Through studying natural language, my research develops computational methodologies that model different dynamics in human morality. You can find descriptions of my research directions below.

Computational inference of societal moral change

Understanding how morals change is crucial for addressing societal challenges and predicting the future dynamics in society. Working with large-scale machine learning algorithms, such as language models and graph neural networks, my research develops frameworks to study the temporal shifts in people’s moral values, and what drives these shifts.

Moral variations in large language models and NLP technologies —— Language models are one of the most exciting advancements in AI and people have started to use them in many different ways. In recent applications, these models have shown the potential to retrieve and describe human moral values. However, the extent of this knowledge is limited, as language models often struggle to capture the nuances in moral values of different cultures and historical settings, leading to misrepresentation of marginalized communities and harmful biases in language generation. My research, thus, studies language models’ performance in various moral scenarios and investigates how their parameters and inner architecture could enable or restrict certain moral and ethical behaviors.

Evolution and acquisition of moral language

I am interested in understanding the origins and the development of moral language in humans. By drawing insights from evolutionary studies, developmental psychology, and computational studies of semantic change, my research examines how humans have developed a moral lexicon over time and how children begin to learn such lexicon to communicate their moral concerns.