The new translated video caption function is great and so useful that it's so fast, we really appreciate this function! We're a global organisation and when testing this function, we have had some international feedback that captions aren't as accurate as they could be. I totally appreciate that this is in BETA and that language is incredibly complex. However, from the feedback we’ve had, it seems that the AI produces English captions (the language the video was recorded in) then splits sentences to fit the right number of words on screen at once. For the translation, it appears that the AI then translates each part of the sentences in isolation. This is an issue because for non-European languages like Japanese, you can completely lose the meaning of the sentence due to the word order (in English, the word order is “Subject + Verb + Object”. But in Japanese, it is “Subject + Object + Verb”). Our Japanese colleagues reported: 20-30% of the subtitle looks fine and makes perfect sense, another 20-30% makes no sense and, like "What the heck?", and the remaining 40-60% is good enough, like "Sounds funny but I can understand what is to be meant". With this in mind, I would suggest a better process would be to produce a transcript of the video when using this function, then translate the transcript (so that sentences aren’t split and meanings lost) and then insert captions from the translated transcript. I appreciate that getting the timings correct using this method could be difficult, but I think this will result in greater accuracy of translation. I hope that this feedback helps and thanks again for creating this function, it really is incredible and will save us hours of work when it’s more accurate!
AdChoices