Experiment: Modifying Language

[slifty]

Hypothesis

Language (words and speech patterns) can signal "tribe" or trigger heightened emotional responses (negative or positive) in a reader, making them more likely to assume bad intent from the author (or otherwise augmenting the risks of dis-confirmation bias) when consuming credible information.

Proposed intervention

Adjust the presentation of credible information by replacing charged terms with terms that will be interpreted neutrally by the reader based on their identity.

American examples: "Guns" vs "Firearms"; "Illegal" vs "Undocumented"

Concierge MVP

1. Select a fact check and write multiple versions of it for specific audience archetypes.
2. Run a study to see if readers are more receptive to fact checks that have been written with their personal predispositions in mind.

Engineered MVP

1. Identify a set of audience archetypes
2. Identify language translations for each archetype
3. Have a user select their archetype
4. Find / replace words and phrases in content based on the dictionary + archetype combinations

Associated Research

{Public health research around the impact of language?}

[crupar]

Linguistics Question: Would this be considered using different 'registers'?

[slifty]

There are some wireframes for this project now; I think they reflect the following high level "chunks" of complexity:

Frontend:

• Site scaffolding

Backend:

• Claim detection integration
• Camp signaling detection
• Judgmental / sentiment analysis
• Entity detection; entity sentiment mapping

[slifty]

Language Check.pdf