I’m really excited to tell you what we’ve been working on at the Consortium. I believe this project, two years in the making, will do a lot to improve the trust between journalists and the communities they cover.
It’s called Authentically, and it’s an AI-powered program designed to make sure that journalists are not inadvertently defaulting to words that convey bias, and in the process harming their relationship with the public.
Kendall Moe is the senior project manager leading this work, so I asked her to tell you all about it for this edition of the newsletter.
Here’s Kendall’s piece explaining Authentically, how it came to be, and how why it’s so important:
In 2022, when the Consortium on Trust in Media and Technology was in its infancy, we had an advisory meeting where a question was raised about the language of journalism.
The basic premise was, do all reporters tend to gravitate toward the same patterns to describe people and their situations?
I have a degree in linguistics, so I was really intrigued by this question. We know that the slightest variations in language can be very powerful. Language is not just a means of communication but a mirror of our identities and ideologies, with the power to convey bias, positioning and blame. In my linguistics studies, I learned that the language a person uses can influence their worldview. Journalists, therefore, could unintentionally distance readers when their language signals a disconnect between our worldview and theirs.
Thinking about this launched me into a two-year research study of news coverage of polarizing events, using a method called corpus linguistics. I looked at stories about the U.S. Supreme Court Dobbs decision, overturning Roe. vs. Wade, and the national protests that followed George Floyd’s murder.
I immediately noticed patterns in the labels that journalists use to describe people. In coverage of Dobbs, we found that people who described themselves as pro-life were likely to then be described by the journalist using words like “proudly,” “unapologetically,” or “adamantly.” People who said they are pro-choice were labeled “necessarily,” or “increasingly.”
The data indicated that the words that signal pride and morality are more associated with pro-life, and the words that signal necessity or urgency are more associated with pro-choice.
In the two months after George Floyd’s murder, I discovered that coverage of demonstrations used verbs with negative connotations. For example, “protests spark,” or “protests erupted,” are phrases that evoke fiery imagery.
I also noticed that the individuals who were demonstrating were described as protestors far more often than they were described as activists. That would be fine, except that when the word “protestor” is used, it is much more likely to be accompanied by emotional and negative modifiers, and when activist is used, it’s more likely to be paired with descriptive or identity-centered language.
For example, the top phrases that included protestor were:
-armed protestor
-unarmed protestor
-peaceful protestor
Phrases that accompanied the word activist included:
-civil rights activist
-left-wing activist
-conservative activist
The activist modifiers were more focused on identity or cause, whereas the modifiers for protestor were more focused on danger and disruption.
Similarly, after the Dobbs decision, some of the verbs most commonly used with the word protestor were “confront,” “disrupt,” and “charge.” Meanwhile, the verbs used to describe activists were “cheer,” “rally,” and “gather.”
We know how this happens. When you look up activist in the thesaurus, protestor is listed as a synonym. In the news stories we analyzed, activist appeared alongside words like advocate, scientist, and supporter. Protestor, on the other hand, had synonyms like: rioter, looter, suspect. While the thesaurus may say these two words hold similar weight, the data tells a different story. There's a disconnect there.
So what can we do about this? We built a tool that can help journalists reset and re-code. It’s called Authentically. Here’s how it works: After a journalist writes a draft of a story, they drop the text into our program and it returns a rating, along with a list of influential words that were written.
The tool is meant to be a guide. A journalist might be trying to convey a certain tone – this may re-affirm that. But it’s possible that the words that are used are sending an unintended message. Often, when this happens, it isn’t until after a story is published and comments, social media posts, and emails start rolling in that a writer realizes the tone their words created.
Authentically will close that gap. It will give reporters the ability to better anticipate and better communicate. And that is more important than ever. News organizations are struggling to stay afloat and a lack of trust is contributing to the bottom line. If we can work on repairing trust, we can work on rebuilding news.
Authentically is currently in a beta phase, being tested by four newsrooms across the country this summer.
Kendall Moe is a Senior Project Manager at the Consortium on Trust in Media and Technology. Her work is focused on conducting and managing research for Authentically, which aims to enhance journalistic integrity through data-driven language analysis.