argia.eus
INPRIMATU
Automatic translation subtitles, new tool, some tests
  • A Basque hacker attempt, Zitale, has put on the net a practical tool for translating subtitles (in SRT format) and other elements, which relates to the neural translator of the Itzulita government. We were informed that the tool was there and offered us the opportunity to test it: we have conducted three tests in three languages (English, French and Spanish) with the films Titanic, Navalny and Blackfish (one historical fiction and two documentaries) and we have made the results available to everyone so that each one can draw their conclusions.
Azpitituluak Euskaraz @azpitituluak Sustatu 2023ko ekainaren 28a

The tool is https://itzultzaile-neuronala.zital.freemyip.com/

If you open a SRT file, you should copy its content to the left reddish part, removing it above the local text:

Select the translation address and press. Quiet, it can take between 5 and 10 minutes. The result will finally appear on the right side with the same synchronization and with the review and editing option.

To test in practice we have taken three films. The 1997 Titanic classic, remembered this week for events, has been translated from a Castilian TRS. Blackfish's 2013 documentary on orcas or ezpalas dams has been brought from French. And we've come back from English Navalny, the Russian policy that was awarded the best documentary in the last Oscar.

In the three cases, along with the film profile, in Azpitituluak.eus we have included the file in original Spanish. We have uploaded them without amendments, nothing has been touched, because we have decided to share the dream with experimental intentions.

The tool works correctly, it is very practical in this sense. As for translation, Translate is not generally ready. The result requires a lot of corrections -- it's generally understandable, a little bit more for documentaries, but it's not at all correct. A human translator should review the result from the top down. How much work is that? Does it save you 30%? We don't know. It also seems dangerous to predict that "this saves translators work", as it risks that the presence of automatic translation is an excuse for reducing working conditions. On the contrary, we believe that with these three examples, in any case, they demonstrate the opposite in 2023: human work is essential to translate the subtitles. To cite a couple of examples, the English plural You is always unique from Itzuli, and in the documentary on orcas (from French) you will not read orcas or shipyards, but killer whales, orcas (sic) and epaulard without translation...

As for the tool, congratulations to Zital. Here's the free code in Github. We are impaired by possible improvements, especially some additional possibility for conservation and treatment when auditing the results -- the code is open to potential developers.

Moreover, there is an interesting tool https://itzultzaile-neuronala.zital.freemyip.com/ not only to translate SRT subtitles, but also a large-format text, XML code and .po localization files. Thank you Zital.

Finally, a few frames:

Killer whales, killer whales.