print PRINT


How Twitter moderates content in Japan

  • December 20, 2021
  • , The Japan Times
  • English Press



In this era of fake news and COVID-19 misinformation, social media giants — and the amplified role they can play in shaping public opinion — have come under intense scrutiny worldwide.


A similar kind of social media reckoning is now catching up with Japan, too, after a recent string of events has accelerated conversations about online harassment and cyberbullying.


While Facebook is the center of the reassessment of social media companies’ role in the United States, it’s Twitter that is seen as the most influential in Japan, where the platform has its second-largest market after the United States.


“Twitter is huge in Japan. As far as platforms for anonymous speech go, it is the most influential,” says Taro Yamada, an internet-savvy lawmaker in the ruling Liberal Democratic Party who has played a key role in the party’s internal discussions about how to curb online defamation. “It is also significantly affecting the way public opinion is shaped.”


Twitter’s dominance has helped make the microblogging service deeply intertwined with the lives of many Japanese, turning it not only into a popular tool of self-expression and communication, but also a critical information hub in the event of natural disasters.


At the same time, however, various opinions and myths have taken root around Twitter Japan, the Japanese arm of the U.S.-headquartered Twitter Inc., including accusations that offensive tweets written in Japanese are not properly policed and removed by the Tokyo-based subsidiary.


For Twitter Japan, that line of criticism is an all-too-familiar refrain. Demonstrators have rallied outside its Tokyo office in the past, denouncing what they claimed to be a lack of oversight over discriminatory tweets.


The perception of Twitter as a hotbed of vitriolic language was further reinforced by the suicide in May 2020 of reality TV show star Hana Kimura, who had been bombarded on the platform with posts wishing her dead or otherwise dehumanizing her.


A campaign has also sprung up calling on Twitter to appoint a full-time human “grievance officer” in Japan to make up for what campaigners allege to be the “low accuracy” of its technology-aided content moderation efforts in Japanese.


With Twitter linked more inexorably than ever to the issues of defamation, harassment and bullying, questions arise: How does Twitter moderate content on its Japanese-language platform? Why are tweets that strike many as offensive and intimidating not being taken down?


Content moderation

According to Twitter Japan, the handling of all reports and complaints from users across the globe is centralized by Twitter Inc., despite the widespread perception that the vetting system of tweets is compartmentalized according to language or region.


“Whether you submit reports in Japanese or whatever language, they all go straight to Twitter Inc.,” says Satoshi Hattori, head of public policy at Twitter Japan, denying that the Japanese arm is exclusively responsible for probing tweets written in Japanese or tweets posted from users in Japan.


Hattori says he is aware that users in Japan sometimes exhort their followers to write their complaints in English and submit them to the U.S. headquarters of Twitter, instead of Twitter Japan, based on an assumption that the former acts more swiftly than the latter to take down problematic tweets.


Given the all-encompassing nature of Twitter Inc., however, the notion that Twitter Japan is somehow more reluctant to take action or operates on a different set of rules and protocols than its peers overseas, Hattori says, is a “total myth.”


Whatever the language is, every single tweet flagged by users as being in any way inappropriate is sent to, and then scrutinized by, an army of human content moderators that the social media goliath has at its disposal.


In recent years, artificial intelligence is increasingly utilized to proactively search for and flag potentially problematic speech. Whereas Twitter’s content moderation used to be entirely passive, in the sense that it only responded to users’ complaints, about 65% of tweets that moderators now handle are those preemptively flagged by AI, Hattori says.


Those moderators, he says, have their own specialized languages, including Japanese, and work from various parts of the globe as part of Twitter’s 24/7 operation.


Twitter Japan doesn’t disclose details such as how large the Japanese-language team is, where exactly they are based or whether they are employees or contractors. But as far as Japanese-language specialists are concerned, “they are a group who are not only proficient in language itself but well-versed in the cultural and societal background (of Japan),” Hattori says.


This is a key aspect of their qualification that is integral to their ability to read between lines and take context into account.


“When one person writes ‘shine (die)’ to another, for example, that could come across as banter if they’re friends. … As a content moderator, you’ll need to go beyond mere linguistic proficiency to be able to discern these nuances and decide whether there is a problem,” Hattori says, adding that for each reported tweet, content moderators are trained to trace back previous exchanges to better understand context.


Another common myth is that accounts are more likely to be frozen if targeted en masse, with users sometimes ganging up on whatever tweet or account they take issue with and uniting in their efforts to have it removed or banned.


“Even if tens of thousands of complaints are filed about one specific tweet, we won’t take action if it’s not ruled to be a violation,” Hattori says. “Likewise, a single complaint can spark action if the tweet in question really constitutes a violation.”


Whether tweets are deemed problematic or not is a decision that is made solely in accordance with the so-called Twitter Rules — a publicly viewable list of prohibited activities that alerts users to bannable content, including violence, child sexual exploitation, abuse/harassment and hateful conduct.


When a tweet is found to breach such rules, Twitter typically first asks the user responsible to delete it. Failure to comply with the request could result in the tweet being made unshareable, removed unilaterally or their account being frozen, depending on the severity of a case, Hattori says.


Twitter scrutiny

How Twitter goes about censoring and policing problematic content is an issue of growing interest in Japan, given the platform now wields considerable influence and lies at the heart of many key events.


According to the firm’s annual report, revenue in Japan generated $547.86 million last year, or about 15% of the platform’s total earnings of $3.72 billion.


In a nation prone to earthquakes and floods, Twitter has also carved out a reputation as a pivotal information-gathering service that both the public and municipalities now heavily rely on to keep abreast of the post-disaster situation as well as stay connected with families and communities. The number of Twitter users in Japan spiked after the devastating March 2011 earthquake and tsunami that triggered nuclear meltdowns in Fukushima.


Its function as a critical piece of information infrastructure aside, the social networking service, Twitter Japan officials say, also has a good affinity with Japanese people both linguistically and culturally.


The unique writing system of the Japanese language — a combination of hiragana, katakana and kanji — makes communication easy even within the 140-letter constraint, says Natsuko Nishimura, head of communication at Twitter Japan.


The lack of different time zones, she says, also suggests people across the nation can “share a sense of oneness” on Twitter as they tweet about the same TV show simultaneously.


A good example of this is a playful tradition dubbed “Balse Festival,” where fans of Hayao Miyazaki’s classic Ghibli film “Castle in the Sky” — whose rerun is routinely broadcast in Japan — make sure to tweet the word “balse” at the exact moment the key incantation is uttered at the climax of the movie. In 2013, the influx of “balse” even broke Twitter’s all-time tweets-per-second record.


It goes without saying, however, that not everything about Twitter is useful and amusing: What the suicide of Kimura underscored is how emotionally devastating a torrent of bitter online comments can be, even if they are technically in compliance with Twitter’s terms of use.


Although the Twitter Rules say “severe, repetitive usage of slurs” are subject to removal, a rebuttal or an expression of criticism is generally tolerated on the platform. The act of egging on other users to target a specific account, for example, is in itself not considered subject to crackdown if it’s done to express “proper criticism or opinions” and in line with the rules, Hattori says.


“Lots of comments hurled at Kimura, for example, were somewhere between slurs and criticism. For attackers, it can feel like a one-on-one exchange with her, but for the receiver, it’s basically her versus hundreds or even thousands of comments targeting her,” says Shinichi Yamaguchi, an associate professor of economic science who studies fake news and social media at the International University of Japan.


“When that happens, the toll they take on the receiver’s mental health is huge, even if they don’t technically violate those rules by Twitter,” Yamaguchi says. “There is room for more discussions about how to tackle a situation like this.”


Twitter, for its part, maintains the position that the company is in principle not responsible for taking down posts as long as they are within the scope of what is allowed under the rules — however ethically questionable they might be.


“If tweets are found to be in violation of our rules, we take action. For tweets that don’t violate the rules, we take appropriate measures based on each country’s law or legal demands made by authorities,” Hattori says.


A gray area is when tweets are neither in violation of the terms of use nor are outright illegal, as in the vilification of Kimura and, more recently, Kei Komuro, who married Japan’s former Princess Mako in October amid a family financial dispute.


“I don’t think it should be our responsibility to decide whether to take action against these gray-zone posts that people may or may not find problematic,” Hattori says.


Legislative action

Given the dominance of Twitter, this stance makes sense, Yamaguchi says.


“Giving Twitter the power to determine the validity of speech, including tweets that contain more criticism than slurs, so it can go on a removal spree as a result, means it will be allowed to control the space of speech that is now very public in nature,” he says.


Concerns over such an aggressive crackdown have already come to a head in Germany, where the 2017 Network Enforcement Act has obliged social media giants, including Facebook, Twitter and YouTube, to remove content that is “clearly illegal” within a 24-hour deadline after a user’s complaint, or face fines up to €50 million (¥6.4 billion).


A desire to avoid such fines can be so strong that social media companies may be incentivized into “overblocking” content rather than protecting free speech, Yamaguchi says, arguing Japan shouldn’t follow in Germany’s footsteps.


Instead, one alternative is for the Japanese-language version of Twitter to adopt a self-moderation prompt that urges users to rethink a reply deemed by its technology to be potentially offensive before they hit send, Yamaguchi says.


Internal testing of the prompt, which is already available in English-language settings, shows that around one-third of all users typically revise their initial post or decide not to post a response at all, according to Twitter. Twitter Japan says there is no plan for the feature to be introduced to the Japanese-language platform at this time.


Meanwhile, the government has taken some steps to crack down on offensive messages that are shared on social media. An amendment to the provider liability law was passed in April to simplify the process of identifying those responsible for posting anonymous offensive messages online — a step that is essential to initiating civil lawsuits against them.


Separately, the government is also looking to amend the Penal Code to stiffen penalties with respect to online insults, envisaging greater fines and the introduction of prison sentences.


While hopes are high that such stricter penalties could act as a deterrent, LDP lawmaker Yamada says the proposed measures aren’t all they’re cracked up to be. They’re no panacea, he says, considering the crime of insult in principle only applies to cases where one’s public reputation has been damaged.


“Right now, I see things like defamation, fake news and bullying all talked about in the same breath,” Yamada says. “But it’s not like the crime of insult alone can cover all of these issues.”


While political interest in the issue of online defamation is on the rise, it has, so far, failed to extend into a broader dissection of the power of social media behemoths in Japan.


Recent years have seen the advent of social media-driven hashtag movements zeroing in on political issues, the most prominent example being last year’s Twitter-based crusade against the government’s attempt to extend the retirement age for prosecutors.


“These campaigns are starting to have an impact on Japanese politics. That’s for sure,” Yamada says.


But in a nation where few politicians are Twitter-savvy, “I’d say that neither the government nor the ruling party have this perception that social media companies are becoming a threat to Japanese politics. Their behavior, I think, is still seen as pretty neutral in Japan,” he says.


“In future, though, if the issue of fake news or anonymous messages worsens to the point where Japanese politicians feel they’ve had enough, a day might come when they decide these social media companies can’t be left unchecked and that their responsibility as platform operators must be scrutinized.”

  • Ambassador
  • Ukraine
  • COVID-19
  • Trending Japan