When Correct Isn’t Fluent: Translationese in Turkish-to-English Academic Writing and Its Implications for EAP

Turkish scholars, like many non-native English speakers, routinely translate manuscripts written in their first language before submitting them to international journals, a practice that leaves linguistic traces even in grammatically sound texts. This study examines that residue, commonly termed translationese, in the introductions of 40 bilingual articles from Education and Science, an SSCI- and Scopus-indexed Turkish education journal, using a reference-free neural quality-estimation metric (COMET-QE) to score each English introduction against its Turkish source. After excluding five articles for alignment issues, scores across the remaining 35 clustered within a comparatively narrow band (M = 0.68), suggesting source-language interference is pervasive rather than occasional in this corpus. Contrary to expectation, neither how much a translation expanded relative to its source nor its raw length showed a significant relationship with its score. Close analysis of the lowest- and highest-scoring texts instead pointed to specific choices, such as calque, repetitive noun-phrase reference in place of pronouns, and clause structures that preserved Turkish syntax at English's expense, as more reliable markers of quality than any global text property. For English Language Teaching, particularly academic writing instruction for Turkish-speaking researchers, these findings suggest that raising awareness of specific interference patterns may serve learners better than generic fluency advice, and that quality-estimation tools could support self-editing before submission.