Metaphorical and fixed Expressions in Students’ Translation: A Systematic Self-Review of Studies on Error Types, Strategies and Error Sources

Abstract

This study sought to conduct a systematic review (SR) of the author’s research program on student translators’ difficulties in translating metaphorical and fixed expressions between English and Arabic. The corpus comprised 12 studies categorized into 3 thematic clusters: General and technical language metaphorical and fixed expressions, and human vs AI translation weaknesses.  Across the 12 studies with expressions involving impossibility, kinship (ibn/bint & Om/Abu), numeral, time, dar/bayt, color, binomials, or word + preposition collocations, the students translated fewer than 35% correctly, left many items on the test blank, and relied heavily on literal translation. Expressions similar in English and Arabic were easy, whereas opaque and culture‑specific expressions were difficult to translate. In the study on common names of chemical compounds, the students translated fewer than 20% correctly and defaulted to transliteration or literal equivalents. AI outperforms students in accuracy, but also defaults to literal translation, especially when expressions require cultural grounding or conceptual alignment. The translation weaknesses observed across all studies reveal deeper linguistic, cognitive, and cultural factors. Student translators rely on surface level lexical processing rather than conceptual mapping, show limited awareness of how metaphors, metonyms, and set phrases operate in English and Arabic, and demonstrate low tolerance for ambiguity and weak inferencing skills when confronted with opaque expressions. Cultural literacy gaps further hinder performance, as many expressions require familiarity with religious, historical, scientific, or social references. Structural complexity (as in binomials & collocations) interacts with semantic opacity, compounding the difficulty and affecting both meaning and syntactic accuracy. Although AI is more accurate overall, it still struggles with conceptual alignment, often prioritizing direct lexical equivalence and producing culturally awkward or semantically distorted translations. Overall, the corpus indicates that translation difficulties stem not from isolated linguistic gaps but from a systemic lack of conceptual, cultural, and cognitive readiness to handle non literal meaning. The significance of this SR lies in providing the first unified, empirically grounded account of how Arabic English student translators, and AI systems, struggle with non literal meaning across multiple figurative domains, offering a foundation for redesigning translator training, curriculum development, and AI assisted pedagogy.