With rising demand for precise translation in specialized STEM fields, this study compared human translations (third-wave globalization, 1990–2019) with Matecat-assisted outputs (fourth-wave, AI-integrated workflows post-2019) of English-to-Persian scientific article introductions using Reiss’s (2000) linguistic-functional model, emphasizing semantic accuracy, lexical appropriateness, and grammatical correctness. A convergent mixed-methods design was employed: the corpus included eight English introductions (two each from Chemical Engineering, Chemistry, Industrial Engineering, and Physics), produced as archived human versions (2013–2017) and 2025 Matecat translations. Five MA students in Translation Studies blindly rated all 16 texts on a 5-point Likert scale (240 ratings total) with optional comments; quantitative analysis encompassed descriptive statistics, ICC, and Kendall’s W for reliability, and a Wilcoxon signed-rank test, supplemented by exemplar-based textual analysis. Matecat translations achieved significantly higher mean quality scores (T = 0, p .01, r = .63), excelling in grammatical correctness and fluency, whereas human versions showed greater inter-rater consistency and stronger terminological stability and semantic nuance despite lower fluency. Qualitative feedback revealed Matecat’s lexical variability as a source of evaluator divergence. These results highlight the potential of hybrid human–machine workflows for specialized translation and offer implications for translator training, CAT-tool enhancement, and quality assurance in multilingual STEM contexts.
Baghshahi et al. (2025) studied this question.