The parallelism doesn't really work, I'm going to try to stretch it to make a point though.

Imagine that we were at a stage in which LLMs didn't really make sentences, only output like "Potato rainbow screen sunny throat", then we studied which words are used. There's really not much value to the words at all, we could maybe see which words are bundled together, we could try to ascertain what kind of words are used more, but in wanting to study the coherence of it all, it just holds very, very little value.

Chords by themselves hold very little meaning. The sensations evoked come from chords in a context and the progression provides very valuable context. Talking about a chord in a song is like talking about a word in a book, it's never really about that piece of the puzzle appearing, it's about how that piece is used in the puzzle.

It does work, particularly the emphasis on causal sequences being wholly inefficient to represent multidimensional and abstract concepts such as those that exist in both language and music.

The fact that you never refer to "syntax" even in this attempt at high level reasoning gives me pause and I cannot help but to conclude that you are making arguments in bad faith.