This directory contains the code, the data, and the results for the paper “Do Transformer Networks Improve the Discovery of Rules from Text?” The paper can be accessed here.

Due to the heavy computational nature as well as huge amounts of memory required to execute the code, we ran it on University of Arizona high performance computing Puma cluster which uses a Slurm job scheduler. We release the HPC execution scripts here as well. If you want to run the code on a different HPC setting, you need to change the scripts accordingly.

The contents of this directory are as follows:

• setup: contains the instructions to create the environment for executing the code.

• code: contains the source code as well as the data. The HPC execution scripts are also included.

• results: contains the experimental results from the paper.