An Experimental Study of Neural Approaches to Multi-hop Inference in Question Answering

This package provides the experimental configurations, data, results, and implementations required to perform an experimental approach to neural inference in question answering.

The folder structure is as follows:

- Configurations: provides JSON files with all of the configurations that must be tried out during grid search.
- Data: provides the 01K and the 10K bAbI repositories in English (en), Spanish (es), Hindi (hn), and Portuguese (pt).
- Results: provides the models learnt in H5 format and the experimental results.

The main script is "generate-script.cmd". It is a factory that generates a command-line script called "run-experiments.cmd" that runs all of the experiments in sequence. The remaining are Python scripts that implement our proposal.

To run the experimentation, please, create a Python enviroment and install the requirements in file "requirements.txt" using PIP. Activate the enviroment and the execute script "run-experiments.cmd". Please, realise that the original experimentation took roughly 20 months to complete on a high-end server.