This is the code repository for our paper https://aclanthology.org/2021.emnlp-main.119.pdf. For plots in the paper:

type can take following values:

sra: This corresponds to attractors being semantically related to the target/background.

sua: This corresponds to attractors being semantically unrelated to the target/background.

inter: This corresponds to attractors being semantically related to the target/background and being present between key entity and critical fact.

T: This corresponds to attractor being semantically related to target word.

B: This corresponds to attractor being semantically related to background word.

Accuracy plots:

python plotDistanceVsAccuracy.py type e.g python plotDistanceVsAccuracy.py T

Relative Prob plots:

python plotRelativeProbVsAccuracy.py type

Accuracy plot for Varying Key entity position:

python plotDistanceVsAccuracyKeyEntityLater.py