Moral content exploration in psychotherapy sessions

Keywords: Moral Foundations Theory, value conflict, psychotherapy, large language models
Project partner: Dr. Helene Seaward
CeDA collaborator: Rodrigo C. G. Pena
Repository: moral-content-search

The Moral Foundations Theory (MFT) organizes moral intuitions into distinct "foundations," each thought to have evolved to address recurring social challenges:

• Care/Harm: highlights the importance of avoiding harm and promoting the well-being of others. Associated with virtues such as kindness, nurture, and compassion.
• Fairness/Cheating: focuses on the concepts of justice, equity, and reciprocity. Associated with virtues such as fairness, justice, and equality.
• Loyalty/Betrayal: highlights the importance of loyalty to one's group and disapproval of betrayal. Associated with virtues such as loyalty, patriotism, and self-sacrifice.
• Authority/Subversion: promotes respect for legitimate authority, social order, and hierarchies. Associated with virtues such as leadership, respect, and obedience.
• Purity/Degradation: focuses on ideas of cleanliness and contamination, often extending to moral concepts. Associated with virtues such as chastity, wholesomeness, and piety.

Identifying these dimensions in text is usually modeled in the literature as a multi-label classification problem over context windows.

In this project, we are working with the following definition of value conflict: segments in which participants in the therapeutic interaction express differing moral foundations.

• Automatically identify value conflict passages in psychotherapy transcripts.
• Build a tool to explore moral content and value conflict passages in transcripts.
• Adapt the automatic value conflict identification and exploration to run on a private dataset, using simple hardware.