured out by experimenting, like:

I will sometimes start the subtitle a little earlier than they actually start speaking in order to give more time onscreen to read it.

I do the EXACT same thing -- somewhere between 200-300 milliseconds still retains its fluidity with the dialogue, sometimes when I'm desperate for more time, I'll bump it up to 400.

I should note that I merged many lines during my translating and editing of this film, and I extended the times wherever possible, which was extensively. But I can certainly see here I could have merged more. I also saw one other minor error where I used a definite article where I should have used an indefinite article, so maybe I should make a few more fixes and reshare them.

I also was very conscientious of the letters per second, and as you'll see in my next write up, that was an area I struggled with. Polish has this natural manner of being capable of great brevity, and required many more words to convey meaning in English without sounding off. endurro would point out some of these phrases were short for much longer phrases. His help was great in these areas! And I thought English could be brief! Some of these explanations, where he took the phrase in Polish and inserted the dropped words to show what the full sentence looks like, resembled precalculus problems where you would substitute out numbers for variables, but instead of a simple number substitution, it was whole equations, with more variables, division signs, square root signs, and exponents.

In these areas, I pondered much how to reword lines naturally, and like Night, I was pretty fussy over retaining all meaning. Occasionally, I was able to drop words easily, other times, I had to muse over while away from the computer. I marvel at what other subtitlers have been capable of while watching my shows.

And with a minimum gap between subtitles of 24 milliseconds, it only lengthened the "fixable" ones by 1 millisecond!

I noticed this very thing when I used the "Fix Common Errors" tool. Most lengthened times only resulted in one extra millisecond that, for some reason, it doesn't allow me to reach when I am dragging the time-lines over the waveform. I still use it, though, but like you mentioned, some of what it catches are not corrections one would want at all! I had to carefully look over each line it would affect.

[Night457]

I also saw one other minor error where I used a definite article where I should have used an indefinite article

INEXCUSABLE!!!

(It has been so long since high school English that I don't even remember the difference.)

I know the pain of struggling for hours over complicated improvements and thinking it is all done, then seeing yet another simple obvious fix. I assumed that with the time you spent on this, that it was well-combed-over and that you WERE paying attention to sentence length and and time length, so no criticism was intended. (And I have not yet watched the movie with the subs, though it is on an actual written list for me to do so.) I never would have thought to compare translation to calculus.

Most lengthened times only resulted in one extra millisecond that, for some reason, it doesn't allow me to reach when I am dragging the time-lines over the waveform.

I have noticed that the waveform has a grid background and you can only make manual changes that are the size of one block. If it was zoomed in close enough so that one block equals one millisecond, then the wave would extend for a THOUSAND blocks for ONE second, with many pages of scrolling to the left or right! That is simply unusable. The zoom options under the wave only extend from 10% to 250%, which is effective to get a broad enough or close enough view as needed. If I need finer tuning, I adjust the numbers for the "Start time" and "Duration" to the left of the "Text" editing box. I am sure that I would never be able to hear the difference of a millisecond, however, and it would not aid in my reading.