AI-based translation systems have advanced substantially; however, accurately conveying cultural and pragmatic nuances remains a major challenge in English–Chinese translation. This study presents a systematic comparison between traditional Statistical Machine Translation (SMT) and modern Neural Machine Translation (NMT) approaches, with a specific focus on cross-cultural communication effectiveness. We evaluate a Moses-based SMT system and Google Translate as a representative production NMT system using a curated diagnostic corpus of 200 sentences across four genres, complemented by an external publicly available English–Chinese benchmark (IWSLT17) for larger-scale validation. Performance is assessed using automatic metrics (BLEU, TER) and human evaluations of adequacy, fluency, and cultural appropriateness, complemented by a fine-grained error analysis. Experimental results show that NMT consistently outperforms SMT across all metrics, achieving a higher BLEU score (41.23 vs. 32.45), improved cultural appro- priateness (3.78 vs. 2.45 on a five-point Likert scale), approximately 75% fewer syntactic errors, and more effective handling of culturally adaptive translation strategies. Despite these gains, both systems exhibit limitations in addressing deeper pragmatic and context- dependent phenomena. The findings highlight the strengths and remaining gaps of contemporary NMT systems in cross-cultural translation and establish a reference point for future research on culturally aware and context-sensitive translation models.
Zhang et al. (Wed,) studied this question.