Do conservatives and liberals speak different languages? Given the heated nature of political debate these days, it certainly seems like it. But how do you find common ground when people from different sides of the aisle are trying to communicate?
This question was one of the motivations behind Partisan Thesaurus, a project that enables you to type any word and see the other words with which liberals and conservatives mostly commonly associate it. Type a politically charged word like “immigrant,” and see that liberals associate it with “undocumented,” “innocent,” and “unarmed,” while conservatives associate it with “alien,” “outcast,” and “obscure.” The top three words for “jobs” in the liberal thesaurus are “businesses,” “mortgages,” and “products,” while on the conservative side they’re “families,” “workers,” and “votes.”
Built by Melanie Hoff, a graduate student at NYU’s Interactive Telecommunications Program, Partisan Thesaurus uses one machine learning algorithm that’s been fed two separate bodies of text. One corpus is composed of liberal texts, including writing by Joseph Stiglitz and Jon Stewart and interview transcriptions from Hillary Clinton and Bernie Sanders, while the other is made up of conservative texts, including books by Ann Coulter, Ayn Rand, and Ronald Reagan, as well as transcriptions from interviews with Donald Trump and Chris Christie. Hoff says she chose which texts to include in each corpus after consulting with political historians and reading lists that coincide with different political ways of thought. “I think of the project as a crude model of someone’s mind if they’d only ever been exposed to liberal media or conservative media,” Hoff says.
The differences are striking when placed next to each other against a bold red and blue background that corresponds to political party. “A lot of miscommunication happens across party lines because we’re semantically speaking different languages,” Hoff says. “We may think when we talk about the word democracy, we mean the same thing. But in this project I’m trying to show that we may not even agree on what we disagree on.”
Partisan Thesaurus, while politically aimed, also underlines how easily machine learning algorithms reflect the biases of their designers. Outputs are different depending on which texts the algorithm has been fed–a discrepancy that shows just how vulnerable technology is to human prejudices, even when built with the best intentions. Rather than ignoring the issue of bias, Hoff decided to use it as a tool to reveal inconsistencies in political discourse as well as the fallibilities of machine learning.
While the Thesaurus in its current stage is a simple tool, Hoff has plans to take it much further. She first hopes to expand the corpus by adding text from left- and right-leaning news sources, since right now the Thesaurus is laden with big words, as you’d expect from the political candidates’ interviews and books it’s been fed, but can’t handle swear words or colloquialisms. “They’re not going to talk about poop, but I want to know what the right associates with poop versus the left,” she says.
Eventually Hoff hopes to turn Partisan Thesaurus into a full-fledged translation engine. Imagine translating a Breitbart article into liberal language, or even seeing what the liberal and conservative versions of a children’s book would look like. From there, Hoff wants to bring that intra-language translation tool into your browser with a Chrome extension that would enable you to toggle between conservative and liberal language for any website on the Internet. She hopes these kinds of tools would bring more awareness about how people’s differing life experiences change what being a citizen means to them–and hopefully encourage compassion.
Hoff believes that Partisan Thesaurus might ultimately be used to find similarities instead of just reveal differences. It could provide a common language, and with it, common ground. That’s when the real work could begin.