From: Sandra B Fan (sbfan_at_cs.washington.edu)
Date: Mon Dec 08 2003 - 11:03:29 PST
"Proverb: The Probabilistic Cruciverbalist"
Greg A. Keim, Noam m. Shazeer, Michael Littman, et al.
One line summary: The authors combine a variety of puzzle-solving
techniques and with probabilistic methods to create a two-step
architecture cruciverbalist that performs at least as well as the casual
human player, but not competitively against the best.
Important points: This project was a good example of how to combine
different AI techniques using a probabilistic model. Hey, if you can't
figure out the *best* way of doing something, do them all! Also, they've
demonstrated that this is a very useful AI problem to solve, because it
encompasses a wide enough range of knowledge and techniques, but isn't as
wild and hairy as the real world--you still have your constraints of the
grid, and the alphabet, etc.
Flaws: They said they ignored special formatting and unusual marks that
may have appeared in the crosswords. I think that may have contributed to
some of Proverb's unsolved targets. Secondly, there were a couple of
distracting typos. Also, I'm not sure why they picked "10" as the max
distance for Encyclopedia, it seems sort of arbitary.
Further directions: Inclusion of the aforementioned ignored symbols might
be one step in improving Proverb's performance. Application of Proverb's
two-step probabilistic architecture to other problems may also be
interesting (though I think it's been done).
This archive was generated by hypermail 2.1.6 : Mon Dec 08 2003 - 11:03:29 PST