The main reason I use semantic line breaks, not explicitly mentioned in this article, is that it minimizes reformatting when editing. Only the subclause being edited is reformatted, while the rest of the paragraph remains as-is. This also minimizes the changes in line-oriented diffs.
While one could rely on automated line-wrapping instead of using hard line breaks that require reformatting, it isn’t usefully available in all environments, in particular for indented paragraphs and when having elements like ASCII art or code that shouldn’t be word-wrapped, and it makes plain-text diffs larger than necessary when whole paragraphs are on a single source line.
Wouldn't putting a single word per line minimize reformatting and changes in line-diffs further?
From an opposite angle, I imagine you do not use semantic line breaks in ASCII art or code that shouldn't be word-wrapped, so I imagine why not push for that too?
Mostly to point out that this is a micro-optimization for little demonstrable benefit.