May 7, 2005
On the discussion of data
regularity à
On the Stephenson Cyber Attack
Taxonomy à
Note from Amnon, (www.visualtext.com)
As I've mentioned, I'd really like to get the raw data,
such as customs reports, plus some idea of desired outputs, so that I can
tinker with this and show what is possible.
I wrote a few weeks back that I was having trouble with the particular
zip format that was used by them [1]. (link)
I
believe that in AI, NLP and the like, engineering comes first, with theory
lagging far behind. Bottom up construction of working systems is my
approach. (As one example, I can point
to our work in MUC-3, MUC-4 with INLET/TexUS, followed by Hobbs writing a paper
on what Information Extraction systems should look like in MUC-5, implemented
in the FASTUS system.)
Anyway, in spite of my engineering orientation, my
orientation aligns well with the 2nd school, in which one heads toward
automation, but with a man-in-the-loop as long as needed.
If you will lay out a practical path or set of
requirements, I can implement the VisualText-based components.
[1] Nathan is the one who zipped up this data. I forwarded your note to him but I think he missed the opportunity to fix this so that you could spend some time working with the data. Perhaps he will become aware of the missed opportunity and sent a message to you on this.