An empirical exploration of regularities in open-source software lexicons

The software lexicon is an important source of information during program comprehension activities and it has been in the focus of several recent case studies. Identifiers and comments, which constitute a lexicon in software, encode domain concepts and design decisions made by programmers. The paper presents an exploratory study that investigates regularities in the software lexicons of open-source projects by analyzing distributions of tokens in diverse software artifacts. The study examined source code of 142 systems from different domains, written in 12 different programming languages, as well as bug reports and external documentation. We discover that distributions of lexical tokens in studied artifacts follow the Zipf-Mandelbrot law, which is an empirical law in statistical natural language processing. Furthermore, the study reveals that the Zipf-Mandelbrot law is not confined to program lexicons in object-oriented languages, as shown in the previous studies, but also emerges in source code written using procedural, functional and markup languages, as well as other software artifacts. Our study also indicates that a previously devised software science equation does not hold for describing the program vocabulary-length relationship and more studies are necessary.