Universal part-of-speech tags and universal features in the Czech data have been obtained by an automatic conversion of the PDT morphological tags via Interset.
Interset uses only the tags as the source information. In some cases, the universal POS tags and features were further (automatically) refined based on word forms, lemmas and dependency relations. In particular, the lemmas in PDT themselves may contain additional features; these were removed from lemmas and converted to additional language-specific features in the FEATS column.
The set of universal features was extended with several additional features for Czech. Some of the additional features are language-specific, others are more general and they are needed to preserve information originally encoded in PDT. On the other hand, the u-feat/Definite feature is not used in Czech.