Figure 1 shows the two stages of the map
maker system in action.
First, in the DESIGN INFORMATION CAPTURE stage the designer simply uses the
electronic design notebook as a day
to day tool for conceptual design,
tagging parts of the notebook with visual markers when key ideas (in the
designer's own conceptual framework) are expressed. This additional
participation, not required for low level functioning of the map maker, is
20 or 30 seconds per hour to tag items of interest in notebook pages as
they are generated, and 20 minutes per week to update the idea tag table (a
machine readable outline of what the designer thinks are the current key
ideas in developing the design).
A knowledge engineer (or well motivated designer) then creates a
requirement to text
graphic feature table based on the particular
designer's notebook habits and design ideas. The table is generally quite
short for instance, one screenfull was used to cover the 76 notebook
pages in the example. Updating the table would only take a couple of hours
Then in the DESIGN INFORMATION USE phase, the map making system translates
a user query in terms of design requirements into a text
graphic data base
query, which returns a set of notebook pages dealing with those
requirements. The text graphic data base query is actually an expression in
Professor Michael Genesereth's MRS a lisp syntax logic programming
system with additional predicates that pertain to text graphic features.
A particular logic query will contain predicates to select pages with the
features determined to correlate with material about the requirements and
parameters mentioned in the user query.
In spirit, the map
maker system employs the "gold digger model"
convenient (easy), opportunistic, and initially shallow.
This approach has the following features:
for the designer
in vivo and in sito
"in life" and "in original place" where design
is done, ie the living design notebook
the raw text graphic data on the pages is complete
and informal, but easily lends itself to formal techniques
and models (see the RESULTS section for a list)
does the easy things first (text graphic syntax
driven feature recognition)
breadth first, takes whole notebooks as input
starts with text graphic syntax, but upwardly
compatible with deep reasoning methods.
the more tagging the designer does, the better the performance
of the map maker; and as progress is made in the area of
text graphic understanding, the more knowledge will be able to
be recovered in the future from design documents of the past.