Skip to content

Latest commit

 

History

History
55 lines (36 loc) · 2.48 KB

README.md

File metadata and controls

55 lines (36 loc) · 2.48 KB

Computational Model for Linguistic Humor in Puns

This is a clean python implementation of computational model for puns. The model employs a probabilistic framework to compute two measures, namely ambiguity and distinctiveness, upon which a regression model is fit to generate funniness ratings.

If you find this code useful in your research or want to learn more about the model, please refer to:

@article{kao2016computational,
  title={A computational model of linguistic humor in puns},
  author={Kao, Justine T and Levy, Roger and Goodman, Noah D},
  journal={Cognitive science},
  volume={40},
  number={5},
  pages={1270--1285},
  year={2016},
  publisher={Wiley Online Library}
}

Experiments

Author's original implementation can be found here: [justinek/pun-paper]. This implementation reproduces the results in the paper and I obtain exactly the same ambiguity and dinstictiveness values as there in original result csv sheet data-agg.csv (received from author).

However, I noticed several issues in the original relatedness and unigram data used by author to compute measures. I also found some bugs in the original implementation. Thus, I performed a series of experiments to fix them which I dubbed in this repository as v0, v1 and v2.

v0: Author's original implementation.
v1: Neat re-implementation of v0 (also compatible with python3).
v2: Built on top of v1. Fixes v0 data issues and implementation bugs.

Following is a concise table representing different models:

v0 v1 v2
Reproduces paper results ✔️ ✔️ ✔️
Python3 compatible ❌ ✔️ ✔️
Uses correct data ❌ ❌ ✔️
Uses correct model ❌ ❌ ✔️

Note

▶️ Please use the final version v2 if you wish to reproduce the correct puns model or adapt it for further research.

▶️ Additional notes, introductions to run code and results of each version can be found in its respective directory.

Acknowledgements

  • [justinek/pun-paper] served as an important reference. Most of the data and v0 scripts were borrowed from it.
  • Thanks to the authors, Justine Kao and Noah Goodman, for their help.

License

BSD