Linguists have acknowledged for a long time that words have their own history that does not necessarily fully align with the history of the languages as a whole. Recent approaches in computational historical linguistics have started integrating this fact into their models of language change, yet without modelling linguistic processes like word formation in detail. Inspired by advances in evolutionary biology, we propose a framework for annotating etymological relationships within word families as word trees to make this kind of data available for quantitative and qualitative studies. We test our dataset on a data sample of etymologies and include a small selection of Python scripts that enable checking the annotation for consistency and deriving basic statistics.