Online literature has seen explosive growth in China in recent years. More than half of China’s internet population, equivalent to some 455 million people, are reading books online, according to the Chinese Academy of Social Sciences. More than 17 million people are publishing novels on the internet. Popular works span urban romance, history, science fiction and fantasy.

Chinese companies are hoping that level of popularity can translate elsewhere. And there are signs of potential: some Chinese web literature already has a cult following outside China, particularly the wuxia and xianxia martial arts genres. In 2019, there were 31 million overseas readers of Chinese web novels, iResearch estimated.

Not all of these readers are consuming AI-translated copy, though. Some wuxia fans dedicate themselves to manually translating novels and putting them up on sites like Wuxiaworld. Many take pride in the quality of their work, something that they doubt algorithms can replicate soon.

“Perhaps one day computer/machine/AI translation will be able to handle literature. But not right now,” translator Jeremy Bai said in an email. Bai has been translating works for Wuxiaworld under the nom de plume Deathblade since 2015, specialising in Chinese fantasy.

But as dedicated and passionate as translators like Bai are, manual translation does take a lot of time. And Chinese platforms need a much faster process for the amount of literature they aspire to put out.

That’s where companies like Funstory.ai come in. The start-up is working with more than 60 web novel platforms in China, translating and distributing their content globally on a variety of international platforms, including Amazon Kindle and Apple Books.

Tencent’s China Literature, which dominates China’s online literature industry, also offers its own AI-translated works on its English platform called Webnovel.

“For anyone working in publishing, the temptation is to entirely dismiss AI involvement in any literary endeavour, whether it is within the act of creative writing or in literary translation,” said Jo Lusby, co-founder and CEO of Pixie B, a Hong Kong-based consultancy that specialises in China’s creative sector. “The reality probably is far more nuanced than that. I can well believe that I could read something literary and not know that it was done by an algorithm.”

Indeed, some readers have said they were impressed with the quality of AI translation.

“I barely noticed any grammar issues at all,” said one user commenting on an AI-translated novel on Webnovel titled Chief honey pet: raise a little warm wife from a famous family.

“In fact, I forgot about spotting slips as the story kept pulling me down the page.”

While grammar might not be an issue with these AI-translated works, their phrasing often feels like they are written by a non-native speaker.

An extract from one of the web novels read, “The Du Family’s recruiting for guards, this will be a good opportunity. If one can be accepted to be a guard of the Du Family, they may have the chance to learn Du Family’s martial skills.”

AI might work to a certain extent, but whether it can replace literary translators is a different question entirely, said Lusby.

One advantage of human translators is that people can decide when a literal translation works and when it doesn’t. Chinese web novels are not known for having high-quality prose, according to Bai. But a good translator can edit them in a way that improves the text.

“Even if they don’t, the resulting product will be roughly the same quality as the original,” Bai said. “But machine translation cannot make the resulting work better, nor will it be the same quality. It will be worse. At best it will be ‘Chinglish’, and oftentimes it will be incomprehensible.”

Bai used the phrase “zhuwei pingshen” as an example. The phrase is used in Classical Chinese to ask people to get up from kneeling. Bai said he translates it as “Rise, everyone”. But anyone using Google Translate will see “flat body”, while Baidu Translate turns out “all right”.

“Granted, computer translation software can have their dictionaries updated to provide correct translations in cases like these,” Bai said. “But that is a manual process that takes a lot of time and skill.”

This is why not everything at Funstory.ai is automated. Brief, the company’s CEO, said people can prime a text to make it ready for AI translation. Once people “process relevant data”, the computers can better understand the literature, he said.

One way to do this is manual categorisation. For example, an AI might not recognise Xiang Long Shiba as a martial art technique, so the term needs to be manually categorised as such before translation beings, Brief explained. And the Huashan Sect, a fictional martial arts sect in the wuxia genre, also has to be manually categorised. Sometimes, people also have to help break down long and complex sentences before a novel is translated by AI.

A whopping 90 per cent of the novels translated by Funstory.ai have human intervention before AI takes over, Brief said. Once translated by AI, those novels go straight to release. For the other 10 per cent, people will review the translation after the fact.

“Computers’ understanding of literature is still in the very early stages, even though we’re already ahead globally,” Brief said. “But because the Chinese language really is broad and profound with complex sentences and various possibilities, computers may still get confused and freak out sometimes.”

Still, Brief said that people read web novels for fun and to kill time. So readers value the plot more than the literary quality of the work, he said.

“They don’t have to feel beauty, they just need to feel entertained,” Brief said.

Lusby, from Pixie B, agrees that online literature is plot-driven, with text often structured to act as literary clickbait. It’s also intended to be read in short bursts and is highly commoditised, she said.

But she doesn’t think AI can currently offer the surprise and unpredictability achieved through nuanced language. She said that traditional translations retain some of the “tone and flavour” of the writer doing the translating, but AI risks creating a “stale and monotone” work.

“While I can see AI taking a role in translation work … I cannot imagine the most popular or successful long-form works achieving significant success without the intervention of a human eye to add the spirit and personality that makes a work popular,” Lusby said.

By pushing out copious volumes of translated web novels, companies hope that online literature can help promote Chinese culture overseas. China Literature has said it wants to become “China’s Marvel”.

But while Funstory.ai is supposed to help the industry grow, some think there is already too much content, and that makes finding the good stuff harder than ever.

“There is a lot of interest and demand in Chinese web novels,” Bai said. “That said, the supply has skyrocketed, and the demand has not. Although there is some growth, the current market is completely saturated with content.”

“These machine translation sites are ‘poisoning the well’ by putting out so much bad content,” he added.