From: Katarzyna Wilamowska (kasiaw@washington.edu)
Date: Mon Dec 06 2004 - 12:02:38 PST
PROVERB: The Probabilistic CruciverbalistKeim, Shazeer, Littman, Agarwal, Cheves, Fitzgerald, Grosland, Jiang, Pollard, WeinmeisterOne-line summary:The paper discusses the design and testing of a crossword puzzle solver, as well as some problems the author's faced along the way.Ideas in the paperThe first part of the paper discussed in depth the difficulties that one comes across when having to solve the clues given. The authors were able to classify most of the clues into specific categories and access their difficulty relative to other types of clues.In the second part of the paper concentrated on the model, but mostly on explaining some of the modules that were used within the model. I thought that the idea of looking for transformations was pretty cool.Flaws in the paper I guess it was counter intuitive for me to think of excluding CWDB data from the modules, since it would make sense that crossword clues are a special type of word format. Research questions on the topicWhy did the authors not use any data to train or test PROVERB from Dell or Riddler?What was the exact cause of overlap between modules? And how can this information be used to decrease the number of modules needed and remain with the same accuracy?
This archive was generated by hypermail 2.1.6 : Mon Dec 06 2004 - 12:02:39 PST