NL2FOL: Translating Natural Language to First Order Logic for Logical Fallacy Detection

Dependencies

• transformers
• torch
• accelerate

Run Instructions

For converting natural language to first order logic on the given dataset, run:

python3 src/nl_to_fol.py --model_name <your_model_name> --nli_model_name <your_nli_model_name>  --run_name <run_name> --dataset <logic or logicclimate> --length <number of datapoints to sample from dataset>

For converting first order logic to SMT files and generate results, run:

python3 fol_to_cvc.py <file containing fol translations>

For getting the final result metrics, run:

python3 get_metrics.py <path to results csv>

To interpret the SMT results, run:

python3 interpret_smt_result.py <output_of_smt_file_path> <json to relevant sentence data with Claim, Implication, Referring expressions, Properties and Formula>

Citation

Paper