As far as I understand sentiment analysis is more of a pragmatic pursuit, and the prototypical use case is reviews of things or creations. It's a business tool and not science. As general language/literature has no set object, I don't see how we could even interpret long term "analyses" of it.

It's good that the authors are at least trying to create some generally interpretable categories mapped into language, even if they may not be (yet) fully convincing, as other commenters point out.

I'd be curious to see how sentiment analysis of contemporary writing has changed over time. That would be easier to interpret than this metric too.