English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

The impact of AI models on marginalized communities has traditionally beenmeasured by identifying performance differences between specified demographicsubgroups. Though this approach aims to center vulnerable groups, it risksobscuring patterns of harm faced by intersectional subgroups or shared acrossmultiple groups. To address this, we draw on theories of marginalization fromdisability studies and related disciplines, which state that people fartherfrom the norm face greater adversity, to consider the "margins" in the domainof toxicity detection. We operationalize the "margins" of a dataset byemploying outlier detection to identify text about people with demographicattributes distant from the "norm". We find that model performance isconsistently worse for demographic outliers, with mean squared error (MSE)between outliers and non-outliers up to 70.4% worse across toxicity types. Itis also worse for text outliers, with a MSE up to 68.4% higher for outliersthan non-outliers. We also find text and demographic outliers to beparticularly susceptible to errors in the classification of severe toxicity andidentity attacks. Compared to analysis of disparities using traditionaldemographic breakdowns, we find that our outlier analysis frequently surfacesgreater harms faced by a larger, more intersectional group, which suggests thatoutlier analysis is particularly beneficial for identifying harms against thosegroups.

Centering the Margins: Outlier-Based Identification of Harmed Populations in Toxicity Detection