Authors:

Karsten Assoua

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

as textbooks used in different U.S. states and school systems. Building on this foundation, our project investigates moral messaging in a different context: fictional children's books. Unlike textbooks, these stories often convey values implicitly, e.g. through narrative structure, character development, and subtext, rather than through explicit facts or named entities. Our goal is to evaluate whether small language models (SLMs), specifically DistilBERT and Longformer, can be fine-tuned to accurately and consistently identify moral themes embedded in such stories. We focus on a curated set of eight core moral values and train our models to detect their presence or absence in a diverse corpus of children's literature. Our approach emphasizes thematic comprehension over surface-level prediction, aiming to surpass generic large language models in both nuance and specificity. By iteratively refining the model over multiple epochs, we aim to develop a lightweight, efficient system capable of moral theme extraction that reflects deeper narrative understanding and paves the way for ethical AI tools in literary analysis and moral psychology. 137 Katelyn Williams:

Source:

Brown / SPRINT|Undergraduate Teaching and Research Awards (UTRA)

Topics:

No topics listed

Co-authors:

Karsten Assoua