Linking Program Comprehension to Neural, Behavioral,and Psycho-Physiological Correlates

软件开发人员将大部分时间花在阅读和理解源代码上。早期的程序理解理论提出了假设驱动(自上而下)的机制和逐行(自下而上)的代码理解机制。然而，自上而下和自下而上程序理解的潜在认知过程在本质上仍然不清楚。在第一个项目阶段，我们已经着手通过从功能磁共振成像(FMRI)获得的客观测量来直接研究这些过程。我们已经确定了参与者在试图理解源代码时特别涉及的一组大脑区域。激活模式表明语义加工与左侧有关，即语言半球，以及注意力和工作记忆功能。促进自上而下的理解，例如，通过信息信标，已经导致了这些大脑区域中的一些区域的激活减少，这表明认知努力更少，我们已经从行为数据中证实了这一点。我们的结果为其他研究人员使用我们的神经成像方法进行程序理解的后续研究铺平了道路。第二个项目阶段将完善我们的实验框架，包括指定视觉注意的时间进程的眼睛跟踪，以提供更具体的fMRI数据分析，以及识别认知负荷变化的心理生理指标(瞳孔测量、皮肤电导、心率和呼吸)。更具体的控制条件将增强fMRI对比度，使我们能够研究代码方面(包括信标、计划、模式和启动)对识别的大脑区域中的大脑激活的细粒度影响。此外，我们将研究结构代码元素(IF-THEN-ELSE语句、循环、递归)以及编程经验对自上而下的程序理解以及伴随而来的大脑活动、心理生理和行为的客观测量的影响。最后，基于神经科学在对象处理的神经基础上的进展，我们将解决长期存在的问题，即面向对象编程和函数式编程之间的差异。

Software developers spend most of their time with reading and understanding source code. Early theories of program comprehension have proposed hypothesis-driven (top-down) mechanisms and line-by-line (bottom-up) mechanisms of understanding code. However, the underlying cognitive processes of top-down and bottom-up program comprehension are still essentially unclear.In the first project phase, we have set out to directly study these processes by means of objective measures obtained from functional magnetic resonance imaging (fMRI). We have identified a set of brain regions specifically involved when participants are trying to understand source code. The pattern of activation was indicative of semantic processing tied to the left, i.e., the speech hemisphere, as well as attention and working memory functions. Promoting top-down comprehension, e.g., by informative beacons, have led to reduced activation in some of these brain areas, suggesting less cognitive effort, which we have confirmed by behavioral data. Our results have paved the way for other researchers using our neuro-imaging approach for follow-up studies on program comprehension.The second project phase will refine our experimental framework by including eye tracking for specifying the time course of visual attention to inform a more specific analysis of the fMRI data as well as psycho-physiological measures (pupillometry, skin conductance, heart rate, and respiration) to identify changes in cognitive load. More specific control conditions will enhance the fMRI contrast, allowing us to study the fine-grained effects of code aspects (including beacons, plans, patterns, and priming) on brain activation in the identified brain areas. Moreover, we will study the impact of structural code elements (if-then-else statements, loops, recursion) as well as programming experience on top-down program comprehension and concomitant objective measures of brain activity, psycho-physiology, and behavior. Finally, based on the progress in the neurosciences on the neural basis of object processing, we will tackle the long-standing question about differences between object-oriented and functional programming.