Abstract:

Students in higher education regularly engage with foreign-language and linguistically complex scientific literature, which imposes significant comprehension challenges, particularly for non-native speakers. Conventional translation tools address surface-level language barriers but do not support the deeper interpretive processes required for genuine academic understanding. This paper presents Lingua, an AI-powered language assistance prototype based on GPT-4-Turbo, developed through iterative prompt engineering to support students in comprehending foreign-language and complex academic texts. Tool design was grounded in Cognitive Load Theory (CLT) and an extended Technology Acceptance Model (TAM) and informed by a quantitative preliminary needs assessment. A controlled between-subjects laboratory experiment (N = 66) compared student perceptions of Lingua with the established machine translation system DeepL across five TAM-based acceptance dimensions: Perceived Usefulness (PU), Cognitive Relief (CR), Perceived Ease of Use (PEOU), Output Quality (OQ), and Behavioral Intention (BI). All measurement scales demonstrated good-to-excellent internal consistency (α = .81–.91). Lingua significantly outperformed DeepL across all five dimensions (p ≤ .007), with medium-to-large effect sizes (Cohen’s d = 0.69–1.40). The findings demonstrate that task-specific AI configuration via prompt engineering can produce comprehension-oriented language tools that substantially exceed general-purpose translation systems in student acceptance.

Keywords:

AI language assistance, academic reading, technology acceptance, cognitive load, higher education, GPT-4, prompt engineering.