Lexical compositionality has received some attention in historical linguistics, linguistic typology, and cognitive linguistics, but so far studies have mostly concentrated on the morphological complexity of individual words and languages, while the fact that words form families which interact during language change and language use has been typically ignored. As a result, we do not know how word families evolve along language phylogenies, which semantic processes underlying word family formation are universal, and to what extent human cognition influences the productivity of lexical roots to form families. In order to tackle these questions in future work, I propose some new ideas that help to annotate cross-linguistic data on word families. These ideas include concrete suggestions for the handling of non-concatenative word formation in comparative wordlists, new techniques for the annotation of regular sound correspondences, and some initial thoughts on the consistent annotation of semantic relations among word family members.