Author(s): Loureiro, Albina Silva ; Ferreira, Patrícia
Date: 2025
Origin: Sensos-e
Subject(s): AI assisted translation; Crosslinguistic influence; Translator training; Error analysis; Portuguese/English translations; AI assisted translation; Crosslinguistic influence; Translator training; Error analysis; Portuguese/English translations
This study investigates whether AI-powered translation systems—specifically ChatGPT-3.5 and ChatGPT-4—reproduce patterns of Crosslinguistic Influence (CLI) in translations between European Portuguese and English. Two research questions guided the analysis: (1) To what extent do AI-generated translations reflect negative CLI? and (2) How does identifying and correcting such phenomena support the development of critical AI literacy and translation competence among students? A qualitative case study was conducted with 101 higher education students, all native speakers of Portuguese with C2 proficiency in English, enrolled in a Translation Practices course over two academic years. Students individually analysed AI-generated translations of satirical and opinion texts, identifying errors related to morphology, syntax, semantics, and cultural non-equivalence. Findings revealed recurrent patterns of negative CLI, such as incorrect verb moods, literal word order, misleading lexical choices, and culturally inappropriate expressions. While ChatGPT-4 produced more fluent texts than ChatGPT-3.5, both models reproduced similar CLI-related patterns. Pedagogically, these results underscore the value of integrating AI translation outputs into translator training as resources for error analysis, thereby fostering students’ critical engagement with technology and strengthening their professional competences.
This study investigates whether AI-powered translation systems—specifically ChatGPT-3.5 and ChatGPT-4—reproduce patterns of Crosslinguistic Influence (CLI) in translations between European Portuguese and English. Two research questions guided the analysis: (1) To what extent do AI-generated translations reflect negative CLI? and (2) How does identifying and correcting such phenomena support the development of critical AI literacy and translation competence among students? A qualitative case study was conducted with 101 higher education students, all native speakers of Portuguese with C2 proficiency in English, enrolled in a Translation Practices course over two academic years. Students individually analysed AI-generated translations of satirical and opinion texts, identifying errors related to morphology, syntax, semantics, and cultural non-equivalence. Findings revealed recurrent patterns of negative CLI, such as incorrect verb moods, literal word order, misleading lexical choices, and culturally inappropriate expressions. While ChatGPT-4 produced more fluent texts than ChatGPT-3.5, both models reproduced similar CLI-related patterns. Pedagogically, these results underscore the value of integrating AI translation outputs into translator training as resources for error analysis, thereby fostering students’ critical engagement with technology and strengthening their professional competences.
