I should add that I'm looking forward to someone with more understanding of this field than I to share their thoughts on the paper - it caught my attention because Steven Pinker just called it "the most important paper in cognitive neuroscience in many years."

Implications are that left mid-superior temporal cortex (LmSTC) carries information about sentence-level meaning, which had kind of already been implied in prior experimentation.

The discussion states that they "Provide preliminary evidence for a long-standing theoretical conjecture of cognitive science: that the brain, on some level, functions like a classical computer, representing structured semantic combinations by explicitly encoding the values of abstract variables", which is little more than hyperbole.

It's interesting, insofar as they identify a very specific region of the brain dedicated to a specific aspect of understanding language, and which plays a small role in a larger network of regions, but it's a bit of a stretch to imply that it's the "most important paper in cognitive neuroscience in many years". Pinker is quoted because this provides some empirical evidence for some of his theories.

I should also add when they mean "architecture" they mean it in the literal sense -- the physical structure in the brain in which encoding sentence meaning takes place.