Word sense disambiguation is a common problem in the field of English language processing. In this paper, we use the Transformer model and LSTM model to construct a fusion model for word sense disambiguation, which provides a method to solve the problem of low accuracy in English vocabulary disambiguation. This paper first introduces the construction of a Transformer-based context embedding model to achieve word sense disambiguation, which effectively captures semantic and sequential information in the context. On the basis of the Transformer lexical disambiguation model, the fusion model of lexical disambiguation is proposed by integrating the LSTM network, and the efficiency of lexical disambiguation is improved by taking advantage of the long and short-term dependency properties of the LSTM network and the parallel processing mechanism of Transformer network. The average disambiguation accuracy of this model reaches 75.24% in English word disambiguation, and the average disambiguation accuracy increases and decreases by less than 5% in different language scales, and the F1 scores of words with different lexical natures are more than 80. The average disambiguation accuracy of this model is higher than those of the comparative LSTM models in different disambiguation features. The English word disambiguation model’s overall performance in this paper is satisfactory.