A semantic (typically XML based) format would represent a poem internally as
Code:
<poem>
<title>The blah poem</title>
<stanza>
<line>blah blah</line>
<line>...</line>
</stanza>
</poem>
PDF would represent it as
Code:
<bold, font size=20pt, centered>The blah poem</b, f, c>
<blank>
blah blah<line end>
...<line end>
<blank>
You see the difference, I hope? But forget about poems and stanzas. Say you're using your reader on a train and you enter a dimly lit tunnel, this makes the words on the screen harder to read so you want to increase the font size. PDF no can do, a semantic format, easy.
The conversion process you describe will work well only for very simple PDF documents. Whereas if the source file was in something like HTML you could reliably, with high fidelity convert a far larger set of typographical features.