Hate Speech: From automatic classification to understanding emotional dynamics

• Applicant: Professorin Dr.-Ing. Tanja Schultz
• Subject Area: Social Psychology, Industrial and Organisational Psychology ; Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 465129985

Hate speech, i.e., verbal aggression targeting members of various social groups, is widespread in everyday life. It has detrimental consequences for everybody exposed to it. Our objective is to understand both, the underlying properties of hate speech, and its impact on listeners. We will develop an audio database of hate speech and develop acoustic, linguistic, and paralinguistic features, relevant to fully automatic identification of hate speech from audio recordings. We hypothesize that auditory hate speech will be a more powerful elicitor of emotional responses than its textual counterpart. We further hypothesize that auditory hate speech will result in slower desensitization. Finally, we hypothesize that engagement with auditory hate speech, such as voluntarily terminating it or talking back, will help overcome desensitization. The proposed project will blend two major methodological approaches. First, experimental studies will manipulate the content of stimuli, their modality, and participants’ response options. Dependent measures will include questionnaire-based self-reports, psychophysiological recordings (electrodermal and cardiac activity, electrical activity of the brain), and expressive behaviors (facial activity). Second, signal processing and machine learning methods will be leveraged to identify and differentiate hate speech from other types of speech for both speakers and listeners. This will be achieved through annotation of data of hateful utterances, and subsequent development of automatic learning classifiers, which will draw on the acoustic, linguistic, and paralinguistic features of hate speech. Annotation of the database will initially be performed manually, and rapidly extended by means of automatic transcriptions based on automatic speech recognition (ASR). Recent advances in automated text-based sentiment analysis have proven the value of "big data" methods. However, many types of auditory hate speech are not available in annotated form. The HateSpeech project might be a game changer. We expect that other researchers in the social sciences will be highly interested in using the outcomes of the proposed project for their own work. Conversely, for the speech community in computer science, the manually annotated database will provide a highly useful ground truth for the study and classification of extreme emotions on the Internet. The proposed project is both highly relevant and highly suitable for scientific collaboration between Poland and Germany. It builds bridges across nations and disciplines to tackle the global challenge of hate speech and searches for answers that are not only relevant for Poland and Germany but of high interest to many countries of the world. For these reasons, we will address the issue of potential similarities between hate speech-specific linguistic and acoustic features across both languages to identify features that are language independent.

• DFG Programme: Research Grants
• International Connection: Poland
• Partner Organisation: Narodowe Centrum Nauki (NCN)
• Co-Investigator: Dr. Dennis Küster, Ph.D.
• Cooperation Partner: Dr. Aleksandra Swiderska