Abstract: Dated language phylogenies have wide applications in historical and anthropological research. However, the quality of such applications depends on adequate models of language evolution. When a language splits into different speaker communities, the speed of language change increases such that a significant proportion of lexical disparity can be attributed to the splitting events. This could be explained by “schismogenesis”, i. e. the exaggeration of language differences to mark social boundaries. Surprisingly, this result has not found its way back into the clock models used in phylogenetic inference. Here, we develop an extension to existing phylogenetic clock models in BEAST2 to account for bursts at language splits. We validate the mode in a simulation studies and fit it to four major language families of the world: Austronesian, Bantu, Indo-European, and Sino-Tibetan. In each of the four families the burst clock model fits the data significantly better than a strict or relaxed clock without bursts. Astonishingly, the estimated bursts lead to changes equivalent to about 100 years of independent evolution. This makes bursts at splits a major contribution to language evolution, and our model may even be applicable to some cases of biological evolution.