Research map / Label-efficient and robust learning / Recognition and long-tail
Quantifying Political Polarization through the Lens of Machine Translation and Vicarious Offense
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
An AAAI 2024 invited talk survey of three research threads on US political polarization: machine-translation-based polarization measurement, analysis of cable news coverage of policing, and vicarious offense across political identity groups.
Problem
Quantifying political polarization in large-scale multi-issue social web discourse, understanding disparate news media portrayal of policing, and measuring how well political groups can predict what content other groups find offensive.
Approach
Treats sub-communities as speaking different "languages" and uses word-level translation alignment between them; words that translate to different words signal political misalignment (disagreed pairs), giving an unsupervised polarization measure. Also introduces vicarious offense: whether annotators from one political group can predict offensiveness judgments of another.
Key results
- Vicarious offense study on 2,310 social media posts finds Republicans are the least understood political group and struggle most to understand others.
- Independents are found to be the most tolerant in terms of web censorship.
- Hot-button issues such as reproductive rights or gun control/rights have a profound effect on annotator disagreement on offensiveness.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Evaluate Republican-leaning users’ perceptions of what others find offensive | Vicarious offense annotation study · 2,310 social media posts | understanding of other political groups | Hurt |
| Compare independents’ judgments of offensive content with other political groups | Vicarious offense annotation study · 2,310 social media posts | tolerance for web censorship | Helped |
| Focus on hot-button issues such as reproductive rights or gun control/rights | Vicarious offense annotation study · 2,310 social media posts | annotator disagreement about offensiveness | Mixed |
Surprising findings
- Republicans are the least understood political group while also struggling most to understand others.
Experimental setup
Datasets: Fox News and CNN viewer comments on YouTube news coverage of George Floyd, Capitol riot, and Chauvin trial from cable networks 2,310 annotated social media posts (vicarious offense study)
Builds on: word alignment
Closest papers by meaning
- Linear Representations of Political Perspective Emerge in Large Language Models (ICLR 2025 · similarity 0.87)
- When Neutral Summaries Are Not That Neutral: Quantifying Political Neutrality in LLM-Generated News Summaries (Student Abstract) (AAAI 2025 · similarity 0.86)
- Examining Alignment of Large Language Models through Representative Heuristics: the case of political stereotypes (ICLR 2025 · similarity 0.85)
- Innocuous-Seeming Data, Latent Ideology: Ideological Generalisation in Finetuned LLMs (NeurIPS 2026 · similarity 0.84)
- Political Neutrality as Balanced Approval: A Large-Scale Human Evaluation of AI Responses (NeurIPS 2026 · similarity 0.84)
- Merging AI Incidents Research with Political Misinformation Research: Introducing the Political Deepfakes Incidents Database (AAAI 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.