Can Language Models Reason about Individualistic Human Values and Preferences?

Authors: Liwei Jiang, Taylor Sorensen, Sydney Levine Yejin Choi,

This is the codebase for the paper.

IndieValueCatalog Data

We transformed the World Values Survey into natural languagestatements that describes people's values and preferences.

Raw World Values Survey Data

For downloading the raw World Values Survey data (Wave 7), please visit the official WVS website and follow their downloading instructions.

Data Conversion

• data/WVS_conversion.py contains the code for data conversion.

Meta Data

• data/demographics_meta_data.csv contains the meta data of the demographics questions.

• data/refined_statements_meta_data.csv contains the meta data of the refined statements.

• data/statements_meta_data.csv contains the meta data of the polarity-grouped statements.

Human Label Data

Please find the converted human label data here.

• demographics_in_nl_refined_statements_combined_full_set.csv contains converted human label data with refined statements.

• demographics_in_nl_statements_combined_full_set.csv contains converted human label data with polarity-grouped statements.

Data Conversion Scripts

• src/prepare_data/compile_wvs_statements.py contains the code for data conversion.

Citation

If you find our work helpful, please feel free to cite our work!

@misc{jiang2024indievalue,
      title={Can Language Models Reason about Individualistic Human Values and Preferences?}, 
      author={Liwei Jiang and Taylor Sorensen and Sydney Levine and Yejin Choi},
      year={2024},
      eprint={2410.03868},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2410.03868}, 
}