TuffyLite Build Status

TuffyLite is a modified version of the open source Tuffy solver for performing inference with Markov Logic Networks (MLNs). Specifically, it is an inference-only variant of Tuffy version 0.3 implementing the changes described in the following paper:

Markov Logic Networks for Natural Language Question Answering. Tushar Khot, Niranjan Balasubramanian, Eric Gribkoff, Ashish Sabharwal, Peter Clark, Oren Etzioni. StarAI-2015, 5th International Workshop on Statistical Relational AI, Amsterdam, The Netherlands, July 2015.

For more information about MLNs and the kinds of problems Tuffy is suitable for, please refer to the original Tuffy solver. The main changes to Tuffy implemented in TuffyLite are described below.

This project also includes minor modifications to the Glucose 3.0 SAT Solver to identify variables with fixed values by unit propagation, backbone detection, or equivalence detection. This information is (optionally) used by TuffyLite iteratively to simplify and often vastly reduce the size of the grounding of hard and soft constraints by exploiting structure imposed by hard constraints.

List of Modifications

The main additions to Tuffy include:

On the down side, TuffyLite focuses mainly on the inference aspect, especially marginal inference, and currently does not support the following:


Please contact Eric Gribkoff or Ashish Sabharwal if you have any questions or comments.