Il post-editing di una traduzione automatica in direzione russo-italiano: un esperimento con apprendenti italofoni di russo L2

This paper presents the results of a pedagogical experiment aimed at examining the revisions produced by two groups of Italian learners of L2 Russian during a post-editing task. Participants were asked to revise a text automatically translated from Russian into Italian without prior training in post-editing. The source text – a news article from BBC Russia – had been translated into Italian using DeepL. The study is part of a broader project designed to integrate AI-based machine translation (MT) tools into the teaching of Russian as a foreign language and translation training. It follows a preliminary experiment conducted in March 2024, which assessed students’ ability to distinguish between a human translation, a raw machine translation, and a post-edited text (Biagini, Bonola, Noseda 2024). The present experiment involved 20 second- and third-year undergraduate students enrolled in Russian language courses at the Faculty of Foreign Languages and Literatures at Università Cattolica del Sacro Cuore in Milan and at the Department of Interpreting and Translation at Università di Bologna (Forlì Campus). The revisions were classified into several categories: formal adjustments, morphological, syntactic and lexical revisions, explicitations, omissions, and corrections of MT errors. The analysis shows that most interventions concerned the lexical level. Only in a limited number of cases, did syntactic revisions prevail, despite the fact that MT output often reproduces Russian syntactic structures, which differ substantially from Italian. The findings indicate that students possess adequate lexical revision skills, whereas their competence in contrastive textual analysis remains insufficient for more comprehensive intervention. Further training in human translation, combined with the development of contrastive linguistic awareness and systematic comparison between human and machine-translated texts, appears necessary to foster the critical competence required for the effective use of MT systems and large language models.