Vector representation of words is used for various tasks of automatic processing of natural language. Many methods exist for the vector representation of words, including methods of neural networks Word2Vec and GloVe, as well as the classical method of latent semantic analysis LSA. The purpose of this paper is to investigate the effectiveness of using network vector methods LSTM for non-classical pitch classification in Russian and English texts. The characteristics of vector methods of word classification (LSA, Word2Vec, GloVe) are described, the architecture of neural network classifier based on LSTM is described and vector methods of word classification are weighted, the results of experiments, computational tools and their discussion are presented. The best model for vector word representation is Word2Vec model given the training speed, smaller word corpus size for training, greater accuracy and training speed of neural network classifier.