N-back Training Stimulus Generation and Paradigm

Programmer: Farhan Bin Faisal
Lab: Meltzer Lab
Date Created: 10 November 2022

This repository contains the full stimulus-generation pipeline and experiment link for a verbal N-back training paradigm. It documents how word and pseudoword lists were created and how they are integrated into the PsychoJS/Pavlovia task.

---

Table of Contents

• Overview
• Repository Structure
• Research Poster
• Experiment Paradigm
• Stimulus Generation Pipeline
  • 1. preprocess.ipynb
  • 2. wordListMaker.m
  • 3a. wuggy/postSOS.ipynb
  • 3b. Manual pseudoword generation with Wuggy
• Dependencies
• Attribution

---

Overview

The aim of this project is to generate frequency- and syllable-matched word and pseudoword lists for a verbal N-back task, and to use those stimuli in an online training paradigm implemented with PsychoJS and hosted on Pavlovia.

The pipeline:

1. Preprocesses a large English word list (filtering by frequency, syllable count, profanity, etc.).
2. Uses SOS to generate matched word lists.
3. Uses Wuggy to generate pseudoword lists corresponding to each word list.
4. Integrates the resulting stimuli into the N-back experiment.

---

Repository Structure

• preprocess.ipynb
  Jupyter notebook for cleaning and filtering the initial word list, lemmatizing items, and preparing SOS-compatible input.

• wordListMaker.m
  MATLAB script that interfaces with SOS to generate multiple matched word lists from the preprocessed vocabulary.

• wuggy/postSOS.ipynb
  Jupyter notebook that calls Wuggy output and assembles nonword lists corresponding to each word list; flags items that require manual pseudoword generation.

• generateConditions.ipynb
  (If used in your project) Notebook for assembling final condition files for the N-back task from the word and pseudoword lists.

• wordLists/
  Directory containing the word lists generated by SOS.

• nonWordLists/
  Directory containing pseudoword lists (generated by Wuggy and/or manually).

---

Research Poster

Below is the poster summarizing the N-back training project:

---

Experiment Paradigm

The N-back training task was implemented in JavaScript using PsychoJS and hosted on Pavlovia.

A demo version can be accessed at:
👉 Run the N-back task on Pavlovia

For the demo, please use the following credentials:

• session: 1
• participantID: 19995

---

Stimulus Generation Pipeline

1. preprocess.ipynb

This notebook prepares the word list for SOS:

• Loads a CSV file of candidate words into a pandas DataFrame.
• Filters rows by frequency:
  • Retains words with 4.0 < Zipf value < 5.0.
• Computes syllable counts:
  • Uses cmu_dict from NLTK.
• Filters by syllable count:
  • Retains words with 1 or 2 syllables.
• Lemmatizes words:
  • Uses WordNetLemmatizer to replace each word with its lemma.
• Discards:
  • Words not found in cmu_dict.
  • Profane words and proper names.
• Formats the final DataFrame into SOS-compatible input.
• Outputs a tab-delimited text file:
  • Output: sos_input.txt

---

2. wordListMaker.m

This MATLAB script uses SOS to generate matched word lists:

• Takes sos_input.txt as input.
• Uses SOS to generate 18 lists of 10 words each.
• Matching criteria:
  • Lists are matched on Zipf frequency and syllable count.
• Constraints:
  • Soft constraints: used to optimize matching on Zipf value and syllables.
  • Hard constraints: enforce minimum syllable count and frequency thresholds.
• Output:
  • Word lists saved in the wordLists/ directory.

---

3a. wuggy/postSOS.ipynb

This notebook generates pseudoword lists for each word list:

• Iterates through each file in wordLists/.
• Uses Wuggy to derive nonword (pseudoword) lists corresponding to each word list.
• Outputs:
  • Nonword lists saved to nonWordLists/.
• Logging:
  • Prints the filenames of any word lists containing items that cannot be converted automatically into pseudowords.
  • These flagged words must be handled manually.

---

3b. Manual pseudoword generation with Wuggy

For words that Wuggy cannot automatically convert:

1. Download Wuggy from:
   http://crr.ugent.be/programs-data/wuggy

2. Recommended settings:

   • Language: Orthographic English
   • Match syllable length
   • Match word length
   • Match transition frequency
   • Match 2 out of 3 segments

3. Manually input the flagged words and export pseudowords.

4. Replace the missing entries in the corresponding nonWordLists files.

---

Dependencies

To run the full pipeline, you will need:

• Python

  • pandas
  • nltk (with cmudict and WordNet corpora downloaded)
  • Jupyter Notebook / JupyterLab

• MATLAB

  • SOS (for stimulus optimization) configured to run with wordListMaker.m

• Wuggy

  • Standalone application from UGent for pseudoword generation

• PsychoJS / Pavlovia

  • For hosting and running the N-back experiment online

---

Attribution

This project was developed in the Meltzer Lab by Farhan Bin Faisal as part of an N-back training study.
Please cite or acknowledge the Meltzer Lab and this repository if you reuse the paradigm or stimulus pipeline in your own work.