Title: Can Professional Translators Detect Text Created by AI?

Abstract:

This research explores the capacity of professional translators, who lack specific training in this area, to accurately recognize short stories in Italian that were produced by artificial intelligence (AI). The study involved an in-person experiment with 69 translators, who evaluated three anonymized narratives: two generated by ChatGPT-4o and one authored by a human. Participants were asked to judge the probability of AI involvement for each story and to explain their reasoning. Although the overall averages did not yield definitive results, a statistically significant minority (16.2%) correctly identified the synthetic texts, indicating that their assessments were based on analytical capabilities rather than random guessing. Conversely, a comparable number of participants made the opposite error, frequently depending on subjective feelings instead of objective criteria, which may highlight a readerly bias toward AI-generated content. The analysis identified low burstiness and narrative inconsistencies as the most dependable signs of machine authorship, alongside the presence of unexpected calques, semantic loans, and syntactic transfers from English. On the other hand, attributes like grammatical precision and emotional resonance often resulted in misidentification. These outcomes prompt critical inquiries regarding the application and limits of editing synthetic texts within professional environments.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC