With the growing demand for multilingual dissemination of online literature, neural machine translation must cope with pervasive colloquial expressions, long-span sentences, and plot-driven variations of recurring character titles and terminology. To address these challenges, we propose an enhanced PPBD-Transformer (Phrase-Prior and Bidirectional Decoding Transformer)for Chinese–English and English–Spanish online-literature translation, targeting multilingual scenarios in cultural exchange and digital literary communication. We introduce phrase-aware adaptive training and a phrase dropping regularization mechanism to strengthen phrase-level alignment and cohesion, while incorporating a backward decoder that is jointly optimized with a forward decoder under a shared encoder to better capture long-distance dependencies and discourse cues typical of narrative text. Experiments on multiple corpora show that, compared with the standard Transformer baseline and several competitive phrase-aware variants, PPBD-Transformer consistently achieves the best performance on the Chinese–English and English–Spanish test sets, improving the Bilingual Evaluation Understudy (BLEU) score by 3.64 and 4.97 points, respectively. These results indicate that the proposed approach provides an effective and practical solution for multilingual translation of online literature, and offers technical support for cross-lingual cultural dissemination, educational applications, and research in digital humanities.
Chen et al. (Fri,) studied this question.