Do you tweet formally for a wide audience (and use abbrevs 4 ur peeps)? You may not realize you’re doing it. But a study of hundreds of thousands of tweets showed that Twitter users subtly tailor their language based on who’s reading.
Twitter “is a single platform that serves a huge range of communicative functions,” says Jacob Eisenstein, who leads a computational linguistics lab at Georgia Tech. With the same 140-character messages, a user can participate in a mass social movement or gossip with close friends. This makes Twitter different from other forms of written communication.
Yet while you can’t control how far your tweets fly, there are ways to narrow or widen your potential audience. If you begin a tweet with another user’s @handle, for instance, only people who follow both of you will see it. You’ve shrunken your readership (at least until people start retweeting your brilliant remark). On the other hand, using #hashtags broadens a tweet’s potential audience. Anyone following or searching for a hashtag on their favorite topic will see your tweet.
Eisenstein wanted to know whether we tweet differently depending on the size of our audience. He and graduate student Umashanthi Pavalanathan combed through a set of 114 million tweets to find out.
The researchers started with tweets that were geotagged and came from the U.S. between June 2009 and May 2012. They removed all retweets. Then, in an aggressive bid to eliminate bots and marketers, they also removed any tweet that included a URL, as well as any users with more than 1,000 followers or followees. This gave them “a subset of messages that is highly likely to be originally written by the author,” they write.
Within the remaining tweets, they looked at two variables. The first was “tweetspeak,” or language that shows up often on Twitter but not in standard writing. For every tweet that used one of their 94 chosen tweetspeak terms, the authors selected another tweet from the same author that did not use tweetspeak. (This would balance the dataset so that certain people or regions didn’t bias the results.) They ended up with 188,000 tweets.
Eisenstein and Pavalanathan were also curious about Twitter slang that’s specific to certain parts of the country. They found 120 terms that were especially likely to come from one metropolitan area. Again, they built a dataset by balancing tweets from the same user that did and didn’t use these terms, and ended up with 224,000 tweets.
Finally, the researchers looked at the potential audience sizes for all these tweets. If a tweet included someone else’s handle (with the @ sign), they considered it to have a smaller intended audience than usual. Tweets that included hashtags had a larger audience.
The wider the audience, the less tweetspeak there was. The researchers saw that tweets with hashtags were less likely than usual to include tweetspeak. But these jargon terms were more common in tweets with @ signs. (This was even true if the other user’s handle was in the middle of the tweet, which doesn’t limit who reads it but might indicate a conversation.)
Geographical lingo was also more common in tweets that @-mentioned someone from the same region as the person tweeting. Regular tweetspeak was more common in these tweets, too. In other words, the more intimate our audience, the more coded our language is.
In some ways, this is similar to how we speak. “Speakers tend to modulate their speech based on whom they imagine to be in the audience,” Eisenstein says. “Like me using ‘whom’ instead of ‘who.'”
When we’re speaking, though, we’re usually aware of how big our audience is. On Twitter, the difference between a tweet that reaches only a few friends and one that joins a national forum is subtle. But to his surprise, Eisenstein says, “our study shows that people are well-attuned to these differences.”
Even though Twitter’s LOLs and IMOs and SMHs may distress some language purists, Eisenstein says Twitter is just a new twist on a classic story. We shape our language to our audience, no matter how few characters we have.
Image: by Beau Giles (via Flickr)
Pavalanathan, U., & Eisenstein, J. (2015). AUDIENCE-MODULATED VARIATION IN ONLINE SOCIAL MEDIA American Speech, 90 (2), 187-213 DOI: 10.1215/00031283-3130324