A Tractable First-Order Probabilistic Logic

Tractable subsets of first-order logic are a central topic in AI research. Several these formalisms have been used as the basis for probabilistic languages. However, intractable, losing original motivation. Here we propose first non-trivially tractable language. It is subset Markov logic, and uses class part hierarchies to control complexity. We call it TML (Tractable Logic). show that knowledge bases allow efficient inference even when corresponding graphical models very high treewidth. also how inheritance, default reasoning, other patterns can be carried out TML. opens up prospect large-scale inference.