Skip to content

Latest commit

 

History

History
46 lines (35 loc) · 1.8 KB

README.md

File metadata and controls

46 lines (35 loc) · 1.8 KB

Probabilistic Worldbuilding from Language

If you use this data or code in your research, please cite:

@article{10.1162/tacl_a_00463,
  author = {Abulhair Saparov and Tom M. Mitchell},
  title = {Towards General Natural Language Understanding with Probabilistic Worldbuilding},
  journal = {Transactions of the Association for Computational Linguistics},
  volume = {10},
  pages = {325-342},
  year = {2022},
  month = {04},
  issn = {2307-387X},
  doi = {10.1162/tacl_a_00463},
  url = {https://doi.org/10.1162/tacl\_a\_00463}
}

This code depends on the following repositories: core, math, hdp, and grammar.

To use this code:

  1. Download the dependencies into a single directory.
  2. Download this repository and run make executive_test CPPFLAGS+="-I[deps_directory]" where deps_directory is the folder containing the dependency folders core, math, hdp, and grammar.
  3. Run ./executive_test.

Console

To run the code in console mode, where the user can input custom sentences and inspect the learned theory and proofs, run ./executive_test console.

Experiments

ProofWriter question-answering

Download the ProofWriter data and run

./executive_test proofwriter --data=[ProofWriter data filepath] --out=[output predicted answers filepath]

ProofWriter data filepath points to OWA/birds-electricity/meta-test.jsonl within the ProofWriter data.

FictionalGeoQA question-answering

Download the FictionalGeoQA data and run

./executive_test fictionalgeoqa --data=[fictionalgeoqa.jsonl path] --out=[output predicted answers filepath]