mathworld

World Models for Math Story Problems

This repository contains code for the paper:

World Models for Math Story Problems (Accepted at ACL Findings 2023)

Andreas Opedal, Niklas Stoehr, Abulhair Saparov and Mrinmaya Sachan

Start by installing all packages:

pip install -r requirements.txt

Citation

Please cite as:

@inproceedings{opedal-etal-2023-world,
    title = "World Models for Math Story Problems",
    author = "Opedal, Andreas  and Stoehr, Niklas  and Saparov, Abulhair  and Sachan, Mrinmaya",
    booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.findings-acl.579",
    doi = "10.18653/v1/2023.findings-acl.579",
    pages = "9088--9115",
}

Demo

You will find a demo of our world model code in mathworld_demo.ipynb. It shows some of the basics, like how to

load an annotated world model and view its metadata
visualize a world model
get the logical forms of a world model
apply the reasoning algorithm on a world model
create a new world model from logical forms

Data

The data with train and test splits can be found in output_files/data. You will find the annotated world model graphs stored in .json format in output_files/data/{dataset} (with dataset $\in$ [asdiv, mawps, svamp]). If you are only interested in the logical forms you have those available in .csv format in output_files/data (along with one sentence history for convenience).

Code

All code associated with building and manipulating MathWorld world model objects is located in the worldmodel directory. The functions used to visualize world models is found in utils/viz_helper.py.

The rest of the code is specific to our paper. preprocessing/ contains the code we used to preprocess the data from the various sources. experiments/predictor contains code used for the semantic parsing / solving (sec 5.1) and the generation (sec 5.3) experiments presented in the paper and experiments/probing contains the code used to run the knowledge probing experiments presented in the paper (sec 5.2).

eth-nlped/mathworld