software translating subtitles

[SmallBrother]

(KomPair)
It can translate using google translate, but as mentioned in this thread before, this is just for a starting point.
After that you must polish the translation as best as you can.

 
I appreciate your well-intended efforts, but I am afraid I do not agree with your opinion.
Using a machine translation as starting point SEEMS a good idea, but it is NOT. On the contrary. It is a beginners trap.
Especially if the source is an English subtitle for an English spoken movie or tv show episode.

Let me explain you why.

Linguistical problems

First of all, and I think we already agree about this, translation will be inferior. Machine translations lack interpretation of context and feeling. Since the sentences will seem to be 'okay', it is very likely that a human check will oversee these mistakes. But for now, let's assume translation is perfect.

Technical problems

Next, you must realize that subtitling is not the same as translating. Subtitling involves much more than just translating. Even if the translation is perfect (which I doubt already), crucial aspects will be lacking. For example, English subtitles for English spoken movies are typically 100% verbatim, including al the "you know", "I mean", "hm-mm", "wow", "yeah", and repetitions and stammerings like "I-I didn't... didn't know" or "no, no, no, no, no". Typically, gaps are 1 or even 0 ms. Line breaks are considered only by mathematical line length instead of word groups. Durations are typically very short. CPS-ratio is not considered at all (which is impossible anyway, if it must be verbatim). Synchronisation is usually only so-so. Etc. etc. etc. All this is not *BAD* for English subtitles, it is how they are made and how they should be made. But for other (translated) languages it *IS* bad. After all, for understanding, the viewer relies totally on the subtitles. On top of that, note that English is a relatively 'short' language. For example Dutch, German. Latin and Scandinavian languages will need approximately 20-30% more text (number of characters). That means that lines will be longer (too long) and CPS-ratio will be higher (too high).

Fixing all this will take ages

So, using a machine translated subtitle as starting point will not only involve fixing up all the wrong or weird translations and sentence constructions, checking them one by one. On top of that, you will have to spot and delete all the unnecessary texts, once again, one by one. You will have to check and fix every line, one by one, for line length, CPS-ratio, line breaks, gaps, sync and so on. Compress, rewrite, merge, re-break en resync them. One by one.

You may now understand that creating a good subtitle from a machine translation will take more time than creating it from scratch. Trust me, I have tried And still, there is a good chance you will oversee wrong translations and similar mistakes. Basically, you are busy longer, with still a likely lesser result.

Reality: people will not do all that fixing, or no fixing at all

Knowing this, you can also imagine that a lot of people will not go through all this hassle and just delete the obvious "hm-mm", "yeah" and "wow", and obvious twisted sentence constructions. They will not bother with any technical aspects (if they are aware of the necessity of this in the first place). Next they are thinking they just created something decent enough. But they didn't. They didn't FIX the machine translation, they just CAMOUFLAGED it. With the evolution of machine translations, this tendency is going on for a while already in basically every language section.

Also, in reality, many people will just upload the raw untouched machine translation.

Machine translations will push away good human subtitlers

And here is the next point, not to be neglected.
A machine translation takes one minute. Fixing it up a bit will take 15-30 minutes. So someone is able to upload 'something' in no-time. This inferior subtitle will always be the first one.The first subtitle will get far most of the downloads.
A human translation takes way more time, let's say 7 hours for an average tv show episode, 15 hours for a movie. Would you spend all these hours to make something, when 95% of the downloads are already taken by users who downloaded the 'fixed' machine translation? I wouldn't.

The actual result?

So basically, the site will be flooded by raw or 'camouflaged' machine translations, which will only disappoint and irritate the downloaders, frustrate the real human subtitlers and cause unnecessary extra work for admins, who try to maintain at least a bit of quality and protect their users.

So, resuming, in my humble opinion:
providing machine translations is not a service, but a hell.
I am sorry this may sound a bit harsh, but I cannot make it nicer.

Thank you for understanding.

[hector]

I think smallbrother is right.

A Google translate takes two minutes.
A human subtitles takes (at least) 10 hours.

You want it fast and dirty?
Go to McDonalds.
You want it slow and good?
Be patient.

I think smallbrother just gave in the nail. You don't know what I'm
talking about? That's because I just made a literal translation from
Spanish "ha dado en el clavo" (he just hit the nail on the head).
Well, this is a silly example and Google is smart enough to not fall
in the trap. But it doesn't take much more.

I don't know if we'll see the day where you can't distinguish a
machine translation from a human translation. I think nowadays machine
translations are still awful. I would completely erase them from
opensubtitles but if we must live with them at least set the
matchine-translation flag.

Seriously, machine translations are only useful if you want a good laugh.
But you'll only laugh for 10 seconds until you get tired of guessing
the underlying meaning.

The key point is effort. Most people pretend to do in 5 seconds the
work of 5 hours. Don't be silly. It's like someone who thinks he knows
Kant just because he read an article in Wikipedia. Yes, Immanuel Kant
who lived in Königsberg in the 18th century. Of course you know his name
and where he lived. But do you know his thought? You kan't pretend to
understand the work of a lifetime in 5 seconds. Well, actually you
can, but that's just silly.

[Ramney]

I use VEED.IO Subtitle Translator for free and not bad.
Pros:
100 languages
95% accuracy rate
Simple to use
Cons:
Some features are paid

[SmallBrother]

I use VEED.IO Subtitle Translator for free and not bad.

Thank you for your input.

But once again I would like to emphasize: subtitling is NOT the same as translating. For subtitles much more aspects come into play, like reading speed, line length, text formatting, synchronisation, gaps, etc. which are not considered by any translating machine.

So, a good translation doesn't necessarily mean a good subtitle. Often on the contrary, especially if translated from English subtitles for English spoken movies, since those are derived from 'closed captions' meant for hearing impaired, applying TOTALLY different subtitling guidelines. So it doesn't really matter if Google, Bing, DeepL, Veed or Tralala is better in translating. NONE of them know how to make proper subtitles.

My point of view is and remains: machine translations are pretty much guaranteed to fail as subtitles. And so they frustrate any human translators and retail rippers doing their job for the subtitles community. So if you think a machine translated subtitle is sufficient for your purpose of watching a movie or tv-show - be my guest.

But please, oh please, DO NOT upload those to OpenSubtitles.

[oss]

we, developers of opensubtitles still evaluating the AI. For example you can kind of teach ChatGPT to translate subtitles, with all rules you give to human. It ends up with 95% quality subtitles. You should try guys, AI is the way to go in future and probably there will be very very few subtitles translated by humans in few years. That's my opinion. These AI things are completely different than old Translate from google for example.

Try ChatGPT, play with it...

[SmallBrother]

we, developers of opensubtitles still evaluating the AI. For example you can kind of teach ChatGPT to translate subtitles, with all rules you give to human. It ends up with 95% quality subtitles. You should try guys, AI is the way to go in future and probably there will be very very few subtitles translated by humans in few years. That's my opinion. These AI things are completely different than old Translate from google for example.

Try ChatGPT, play with it...

I disagree.
ChatGPT may improve the TRANSLATION qualities, but NOT the SUBTITLING qualities.

This is where for example Netflix is going, including human editing afterwards. The result is eeeeh... disappointing.

Maybe in 10 years.

[oss]

let's see, I give it max 1 year and even best subtitle translators will not recognize it, if it was done by human or software. That being said, to translate subtitle from one language to another WITHOUT knowing what is happening on screen.

And then we can talk about, that it is actually also NOT possible for humans to translate subtitles without watching actual movie to have a GOOD quality.

So then - yes, it will take longer. But in 5 years I believe it will be very possible. The development in AI is exponential, we as humans like to think in linear.

For now, sure best translations are by Human.

[SmallBrother]

Yes, when looking at translating, you might be right. But then we are still talking purely about translating, right?

But that's exactly my point: making subtitles involves much more than just translating. An accurate translation is not enough for a good subtitle - actually on the contrary. Ironically, a 100% accurate translation will often result in a bad subtitle. No, I haven't gone crazy Let me explain why.

First some facts:
- The vast majority of movies and tv-shows are English spoken.
- Usually, English subtitles are the first subtitles to become available.
- These English subtitles must be 100% verbatim: everything what is said, must be subtitled literally word by word.
- Any translation machine will attempt to translate everything 100% accurately.

And this is exactly where it goes wrong.
A small example.

We take the official retail subtitles uploaded by English admin Scooby74 for the movie "Plane" (according to official English subtitling guidelines), with a visual of lines #602-607 (approx. 0:37:10 and following):

English verbatim subtitle:



But... in this we find some 'problems':

The most important one is reading speed. For all these lines the reading speed is too high. Line #605 is even extremely high: 68 characters / 14 words in only 1,7 seconds. This is two times faster than what an average human being can read and three times too fast if you also want to be able to follow what's actually happening on the screen.
In other words, the text will disappear before you are done with reading and the viewer is stressing out trying to follow the movie.

There are also some more technical problems, but while those definitely reduce the reading & viewing experience, they are not really crucial, so let's ignore those for now.

Now, a good subtitler will not only translate, but also solve this reading speed issues by 'compressing' the text: rewrite the text as short as possible, while keeping the essence of what is said.

The result, translated back into English, looks something like this:

Compressed and fixed subtitle:



This is what the subtitler (text-wise) did:
- Compress the text of subtitles 602 and 603 and merge them into a single subtitle.
- Compress the text of subtitles 604 and 605, merge them and adding a dialogue hyphen.
- Compress the text of subtitles 606 and 607, merge them, simplified the idiomatic expression "You know the drill.", remove the meaningless and mumbled "Got it." and remove the dialogue hyphen.

I don't see any Artificial Very Intelligent Translating Machine doing all this text editing and merging in the near future. Until then, it will only stick to pure and literal translations. And so, the result will vary between at least poor and even useless subtitles.

Conclusion:
Machine translations of English subtitles of English spoken movies will result in very poor subtitles - or worse.

I understand: it would be great if we need just one English subtitle and we can then generate subtitles in 250 languages without any human interference. This is the wet dream of every subtitle website programmer. But it's the nightmare of every movie lover.

[oss]

the most important question is - can you create as translator good subtitle without viewing the movie and seeing what is happening on screen ? I dont think so. So thats why AI will lack good translation for now.

For compressing the text and stuff - well, you can just tell the boundaries to AI and it will try keep it, really - try it. I even try to translate some music songs and told AI to rhyme translation, of course it was not best, but good enough.

AI is doing some amazing stuff, it is good to at least try it.

chat gpt.

[rasubet]

You should check TransPerfect which creates the subtitles based on your source content, ensuring the timing and synchronization of the subtitles are accurate. They can handle different subtitle formats and adapt to your specific requirements.

[oss]

we have this on https://ai.opensubtitles.com already

[hendrikeduard1]

I don't have the title "Subtranslater" but so far I have translated 83 subtitles from English and French to Dutch.
I feel like I'm becoming redundant due to the promotion of AI.
I have my doubts about this phenomenon; currently I am translating a very difficult title.
Many times the durations in the English file are incorrect. Sentences of 90 characters are pasted in 1.5 seconds and sentences of 15 characters in a time of 35 seconds!
So my job is to rewrite half the script and ensure that the reading speed is acceptable. My goal is 16-18 CPS with a max of 20 CPS.
Furthermore, not to keep the gap between the titles too small, etc.
One thing that AI probably doesn't do is check the movie to see if everything is correct and also translate the titles that appear on the image, such as translations from a foreign language, i.e., so text not included in the original sub file.
My humble opinion is that to make a good subtitle, about 25 hours is not unusual. AI may translate perfectly in 1 minute, but can never adapt the texts to the reading speed and keep the content relevant to the film.
If AI is going to be used from now, I'll finish what I'm working on and stop doing it....

[oss]

thats why we REQUIRE any uploads which are AI generated to mark it so.

If admin allows that.

It is always big discussion, there are users who saying even not perfect subtitles are better than none. And then there is other users and views on this point.

We try to be neutral.

[goodfella]

I would like to recommend you to try this product, and I am the developer :
https://blog.opensubtitles.com/opensubt ... h-quato-ai

If need any help, just notice me in PM.

[ai.opensubtitles.com]

Hi, I'm the developer of AI.opensubtitles.com. It is an online platform that allows users to transcribe and translate their audio and video files using artificial intelligence. Users can upload their files and choose from a variety of languages and different AI models to provide fast and accurate results. The platform charges users based on the number of credits they use, which depend on the AI model and the length of the file. Users can buy credits that never expire and use them whenever they need them.

https://ai.opensubtitles.com/