feature structures vs. feature structure descriptions

[Carl Pollard]

I suppose many points of view are possible. The way I think of it is
that feature structures are models of linguistic expressions and their
parts, and feature logic is an (object) language for talking about
them. If by "feature descriptions" you mean attribute-value matices,
then I would say they are informal substitutes for formulas of
feature logic. A more precise expression of this point of view
is given in my paper `Strong generative capacity in HPSG' in the
Kathol, Koenig, and Webelhuth volume.

Carl