BLEU Function Analysis of Machine Translation Based on Transformer Model

The internal structure and functions of the encoder and decoder of the Transformer model are reviewed, and the applications of self-attention mechanism, feed forward neural network, residual connection, and layer normalization are analyzed in the field of natural language processing. The Transformer model is introduced into machine translation as an engine, then the translation experiment is conducted, and the translation quality of the Transformer model is evaluated with the BLEU function. The findings reveal that, within the similar Transformer architecture, translations generated by GPT-4o as the translation engine exhibit higher quality compared to those produced by Youdao translation engine. However, the BLEU scores for both engines are relatively low, indicating a significant gap from the reference translation. This underscores the necessity for expert post-editing to further enhance quality, highlighting that human-machine interaction is currently an essential pathway for improving machine translation quality.