Tristan “Akada” Grant is Charmed The Rise of the Bookworman isekai light novel series currently being translated Japanese novel club. But when Volume 4 Part 5 ended on a cliffhanger, Gallant couldn’t wait for the official release. Instead, he inserted the next volume into ChatGPT.
© 2024 Gong Yishu
Galante said in a statement that the final text was “readable and, for the most part, understandable.” Glass Reflection YouTube Video This experience provides a hilarious illustration of ChatGPT errors and ignored instructions. “Do I regret it? [it]? Not at all,” he said. “But if I had to choose… I would still choose human translation without hesitation. ”
There’s no comparison: human translation is more accurate, more detailed, and, humaninstead of ChatGPT. But when fans like Gallant need to choose between a “correct” translation and a “now” translation, they sometimes choose machine translation. As machine translation tools under the artificial intelligence banner become increasingly accessible, the light novel landscape is at a crossroads. Will light novel publishers sacrifice their products for faster speeds? If so, how much was sacrificed?
One of the most popular terms in the field of light novel translation today is machine translation post-editing (MTPE). With MTPE, machine translation tools generate raw translations that are then smoothed and localized by human translators. As an insider at a major light novel publisher told Anime News Network, “Of course, we are working on MTPE, everyone is working on it.” But is this true? More importantly, do AI translation tools really work?
MTPE: A happy medium?
Artificial intelligence (AI) has become a hot issue in the translation world. But what we call artificial intelligence is actually a number of different technologies. Currently, different companies in the animation, manga, and light novel industries are adopting different types of AI, but with mixed results.
One of the most controversial technologies is large language models (LLM) like ChatGPT. Users enter a prompt into LLM, such as “Please translate this light novel into English,” and LLM responds by searching the Internet for an approximate answer. recent, crunchy Viewers were left in limbo when they noticed the film’s German subtitles Necronomicon and Cosmic Horror Show That Seems to be cut and pasted from ChatGPT.
Machine Translation Post-Editing (MTPE) is considered a step in between. MTPE can be described as a partnership between a human translator and one or more artificial intelligence tools. One of the comics publishing companies most vocal about using MTPE is Orange Company (Do not compare with orange studio of beast star and Three gun attack). Chatting with Deb Aoki on Comics BeatRei Kuroda, Vice President of Product and Head of International at Orange Orange Company Several types of translation processes are used, one of which sounds a lot like MTPE:
“In some cases, we start with an AI-generated translation, which a human translator will review as a starting point and edit or adjust if needed.” Kuroda immediately added Orange Company “Never publish any translations generated solely by artificial intelligence.”
Kuroda’s strong emphasis on this sentiment shows that anime viewers, manga and light novel readers view AI tools negatively. When anime news network conducts A survey on the use of artificial intelligence in animationOf the 550 respondents, 68% said they “do not accept at all” the use of artificial intelligence for manga or anime translation. Concerns include mistranslations (90%), loss of cultural references (83%) and human translators losing their jobs (79%).
“Fans won’t pay for something they can already get for free,” says comic translator Katrina Leonudakis. “They are paying for the quality provided by professional translators who understand the words, paragraphs, culture, narrative style and expectations for both Japanese and English writing. If LN Publishing wants to continue to survive in this market, they need to continue to use human translators, even if takes longer and higher cost“.
In other words, it’s not readers clamoring for MTPE. Companies are looking for a faster and cheaper solution than hiring human translators. But can artificial intelligence be achieved?
“Machine learning – the real artificial intelligence behind artificial intelligence – has a lot of great uses, especially in medical and engineering fields. The problem for consumers is that it is used to cut costs and create mediocre products,” Leonoudakis said. “This is a best guess at best, and as we see in jokes on the internet, AI has a track record of guessing wrong.”
Light novel translation: How sausage is made
While Gallant is busy translating the next volume The Rise of the Bookwormwhat are the English publishers of this book doing? according to Sam PinanskiCEO and Founder of M12 Media Japanese novel club (JNC) stated that their internal translation process “originally was modeled not on traditional publishing houses but actually on the way anime subtitle groups were organized in the early 2000s” and has changed little since.
© J-Novel Club
“JNC translators submit a small number of manuscripts each week for 8 to 14 weeks,” says Andrew SchubauerProject Manager at JNC. “Every week, our editors fix grammatical errors, polish prose, and make the text sound beautiful under the supervision of a translator.”
Schubauer says the translation/editing phase of most books takes between nine and 15 weeks, plus an additional four weeks for quality assurance. JNC uses in-house tools, including file converters and spelling and grammar checkers, but not artificial intelligence.
Pinansky said that he has tried artificial intelligence tools in his spare time, but does not believe that current machine translation technology can be applied to the JNC workflow, because the JNC workflow relies on back-and-forth conversations between translators and editors. He believes that “sooner or later” fully automated translation will be possible, but it will require significant upfront investment.
“The main reason it hasn’t been implemented yet is probably that typing in 200,000 characters and outputting 100,000 words is not a profitable use case for large tech companies, where translating business PowerPoints one at a time is enough,” Pinansky said.
Pinansky said that once machine translation does become comparable to JNC’s process, the community will have some interesting choices.
“Will people accept a ‘good enough’ fully automated translation? Will they pay to buy the original Japanese version and translate it themselves? Will JNC eventually need to compete not with piracy itself, but with original online novels released by authors for free? Isn’t it better in the sense that original authors can now showcase their online novels to the world, not just Japanese readers? … If the inevitable decision from consumers is “we don’t need you anymore,” is it worth it for companies like JNC to pursue machine translation to compete?
Schubauer is not sure whether machine translation can compete with human translation.
“Light novels may not be high art, but that doesn’t mean they were created without love and care,” he said. “It’s an honor to be trusted to create a story that the author poured his or her soul into. Can a machine translate soul?”
Leonudakis said that if machine translation is meant to produce a “good enough” interpretation of a light novel, it makes no sense for a translation company to sacrifice human quality for speed because fans already know where to find it.
“Viewers who buy light novels know where to get light novels for free online. These ‘free’ English translations are often just Narou/machine translations of public chapters polished by savvy fans,” she said. “If publishers want to start using machine translation – even with MTPE – then congratulations. You’ve been defeated by the fan community.”
“How on earth am I going to compete with that?”
Of all the people I interviewed, both publicly and privately, for this article, the most optimistic about the capabilities of machine translation were none other than the translators of Machine Translation. The Rise of the Bookworm himself. WhatTrue to his pros, he said he’s well aware that fans will turn to AI while they wait for him to finish translating.
“This is a real, serious failure of humanity that took us 5 years to translate [Ascendance of a Bookworm,] A series of 33 books that the AI could do in just one afternoon,” he said.
What Said he wouldn’t accept the MTPE job because “I don’t think I have enough Japanese and English skills that MTL tools can currently improve my output. It would make my workflow more difficult because as someone who is used to delivering 97% or higher accuracy, I would feel the need to fix the AI’s bugs to that high of a standard, which would slow me down significantly.” But over the years, What He has a habit of checking his own translations using artificial intelligence tools. Google Translate never bothered him, but he was shocked by how well ChatGPT matched his work. Machine translation is far from perfect, but What It is argued that what readers are looking for is not a perfect translation.
“The hard truth that publishers know is that high translation quality is basically never a factor in product sales,” he said. “A particularly bad translation can be controversial and turn people away, but you don’t have a flood of buyers rushing to buy something because it’s translated well. You just need a decent translation.”
also, What Says a company has to make a big mistake to get readers to notice in the first place, pointing to the example of multi-billion dollar company Hoyoverse accidentally leaving information in Claude AI in the French translation of one of its flagship titles, Collapse Star Trail.
© Cognitive Circle
“What’s particularly noteworthy about these cases is how they would have been noticed without these obvious bugs. Users here only noticed ‘some recent quirks,’ rather than ‘a huge drop in quality.’ Until then, people must always question how much of the stuff they consume has been unknowingly replaced by artificial intelligence.”
Actually, to booksJapanese publisher The Rise of the Bookwormwhich began a partnership with artificial intelligence tool Corona EX last year. These AI-translated chapters are directly related to Whatlivelihood.
“I do think machines will take my job,” What explain. “Imagine a world where novels are released globally in every language at the same time, because obviously once a writer completes his MTL, he’s immediately given a translation into every language. How on earth would I compete with that?”
Despite scratching Gallant’s with an LL.M. The Rise of the Bookworm Out of short-term cravings, he ended up buying Part 5, Volume 5. At least for Gallant, there’s a clear difference between the AI translation he read first and the AI translation he read. WhatLater official version.
“Humanly translated version by What Clearly a better translation,” he said. “Not even close. even when using [machine translation] That’s what I thought, and that’s still what I thought. “
