<p>Emotion mining is one of the relevant areas of natural language processing. Nowadays, the problem of emotion extraction from text is not completely solved. The paper deals with the main problems of automatic emotion mining related to language features as well as addresses methods to solve the said problems. The article describes an experiment of analysis of the system for automatic sentiment analysis Indico for Russian and English languages on the example of automatic extraction of the emotion “joy.” Testing was carried out on the parallel corpus of Russian and English political texts. A comparative analysis of results for both languages and selective manual analysis of the texts was conducted, and the main errors were described. According to the results, it can be concluded that the quality of the automatic sentiment analysis is not high enough. In conclusion, the possible causes of the system's errors were considered.</p>