اللسانيات التطبيقية
Volume 10, Numéro 1, Pages 50-61
2026-06-04

The Translator’s Psycho-cognitive Competence Versus Neural Machine Translation (nmt) Performance: The Inevitability Of Data Curation In Translation Training

Authors : Alloune Nadir .

Abstract

This research investigates the widening psycho-cognitive gap encountered by trainee translators when engaging with highly technical institutional texts in the era of artificial intelligence. Centered on the transition from traditional Translation Memories (TM) to advanced Neural Machine Translation (NMT) architectures, the study employs an experimental methodology involving 47 Master’s students (MCIL) from the University of Algiers 2. Their translation outputs of technical economic reports from the Arab Monetary Fund (AMF) were compared against a generic NMT base model and a specialized system fine-tuned using a manually curated 1,200-sentence parallel corpus. The findings reveal a persistent state of "cognitive saturation" among students, who frequently defaulted to "automation bias" when utilizing general-purpose Large Language Models (LLMs). Conversely, the fine-tuned NMT model, anchored by specialized training data, achieved 100% terminological accuracy. The study concludes that the future of professional translation teaching lies not in passive tool adoption but in the mastery of "Data Curation" and "Corpus Engineering," effectively shifting the translator’s role from a linguistic producer to a high-level digital overseer.

Keywords

psycho-cognitive competence ; Marian-NMT ; fine-tuning ; data curation ; economic translation ; university of Algiers 2 ; post-editing