home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!stanford.edu!bcm!convex!news.oc.com!eff!sol.ctr.columbia.edu!spool.mu.edu!yale.edu!jvnc.net!darwin.sura.net!paladin.american.edu!auvm!PARC.XEROX.COM!SIBUN
- X-Delivery-Notice: SMTP MAIL FROM does not correspond to sender.
- X-else-reply-to: sibun@parc.xerox.com
- Fake-Sender: sibun@parc.xerox.com
- Message-ID: <92Nov9.163129pst.29193@hmmm.parc.xerox.com>
- Newsgroups: bit.listserv.csg-l
- Date: Mon, 9 Nov 1992 16:31:23 PST
- Sender: "Control Systems Group Network (CSGnet)" <CSG-L@UIUCVMD.BITNET>
- From: Penni Sibun <sibun@PARC.XEROX.COM>
- Subject: Re: Marking and models
- In-Reply-To: "William T. Powers"'s message of Mon,
- 9 Nov 1992 10:44:46 -0800 <92Nov9.130140pst.11863@alpha.xerox.com>
- Lines: 73
-
- (penni sibun 921109.1700)
-
- [From Bill Powers (921109.1100)]
-
- Avery Andrews (921109) --
-
- Your comments on "marking" plus some by Hank Folson and Dag Forssell
- have brought my feeling of "something missing" into sharper focus.
- What's missing in linguistics is a model.
-
- When you talk about marking things in the knowledge base as said or
- not said, you're IMPLYING a model without saying what it is.
-
- i think this discussion of marking is a serious sidetrack. the
- already-said? mechanism was basically a kludge to allow me to complete
- my thesis in finite time. i actually think redundacy is fascinating
- (it happens all the time) but it wasn't my focus. i have a friend in
- computational linguistics who is doing a thesis on just a class of
- redundancies (and i may be able to incorporate her work into salix).
-
- in any event, here's all that already-said? ever claimed to do
- (Locally Organized Text Generation. COINS Technical Report 91-73,
- Department of Computer and Information Science, University of
- Massachusetts, 1991. Also Report SSL-91-21/P91-00159, Xerox Palo Alto
- Research Center, 1991, p. 41) (``{\em'' indicates emphasized text):
-
- Each object [in the KB] has an {\em already-said?} field. This field is
- initialized to {\em nil}; when an object has been explicitly mentioned
- in the text, this field is set to {\em t}. The strategies which find
- something to say next will not be triggered by an object that is {\em
- already-said?}. An object which is {\em already-said?} may be
- mentioned again, but only in relation to another object. For example,
- in the following text:
-
- (1) there's Penni
- (2) and then there's Barbara
- (3) who is Penni's sister
-
- In (1), Penni is explicitly mentioned, and marked as {\em
- already-said?}; in (3) she is mentioned in passing, in relation to
- Barbara.
-
-
- The reason I asked about the nature of the knowledge base and how
- marking might be accomplished is that I was trying to get at your
- underlying model. What, literally, is the form of the knowledge base
- in a human being?
-
- i'm not even sure that there's a kb-analog in the brain; i bet there
- is no ``marking.''
-
- can see what its other implications are (other than serving the
- immediate purpose of explaining how we know something has been said).
-
- ``marking'' doesn't have explanatory power. it's a mechanism (and a
- kludgy one at that, as i said). again, we seem to be getting a
- conflation of programs and theories.
-
- But most phenomena seem at first to be outside the
- model only because we're already interpreting them in terms of a
- different model. "Marking" is an example. Behind that interpretation
- are images of the way we would use a pencil-and-paper checklist. But
- that image is inadequate, because it implies operations and
- perceptions that are not in the model -- and that are not plausible
- additions to the model.
-
- sigh. right. that's why i'm sorry you got that extra-model
- interpretation and hope everyone drops it!! the ``checklist model''
- is not my creation; it's not in *my* thesis!
-
- cheers.
-
- --penni
-