Expert Programmers Have Fine-Tuned Cortical Representations of Source Code

Abstract

Expertise enables humans to achieve outstanding performance on domain-specific tasks, and programming is no exception. Many studies have shown that expert programmers exhibit remarkable differences from novices in behavioral performance, knowledge structure, and selective attention. However, the underlying differences in the brain of programmers are still unclear. We here address this issue by associating the cortical representation of source code with individual programming expertise using a data-driven decoding approach. This approach enabled us to identify seven brain regions, widely distributed in the frontal, parietal, and temporal cortices, that have a tight relationship with programming expertise. In these brain regions, functional categories of source code could be decoded from brain activity and the decoding accuracies were significantly correlated with individual behavioral performances on a source-code categorization task. Our results suggest that programming expertise is built upon fine-tuned cortical representations specialized for the domain of programming. Significance StatementThe expertise needed for programming has attracted increasing interest among researchers and educators in our computerized world. Many studies have demonstrated that expert programmers exhibit superior behavioral performance, knowledge structure, and selective attention; but how their brain accommodates such superiority is not well understood. In this paper we have recorded brain activities from subjects covering a wide range of programming expertise. The results show that functional categories of source code can be 1 decoded from their brain activity and the decoding accuracies on the seven brain regions in frontal, parietal, temporal cortices are significantly correlated with individual behavioral performances. This study provides evidence that outstanding performances of expert programmers are associated with domain-specific cortical representations in these widely distributed brain areas.

Reading and Calculation Neural Systems and Their Weighted Adaptive Use for Programming Skills

“…Review of fMRI Studies of Programming Skills. To further evaluate the hypothesis that program understanding shares, the same resources as reading and/or calculation, one takes into account the results from the available fMRI studies using a program understanding task [7,24,30,32,33,[51][52][53]. These are, to our knowledge, the only studies available in the literature about the neuronal correlates of program understanding.…”

“…These are, to our knowledge, the only studies available in the literature about the neuronal correlates of program understanding. The one from Castelhano et al uniquely reported functional and effective connectivity, but the amplitude findings of the available articles provide relevant insights on the relative weight of each network in programming tasks [2,30,32,33]. The work from Siegmund et al used detection of syntax errors as contrast condition to investigate the cognitive process of programming/source-code comprehension.…”

“…They found that code writing involves the right hemisphere brain regions involved in spatial ability and planning and present evidence suggesting that code and prose writing are quite dissimilar at the neural level. Ikutani et al [33] showed a fine-tuned representation of source code in the brain while Ivanova et al's work [32] report code comprehension activations in particular differences in BOLD responses to code problems with responses to contentmatched sentence problems. All the available fMRI studies of programming (a still new field with relatively few studies (N = 7)) are included in the meta-analysis (including contrast and conjunction comparisons with reading and calculation).…”

“…Additionally, a closer look into the programming studies identified a set of regions (Figure 1; Table 3; Figure A2) involved either in reading, calculus/math, or both: [51,52,54]. The medial frontal cortex, including the cingulate cortex and, most importantly, the anterior insula were also activated [7,24,30,32,33]. In the study by Castelhano et al, (blue regions represented in Figure 1) we investigated the neural underpinnings of programming by using fMRI while subjects performed a bug-detection task, which requires deep program understanding.…”

“…In spite of the evidence for a network involved in programming skills [7,30,32,33], this does not necessarily imply a novel form of brain specialization but might emerge from a new form of top-down controlled brain connectivity with distinct weights of reading and calculation systems.…”