This repository contains the code developed for "Unraveling Disagreement Constituents in Hateful Speech". This repository contains several scripts to reproduce the results presented in the paper. The scripts allow both to estimate the Constituent Disagreement Scores for each constituent in the training dataset, and to estimate Sentence Disagreement Scores (SDS) according to the four proposed strategies: Sum, Mean, Median, and Minimum.
The datasets are reserved for the participants of Task 11 @SemEval2023 on Learning with Disagreements (LeWiDi), 2nd edition. Please refer to the paper for additional information. To request the datasets, please refer to the official LeWiDi page. The proposed code involves positioning the dataset in the main directory. To execute the code, please place the datasets there or change the reference paths.