November 13, 2018
Haines 215

Comp Soc is having a paper workshop for:

Alina Arseniev-Koehler (UCLA): "Measuring Stigma in Language, with Word2Vec and Human-Rating data"

Abstract: This research examines stigma in news discourse, and ultimately aims to explain why stigmatized conditions (e.g., obesity, schizophrenia, illiteracy, lung cancer) receive different amounts of stigma. For this CompSoc session, I'd love to brainstorm strategies to measure how stigmatized a condition is in news discourse


My data includes 1) human-ratings on how stigmatized 93 conditions are, and 2) quantitative representations for each of these 93 conditions, which represent their meanings in Google News. (Specifically, these quantitative representations are Word2Vec word-vectors corresponding to each condition). 


Aspects of measuring stigma in language, that I've already piloted:

  • Previously, I developed methods to measure the sentiment of a word (ranging from positive sentiment to negative) and the morality of a word (ranging from moral to immoral) in a given Word2Vec model. Conditions which receive more stigma according to human-ratings, also tend to have slightly more negative sentiment and be more immoral in news discourse. This suggests that sentiment and morality operationalize some, but not all, aspects of stigma. What is missing?
  • The raw quantitative representations of the 93 conditions are (weak) predictors of stigma level in human ratings, suggesting that stigma is likely encoded in the way that these conditions are discussed in GoogleNews.