In this project, we built reliable classifiers for sentiment analysis using Czech data. However, the project is also aimed at the area of pragmatics, namely at various ways of expressing evaluative meaning in Czech. Evaluative language is a type of meaning expression which applies across different genres, from everyday conversation up to scientific discourse. Though there are many resources of evaluative texts, and corpus linguistics offers both data and tools to handle it, there is still lack of systematic approaches to Czech evaluative language. We observe the ways the evaluative meaning is modelled in Czech using means from different layers of linguistic description. We describe the mutual interaction of lexical and structural meaning and the role of specific areas of research, such as valency, idiomatics, or coreference in the system of evaluative expressions. Moreover, we are interested in the influence of text domain on the choice of language means. The formulated linguistic hypotheses are tested using natural language processing methods, and the resulting model will be compared to other available models for different languages.