The choice wasn't unconstrained tag-soup markup vs XML. HTML was originally spec'd using SGML, and SGML is a proper superset of XML. However, reporting the exac

The choice wasn't unconstrained tag-soup markup vs XML. HTML was originally spec'd using SGML, and SGML is a proper superset of XML. However, reporting the exact place of an error in a large SGML document in the presence of tag inference/omission is challenging or outright impossible. So W3C spent a significant amount of work to bring forward XML as a simplified SGML subset starting in 1997, with the intention that XML eventually replaces SGML as the basis for HTML. Obviously that didn't happen; but that's no reason to give up on structured markup altogether. SGML is perfectly capable to validate and canonicalize almost all HTML 5 documents (save for a few not-so-well-thought-out design decisions from the Hickson era). The same can't be said for the "official" HTML5 validator; even the specification text itself contains content model errors.