It's not validation in this case, it's finding them in a body of text. For validation you generally have it easier because you often only deal with a single datum in a field (and thus string).

The CLDR provides locale-specific formats for some datum types (IIRC numbers, dates, and durations). These formats can probably be reversed into the corresponding regular expression in order to perform locale-aware data detection.

Of course further complexity is added in people being able to provide partial and context-dependent information, which is much harder to detect e.g. "December 3rd"