<p>The history of Chinese and foreign legal systems and legal language shows that the long-term development of each legal system, and its language expression system is inseparable from the absorption of the advantages of other legal systems through translation. With the aid of the "translation stream," the modern socialist legal system with Chinese features is progressively reaching the vast nirvana that incorporates the benefits of diverse legal systems. In order to obtain more efficient and intelligent legal language translation, the study conducts comparative analysis at the theoretical level of words, sentences and texts, proposes translation principles and methods with legal characteristics, and develops a multi-level knowledge-based neural machine translation model. The research results show that after training, the MKNMT model only iterates 308 times, and the loss value can reach the minimum; the MKNMT model only iterates 316 times to achieve the target accuracy. In different training sets and translation tasks, the BLEU value of the MKNMT model is the highest. From the overall analysis, the lower the perplexity of the model, the higher the BLEU value. To sum up, the translation performance of the MKNMT model proposed in the study is very good, which successfully raises the machine translation model's translation quality, obtains better application results in the prevalent English–Chinese legal language, and increases the interest of legal readers in reading.</p>