Chiara Roberta Casale
Analysis of Large Language Models' Narrative Construction Across Languages: A Comparative Study of English and Italian Event Descriptions.
Rel. Antonio Vetro', Marco Rondina. Politecnico di Torino, Corso di laurea magistrale in Data Science And Engineering, 2026
|
Preview |
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (5MB) | Preview |
Abstract
Large language models are increasingly used as sources of information, and while prior work has extensively examined their biases, less is known about how consistently they describe recent political events across languages, prompts, and models. This thesis investigates how depictions of politically contested protests vary across promps, languages, event cases, and models. More specifically, it examines anti-ICE protests in the United States, protests and uprisings in Iran, and the Askatasuna protests in Italy. Responses were collected from the ChatGPT, Mistral, and Qwen web interfaces in English and Italian using neutral, protester-critical, authority-critical and headline-generation prompts. The 72 conditions were each repeated ten times, producing a corpus of 720 responses.
The preselected first response from each condition underwent reflexive thematic and source analysis, focusing on event framing, the actors involved, protesters’ demands, violence, factual content, and source use; semantic similarity with the nine subsequent responses was calculated to describe output variation
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
URI
![]() |
Modifica (riservato agli operatori) |
