The Neuroscience of Program Comprehension

Resources for the paper Comprehension of computer code relies primarily on domain-general executive brain regions by Anna A. Ivanova, Shashank Srikant, Yotaro Sueoka, Hope H. Kean, Riva Dhamala, Una-May O'Reilly, Marina U. Bers, Evelina Fedorenko

Paper preprint: https://www.biorxiv.org/content/10.1101/2020.04.16.045732v1

Published in eLife, December 2020: https://elifesciences.org/articles/58906

The labs involved:

https://evlab.mit.edu/

https://sites.tufts.edu/devtech/

https://alfagroup.csail.mit.edu/

For additional information, contact [email protected], [email protected], or [email protected].

Stimuli

Python (Experiment 1)

All stimuli related to our Python experiments (Experiment 1) are available in ./stimuli/python/.

We have made available stimuli pertaining to the three conditions we test in our work -

• EN: Codes with variable names in English.
• SENT: English descriptions of these codes.
• JAP: Codes with obfuscated variable names.

We have shared three files corresponding to these conditions, each containing all 108 stimuli, as well as three folders each containing individual files for the 108 stimuli across the three conditions. The stimuli vary by problem content (math vs. string manipulation) and problem structure (sequential/for loop/if statement), yielding 6 problem types in total. Note that only 72 stimuli (the first 12 for each of the 6 problem types) were used in the study.

ScratchJr (Experiment 2)

The stimuli used in the ScratchJr experiment (Experiment 2) are available in ./stimuli/scratchjr/. Each folder item contains the .png image of the code snippet, two videos (match and mismatch; only one of the videos was shown to each participant), and the audio of the sentence describing the video (the audio was not used for the current experiment).

Data

We make available the data we used for statistical analyses and plotting:

• behavioral responses
• responses to conditions of interest within MD and language fROIs (data_fROI*)
• responses to conditions of interest within parcels defined using the code problems > sentence problems contrast (data_GSS)
• voxelwise spatial correlation between responses in language fROIs (data_spcorr)

Original .nii files are available upon request.

Analyses

We provide the R code that was used for statistical analyses and plot generation.

Parcels

We make available the brain parcels used to define functional systems of interest (MD and language). The parcels were defined using the GcSS method using previously collected data from 197 (MD) and 220 (language) participants.

Python proficiency test

For Experiment 1, we developed a custom Python proficiency test (22 short response questions and 1 long response question, intended to be completed within 1 hour). There was no overlap between the proficiency test questions and the questions presented to participants in the scanner.

Programming background survey

Subjects also responded to a survey at the end of their scans. They reported their proficiency and familiarity with a number of programming related topics. ./survey/ has a PDF of these questions.