Lost in Translation? Evaluating Machine Translation for Psychoeducational Reports
摘要
Multilingual (ML) students represent a growing portion of the U.S. public school population, yet school psychologists often lack the linguistic resources needed to provide translated psychoeducational materials. Machine translation (MT) tools such as Google Translate (GT) and ChatGPT-4o (GPT) offer potential solutions, but their effectiveness for translating psychoeducational reports remains underexamined. This study evaluated the fluency, accuracy, and error patterns of English-to-Spanish translations generated by GT, GPT, and a professional human translator (HT) using a fictional psychoeducational report summary. Two bilingual graduate students rated 71 translated sentences using fluency and accuracy rubrics, with disagreements resolved by a bilingual school psychologist. Results indicated no significant differences in fluency across systems. Accuracy differed significantly across systems overall, although follow-up pairwise comparisons were not statistically significant after correction. Analyses also revealed significant differences in the total number of errors produced across translations, but no significant differences in the distribution of specific error types. Across systems, errors included mistranslations, grammatical errors, untranslated words, and critical inaccuracies that affected interpretability. Overall, no translation method produced a perfect translation. However, results remain preliminary and require further study. Implications for school psychologists include the need to improve the readability of psychoeducational reports and ensure that translators’ workflows are supported by appropriate training and oversight.