Error Analysis in Digitally-Mediated Communication: A Corpus-Based Study of Student-Chatbot Interactions within the UNITE Project

This thesis analyzes the types and distribution of errors made by Italian university students interacting with AI-powered chatbots in English, using the error-annotated corpus collected as part of the UNITE project. The study is guided by three research questions: the general distribution of error categories, the relationship between self-reported proficiency and linguistic accuracy, and the specific challenges within Grammar and Lexis. The results show that Digitally-Mediated Communication errors are the most frequent, followed by Grammar and Form errors that reflect more genuine linguistic difficulties. A clear link between proficiency level and accuracy was also found, with more advanced learners producing longer and more accurate output. The qualitative analysis points to L1 interference, monitoring issues, and incomplete rule acquisition as the main causes behind recurring errors in areas such as articles, verb tense, noun number, and fixed word combinations.