At Reuters, Li Hui and Megha Rajagopalan look inside Sina Weibo’s Tianjin censorship center, where 20-something college graduates plow through 3,000 flagged posts per hour for the equivalent of $490 a month.
“People are often torn when they start, but later they go numb and just do the job,” said one former censor, who left because he felt the career prospects were poor. “One thing I can tell you is that we are worked very hard and paid very little.”
[…] “Our job prevents Weibo from being shut down and that gives people a big platform to speak from. It’s not an ideally free one, but it still lets people vent,” said a second former censor.
[…] “The most frequently deleted posts are the political ones, especially those criticising the government, but Sina grants relatively more room for discussions on democracy and constitutionalism because there are leaders who want to keep the debate going,” said the first former censor.
The censor’s claim that critical posts are more likely to be deleted conflicts with findings from a Harvard team led by Gary King. In one study published last summer and another this year, they found that deletion was incurred not by criticizing authorities, but by touching on topics deemed to have “collective action potential.” King explained on NPR’s Morning Edition this week:
King offered a couple of examples of how the censors work: A Chinese mother once protested a local official outside his hotel. Her demonstration led to sympathetic outrage on social media sites, but the action was almost entirely online — and that flurry of posts went uncensored, King said.
By contrast, after an earthquake damaged nuclear reactors in Japan, there was a run on salt in China, King says, because people believed — wrongly — that eating salt could protect them against disorders linked to radiation. People physically mobilized around the issue, and media posts that cataloged these activities were quickly censored, King said, because the online commentary corresponded to a physical, public presence.
King also looked at messages with a pro- or anti-government tilt that attempted to mobilize people: “If … you say, ‘Hey, let’s go protest,’ and have a whole bunch of people march on some government office, it will be censored,” he says. “But at the same time, if you say, ‘Let’s have a big party for all the government officials who are doing such a great job,’ and you are also able to move people, you will also be censored.” [Source]
Blocked on Weibo author Jason Q. Ng commented on his blog that “either King’s conclusions really don’t apply to Weibo or this one censor doesn’t really have a good sense of what his team is actually censoring [….] My gut is that the King conclusion may not apply well to Weibo or is perhaps an overreach, but as far as folks can tell, their data seems pretty irrefutable [….]”
For their more recent study [PDF], King and the others went as far as to set up a China-based social media site to get an inside view of censorship requirements. Their description of the mechanics of censorship—in which keyword filters are paired with human review—closely matches that of Reuters’ Sina censors.
Another detail in the Reuters report that drew attention was the number of censors Sina employs: apparently around 150. Some observers, including journalist and blogger Bei Feng, felt that this figure was too low. Jason Ng proposed an alternative explanation:
The number of censors cited in the article (“40 censors work in 12-hour shifts” and “100 people worked non-stop for 24 hours” during intense periods) fits with the theory that much of the censorship is computer assisted. Without using algorithms to flag posts, Zhu et al’s paper (The Velocity of Censorship) estimated that you’d need 4,200 censors to to manually read every new post on Weibo, which would be inefficient. But is this Tianjin office the only such facility Sina Weibo has, or is it just one of many? Article doesn’t make that clear.
[…] For the most part, the details revealed about how the censorship works mostly gibes with the research out there (high five researchers!). For instance, Zhu et al estimate of how efficient censors are (can read 50 posts per minute) is exactly what the censors report (3000 posts per hour). [Source]