6.3.2. Solution

Think hard about what you want a word to be and what separates one word from the next, and then write a regular expression that encodes your decisions. For example:

/\S+/               # as many non-whitespace characters as possible
/[A-Za-z'-]+/       # as many letters, apostrophes, and hyphens

6.3.3. Discussion

Because words vary between applications, languages, and input streams, Perl does not have built-in definitions of words. You must make them from character classes and quantifiers yourself, as we did previously. The second pattern is an attempt to recognize "shepherd's" and "sheep-shearing" each as single words.

Most approaches have limitations because of the vagaries of written language. For instance, although the second pattern successfully identifies "spank'd" and "counter-clockwise" as words, it also pulls the "rd" out of "23rd Psalm". To be more precise when pulling words out from a string, specify the characters surrounding the word. Normally, this should be a word boundary, not whitespace:

/\b([A-Za-z]+)\b/            # usually best
/\s([A-Za-z]+)\s/            # fails at ends or w/ punctuation

Although Perl provides \w, which matches a character that is part of a valid Perl identifier, Perl identifiers are rarely what you think of as words, since we mean a string of alphanumerics and underscores, but not colons or quotes. Because it's defined in terms of \w, \b may surprise you if you expect to match an English word boundary (or, even worse, a Mongolian word boundary).

\b and \B can still be useful. For example, /\Bis\B/ matches the string "is" within a word only, not at the edges. And while "thistle" would be found, "vis-à-vis" wouldn't.