\W (for ASCII)

Eh, no, well, maybe, perhaps, who can tell?

\W will be Unicode semantics if the string is internally in UTF-8 format, if the pattern contains non-Latin-1 characters (well, sometimes, not always), or, in 5.14, if you use the /u modifier. Otherwise, if locale is in effect, or, in 5.14, if you use the /l modifier, it will be using the semantics of whatever locale is in effect. The /l modifier will overrule the heuristics that would otherwise trigger Unicode semantics. And otherwise, or if either the /a or the /aa modifier is in effect, ASCII semantics will be used.

Oh, and then there's use feature 'unicode_strings'; that may trigger the first case when you don't expect it. use 5.012; will enable it.

[^:alpha:] (for Unicode?).

[:alpha:] is a POSIX class. It should never match any code point larger than 255 - but Perl has f*cked this one up. I think this is fixed by using one of the new 5.14 regexp modifiers, but I'm not quite sure.