From: Kevin Wampler (wampler@cs.washington.edu)
Date: Sun Dec 05 2004 - 18:44:44 PST
In "PROVERB: The Probabilistic Cruciverbalist" Keim et. at. describe the
design and performance of their crossword solving program (PROVERB). In
concert with the diverse nature of the skills needed to solve a
crossword, this program combines many artificial intelligence
techniques. For example, candidates are chosen by expert modules (which
may employ natural language processing techniques), are merged with a
probabilistic model, and then filled in a grid by solving for a
optimization with constraint satisfaction problem. The reasonable
success of this program provides a good illustration of how some
difficult problems can be solved by combining multiple AI concepts.
This strength of showing the power of combining multiple AI methods is
also one of the weaknesses in the solver. The actual construction of
the solver, although it uses many elegant methods, is rather ad hoc.
This is particularly evident in the expert modules, but many of the
choices made in other parts of the solver also reflect this trait. This
means that although the system does illustrate a combination of
techniques well, it is not particularly useful in solving other AI
problems. This would likely be the most fruitful path for future work
(difficulty aside). It would be particularly useful to show a single
unified approach which could solve crossword puzzles, as it would likely
be applicable to a variety of other problems.
This archive was generated by hypermail 2.1.6 : Sun Dec 05 2004 - 18:44:45 PST