Psycholinguistic Research in Parsing

As an NL Engineer, I'm not really impressed with the connectionist parsers (including my own).
They're more a proof of principle.
Psycholinguistic parsers on the other hand seem to be useful as (albeit slow) systems.
A lot of work has been done with stacks and resolving ambiguity.
However, another method of parsing is with a memory based model.
Memory items are activated, decay and may be reactivated.
The model I'm really keen on here is Rick Lewis's model (Lewis and Vasishth 2005).
It's based on the ACT-R cognitive architecture.
The idea is that there are memory items (phrases) that are stored.
These have an activation associated with them that varies over time.
This all comes from the architecture so is not ad hoc.
Active items can be combined by grammar rules.
Rick did NL-Soar and was a PhD student at Carnegie Mellon when was finishing at Michigan. This really fits the psycholinguistic bill.
My parser is, in essence, a neural implementation of this.