Contact Me
If you have any questions, comments or suggestions please fee free to share your feedback by filling the form below in Feedback & Comments section.
Note: By default all of your comments and feedback are made public and visible on this page. If you want to share something in private, you can email me at gakbar {at} gmail {.} com, or you can mention this in your comments text.
Hi Akbar, translating an Italian source file to German and English with OpenAI, I got the following error message: “Unfortunately, the translation process failed with the following error:
Error: This content doesn’t comply with OpenAI’s usage policies and thus can’t be translated. Moderation Category: hate/threatening”
Of course, the source file doesn’t contain any hate or threatening speech. There is also no problem translating it with Deepl or Gemini. Just OpenAI pulls this error message. Please, feel free to have a look at order #W3dA2uXcEw, gd5LlwmHHM4P, WRJgT9L76s. None of them were charged. Thanks!
Thanks for getting back to me and sorry for the trouble there.
Unfortunately, OpenAI is very picky about what content it process and sometime even a single word or sentence would trigger its moderation filtering. I think in this case, it’s tripping over the subtitle line#9 which is just a bit hard language/tone.
It’s annoying for sure, but this not something under the direct developer control – you can’t bypass these moderation filters or risk being flagged by their system for violating their usage policieis.
Gemini also has a moderation layer, but it’s far more lenient, and I don’t think DeepL performs any moderation at all (at least not officially). Which explains why you were able to translate using these tools. So I guess when you run into any moderation issues, the only workaround is to try other translators. I hope in the future they add improve their moderation policies to also take into account the context and type of dialog to make it better.
Thank you for your understanding.
Regards, Akbar
Thank you very much for your feedback and considerations! Good to know about these restrictions in the case of OpenAI! I’ll stick to Gemini and Deepl in this case. I could try to eliminate the word that’s probably provoking the moderation, but I really do not have any hence which word or sentence it could be. Line #9 seems pretty normal to me, a sentence about “Kindergarden”… Anyway, thanks for your reply!
Hi Johannes,
Sorry, I was wrong! Doing some more experimentation, it looks like it’s actually some content between line#19 to 25 which is triggering that moderation category. Though that said, I found that in this case it was actually tripping due to a new moderation category “hate/threatening” added recently by OpenAI, and which I was not fine tuning properly. I have now applied that filters and hopefully it will now let you translate using OpenAI as well.
Thanks you for bringing this to my notice and for patience while I keep improving this tool.
Regards, Akbar
I am having a similar problem to others. Using gemini, paragraph mode some lines are not translated and shifted. Looks like the model is still tripping some orienting paragraph info, order 7NjYPjuNkXXXX
Hi Patrick,
Thanks for reporting this back and sorry for the trouble. Yeah, sometime these models get a batch of the lines so wrong that it’s completely messed up. Unfortunately, there is not a simple workaround to avoid this as due to some inbuilt randomness in the model, it removes the line/row markers, thus making it very hard to restore the correct line/timings back.
The only workaround/fix is to manually review the translation and then shift/re-translate line batches. You can read more about these here:
https://blog.syedgakbar.com/subtitles-translator-help-tips/#retranslate-block
Regards, Akbar
Hi Akbar! Translating from Italian to German with Deepl, I run into some issues: The translator doesn’t translate line per line when “paragraph” is selected in the configurations. Often, he translates several lines from Italian into one single in German. Consequently, leaves lines empty in the translated file. Neither does he keep the line breaks in every single line I checked this twice with the same source file and got the same results, see order #X3yFvDBbrXXXX and #pDcSnZ07XXXXX). Translating the same source file from Italian to English with Deepl, works just fine (#BbTX5aCFXXXX). Why does Deepl present this strange behaviour just with Italian to German, the “mother language” of Deepl?
Hi Johannes, thanks for trying these out and sharing the issue with some example – it was very helpful. Apparently, DeepL translator was skipping/ignoring the splitting tags which were in between a sentences when processing certain language pairs. I guess this problem started happening with the latest language model optimisations. Thanks to your good example, I’m able to further tweak the splitting logic on my end to improve the translation in the paragraph mode for DeepL translator. This now seems to be working reasonably well for me. Can you please give this another try when convenient?
Regards, Akbar
Hi Akbar! Thank you very much for your reply and your efforts to tackle it down.
I checked it with the original Italian subtitle file and also with another one in Italian. Really, I think you solved the problem! It’s working perfectly now. If you’d like to have a look at this two last translation tasks, here are the order numbers:
#BPlgZVrXXXX
#WLMegfgXXXX
Again, thanks a lot. Really great work your site!
Kind regards.
Johannes
Thanks Johannes! I’m glad it fixed the paragraph issues very well.
Regards, Akbar
Akbar, I have to correct myself: Concerning the splitting logic in paragraph mode, it really seems to work well now (Italian to German), but Deepl apparently still ignores line breaks inside the paragraphs (if a subtitle have two lines). The line break tag \N get lost in the translation files in case of German. If translated Italian to English, it’s a bit better, but sometimes Deepl keeps the tag, sometimes only part of it (only the back slash \) and sometimes ignores it completely. Also hyphens are sometimes ignored (in case of dialogue). See line 74 in order #BbTX5aCXXXXX (this order contains two translation tasks, one to German and the other to English, see the English one).
Thanks Johannes! This is an interesting one. There are two problems here. First is that I’m not sure that “\N” for line-break is the standard SRT specification. I tried exploring various standards and extensions of the SRT format, and none clearly documents this line splitting behaviour:
https://en.wikipedia.org/wiki/SubRip
That said, if the DeepL doesn’t handle this line-break correctly, I’m not surprised. It can be only configured to ignore certain types of XML tags – but not any individual string or character patterns. So I’m out of luck there as well.
However, I went with a intermediate workaround for now and that’s I’m replacing the \N with actual line-break character before translation. This seems to keep the DeepL translator happy and it brings in correct translation in most of the cases. The only concern I have is that it may cause problem for some other content which needs this \N character in place, but I guess I can see when that’s needed and then use any other encoding technique. For now this will at least get you some good results.
Thanks again and as always if you run into any further issues, happy to help.
Regards, Akbar
Great, thanks a lot for your workaround! About the “\N” question: I do not know either what’s really the standard specification here, Aegisub is using this string for line breaks, other tools like Tero Subtitler or youtube do use some line break tag, but not visibly, so I do not know what’s the code behind of it to keep manually set line breaks in place. Meanwhile, I guess, for my purposes, concerning the Italian-German language pair, Gemini 1.5 Pro gives the best translation results in accuracy and also fidelity. It’s really amazing to see the level in which Gemini 1.5 is considering inherent context of the source text. GPT 4o seems to be on a similar level. If I would do an actual ranking list for myself, Deepl would get on the last place…
Thanks Johannes for the feedback and sharing your thoughts. Yeah, it’s amazing to see how the machine translation have evolved – especially since introduction of the large-language-models in the recent years. When I started working on this hobby projects nearly 10 years ago, the translation for most of the Asian language was really bad – it was mostly word-by-word translation, without keeping any context and/or language flow. And now we have come to a point where these machine translations are almost 80-90% accurate for most of the populate languages. And I can only see these improving insanely fast now. A bit scary, but this is an era of AI 🙂
Regards, Akbar
hello akbar,
i am enjoying translating using your subtitle tool.
a little problem though.
i am translating english to hebrew. and the second line in the translation does is not so successful like the first one-Punctuation marks such as exclamation marks, periods and commas are not where they should be – from right to left, but appear in reverse from left to right.the first line is always ok!
would love your help with this.
thank you.
Hi Effi, thanks for the feedback. Regarding your issue, can you please confirm if the “Embed Language Bi-Directional codes” is turned ON you in the settings (upper right-corner of the web-page)? Also, I would suggest, if not already doing, is to change alignment of the translated subtitles to right. If still run into any issue, a 2-3 line samples of it would be helpful in isolating and fixing the problem properly.
Regards, Akbar