Compounding and derivation are frequent in many language families. As a consequence, words in different languages are often only partially cognate, sharing only a few but not all morphemes. While partial
cognates do not constitute a problem for the phonological reconstruction of individual morphemes, they
are problematic when it comes to phylogenetic reconstruction based on comparative wordlists. Here, we
review the current practice of preparing cognate-coded wordlists and develop new approaches that make
the process of cognate annotation more transparent. Comparing four methods by which partial cognate
judgments can be converted to cognate judgments for whole words on a newly annotated dataset of 19
Chinese dialect varieties, we find that the choice of the conversion method has an impact on the inferred
tree topologies that cannot be ignored. We conclude that scholars should take cognate judgments in
languages in which compounding and derivation are frequent with great care and recommend to assign
cognates always transparently.