SIGGEN 
David McDonald 
mcdonald AT cs.brandeis.edu 


As in the past, SIGGEN's activities this past year have revolved around its
moderated bboard, the workshops of the membership, and initiatives being worked
on by its five person organizing committee.

Our bulletin board resides at "siggen@benus.bgl.ac.il". Queries, announcements,
and other messages to the group can be sent there and will be collected and
forwarded to the whole group. 

Two workshops on language generation are being held this year. SIGGEN acts just
as a mediator in coordinating workshops, though this year many on the organizing
committee were also principals in these workshops' organizing committees. The
Seventh International Workshop on Natural Language Generation will be / was held
this June in Kennebunkport Maine. [Details to go into Finite String report] An
invited workshop on generation will be held in Dagstuhl, Germany in July. 

We have started three new projects via SIGGEN and anticipate beginning a fourth.
We are trying to develop a comprehensive description of the field of natural
language generation in terms of a set of well defined and balanced subject
keywords with pointers back to the existing literature. An initial draft of this
list should be available in the early fall for community discussion. The aim is
to be inclusive and non-presumptive, so we are seeking creative ways to present
the keyword and cross-index set that will make this possible.

We are trying to make the unpublished papers of past generation workshops
available in some form over the net. The technical means for accommodating this
is still being worked out since our authors use a wide variety of editors and
formats and the search for a workable common denominator that is not too low is
still ongoing. 

We are also compiling a description of the steps that go into organizing our
workshops and compiling a set of materials that should be reusable so that the
next organizers will not have to start essentially from scratch as has been the
case in the past. 

Finally, with the evaluation of natural language systems having become so
important and with the long-standing difficulty of even defining what it would
mean to compare two generators (different assumptions about the boundaries of
the process, different representations of the source and different kinds and
amounts of information in it, differences in the task and setting, etc.), we
anticipate that we will also initiate a community discussion of evaluation via
our query board, and solicit careful descriptions of common technical problems,
linguistic or semantic phenomena that are problematic for generators (or
presumed to be simple), etc. The result would be a shared and evolving workbook
can be accumulated for researchers to consult.