I do indeed admire johngg 's regex approach (and have ++ed it), but at the same time, hesitate to walk away without pointing out that it has NO capacity to flag mis-nesting (mis-nesting by .html or .xml standards, that is) and suspect that at some point bwgoudey's input data may have an anomaly or two.

Suppose the $html in johngg's Re: Split tags and words nicely were changed to:

q{<tag ref=1><tag ref=1a>Start<tag ref=2>and </tag><tag "ref=3">more</
+tag>and end};
[download]

Note unbalanced opens (4) and closes (2)

Leaving all else alone, output becomes:

<tag ref=1>
<tag ref=1a>
Start
<tag ref=2>
and
</tag>
<tag "ref=3">
more
</tag>
and end
[download]

... which offers no ready hint or markup or warning that the tags were mis-nested.

This is part of the reason that so many monks will advise against trying to parse the likes of .html or .xml with regexen and advocate the use of some of the modules mentioned above.