1-4 August, 2000
Frank Cornelissen
Visit to DFKI Saarbruecken

The purpose of this visit was to get some idea of possible
collaborations between our groups, and to get our hands on their
planner (and get some hands on experience with this).


The first day i gave a talk about our use of communicative devices for
generating presentations from rhetoric. This talk was reasonably well
received, but although i emphasized that we are *not* in the business
of information retreival, a lot of questions afterwards where about
this topic. The use of these devices was not questioned, but a remark
was that their will not be a "best" mapping from the rhetoric to
communicative devices. This is of course not the real purpose of these
devices; their use should however be consistent within one
presentation (or preferably presentation system) so that the user
might recognize this. The slides used for this talk can be found at
/hosts/multimedia/ins2/talks/dfki-2000-08-01.(pdf)|(ps) (pdf for
viewing/presenting, ps file for printing).


The next day i talked with Alexendar Kroener about his research in
automatic layout systems. He has implemented such a system for html
pages which will determine an absolute layout for a page, also
including links if necessary. I did not play with his system, but it
is available from the web at http://www.dfki.de/~kroener/DC.htm
This system looks like it is doing something like we are trying to do
for smil presentations, but is aimed at static pages. It (or it's
predecessor?) has also been used to generate yellow--pages paper
output. He is also currently working on applying his system for
realtext layout, which will enable one to absolutely determine the
bounding boxes of a given text at a given font size (which is *very*
hard at this moment, since realtext does a lot of weird things which
his system is trying to emulate in order to do these bounding box
calculations). 

Most of the time however i worked with their planning system which
will be installed here as well. This planning system, named ckuckucq
(for no apparent reason, i believe) is based on the basic planning
system developed by Elisabeth Andre, tailored to smil generation. It
combines planning operators (presentation strategies) with temporal
and spatial constraints. It allows you to specify constraints on the
elements of each plan operator (for example, in a ``contrast'' plan
operator you could specify that the elements should be left of each
other, or whatever you want). However, their are several limitations
to the planning system which will limit it's usefullness to us as is.

The planning system is very deterministic. Once the plan operators
have been determined the constraint satisfaction solving process is
started. However, when the csp is unsolvable, the system can't
backtrack over the chosen plan operators from here. This means that
something like the interdimensional constraint resolution for
sequences is impossible.

The handling of sequences is weak. The system can only display
sequences of items temporally, or one has to enhance the data to imply
that it is part of a sequence and adjust the strategies accordingly,
which makes the strategies less generic. 

Another problem (which our system also has currently) is that all the
media--items are in a hierarchy (that of the plan-operators), and that
constraints between these media--items (specified within the plan
operators) must follow this hierarchy. However, one would like to
specify constraints that cross the hierarchy (for example the spatial
constraint communicative device stuff).

All in all, it was an interesting visit, and i learned a lot about
their planning system and the way they use it. Above are the quirks i
found when using the system and comparing it to an ideal system we
would want to make, and it may sound a bit negative. On the positive
side however is that i got something out of it rather quickly, and it
was reasonably intuitive (save some archaic syntax for the
plan-operators that is a left over from a previous version of the
planner in lisp; it is now implemented in java. The plan is to change
this syntax to some kind of xml input). I liked their planner,
although it is rough around the edges.