Nonnative Mandarin speakers always have some unnatural pauses when speaking Mandarin due to their native pronunciation habits. Accurately predicting the prosodic structure of Chinese sentences is the key to improving fluency in Mandarin for nonnative speakers. This paper investigated the influence of the Chinese prosodic boundaries on the Mandarin spoken by international students. First, we proposed a new method to predict the prosodic word and prosodic phrase boundaries from Chinese sentences to obtain the prosodic boundaries automatically. Then, we used the predicted results to improve the Mandarin spoken fluency of international students. To train the prosodic boundary prediction model, we firstly constructed a Chinese prosodic boundary corpus that includes 100,000 Chinese sentences with manually labeled prosodic boundaries under the guidance of a linguist. We also proposed an end-to-end Chinese prosodic boundary prediction model based on the sequence-to-sequence model with a new feature named number of syntax hierarchy (NSH). Finally, we assess the fluency score of Mandarin using 1300 utterances recorded by six international students and a native Mandarin speaker. The utterances are recorded without/with the predicted prosodic boundaries. The experimental results show that the F 1 scores of the prosodic word prediction model and the prosodic phrase prediction model are 98.14% and 85.24%, respectively. The fluency assessment results show that the fluency score labeled with the prosodic boundaries is higher than the fluency score of the international students when they read freely. The improvement of the score is between 7 and 16. Therefore, our method can be applied to the Mandarin education system to improve the spoken Mandarin fluency of nonnative speakers.