Echoes of power: language effects and power differences in social interaction

Cristian Danescu-Niculescu-Mizil, Lillian Lee, Bo Pang, Jon M. KleinbergWWW 2012 · OpenReview

TL;DR

This paper introduces a novel method for detecting power dynamics in social interactions by analyzing "linguistic coordination" – how closely people mimic the function-word usage of their conversation partners. They demonstrate that this subtle linguistic mirroring reliably reveals both static (status) and situational (dependence) power differences across diverse settings like Wikipedia discussions and Supreme Court arguments, offering a domain-independent way to understand social hierarchies.

What Problem Does This Paper Address?

While much research on online communities focuses on the structure of interactions (who talks to whom), the content of those interactions, specifically language, has been a less explored frontier for understanding social dynamics. Researchers have struggled to find domain-independent linguistic signals that reveal social properties like power, status, or dependence. Traditional methods often rely on specific keywords or structural features that don't generalize. This paper aims to bridge this gap by showing that subtle, unconscious linguistic mimicry can serve as a powerful, generalizable indicator of power imbalances in conversations.

Methods

The core of the paper's methodology is linguistic coordination, defined as the degree to which a speaker's function-word usage mirrors that of their conversation partner. They focus on function words (articles, prepositions, pronouns, etc.) because these are thought to be processed unconsciously and thus reflect underlying social dynamics rather than deliberate content choices.

They developed a framework to measure this coordination:

  • Linguistic Style Markers: They use 8 categories of function words (e.g., articles, prepositions, personal pronouns) derived from the LIWC (Linguistic Inquiry and Word Count) lexicon.
  • Coordination Measure: For a speaker (b) and a target (a), coordination is calculated as the difference between the probability of the speaker using a specific function word category given the target used it, and the speaker's baseline probability of using that category. This captures how much the target's usage influences the speaker's immediate reply.
  • Group Coordination: This individual measure is extended to group-level coordination by averaging across participants.
  • Aggregated Measures: To handle cases where coordination for specific markers is undefined for some users, they employ three aggregation strategies to create overall coordination scores.

Datasets

The study utilizes two distinct datasets of goal-oriented conversations:

  • Wikipedia Discussions: Over 240,000 conversational exchanges from Wikipedia talk pages. This dataset is valuable for its large scale and the availability of rich metadata, including user status (admin vs. non-admin) and the exact timing of status changes.
  • U.S. Supreme Court Oral Arguments: Transcripts of 50,389 conversational exchanges from 204 Supreme Court cases, involving Justices and lawyers. This dataset provides a contrasting, more adversarial setting and allows for the study of dependence-based power dynamics.

Key Insights and Ideas

  • Unconscious Mimicry as a Power Signal: The most striking insight is that subtle, unconscious linguistic mirroring (coordination) is a robust indicator of power. This suggests that social hierarchies manifest not just in what we say, but how we say it, in ways we may not even be aware of.
  • Domain Independence: The proposed linguistic coordination framework proves effective across vastly different domains (online community vs. formal legal setting), highlighting its generalizability.
  • Distinguishing Status and Dependence: The research successfully differentiates between power derived from static status (e.g., admin vs. non-admin) and situational dependence (e.g., needing to convince an opponent), showing that coordination reflects both.
  • Power Dynamics are Bidirectional: While the primary focus is on how lower-power individuals coordinate to higher-power individuals, the study also shows how higher-power individuals' coordination patterns change, suggesting a more complex interplay.
  • Subtlety Over Content: The findings emphasize that function-word coordination is more revealing than content-based analysis (like bag-of-words) for inferring power, especially in cross-domain settings.

Key Results

  • Ptarget Hypothesis Supported: In both datasets, individuals generally coordinate more with higher-powered individuals (e.g., lawyers coordinate more with Justices; users coordinate more with admins).
  • Pspeaker Hypothesis Nuanced: While the hypothesis that high-powered individuals coordinate less was contradicted in initial observations (admins coordinated more than non-admins), further analysis controlling for personal characteristics revealed that after gaining power, individuals do coordinate less.
  • Status Change Effects: On Wikipedia, users showed a decrease in coordination after becoming admins, and this change was observable even in the months following the promotion, suggesting an acclimation period.
  • Dependence Drives Coordination: In Supreme Court arguments, lawyers coordinated more with "unfavorable" Justices (those who would eventually vote against their side, creating dependence) than with "favorable" Justices. Conversely, unfavorable Justices coordinated less with lawyers.
  • Cross-Domain Generalization: Coordination features significantly outperformed stylistic features and bag-of-words in predicting status across domains, achieving 57.1% accuracy when trained on one domain and tested on another, compared to chance (50%) or much lower performance for other methods.

Limitations and Future Work

  • Personal Characteristics (Hypothesis B): The study acknowledges that individual traits (like sociability) also influence coordination levels, and disentangling these from power effects remains an ongoing challenge. The authors suggest that while they've controlled for some aspects, further work is needed.
  • Gender Differences: The paper notes interesting gender-based coordination patterns in the Supreme Court data, but acknowledges that separating power-based explanations from cultural or gender-specific communication theories is complex and requires further investigation.
  • Other Coordination Forms: The authors mention that language coordination is just one form of coordination (e.g., posture coordination), and exploring how these interact could be fruitful.
  • Compact Representations and Privacy: The reliance on function words opens up possibilities for creating compact, privacy-preserving representations of conversations by redacting content words.

Summary updated

Abstract

Understanding social interaction within groups is key to analyzing online communities. Most current work focuses on structural properties: who talks to whom, and how such interactions form larger network structures. The interactions themselves, however, generally take place in the form of natural language --- either spoken or written --- and one could reasonably suppose that signals manifested in language might also provide information about roles, status, and other aspects of the group's dynamics. To date, however, finding domain-independent language-based signals has been a challenge. Here, we show that in group discussions, power differentials between participants are subtly revealed by how much one individual immediately echoes the linguistic style of the person they are responding to. Starting from this observation, we propose an analysis framework based on linguistic coordination that can be used to shed light on power relationships and that works consistently across multiple types of power --- including a more "static" form of power based on status differences, and a more "situational" form of power in which one individual experiences a type of dependence on another. Using this framework, we study how conversational behavior can reveal power relationships in two very different settings: discussions among Wikipedians and arguments before the U. S. Supreme Court.
BibTeX
@inproceedings{danescuniculescumizil2012echoes,
  author = {Cristian Danescu-Niculescu-Mizil and Lillian Lee and Bo Pang and Jon M. Kleinberg},
  title = {{Echoes of power: language effects and power differences in social interaction}},
  booktitle = {The Web Conference},
  year = {2012},
  url = {https://openreview.net/forum?id=B1gejb-_WH},
}
Figure