This project investigates the extent to which LLMs’ sentence processing aligns with human cognition, focusing on the psycholinguistic phenomenon of "Memory Encoding." We compare the performance of BERT and Pythia on sentences containing at least two nouns which describe humans/objects individually, asking questions to assess the models’ comprehension.
- To get scores of table of sentences- run apply_scores_as_new_col.py with the desired models configuration.
- Scripts for analyzing the results are at analysis/analysis_scripts