Cue vs. Target Word Memorability Analysis in Cued Recall

Project Overview

This project investigates the nature of word memorability within a cued recall paradigm. It challenges the view of memorability as a purely intrinsic, stimulus-driven property by demonstrating its dependence on the functional role a word plays – either as a retrieval cue or as a target to be recalled. Using data derived from Cox et al. (2018), this analysis explores how different psycholinguistic properties predict memorability for cues versus targets and examines memorability differences across semantic categories.

Dataset

The primary dataset used is based on the cued recall data from:

• Cox, G. E., Hemmer, P., Aue, W. R., & Criss, A. H. (2018). Information and processes underlying semantic and episodic memory across tasks, items, and individuals. Journal of Experimental Psychology: Learning, Memory, and Cognition, 44(9), 1367–1379.

Word properties were compiled from various sources and merged with the memorability scores derived from the Cox et al. data.

Key data files involved:

• Cox Data Final.xlsx: Combining entire test phase data across all cued recall sessions.
• Cox Data Mem Combined.xlsx: Combining cue and target memorability scores across the 924 words.
• Cox_Word_Properties.xlsx: The final output file containing words, their calculated properties, and memorability scores, used for regression. Generated from Word_Properties_and_VIF.ipynb.

Code Structure & Files

• Split_half_Cue_Mem_Cox.ipynb: Get cue memorability scores and perform Split Half Analysis to get CC correlation.

• Split_half_Target_Mem_Cox.ipynb: Get target memorability scores and perform Split Half Analysis to get TT correlation.

• CC_TT_CT.ipynb: Get cue and target memorability scores, get CT correlation and compare with CC and TT split half scores.

• Word_Properties_and_VIF.ipynb:

  • Loading data (Cox Data Mem Combined.xlsx).
  • Computing word properties for the Cox dataset words using external datasets (SUBTLEXusExcel2007.xlsx, BOI means.csv etc. as described in the paper) and nltk.
  • Merging and saving the final file as Cox_Word_Properties.xlsx.
  • VIF analysis to obtain the final word properties to create Regression models(for both cues and targets).

• Regression_stats.ipynb: Final regression table creation calculating all statistics.

• SVR_training.ipynb: To save the trained SVR cue and target models respectively.

• LIWC_GI_file_gen.ipynb:

  • Predicting word level and category level memorability scores on LIWC and GI words using the saved SVR models.
  • Saving the files.

• Category_level_analysis.ipynb:

  • Loading the category level and word level files.
  • Performing dichotomy analyses and category vs individual word level correlational analyses.

Requirements & Dependencies

• Python 3.x
• Pandas
• NumPy
• Matplotlib
• Seaborn
• SciPy (specifically scipy.stats for t-tests and percentile calculations)
• Statsmodels (for OLS regression, VIF, ANOVA)
• NLTK (Natural Language Toolkit)
  • Requires downloading specific NLTK data packages: punkt, wordnet, cmudict (using nltk.download())
• adjustText (for non-overlapping text labels on plots)
• openpyxl (or similar engine for reading Excel files with Pandas)