For today I have a small collection of words that pertain to how we’ve learned to modify the way we write in order to accommodate censors, whether those censors are computers or humans.
To begin, the word algospeak, which is a portmanteau of algorithm + speak. This word broke big earlier this year due to an article in the Washington Post by Taylor Lorenz that explained the idea of algospeak and why it had emerged, and that provided many examples.
Algospeak is the use of euphemisms to get around algorithms in social media platforms that downrank or remove posts that include certain keywords. From the perspective of the social media company, they’re trying to limit the spread of misinformation or of other controversial content. Some of the topics that seem to be off-limits are suicide, anti-vaxxer sentiments, sex work, and ethnic or other slurs. Cory Doctorow reports that an in-company chat system at Amazon is designed to ban words like union and grievance.
I say seem to be off-limits because the companies don’t publish lists of terms that the algorithms focus on; instead, people discover these terms through experience, i.e. by having content modded. Once people know that a term is on the hotlist, they invent workarounds. Sometimes these are recognizable versions of the verboten term. The WaPo article lists examples like unalive for “dead” or “kill” and SA for “sexual assault,” and the distant but still recognizable leg booty for “LGBTQ.” At other times, you’d never guess the meaning of an algospeak workarounds. An example is swimmer, a term used in some anti-vaxxer forums to refer to vaccinated people, or sex workers referring to themselves as accountants.
For people who rely on TikTok as an influencer platform, having content demoted can have commercial implications. The WaPo article cites a woman who has to use algospeak in her TikTok videos about women’s health to avoid being demoted for using words like sex and period. Similarly, Facebook doesn’t let people share or sell prescription medicines, so in communities for diabetics, users have learned to use algospeak to tell others that they need insulin or can provide it for an emergency situation.
The WaPo article wasn’t the first to use the word algospeak, but it’s pretty new. A post on the KnowYourMeme site traces the term to a Twitter quote retweet from December, 2021:
I poked around a little to see if I could find an earlier cite, but I didn’t come up with anything. (I didn’t try that hard.)
Ok, I said at the beginning that there were multiple terms that pertained to working around algorithms. The discussions I’ve found of algospeak sometimes mention Voldemorting. “Voldemorting is the anti-SEO, the anti-keyword, and the anti-hashtag,” says Gretchen McCulloch in an article in Wired magazine. For example, perhaps you don’t want to give a particular celebrity the boost of mentioning their name on Twitter, or you don’t want a company to know that you’re talking about their product. So instead you use a made-up name. The term Voldemorting comes from Harry Potter and the idea that Voldemort, the archivillain in that series, is He Who Must Not Be Named, because “a ‘taboo’ spell is placed upon the name, such that Voldemort or his followers may trace anyone who utters it” (Wikipedia). That seems like an uncannily accurate description of how our words can be traced back to us online.
Some articles about algospeak also mention leet or leetspeak, which is a form of writing words in which letters are replaced with numbers or characters—for example, 1337 or l33t for “leet”, or n00b for “newbie.” There is a certain element of fun to leetspeak, but there are also suggestions that it was used as a form of algospeak on old bulletin-board systems that forbade certain topics. In the early days of online life and of simple filters, a leet term like haXOr might be enough to get around a filter that was looking for the word hacker.
Update (24 Jul 2022): By coincidence, the Language Log blog has a post by Victor Mair about how people in China evade censorship by using a form of “typos” known as the Grass-Mud Horse Lexicon. The post has a number of examples. (Also, a somewhat frightening description of the thoroughness of Chinese censorship.)
And speaking of primitive filters, there’s also the Scunthorpe problem, which is a form of unintentional filtering. You sign up for a website and try to register under your name, which happens to be Cummings. Nope, says the site, we don’t allow names that include profanity. Huh? In this case—and for names like Dick, Wang, or Cockburn—a site filter has been a little too zealous in looking for unacceptable names. The term Scunthorpe problem is based on a problem back in the 1990s when AOL wouldn’t allow residents of the town of Scunthorpe to register because of the, ahem, profanity in the town’s name. For this problem, of course, the user has to find an alternative name. That, or try to convince the site owners that their filter is being dumb.
The WaPo article about algospeak also introduced me to another, pre-computer term for avoiding censorship: Aesopian language. This refers to language that’s used in subversive texts to get around censorship, or as it says in Wikipedia, for “… communications that convey an innocent meaning to outsiders but hold a concealed meaning to informed members of a conspiracy or underground movement.” In this sense, Aesopian language is a kind of argot or cant, both of which can also refer to language that hides meaning from those who are not in the in-group. The term Aesopian was used by a Russian novelist in the 19th century to describe how he wrote satires of Tsarist society but couched them in a way that would pass the censors. He used the word Aesopian is based on Aesop’s fables, which similarly embed messages in simple-seeming stories.
Many people who write about algospeak observe that even if algorithms watch what we write, people find ways to get around the restrictions. Censorship is hard. Moreover, and counter to Orwell’s idea of Newspeak, limiting the terms that people can use doesn’t seem to prevent them from thinking and discussing those ideas. Overall, that’s probably a good thing, right?
Like this? Read all the Friday words.