The talk summarizes findings of a three-year project on cross-lingual semantic similarity, captured as synonym sets (classes). Our attention is aimed at verbal synonymy in a bilingual (Czech-English) context and based on both syntactic and semantic criteria, focusing on the relation between valency behavior and semantic roles of verb senses of generally different verbal lexemes in translated texts. The research is approached from the Functional Generative Description point of view and supported by the textual evidence taken from the Prague Czech-English Dependency Treebank. The resulting synonym classes are captured in a bilingual synonym verb class lexicon, called CzEngClass. The individual occurrences in the parallel corpus are being linked to the CzEngClass entries, and through them, for comparison and further exploration, to other existing lexical resources, namely FrameNet, VerbNet, PropBank, and English WordNet, while the original links to the Prague Dependency Treebank-style valency lexicons remain in place.