Linguistic phylogenies are standardly inferred from lexical cognate relationships (e.g., Bouckaert et al. 2012, Chang et al. 2015, Sagart et al. 2019). Despite the prevalence of this practice, it suffers from well-known drawbacks. First, it disregards the phylogenetic signal that exists in the form of the words themselves. Second, it limits the modeling possibilities since it relies on an arbitrary coding of the data. In this talk, I introduce a novel framework for linguistic phylogenetics that overcomes both of these shortcomings. The heart of this framework is the TKF91 model (Thorne et al. 1991), which allows phylogenetic inference to be carried out directly from cognate word-forms. This model not only opens up a new horizon in the study of linguistic phylogenetics, but allows historical linguists to investigate questions of sound change that were previously out of reach.