Before reading this please note that I am not advocating for or against XML.
A well reasoned XML instance can replace HTML for most things from presentation, accessibility, interaction, content negotiation, and so forth. You can apply JavaScript and CSS immediately to an XML instance. You don't get any of that capability with JSON, YAML, or anything similar.
I really think the people who hate XML the most are those are those who lack the imagination to move beyond raw data and query relationships. I agree XML isn't good at this, which is why it completely offloads that capability to a sibling technology: DOM.
I have also never seen anybody with a comfortable understanding of the DOM believe JSON, YAML, and the like fill that void, but then these other technologies do not primarily exist to structure human consumable content directly for human consumption. The most important part of the article you linked to describes markup relevance as a percentage of syntax overhead. The more the described content becomes an extraction of computer oriented data the more that percentage goes up thus making XML progressively a bad decision, but the opposite is also true.
The article really nails its bias in this regard by fixating on data as a facet opposed to information as a structure. When the goal is to provide primarily a structure, as opposed to fundamentally offering a syntax, the cost to scale grows inversely to the quantity of content provided. That is largely thanks to lexical scope, which allows a richer interpretation of context without any additional or specified syntax. That is the nature of information versus data. Conversely, data conveyance schemes that exist to primarily offer a syntax scale proportionally to the content provided because the ratio of syntax to content is static without any additional meaning.
For some clarity on the difference between data, information, and knowledge I suggest the DIKW model: https://en.wikipedia.org/wiki/DIKW_pyramid