by Danguolė Kalinauskaitė & Tomas Krilavičius, Vytautas Magnus University, Lithuania.
In March 2021, four Lithuanian institutions, namely, Vytautas Magnus University, Lithuanian Human Rights Center, European Foundation of Human Rights, and Department of National Minorities under the Government of the Republic of Lithuania, launched a joint 2-year project #BeHate-Free: Building Hate-Free Communities in Lithuania. Within the scope of the project, we addressed multiple challenges posed by dealing with hate speech online, such as lack of capacity to identify and report hateful content, and tolerance towards hate speech, to mention a few. Accordingly, all project activities were systematic efforts to give an effective response to an increasing prevalence of intolerable negative content directed at various vulnerable groups of society in Lithuania.
Ethnic, religious, racial, linguistic, cultural, gender and sexuality minorities, migrants, refugees, women and any so-called “other” become the targets of hate speech that is spread by those who are intolerant of any differences between “them” and “others”. By its very nature, hate speech belongs to an umbrella of related negative concepts, such as abusiveness, aggressiveness, racism, etc. Moreover, online hate speech is seen not only as a linguistic and social phenomenon with various tones and forms but also as a cyber threat with challenges faced by automated approaches for its identification. Hateful messages online can be produced and spread through comments on news portals or posts by users/groups on various platforms, they do not go through an editing process, can reach very different levels of exposure depending on the channels and means used, even go viral cross-nationally and, furthermore, they can be generated and posted automatically, i.e. using bots. This means that the elements contained in such messages, including manipulation techniques and rhetoric used to spread hate, can evolve and peak very quickly. Hate speech online can also be available for longer, as compared to hate speech offline, connect with new networks or reappear, as well as be anonymous. Needless to say that these form the basis for circulation of misinformation/disinformation, conspiracy theories.
During the #BeHate-Free project, we organised film screenings, public discussions, and creative workshops to address the lack of knowledge about hate speech online in Lithuania and means to tackle it. With regard to the lack of digital education, we also aimed to strengthen digital skills, tolerant online culture among youths and critical thinking abilities of young people from youth centres across Lithuania, as well as to build the capacity of youth workers to educate young people about hate speech online and develop innovative educational tools to analyse hate speech. Training of youth workers and seminars for teachers and youth workers, as well as development of training methodology on identifying and reporting hate speech, served these purposes. An important objective of the project was to raise public awareness about the issue of hate speech online and its implications on the society at large, which was necessary in order to transform the existing online narratives into more inclusive and positive ones.
Finally, a significant stage of the project was dedicated to the research of hate speech on Lithuanian media platforms in order to better understand the phenomenon itself (what discriminatory narratives are constructed, and what triggers them), to define the concept of hate speech, thus distinguishing it from other types of negative content and enabling to recognize hate speech in context, as well as to generally address the main challenges pertaining to automatic identification of hate speech online. The intended result of the research was an AI-based hate speech detection and media monitoring prototype for the Lithuanian language. Although hate speech detection is still an extremely difficult task, methods for identifying hate speech and related abusive behavior are getting better in terms of performance and generalization. Most of the work is now carried out on deep learning approaches. Neural network architectures, such as convolutional neural networks and recurrent neural networks, have been successfully adapted for the task of hate speech detection.
We employed deep learning to develop our solution, which enabled us to create an innovative hate speech detection tool using artificial intelligence. The developed tool was tested in practice by applying it for hate speech monitoring in selected media, namely, various Lithuanian Telegram channels. By doing this, we discovered groups and channels there that were used to spread hatred online. The research also made it possible to find out what discriminatory narratives are constructed. Main topics identified in those narratives are foreigners, LGBT, gender, and jews. After the latter activities, the tool for hate speech detection was prepared for usage to monitor the public online space in Lithuania, analyse it and systemise the gathered data to lift the level of knowledge about the phenomenon. The AI tool development, in turn, has led to a significant increase of the awareness of the challenges in such tool development and improvement. Thus, the research of the Lithuanian online public space served as a diagnostic of the status quo with regards to hate speech online and the baseline to develop effective and appropriate solutions.
Despite focusing on the national context, our project activities were built on the work of the EU High Level Group on combating racism, xenophobia and other forms of intolerance, especially in the areas of ensuring quality, sustainability, coordination and addressing actual or perceived barriers to reporting.
Danguolė Kalinauskaitė, PhD, Researcher, Lecturer, Vytautas Magnus University, Lithuania
Tomas Krilavičius, PhD, Professor, Dean of the Faculty of Informatics, Vytautas Magnus University, Lithuania
The project #BeHate-Free: Building Hate-Free Communities in Lithuania was funded by the European Union’s Rights, Equality and Citizenship Programme (2014-2020).