The application of artificial intelligence and natural language processing for diagnosing depression using lexical markers in text is considered.Natural language processing methods and machine/deep learning algorithms that demonstrate high accuracy in identifying linguistic features of depression are analyzed. The problems and prospects of a multimodal approach and the development of transparent artificial intelligence systems for clinical practice are discussed.