The Expansion of Source Code Abbreviations Using a Language Model

摘要

Programmers often abbreviate identifiers names in source code to represent single words, i.e. unigrams, or phrases, i.e. multigrams. However, the difficulty to retrieve the original word(s) of an abbreviation during the maintenance phase makes the source code more problematic to comprehend. Incorrect abbreviations expansion may lead to introducing defects in the code. There are many approaches that that automatically expand abbreviations to their original words, unfortunately, they are based on predefined patterns and single-words dictionaries which cannot address abbreviations that are expandable to phrases. In this paper, we describe a bigram-based inference model which utilizes unigrams statistical properties as evidence to retrieve the original word automatically. We evaluated our approach on a set of 100 abbreviations randomly picked from eight open source projects and found that our approach correctly expands 78% of the set.