The project addresses improvements in natural language processing with cross-lingual knowledge transfer. For example, cross-lingual predictive models for hate speech detection that use English for learning can also be used in Hindi or Slovenian. Our existing research has shown that by combining linguistic resources from several languages in combination with English, it is possible to create highly successful crosslingual models even for languages with few linguistic resources. The project will build tools for extraction of useful information from social media, such as people's attitudes towards epidemiological measures, political situations and officials. The inclusion of geolocation metadata in the analyses will also allow for inclusion of geographic region. The work will focus on three languages, English, as a language with many linguistic resources, as well as Hindi and Slovene, where the cross-linguistic models will be applied. The developed technologies will improve linguistic and sociological analyses of non-standard communication in social media, and will benefit broader area of digital humanities.