Implicit-structured representations causally effective
Trained networks can transform distributed representations of passive sentences into active voice and vice versa—exhibiting systematicity and showing that the implicit structure of their representations can be causally effective in processing.
David Chalmers (1990b).
