Desiderata for Actionable Bias Research
Abstract
The identification of stereotypical biases in NLP tools is receiving increasing attention, as corpora, metrics and mitigation techniques are being developed. These resources are instrumental to make progress towards harm mitigation. Building on these early successes of bias research, we present some desiderata to move the field forward in three actionable directions: increasing the visibility of bias evaluations, widening studies beyond gender bias and engaging LLM developers with bias mitigation.
Domains
Document and Text Processing| Origin | Files produced by the author(s) |
|---|
