Recognizing Words from Source Code Identifiers Using Speech Recognition Techniques

The existing software engineering literature has empirically shown that a proper choice of identifiers influences software understandability and maintainability. Researchers have noticed that identifiers are one of the most important source of information about program entities and that the semantic of identifiers guide the cognitive process. Recognizing the words forming identifiers is not an easy task when naming conventions (e.g., Camel Case) are not used or strictly followed and-or when these words have been abbreviated or otherwise transformed. This paper proposes a technique inspired from speech recognition, i.e., dynamic time warping, to split identifiers into component words. The proposed technique has been applied to identifiers extracted from two different applications: JHotDraw and Lynx. Results compared to manually-built oracles and with Camel Case algorithm are encouraging. In fact, they show that the technique successfully recognizes words composing identifiers (even when abbreviated) in about 90% of cases and that it performs better than Camel Case. Furthermore, it was able to spot mistakes in the manually-built oracle.