Remember, words are encoded in the system as overlapping CAs.
Overlap is derived from the word net semantic hierarchy.
A couple of the nets in the parser are made for resolving the ambiguity.
Neurons in the nets send activation to the appropriate rule.
The attachment nets are activated by the semantics of the phrases.
As the system doesn't really learn, some of the attachment decisions are stored; the semantics send activation to the correct rules.
Hierarchical storage allows the system to generalise to unstored triples.
It defaults to a Noun attachment.
The test stores 3, inherits 3, and one default. All work.

Neural PP Attachment