Hanashi marches on
-
I haven't read Hanashi's Tsukimichi releases myself, but a friend does and has had similar complaints, like names changing between volumes and characters doing things they should not be capable of doing (kneeling next to themselves) which does feel like a human did not review the text.
As far as the release pace, Seven Seas used multiple translators to get Mushoku Tensei out faster, but that had similar consistency problems with names changing between volumes. The volumes I read were otherwise largely fine and did not cause me to suspect any use of MTL/AI. In other words, it's possible Hanashi is also using multiple translators, but I don't think anyone external to Hanashi would be able to answer definitively.
-
@unknownmat the only thing I’ve read from them that felt like AI were the broken first releases of Another World 3-4. For Tsuki the mistakes seem no worse than many Yen titles aside from one volume where they swapped the hero names a few times. Pet Saintess had that English as a second language thing where people use proper names in place of pronouns - “does Steve like the tea?” When talking to Steve.
-
@HarmlessDave Fair enough. I tend to agree. I'm overall happy with Tsukimichi and New Gate, and in isolation would not suspect them of AI.
Do you have any thoughts on the translation pace, though? I'm pretty torn about this. I really like how frequently we get new drops, but I just can't see it being possible for a human translator. If they have multiple translators as suggested by @myskaros, then why not attribute them in the copyright page or on the website?
-
@unknownmat - my wild and evidence-free guess for the speedy pace of Tsukimichi releases is they found a translator who is working on it full-time as their main job instead of as a weekend side job or as one of several titles.
I'd guess they've found some contractors who are hungry for work and willing to meet 3-month cycles on the translations for other titles, and that Hanashi might be rolling the dice a little by not being as careful about polishing the translations as JNC is.
If so, that might not be sustainable over years if they burn out their translators, so gradually slowing down releases from 3 months to 4 then 5 might make sense.
That's all guessing, I'm a software developer not a publishing insider :)
Back to AI for a moment, I'd be very surprised if AI can do light novel translations of this quality - not for the sentence-by-sentence generation but for maintaining context within and across the novels. If such an LLM exists, I'd pay good money to have someone point it at the Deathbound Duke's Daughter web novels :)
-
@HarmlessDave said in Hanashi marches on:
I'd be very surprised if AI can do light novel translations of this quality - not for the sentence-by-sentence generation but for maintaining context within and across the novels.
That's actually pretty convincing. You're right that Tsukimichi would require a monster of an editor at least to maintain the current level of quality given the typical LLM output.
-
@unknownmat As one of the faster translators in the business, I translate 100 pages (of source text) a week. An average length light novel is 300 pages, so one a month is easy. If that's your only/primary series and you're full time, even translators working at half that speed can easily manage an every other month release.
-
@doceirias That's an excellent data point, thanks.
Having no other basis for comparison I guess I considered @Quof's translation of Bookworm to be the gold-standard for quality translation pacing, at roughly one volume every eight weeks. I had thought that anything beyond that was either unsustainable or superhuman. My mistake.
Also, since I read on my Kindle I don't really have a good sense of novel length. I'm often shocked by a novel's thickness when I see the print version. It's very possible that a typical volume of Tsukimichi is shorter than a typical volume of Bookworm.
-
@doceirias Incidentally, and surprisingly apropos to this discussion, I noticed the next Death March finally got posted to the Yen Press site. I'm very happy to see that it's continuing, I truly am. But it's not coming out until September! Arg! I'm going to be retired before Death March gets fully published in the US. I think you could get away with translating just one chapter per month and still easily outpace the YP release cadence.
-
@unknownmat Yeah, I really have no insight into what's going on with their scheduling. All I can really say is that translators are pretty much always a couple of volumes ahead of release dates unless we've actually caught up.
-
@unknownmat I think there's also an element of how many names and terms you have to keep track of. Bookworm has a lot of characters and a lot of names, so the translator would naturally need to slow down somewhat cause they'll have to cross reference their rosetta stone to keep the names and terms of lesser used people and things in line across all the books.
Plus the amount of time the editor spends going over things could differ wildly between publishers
-
@unknownmat And so my legend dies...
Translation speed is something that can be pushed to pretty significant extremes. I once talked to a translator - a good one, too - who could put on a League of Legends stream on one monitor, bring up a spreadsheet on the other, and then translate for 16 hours straight with no loss of focus or being particularly bothered by the experience. He could translate a light novel in a week without trying hard at all, and for shorter ones finish them in a few days. My 8-weeks for Bookworm was laborious, but nowhere near the peak of human achievement. The human limit is somewhere in the span of hours or days, definitely not weeks.
Page count is pretty unreliable in general, though. (And I'm not saying this to cope). Publishers often need to "correct" page count based on character count to get true length. Like some series are just written in a way where a page will have 2x as much text than another one. So two books could both be 300 pages, while one is really airy while the other is dense enough to "actually" be 600 pages. If one wants real concrete information, you have to go by character count, which is the only impartial and immutable figure out there. (There can be some squabbling about whether to include punctuation, but that's like a difference of +/- 0.5%, rather than potentially +/- 50%. ) Bookworm is one series where the text is rather on the dense side and 300 pages is counted more like 500 due to character count. (But I say that just to be informative; not to compare to any other series in particular.)
Then there's the matter of textual difficulty. Some series will be harder to translate than others. And to be clear, I would often put Bookworm as a series on the easier side to translate since the author has a fairly dry, efficient writing style. The difficulties I had with Bookworm were often more psychological than anything to do with the text itself - e.g., translating the same tea party scene from 3 different perspectives while none of the speakers are clearly tagged or discernable. Anyway, maybe the cliche example of a hard LN to translate would be Monogatari or something. Translating "100 pages" (or lets say 30,000 characters) of Monogatari in a day would be QUITE different from translating 30,000 characters of a more innocuous text. You can force the same speed, but the strain will be a lot different. I translated Lazy Dungeon Master and Bookworm at the same pace on a week-by-week basis, but I always finished LDM much easier and quickly than Bookworm. If I had to translate something a lot harder than Bookworm I would probably translate it at the same speed overall, but each day would take longer.
Anyway all of this is to say that TL speed is quite a nuanced issue. It can be pushed to unimaginable lengths by the right person in the right position, the average length of a page varies by work, and finally, the difficulty of the text has an impact too. I personally don't think my Bookworm speed was particularly impressive, since I know there was so much more I could have been squeezing into that space (after all, I never forgot that guy who was translating entire books in 3-4 days), but if it had a virtue it was maybe just consistency rather than speed. The consistency to translate at a bit of an above-average pace on the same series for six years straight. The same goes for Kier there, too. But forgive me for expounding so much on the matter.
-
@Quof said in Hanashi marches on:
Thanks for the thorough response.
I guess my main take-away from this discussion is that the translation pace of Tsukimichi at one volume per month is well within the bounds of human ability. My fears of LLM augmentation are unfounded, particularly given the consistent quality of the translation and lack of other AI-indicators.
And so my legend dies...
You have nothing to worry about. Among the Bookworm fandom your legend endures.
I obviously can't dispute anything you're saying, but I don't think I'm wrong to observe that the pace of Bookworm translation was very high by industry standards. Excluding Tsukimichi and New Gate (two recent Hanashi titles) it's the fastest I'm aware-of by a significant margin.
Maybe I'm being unfair to Hanashi Media, but I think my concern that they're cutting corners is at least understandable, given how much of an outlier they are. This is somewhat hypocritical, I realize, since I'm also more than happy to complain about a slow release cadence.
... So two books could both be 300 pages, while one is really airy while the other is dense enough to "actually" be 600 pages...
Interesting insight into the differences between Japanese and English. Is it something like writers who use lots of five-dollar words versus something more plainly written?
... if it had a virtue it was maybe just consistency rather than speed. The consistency to translate at a bit of an above-average pace on the same series for six years straight. The same goes for Kier there, too.
Yes, completely agreed. The sustained consistency was definitely noteworthy. It helped to foster an entire community with "Myne-day" being a weekly event we all looked forward to. It remains to be seen how sustainable Hanashi's translation pace of Tsukimichi and New Gate are - I admit that my concerns above are based on a very small sample size.
I also don't mean to undersell @Kier's contributions. For all I know the editor might have more impact than the translator. His contributions are just less visible to me.
