The inner loop in the algorithm described in
section 4 searches for a word 
and a disjunct
of this word whose first left connector matches 
, or whose
first right connector matches 
. If there were a fast way to find
all such disjuncts, significant savings might be achieved. The
fast-match data structure, which is based on hashing, does precisely
this. The speed-up afforded by this technique is roughly the number
of different connector types, which is roughly 30 in our current
dictionary.