This repository contains the code to perform the experimentation of Bias Detection on different pre-trained Transformers-based Language Models. The code includes multiple experiments, investigating different features of the problem and implementing various techniques. Further information can be found in our paper "Discrimination Bias Detection through Categorical Association in Pre-trained Language Models" by Dusi, Arici, Gerevini, Serina, and Putelli (currently under evaluation).
data: folder containing all the input data for the experiments. Documents are divided in two subfolders:properties, for datasets referring to a single property, andcrossed-evaluation, with datasets referring to pairs of properties.src: folder containing all the code for executing the experiments. Code is divided in different software modules; the code is implemented in Python 3.x.cache: folder currently absent. It will be created when the code is executed, and will contain the cached data for executing the experiments (typically, pre-computed embeddings).results: folder currently absent. It will be created when the code is executed, and will contain the results data of the experiments.
[currently under update]