1 Detect the software sentences

1.1 code snippets.

It first identifies code lines by matching a strong regular expression, i.e., ".*?=.*?;$", and then expands the regions with surrounding sentences by many
weak rules.

-> group 1 of weak rules.
a sentence that contains:
"{",
"}",
"class",
"public",
"interface",
"static",
");",
"// ",
"/*",
"=="

-> group 2 of weak rules.
a sentence that matches:
".*?=.*?;$",
"if\\s*\\(.*",
"\\(*.?\\).*?;",
"while\\s\\(",
"=.*?\\(",
"=.*?,",
"\\(.*?,$",
"\\(.*?\\)$",
"return.*?;",
"list.*?;",
"response.*?;",
"\\+\\+.*?;".


1.2 system messages.

It is identified as the continuous sentences containing ":" or "_".
 
A system message is a sentence that matches:
"[a-zA-Z0-9]+[\\s\\pP]*:[\\s\\pP]*[a-zA-Z0-9]+",
"\\w+[_]\\w+".


2 Before the above process

If a line ends with "." or "?", it is usually not a software sentence. We do not need to match these sentences with the above rules.

private boolean isSent(String line) {
	String[] end = new String[]{".", "?"};
	for (int i = 0; i < end.length; ++i) {
		if (line.endsWith(end[i])) {
			return true;
		}
	}
	return false;
}


3 After the above process

If the distance of two regions is within three sentences, we also take the sentences between them as software language.