Best results in identification of machine-generated text 馃
The TALN-UPF team organised as part of the ERINIA project has participated in the AuTexTification shared task. Its goal was to test computational methods of identifying text generated by large language model, such as GPT-4 or ChatGPT. The results show that the submission prepared by our team (Piotr Przyby艂a, Nicolau Duran Silva and Santiage Egea Gomez) achieved the best perfomance in subtask 1 (differentiating between machine and human-authored text), both for English and Spanish language. For subtask 2 (detecting the language model that generated a given text), our team was ranked 4th for English and 2nd for Spanish. We are looking forward to presenting our solution at the 5th Workshop on Iberian Languages Evaluation Forum at the SEPLN 2023 Conference, which will be held in Ja茅n, Spain in September 2023