Through the advance of information extraction and data mining, a number of knowledge bases (KBs) have been created, for instance, NELL and Google knowledge Vault. In line with this, probabilistic extensions of various description logics have been proposed for reasoning in probabilistic KBs. However, most of these languages are not tractable impeding their practical use. Since present-day KBs can be very large, tractable reasoning is essential. In this work, we propose probabilistic extensions of OWL 2 RL and OWL 2 EL by using probabilistic soft logic for which inference is known to be tractable. We show that inference in probabilistic extensions of OWL 2 RL and OWL 2 EL remains tractable. We present experimental results over a YAGO KB that contains hundreds of schema axioms and thousands of instances.