"Doyle & Patil, "Two Theses of Knowledge Representation""

From: Lillie Kittredge (kittredl_at_u.washington.edu)
Date: Wed Oct 22 2003 - 10:18:07 PDT

  • Next message: Daniel J Klein: "Review of Two Theses of Knowledge Representation"

    Lillie Kittredge
    Review:
    John Doyle, Ramesh Patil, "Two theses of knowledge representation"

    This paper argued against a proposed restricted scheme of knowledge
    representation, and made some motions towards outlining a nonrestrictive
    scheme.

    The main points of the paper were the argument against the restrictive
    scheme:

    1. Restricting languages by ommitting some constructs "destroys the
    generality of the language". That is, there are definable concepts that
    become inexpressible. Since some of these concepts, explained in great
    length in section2 of the paper, are necessary to some applications,
    this is unacceptable.

    2. A fully expressive language, however, must include some of the
    constructions that the creators of the restricted lanaguage were trying
    to avoid, specifically the ones which require non-poynomial worst-case
    response time. Though Doyle and Patil point out that worst-case time is
      not always an appropriate metric, they acknowledge that there must be
    a tradeoff between expressiveness and completity, and that because of
    this tradeoff, there can be no general purpose langauge.

    I felt that the largest flaw of this paper was that it really was not
    clear about what an alternative, non-restrictive knowledge
    representation would consist of. Given that almost all of the paper is
    devoted to tearing down the restrictive scheme, I'd be a lot more
    impressed if they presented a clearer view of an alternative. As it is,
    they just say what principles should be important in designing one.

    Also, it was _way_ too wordy.

    Open questions: What would an unrestricted representation look like?
    How would they actually implement all their ideas about what properties
    it should have? One might also try to classify the systems in which the
    restrticted version would work; ones which do not require the assortment
    of defenitions outlined in the paper's section 2 which are inexpressible
    in the restricted version-- what systems can make do without those, and
    would they really benefit in performance from the restricted
    representation?


  • Next message: Daniel J Klein: "Review of Two Theses of Knowledge Representation"

    This archive was generated by hypermail 2.1.6 : Wed Oct 22 2003 - 10:17:47 PDT