We will now outline how the processing of elliptical utterances can be performed using the item sharing approach.
For example, consider the following dialog between person A and the NLS B:
Generation of the elliptical utterance `Mary,too' can be performed by
means of the item sharing method as follows. We assume that the
structure of an item has an additional slot :producer which indicates
whether an
item has been constructed during parsing or generation.
If A's utterance
has been analyzed we assume that a passive item for the VP
`is coming to the party tonight' is constructed. This item is now also
available for generation. However, the value of the slot :producer
is something like :parsing, since the parser has constructed this item.
When the generator is going to use this item to complete its process,
it can use this information through the elliptic generation process,
for example, as a statement for suppressing uttering of this
``re-used'' string.
Clearly, this is a straightforward but naive approach. In general the process is more complicated e.g., if B's utterance were be `Mary and John, too'. However, in combination with the EBL approach mentioned above it might be possible to handle these cases also.
In a similar way it would also be possible to use integrated generation in order to complete elliptic constructions during parsing, by making use of previously analayzed or produced constructions. Of course, this implies that the NLS has at its disposal a discourse and dialog model.
Note that this kind of processing can only take into account grammatical information. Thus it cannot be verified by the system that generated ellipses are not so brief as to be ambiguous or misleading. In order to solve these cases, it is an interesting issue to investigate whether and how this grammar-oriented method can be combined with knowledge-based methods, for instance, the one described in [Jameson and Wahlster1982] (see also section 5.3).