In short sequence translation tasks, there are problems such as data scarcity and contextual information, and traditional translation methods often fail to effectively capture the before and after dependencies in the source language, resulting in poor translation accuracy and fluency. This study introduces a short sequence English translation method based on Bidirectional Recurrent Neural Network (Bi-RNN) to improve the quality and efficiency of short sentence translation. Firstly, this paper utilizes the bidirectional information processing capability of Bi-RNN to enable the research model to simultaneously capture both forward and backward contexts in the sequence. Subsequently, this paper introduces an attention mechanism, which further enhances the quality and accuracy of translation in the model. Finally, the experimental results of this paper show that the Bi-RNN model cited is significantly superior to traditional LSTM (Long Short Term Memory) models and statistical based methods in terms of translation accuracy, fluency, and context dependency processing. Among them, the Bi-RNN model achieved a score of 0.70 in BLEU-4 (Bilingual Evaluation Understudy-4), while the LSTM and statistical models were 0.65 and 0.60, respectively. In addition, the Bi-RNN model also achieved a maximum score of 4.3 in fluency evaluation. In addition, the ratings in all 10 sentences remained at a high level, reaching a maximum of 4.8 points. In the above data conclusions, the translation method combining bidirectional recurrent neural network and attention mechanism can effectively improve the overall performance of short sequence machine translation.