Getting Sick After Seeing a Doctor? Diagnosing and Mitigating Knowledge Conflicts in Event Temporal Reasoning

This is the code repo for the NAACL 2024 Findings paper Getting Sick After Seeing a Doctor? Diagnosing and Mitigating Knowledge Conflicts in Event Temporal Reasoning.

There are two sets of experiments on two dataset, TORQUE and MATRES.

TORQUE

Code for Counterfactual Data Augmentation: cf_gen

GPT-3 in-context learning experiments: gpt3

MATRES:

The code is adapted from https://github.com/why2011btv/Faithful_TempRel.

Example: scripts/train_baseline.sh

Counterfactual generation: cf_gen

GPT3 in-context learning experiments: gpt3.