FairTransNLP-Language: Analysing toxicity and stereotypes in language for unbiased, fair and transparent systems

FairTransNLP-LANGUAGE: Analysing toxicity and stereotypes in language for unbiased, fair and transparent systems (PID2021-124361OB-C33) subproject of the project FairTransNLP-Language: Fairness and Transparency for equitable NLP applications in social media (PID2021-124361OB-C31) coordinated by Paolo Rosso (Universitat Politècnica de València). Funded by: Ministerio de Ciencia, Innovación y Universidades, programa de I+D de Generación de Conocimiento. MICIU/AEI/10.13039/501100011033/FEDER,UE. Participants: Universitat Politècnica de València, Universitat de Barcelona and Universidad Nacional de Educación a Distancia (UNED).

Short summary: The goal of the FairTransNLP research project is to develop tools and techniques to address two critical issues in the design and use of Natural Language Processing (NLP) applications: Fairness and Transparency. FairTransNLP-LANGUAGE focuses on: (1) modeling toxicity language and stereotypes in order to mitigate the undesirable effects of biased data, and (2) building high-quality unbiased and fair datasets following the new paradigm of Learning With Disagreement.