Multilingual Embeddings as Maps of Equivalent Meaning (MEMEM)

Funded by: University of Vienna | HUJI-UNIVIE Joint Research Grant
Duration: 2024-2026
PI: Fabienne Lind
Collaboration: Hajo Boomgaarden • Annie Waldherr • Jana Bernhard-Harrer • Ahrabhi Kathirgamalingam

In this cross-country research project, the MEMEM team will investigate how multilingual embeddings (MLE) respond to linguistic and cultural differences in contemporary AI and computational methods. MLE play a specific role in mapping equivalent contents across languages and contexts. Specifically, the researchers will use MLE to measure gendered biases and discrimination in Austrian and Israeli media discourses about political topics. Further investigations will then show how and when mappings deviate from valid equivalence, and what differences in the underlying data are responsible for these shifts. Finally, MEMEM aims to discuss implications for cross-lingual AI and textual research and propose a research agenda toward advancing the accuracy and accountability of cross-lingual computational text analysis.

Beyond Computational Propaganda and Bot Activism: Investigating Social Media Suppression in Authoritarian Regimes (BeyondCBA)

Funded by: Vienna Science and Technology Fund (WWTF)
Duration: 2024-2028
PI: Hossein Kermani

The ways through which non-democratic forces use social media to manipulate minds and change the online opinions and behavior of citizens are a big concern in contemporary societies. The project Beyond Computational Propaganda and Bot Activism: Investigating Social Media Suppression in Authoritarian Regimes (BeyondCBA) focuses on the women, life, freedom, i.e., #MahsaAmini movement to enhance our understanding of such nefarious activities. The #MahsaAmini movement was a nationwide movement in Iran in protest to the brutality and anti-women discourse and rules in Iran, which became the center of attention worldwide. BeyondCBA investigates how the Iranian regimes employ cyber army and manipulative tactics like sharing disinformation to suppress the biggest anti-regime movement in its history. Combining traditional social science research methods with state-of-the-art computational methods, like automated text analysis, this project sheds light on regime manipulative campaigns on four popular social media platforms: Twitter, Facebook, Instagram, and Telegram.

The project explores the emergence of social issues on social media, which we observe in popular hashtag campaigns such as #metoo or #blacklivesmatter. These issues begin as discussions on social media, and with time, they can become crucial topics that evoke strong emotions and get everyone talking. At a certain point of viral spread, the issues are picked up by professional news media and make their way into the mainstream. It is extremely relevant for democracy and civil society to define which issues are important and need political and societal action, and the continuous development of social media will only increase its importance. However, science still needs a complete explanation of how such issues emerge and grow, as previously, researchers looked at different factors in isolation only.

As a starting point the project team theorizes that social issue emergence involves two parallel and seemingly contradictory processes: first, the viral spread of the issue on social media (for example, via hashtags and reposts), and second, control attempts by groups and individuals (for example, an activist group spreading a certain message or a celebrity sharing a personal story under a particular hashtag). These processes will be studied in-depth to understand their underlying logic and investigate which groups participate in the hashtag debates in which stages and in what way. In order to find this out, the project team will interview actors involved in social issue emergence, such as social media activists. Additionally, content on many social issues from social media, alternative media, and news will be analyzed with the help of machine learning algorithms. Further, the team will examine how issues evolve and debates on them change over time. Based on these empirical insights, a computer simulation showing how social issues develop and spread on social media platforms will be created.