NetNews Usenet Archive 1992 #26/NN_1992_26.iso/spool/bit/listserv/csgl/1405

Text File | 1992-11-08 | 9.3 KB | 175 lines

Comments: Gated by [e-mail]
Path: sparky!uunet!paladin.american.edu!auvm!VAXF.COLORADO.EDU!POWERS_W
X-Envelope-to: [e-mail]
X-VMS-To: @CSG
MIME-version: 1.0
Content-transfer-encoding: 7BIT
Message-ID: <[e-mail]>
Newsgroups: bit.listserv.csg-l
Date:         Sat, 7 Nov 1992 09:25:48 -0700
Sender:       "Control Systems Group Network (CSGnet)" <[e-mail]>
From:         "William T. Powers" <[e-mail]>
Subject:      Knowledge base; hierarchy & lnaguage; address
Lines: 160

 [From Bill Powers (921107.0800)]

 Dick Robertson (921105) --

 Hardly any of your post got through. When you don't use hard carriage
 returns at the end of each line, everything gets cut off after the
 first line until the next hard carriage return. I use my PC to
 generate text and "print" it to a file, so that the carriage returns
 are always there (except when I forget and send the original). I hate
 the editor that the mainframe uses; it's so clunky and non-intuitive
 and difficult to use.
 --------------------------------------------------------------------
 Penni Sibun(921105.1300) --
 Avery Andrews (921106.1348) --

 I have some questions about the "knowledge base." I gather that what
 you mean by a knowledge base is a set of statements describing
 perceptions, rather than the perceptions themselves. I can see how it
 would be possible to mark a list of statements as "said" or "not
 said," but how would you mark the perceptions to which they refer?

 In a lot of the proposals offered here by linguists, I get a sense of
 something missing. PCT itself is about how a brain works (whether
 we've got it right or not). So I'm always asking "how would the brain
 do that?" I can understand that people can make lists of things and
 check them off, but somehow I don't see that as the way language is
 put together inside the brain -- it's more the way a programmer would
 do it, or someone using a pencil and paper. I think of a model as
 showing HOW something is done, even if only in principle; a lot of
 propositions by linguists seem to me to be concerned with WHAT is
 done, or proposing ways to use the brain's facilities to achieve an
 equivalent effect, as in checking off lists of statements.
 ---------------------------
 Avery:
 The discrete nature of symbol-handling isn't a problem for control
 systems. In Rick's spreadsheet analysis, one level of control is
 logical; the perceptual signal, reference signal, and error signal
 are all either TRUE or FALSE. The input function computes a logical
 relationship between continuous variables: A > B, and so on (several
 different control systems working at the same time). If the reference
 signal for this perception is TRUE and the perceptual signal is FALSE
 then the error signal is TRUE. The output function converts a TRUE
 error signal into a setting (or an increment -- I forget) of a
 reference signal for a layer of continuous control systems just
 below; all the control operations in the lower levels of the model
 are continuous. The higher systems alter the reference levels for the
 continuous variables A and B until the condition A > B is perceived
 as TRUE. At that point the perceptual signal equals the reference
 signal and the error signal becomes FALSE. With the error signal
 FALSE, the output ceases to integrate and holds its value, so the
 lower-level reference signals become constant. The lower systems then
 maintain A and B in the states that correspond to A > B (Rick, do I
 have this right?). You could let the output decay with time, so
 eventually the error would become TRUE again. Note that many logical
 control systems are working at once, so there's no simple relationship between
 a higher system's output and the net reference
 signal for a lower system.

 You could have several layers of discrete control. Suppose you had a
 category level that controlled for the perception A OR B OR C. The
 presence of any of the elements A, B, or C would produce a perceptual
 signal indicating presence of the category whose elements are A, B,
 and C. The category control system would have on-off signals in it,
 and the output would alter reference signals for lower levels (by
 some scanning or search method?) until at least one member of the
 category was being controlled by the lower-level continuous system in
 a sufficiently non-zero state. As a result, a signal indicating
 presence of that category would exist and be sent to higher systems.
 This would happen in response to a reference signal, also on-off,
 saying "provide me a perception of this category" (by using lower
 systems to create an instance of it). Categories are intersected
 simply by sending reference signals to more than one category control
 system at the same time.

 If the higher system were a logical control system like the highest
 level in Rick's model, some logical proposition involving many
 individual categories would be perceived. The control systems
 involved would be controlling for the presence of several categories
 which are variables in a logical proposition. The reference signal
 could be TRUE or FALSE. The error would be converted to an output
 that varies reference signals for categories; if one set of
 categories won't produce the required truth-value of the proposition,
 the reference signal outputs to the category level are changed from
 TRUE to FALSE and vice versa until some set of categories is found
 that can be controlled so as to produce the required proposition with
 the required value.

 This illustrates the basic hierarchical principle in HPCT, which is
 that one level is concerned ONLY about perceptions of that type, not
 with the details of how they are derived or what they mean at lower
 levels. A category control system doesn't care which elements are
 present, as long as the presence of something in the category is
 sensed. The logical level doesn't care which categories are involved,
 as long as they make the logical form TRUE or FALSE as specified by
 the reference signal.

 At the logical level, logical forms are perceived through general
 functions with places in them for variables. It doesn't matter where
 the values come from; in the proposition A -> B (A implies B), it
 doesn't matter what A and B represent at a lower level. All that
 matters, if the reference signal is set to TRUE, is that B is not
 FALSE when A is TRUE. If two categories can be found to plug in where
 A and B go, and their values turn out to be anything but A present
 and B absent, the perceived value of the proposition is TRUE and the
 error is zero. To make the value TRUE, the output of the control
 system must produce (via category control systems) two categories in
 the set accepted by that perceptual function that can be controlled
 in, or are already in, the required states.

 When we experience this hierarchy, we see all the levels squashed
 together in a single space. So it seems to us that it DOES matter what
 lower-level variables are involved in a higher-level perception.
 That is partly because we can perceive as from the lower-level point
 of view at the same time as from the higher-level point of view. It
 is also partly because there is more than one higher-level control
 system involved; lower-level perceptions must be brought to states
 that are perceived, in parallel, in many different ways, for each of
 which there is a preferred level of the perception. So a sentence or
 phrase or conversation, which is a series of words at the lower
 levels, is perceived both in terms of higher-level meanings and in
 terms of higher-level forms of speech or writing for which we have
 preferences.

 The hierarchical kind of analysis, I think, can help us separate the
 different levels that are important in language (as in all else). If
 we just take what we hear as it comes to us, all these levels are
 superposed, and it's difficult to see the differences between them.
 Needless to say, these levels are not those of standard linguistics,
 such as phoneme, morpheme, word, sentence, and so on -- although
 there may be some close parallels. The levels are much more general
 than that: intensity, sensation, configuration, transition, event,
 relationship, category ... . In language they relate to the features
 of speech or writing. In other pursuits they relate to other aspects
 of experience.

 I don't mean to suggest that at the "logic" level, the operations are
 all Boolean, or that at the category level they are just an OR. These
 are just illustrations; we don't know what really goes on at these
 levels. The "logical" rules of language are probably not Boolean. I'm
 just trying to get the principle across, that each level is concerned
 only with variables of its own type, and that discrete control
 systems are not hard to understand in general.
 ------------------------------------------------------------------
 Eileen Prince --

 Ed Ford failed to say that yearly membership in the CSG for students
 is $5. That's one reason the fee for working members is $45 --
 there's a subsidy.

 Also, in the address for the Control Systems Group, please be sure to
 say Ridge PLACE, not Ridge ROAD. And be sure to add the County Road
 510 (CR 510), because there IS a Ridge ROAD in Durango West, about 25
 miles from here and on the other side of Durango.

 Ed, please check your address materials.
 -------------------------------------------------------------------

 Best to all,

 Bill P.