Changes in functional brain activity patterns
     associated with computer programming learning in
     novices
     Kenji Hishikawa 
      National Center of Neurology and Psychiatry (NCNP)
     Kenji Yoshinaga  (  yoshinaga.kenji.3y@kyoto-u.ac.jp )
      Kyoto University Graduate School of Medicine
     Hiroki Togo 
      Kyoto University Graduate School of Medicine
     Takeshi Hongo 
      Otsuma Women’s University
     Takashi Hanakawa 
      Kyoto University Graduate School of Medicine
     Research Article
     Keywords: code comprehension, functional magnetic resonance imaging, program comprehension, the
     neuroscience of programming, programming learning
     Posted Date: November 8th, 2022
     DOI: https://doi.org/10.21203/rs.3.rs-2239916/v1
     License:   This work is licensed under a Creative Commons Attribution 4.0 International License.  
     Read Full License
                                               Page 1/18
    Abstract
    Background
    Computer programming, the process of designing, writing, and testing executable computer code, is an
    essential skill in numerous elds. A description of the neural structures engaged and modied during
    programming skill acquisition could help improve training programs and provide clues to the neural
    substrates underlying the acquisition of related skills.
    Methods
    Fourteen female university students without prior computer programing experience were examined by
    functional magnetic resonance imaging (fMRI) during the early and late stages of a 5-month ‘Computer
    Processing’ course. Brain regions involved in task performance and learning were identied by comparing
    responses to programming and control tasks during the early and late stages.
    Results
    The accuracy of programming task performance was signicantly improved during the late stage.
    Various regions of the frontal, temporal, parietal, and occipital cortex as well as several subcortical
    structures (caudate nuclei and cerebellum) were activated during programming tasks. Brain activity in the
    right inferior frontal gyrus was greater during the late stage and signicantly correlated with task
    performance. Learning was also associated with a rightward shift in laterality of the bilateral inferior
    frontal gyri. Although the left inferior frontal gyrus was also highly active during the programming task,
    there were no learning-induced changes in activity nor a signicant correlation between activity and task
    performance.
    Conclusion
    Computer programming learning among novices induces functional neuroplasticity within the right
    inferior frontal gyrus but not the left inferior gyrus (Broca’s area).
    Introduction
    Advanced computer programs have revolutionized many elds, including telecommunications, scientic
    research, commerce, entertainment, manufacturing, transportation, robotics, agriculture, military defense,
    and space exploration among others. Accordingly, computer programming is considered a necessary skill
    in many elds and a valuable academic discipline as it develops logical thinking. For these reasons,
    computer programming is being integrated into educational programs, often at the request of industry
    leaders. For instance, computer programming is a compulsory subject at the secondary education level in
                                    Page 2/18
    the United Kingdom, Hungary, Russia, and Hong Kong (Yamanishi 2015), and the Japanese government
    recently introduced computer programming into elementary education.
    Previous studies have attempted to identify the specic behavioral and psychological characteristics
    associated with programming skills. In the code recognition process, programmers may rely on the
    breadth-rst searching strategy (i.e., searching a data structure node with a given property) (Vessey 1985)
    or a goal-oriented, hypotheses-driven problem-solving strategy (Vessey 1985; Koenemann and Robertson
    1991). Programmers may also use a specic knowledge structure (Fix et al. 1993; Von Mayrhauser and
    Vans 1995); for instance, Fix and colleagues (Fix et al. 1993) suggested that expert programmers conduct
    symbolic operations that determine which inputs are fed into specic parts of the program for
    processing. Moreover, programmers may use specic patters of eye movements to review computer code
    (Uwano et al. 2006; Busjahn et al. 2015). In addition to such specialized cognitive processes and
    behaviors, programming skills may build upon rather conventional intellectual abilities such as executive
    functions, memory, language processing (syntax and vocabulary), mathematics, and reasoning. For
    successful applications, the computer programmer needs to understand computer language rst and
    foremost, which requires memory for codes and algorithm identication. These cognitive skills overlap
    with those needed to understand conventional spoken and written languages. Thus, learning computer
    programming is similar to learning a second language and so presumably depends on a partially
    overlapping set of neural structures and processes.
    Elucidating the neural substrates of computer programming skill acquisition could facilitate improved
    training methods and the further development of the requisite cognitive skills. Several functional
    magnetic resonance imaging (fMRI) studies investigating brain activities during a variety of
    programming tasks have demonstrated specic activation of frontal and parietal lobes, including
    language-related areas (Siegmund et al. 2014; Floyd et al. 2017; Siegmund et al. 2017; Castelhano et al.
    2019; Ikutani et al. 2021). In addition, a recent study examining both expert and novice programmers
    identied seven brain regions widely distributed in the frontal, parietal, and temporal cortices activated
    during programming tasks and associated with programming expertise (Ikutani et al. 2021). However, it
    remains unclear whether these brain regions are selectively recruited by expert programmers or are
    changed functionally by leaning (i.e., through neuroplastic processes underlying other forms of learning).
    To clarify this issue, it is necessary to measure brain activity repeatedly during programming learning.
    Herein, we investigate the brain activity patterns of novice computer programming students during
    programming tasks (requiring only answer selection by button press) performed under fMRI in the early
    phase and again in the late phase of a rst-ever computer programming course. We set three research
    questions (RQ): 1) Which regions of the brain show activity related to a programming task in
    programming learners? 2) Which regions of the brain show activity changes from the early to late phase
    of training due to learning? 3) How does the brain activity of a functionally altered brain region correlate
    with programming task performance? Programming learning can be regarded as the acquisition of a new
    written language, especially in beginners, so we speculated that the neural substrates would overlap with
                                    Page 3/18
    those observed in second language learners, specically within the extended language network including
    inferior frontal gyrus and superior temporal gyrus (Ferstl et al. 2008).
    Methods
    Participants
    Fourteen female university students (mean age 18.6 years, range 18–20 years) without prior computer
    programing experience participated in the present study. All participants were right-handed according to
    self-report and of native Japanese ancestry recruited from the Faculty of Social Information Studies at
    Otsuma Women’s University, Tokyo, Japan. All had normal or corrected-to-normal vision, no hearing
    impairments, and normal cognitive abilities according to Raven’s Colored Progressive Matrices (mean
    score 34.5; range 29–36) (Basso et al. 1987). Participants gave written informed consent according to
    the protocol approved by the Ethics Committee of Otsuma Women’s University (29-002-2) and the
    National Center of Neurology and Psychiatry (A2017-021).
    Experimental design
    All participants took a 5-month programming class using the “Processing” application
    (https://processing.org/) at Otsuma Women’s University. Processing is a Java-based exible software
    package designed for learning how to code programs for the visual arts. Tens of thousands of students,
    artists, designers, researchers, and hobbyists use Processing for learning and prototyping. The
    participants learned the concepts of programming and how to produce graphics and animation during
    the programming class.
    The participants were examined by fMRI twice, once during the mid-term period and again during the last
    term of the programming class. We employed a conventional task–fMRI design with alternating blocks of
    experimental (programming) tasks and control tasks. The experimental tasks were three types of
    programming-related problems (Fig. 1): predicting the output of code execution (code execution),
    completing an imperfect code (code completion), and nding programming “bugs” (bug nding). For the
    code execution task, the participants read a complete source code and predicted the output if executed
    (Fig. 1A). For the code completion task, the participants read a source code with a blank section and
    chose an appropriate option to ll in the blank (Fig. 1B). For the bug nding task, participants detected
    and counted bugs in a source code (Fig. 1C). In the control task, participants were asked to count the
    appearance of specic words in a nonsense code (Fig. 1D) produced by shuing the source code used
    for the experimental task; hence, the same sets of words were used in both source and nonsense codes.
    In all tasks, the participants were asked to read visually presented codes and select one of three or four
    options by a button press within 20 seconds. All participants completed 40 experimental (the code
    execution task: 19; the code completion task: 10; the bug nding task: 11) and 40 control blocks
    presented over 6 runs.
    MRI data acquisition
                                    Page 4/18