proverb review

From: Lillie Kittredge (kittredl_at_u.washington.edu)
Date: Mon Dec 08 2003 - 10:40:57 PST

  • Next message: Jessica Kristan Miller: "Cruciverbalist paper review"

    PROVERB: The Probabilistic Cruciverbalist
    by, like, 20 people. Most of whom have amusing names.

    -summary-
    PROVERB is a crossword-solving system using a bunch of different AI methods all merged into
    one surprisingly effective and relatively non-klugey whole.

    -main stuff-
    I appreciated the readability and clever title. It was good thinking of them to build in the use of
    a variety of expert modules - both because this increases their accuracy, and because it gives
    them more information about what kinds of methods do better. One of the main revelations, to
    me, was that previous crosswords can be such a good predictor of current ones. The CWDB was
    a useful tool for its generation of candidate answers, but also for the information it gave about
    trends in question form. I found it interesting that the different modules tended to do better on
    different forms of questions (like the thesaurus-based ones doing better on one-word clues;
    pop culture databases better on clues with a blank, etc.)

    -flaws-
    They mention at the end how they used probability theory to improve performance, but they
    don’t give any details about how it improved their performance, or how more improvement
    could be made later.

    -the future-
    It'd be interesting to take a completely unusual crossword, in terms of clues and answers,
    perhaps from some radical new school of crossword-writing, and see how PROVERB did on it.
    Specifically, the CWDB would be a lot less useful, so it'd be relying more on the other modules -
    how much would the CWDB's contribution decrease? Would it be useless, or are there some
    things about crosswords that end up requiring the kinds of clues and answers that are typically
    used?

    They mention in a footnote that there’s a commercial AI crossword solver intended as a solving
    aid. It wouldn’t really be research to see how PROVERB does in that arena, but if could be a
    potentially lucrative path to follow. Also, I want to know how many times the word
    “cruciverbalist” shows up as a crossword answer. :P


  • Next message: Jessica Kristan Miller: "Cruciverbalist paper review"

    This archive was generated by hypermail 2.1.6 : Mon Dec 08 2003 - 10:40:58 PST