[REL] Ko zorijo jagode (1978) [Slovenia]

[David32441]

WEB-DL 1080p (Slovenian audio/subs)


[File]

Wow, thanks!

I've tried translating the Slovenian subs using Google Translate (cringe!), DeepL is better but I can't find a way to translate entire .srt files without giving them my credit card for a "free trial". Screw that. I'll have to look for a better site than the ones I've found so far. Suggestions anyone?

You've posted some great 1080p upgrades bloodhound1st. I don't suppose you'd have either Nie bede cie kochac (1974) or a full sized version of Meztelen vagy (1972)? I'm guessing not, but it never hurts to ask.

The old fashioned way is copy paste 50 subtitle lines at a time into Google translate page. Then it's free.
Subtitle Edit has a built in translator that seems to work now for SRT language conversion without having to go 50 lines at a time.
I've forgotten the link of the website where you could drop in mp3's and it' would generate SRT files after sitting in a queue for about a minute. Max audio length was 1hr. You could convert a films audio to WAV. Split in Audacity into <1h and then encode mp3s then drop them in it would save you some SRTs. Then you could open the 2nd file say "add 1h to all times". Combine the 2 files in notepad. Resequence the numbers in Subtitle edit and you have a combined SRT file.

[deadman]

The old fashioned way is copy paste 50 subtitle lines at a time into Google translate page. Then it's free.
Subtitle Edit has a built in translator that seems to work now for SRT language conversion without having to go 50 lines at a time.

That's how I got my original cringey subtitles. Subtitle Edit can apply Google Translate to an entire set of subs in one shot.

I managed to wrangle ChatGPT into translating the whole thing for me. And it did require wrangling. Not to mention multiple free accounts, because they limit how much you can translate at a time if you ain't paying. Anyway, I cleaned up the resulting file a bit and made a copy without the SDH parts. For those who don't know SDH subs are the ones with stuff like [phone ringing] and [sad music playing], for the hearing impaired. I'll put up both versions.
 
 

[Night457]

Thank you deadman, and congratulations on your success with The Chat!

[pillowbaker]

On the topic of subtitle translation pillowbaker, can DeepL be used to translate an entire SRT file all at once WITHOUT having to give a credit card number? Deadman is interested if so. Please see this thread here and see if you can give advice:
viewtopic.php?p=116451#p116451

Well, I wish I could tell you a special workaround to the deepL issue. I havn't found a way around the 50 lines (or 5000) characters rule. In fact, I've maxed out deepL daily allotment several times during my work on the Mariken novel. And that's only using it as a reference. I have yet to give a credit card to anything, though I'll likely try paying for that syedgakbar.com tool.

The way I use deepL is as a reference of comparison between one or two of the GPTs. I have found that it occasionally produces superior (or more pleasing) wording, though not necessarily meaning. At this point, though, I am not sure if GPT has any better experience when trying to translate from languages like Slovenian, Slovak, Hungarian and Latvian. DeepL might be better in some of those languages than GPT, but I really cannot be too sure. DeepL as a reference helps me identify problematic passages during translation, and it's ability to highlight words and provide a very basic dictionary helps me locate words to look up elsewhere on the internet (HiNative, bab.la, reverse contexto, etc). It also sometimes helps narrow down idiomatic expressions.

[deadman]

At this point, though, I am not sure if GPT has any better experience when trying to translate from languages like Slovenian, Slovak, Hungarian and Latvian. DeepL might be better in some of those languages than GPT, but I really cannot be too sure. DeepL as a reference helps me identify problematic passages during translation, and it's ability to highlight words and provide a very basic dictionary helps me locate words to look up elsewhere on the internet (HiNative, bab.la, reverse contexto, etc). It also sometimes helps narrow down idiomatic expressions.

It would be interesting to find a systematic study of different translators, including DeepL and ChatGPT, using many different language pairs. GPT tends to produce fairly nice sounding translations as far as I can tell - which are occasionally quite wrong. Its ability to concoct bullshit that sounds good is one of the chatbot's main flaws generally imo. They're improving it, but still I've seen it reference studies and books that don't exist. Quote famous people saying things they never said. Stuff like that.

Idiomatic speech is always tricky. 1970's Slovenian teenage slang probably isn't well documented enough online for any translator to recognize expressions and avoid translating them literally.

[goku33]

At this point, though, I am not sure if GPT has any better experience when trying to translate from languages like Slovenian, Slovak, Hungarian and Latvian. DeepL might be better in some of those languages than GPT, but I really cannot be too sure. DeepL as a reference helps me identify problematic passages during translation, and it's ability to highlight words and provide a very basic dictionary helps me locate words to look up elsewhere on the internet (HiNative, bab.la, reverse contexto, etc). It also sometimes helps narrow down idiomatic expressions.

It would be interesting to find a systematic study of different translators, including DeepL and ChatGPT, using many different language pairs. GPT tends to produce fairly nice sounding translations as far as I can tell - which are occasionally quite wrong. Its ability to concoct bullshit that sounds good is one of the chatbot's main flaws generally imo. They're improving it, but still I've seen it reference studies and books that don't exist. Quote famous people saying things they never said. Stuff like that.

Idiomatic speech is always tricky. 1970's Slovenian teenage slang probably isn't well documented enough online for any translator to recognize expressions and avoid translating them literally.

Deepl‘s AI also was always strictly trained with the purpose of translations. Gpt was fed the whole Internet..

[Night457]

So if "GPT was fed the whole Internet", does that explain "it's ability to concoct bullshit"?

[goku33]

So if "GPT was fed the whole Internet", does that explain "it's ability to concoct bullshit"?

Very much so. It wasn’t fed too much curated content. (I’m only making assumptions)

Deepl probably god fed proper tests, curated. Gpt has read research papers but at the same time also Reddit and 4chan lol

[Night457]

Reddit has actually turned out to have more useful information than any other site for me lately! I am sure there is lots of garbage there too (I don't have an account). When I desperately Internet-search trying to find a quick answer about how to do something fairly simple with software - particularly command-line software - the answer is invariably on Reddit. It saves me hours of getting bogged down reading the pages of README files on Github and becoming an expert power user when I only want to do a few things. The power users are willing to share their knowledge on Reddit, thank goodness. I realize this has nothing to do with translation, but then GPT does more than just that. I would rather read the original cranky computer programmer's post on Reddit than an AI regurgitation of it anyway.

[goku33]

Reddit has actually turned out to have more useful information than any other site for me lately! I am sure there is lots of garbage there too (I don't have an account). When I desperately Internet-search trying to find a quick answer about how to do something fairly simple with software - particularly command-line software - the answer is invariably on Reddit. It saves me hours of getting bogged down reading the pages of README files on Github and becoming an expert power user when I only want to do a few things. The power users are willing to share their knowledge on Reddit, thank goodness. I realize this has nothing to do with translation, but then GPT does more than just that. I would rather read the original cranky computer programmer's post on Reddit than an AI regurgitation of it anyway.

Very true and the other day I was asking the Ai for the name of a show I was searching from my childhood and it literally made up a name. The title didn’t exist anywhere. So I asked where they got the name from and they apologized for the mistake lol

Reddit is still very useful and You can be sure ai companies scraped it a lot for their models. But it’s a shame Reddit has changed so much (and crippled their APi)