In order to solve the problems of single learning mode, boring pages, lack of interaction and difficult evaluation existing in Oral English online education and traditional education, this paper proposes a model of Oral English learning based on artificial intelligence. The model is preprocessed by the Bert method to extract the embedded word vector of the sentence, and the multi-layer perceptron is used to calculate the weight value of the historical sentence involved in the coding. Because the speaker’s role and conversation habits will have a great impact on sentence analysis, this paper calculates the historical influence vector according to the role information and the weight value, fully extracts the historical sentence information through the BiLSTM network, and finally uses the softmax function to obtain the intention of the sentence and the probability distribution of slot filling. The experimental results show that the accuracy of intention recognition on dstc4 dataset is improved by 0.12% and the accuracy of slot filling is improved by 0.98% compared with DFF. Conclusion: the accuracy rate of intention recognition and slot filling in the dstc4 data set of the spoken English learning model based on artificial intelligence is significantly improved.