French Amazon series pages do not generally have blurbs (<10%). So I have been prompted to dive back into blurbs and will make a few updates.
If there is no series blurb, we have the series be volume blurbs only.
For machine translation, we have been using DeepL. However, Claude Sonnet is 5x cheaper and generally on par or better in my experience. I’m planning on switching to Sonnet, as doing more volume blurbs makes DeepL untenable.
Machine Translation adds a large bolded text at the beginning saying ‘Machine Translation - Needs Review’. Unfortunately, this has proven to be too large of burden for the community (there are a lot of blurbs).. resulting in these tags never going away. I will be removing those tags and replacing them with an icon next to the text. I’lll also be handling the ‘review’ process more concretely - there will be a checkbox on the blurb to mark a translation reviewed.
I know that some folks are very against generative AI but I don’t really see an alternative here, outside basically not having any English Translation at all. Compared to the numerous other community tasks we have, blurbs don’t seem to be very popular to do
DeepL and other traditional translators also make silly mistakes, especially around names and pronouns, so it’s not like they are perfect. Hopefully the gen AI doesn’t hallucinate a plot summary that doesn’t exist. If you let us know when it goes live people will probably keep an eye out for a bit and see how well it’s working.
I’ve fixed a few obvious mistakes here and there, but translation is hard. I’m not surprised that very few people (myself included) want to write a translation from scratch.
Yes it should be going live very soon, probably tomorrow. And I’ll still surface it was done by claude, just as an icon outside the actual text. I’ll keep you all updated.
One very small change that would make editting blurb translations easier is displaying the original blurb while editting the translation. I’ve come across a ton of DeepL blurbs with small, obvious errors I could fix, but have put off because I want to make sure I’m translating the original accurately, and having to flip back-and-forth between the English and the original to check the wording is a pain.
I don’t think this is surprising since translating or correcting AI translations can take more time and effort than other tasks, and requires a certain level of confidence from the user in their language skills.
Blurbs are also generally used before reading the actual book, when a user doesn’t have the knowledge to gauge if it’s accurate, and most users who have finished a book aren’t likely to look at the blurb unless they’re specifically interested in correcting it.
Questionable AI translations are better than nothing.
This is something I’ve noticed with Korean books, too - often blurbs can be huge, but have little to no information about the book itself, but rather details the author’s achievements, the prizes the book has won, praise from critics/reviewers, and so on.
If there’s an official English translation, that can easily replace this kind of useless info, but if not, the user has to compose their own blurb from scratch, which isn’t particularly appealing IMO.
If I recall correctly, book blurbs need to be original-language and official (from the publisher or a bookseller or something), and the English version needs to be either a direct translation of the original blurb OR an official blurb from an English version of the book. The blurbs need to be sourced to somewhere, so you can’t just make up one to add yourself, and I don’t think you can add one only to the English side if there is no blurb on the original-language side.
But if the blurbs aren’t useful, you can hunt down an original-language blurb from somewhere else that has more useful information and change it to that one. If you can’t find any good blurbs, you can always write your own into a review instead.
Yeah, that was the original guideline, however I wonder if it’s a little bit too stringent. We could trust people just a little bit more as long as they mark their source.
Something like - “we always prefer official publisher blurbs if you can find them but if they don’t exist or the quality is severly lacking, feel free to write your own. Do not spoil anything beyond the first few pages”
We do have full audit tracking history to monitor any abuse which there hasn’t been so far.
Are english translated blurbs necessary for anything higher than B1? At least on the romance language side, you might be able to skip an english blurb for anything not marked children’s book or comic.
Should we consider regenerating all the old blurbs which were generated a few years back and where there have been no human edits? Some of the old translations are terrible and I think newer LLM tech would be a big improvement.
Part of the reason I haven’t done too much blurb editing is that so many of them are so bad. For some of them my edit process was “throw away existing content, feed the JP original to a better LLM and then edit that”…
Also for books with no JP blurb could we rerun the “pull blurb from Amazon” process? For instance おまえさん 上 | L30?? has a blurb on Amazon but not on Natively.
By the way, do you guys also remove the “first volume” and “contains xy special illustration” stuff that Amazon blurbs often have. I often cleaned up everything that wasn’t directly related to the plot.
I think outright removing those when we see them is pretty reasonable.
I would personally recommend against people writing their own blurbs and instead stick to translating the original language’s blurb. This is more difficult for sure, and the original blurbs aren’t always good, as you said. But people have wildly different tolerances to spoilers, so I prefer to stick with what the publisher wrote. Also, I’ve seen user written blurbs on IMDb and they are often pretty terrible.
I think the risk of spoilers is too great to allow this. Or simply people including their opinion in the premise instead of an objective description of the book. Not maliciously of course, it’s just that people’s opinions and biases tend to leak in if they’re not especially careful.
Yes, I agree. I’ve definitely seen some that I considered editing, but they were so far off I didn’t bother.
I think that would be nice, if it can be easily sourced.
Thanks for putting the original blurb as reference when editting, it makes checking the English much more friendly!
Just quickly tried fixing one on one of my books, and my only note is that it’s a bit odd that the flag goes from saying “Machine translated, needs review” to just “Machine translated” after a human has fixed the machine’s work. Can I suggest “Machine translated, with corrections” or something to make it more clear that someone fixed it? Or if no edits were made when reviewing the blurb, maybe “Machine translated and checked for accuracy” or something, since the standalone “Machine translated” label doesn’t really suggest one way or another that anyone’s checked it.
Machine translated, needs review
Machine translated, human reviewed (no corrections)
Machine translated, human improved (corrections)
Just a suggestion - saying blurbs are community reviewed in general sounds fine, but an individual blurb it sounds like a group of people worked on it. I may just be being pedantic, though.
If you let us know when it goes live people will probably keep an eye out for a bit and see how well it’s working.
