terça-feira, 3 de janeiro de 2012

Echoes of Power: Language Effects and Power Differences in Social Interaction

Today I've read a very interesting paper entitled Echoes of Power: Language Effects and Power Differences in Social Interaction. In this paper, some researchers from Cornell University and Yahoo! study the effect of people's position (or power) over the way they influence each other in terms of language usage. The basic idea of the paper is that in communication a higher-powered person (e.g., a justice) is more likely to affect the language of others than a less-powered person (e.g., a lawyer). Moreover, a higher-powered person has its language less affected by the others than a less-powered one. The authors show these patterns using the concept of linguistic coordination, which is, basically, the way people tend to mimic the usage of function words (e.g., articles, pronouns) from those who they are communicating with.

Based on the exchange theory framework, the authors consider two types of power in social interactions: status and dependence. Measures of coordination between two groups are defined for a specific linguistic style marker and also for aggregated markers. Such measures are, basically, a function of the increasing of the probability of a speaker 'a' from one group A to use a given marker in a reply to a target 'b' from a group B given that 'b' has used this marker while communicating with 'a'. The author say that, in this case, 'a' coordinated to 'b'.

They study these exchanges in language patterns in two real-life scenarios: Wikipedia and Supreme Court arguments. In the Wikipedia, admins (higher-powered) and non-admins (lower-powered) communicate about article editions. In terms of power, the authors found that users coordinate more to admins than to non-admins, as expected. However, the results also shown that admins coordinate more to other users than non-admins, which is counterintuitive. A further analysis revealed that admins reduce their levels of coordination after they are promoted. Considering dependence, they found that an user coordinates more to thoses users who he/she disagrees with. In the Supreme Court arguments, justices (higher-powered) and lawyers (lower-powered) argue about cases. As expected, they found that lawyers coordinate more to justices than the opposite. This coordination is even more likely for the cases in which the laywer and justice disagreed. Moreover, favorable justices coordinate more.

The authors also show how coordination features (i.e., features that state whether a user x coodinates more to y than the opposite) are useful for the prediction of the most powerful user in social interactions. In fact, while other features, such as textual content, may be more effective in this task, coordination features work in cross domain settings. Finally, the paper also studies activity and gender bias in coordination.

Link: http://www.cs.cornell.edu/~cristian/Echoes_of_power.html

Nenhum comentário:

Postar um comentário