A distributed neural network model celled SPEC for processing sentence
s with recursive relative clauses is described. The model is based on
separating the tasks of segmenting the input word sequence into clause
s, forming the case-role representations, and keeping track of the rec
ursive embeddings into different modules. The system needs to be train
ed only with the basic sentence constructs, and it generalizes not onl
y to new instances of familiar relative clause structures but to novel
structures as well. SPEC exhibits plausible memory degradation as the
depth of the center embeddings increases, its memory is primed by ear
lier constituents, and its performance is aided by semantic constraint
s between the constituents. The ability to process structure is largel
y due to a central executive network that monitors and controls the ex
ecution of the entire system. This way, In contrast to earlier subsymb
olic systems, parsing is modeled as a controlled high-level process Fa
ther than one based on automatic reflex responses.