To generate elementary trees,
we begin with a canonical subcategorization frame.
The system will first generate
related subcategorization frames
by applying LRRs, then select
subcategorization blocks corresponding to the information in
the subcategorization frames, next
the combinations of these blocks
are further combined with the blocks corresponding to
various transformations, finally, a set of trees are generated
from those combined blocks, and they are
the tree family for this subcategorization frame.
Figure shows some
of the trees produced in this way. For instance, the last tree is obtained by
incorporating information from the ditransitive verb subcategorization frame,
applying the dative-shift and passive LRRs, and then combining them with the
wh-non-subject extraction block.
Besides, in our system the hierarchy for subcategorization frames is implicit
as shown in Figure .