Date: Sat 13 Aug 1988 19:22-EDT From: AIList Moderator Nick Papadakis Reply-To: AIList@mc.lcs.mit.edu Us-Mail: MIT LCS, 545 Tech Square, Rm# NE43-504, Cambridge MA 02139 Phone: (617) 253-6524 Subject: AIList Digest V8 #49 To: AIList@mc.lcs.mit.edu Status: R AIList Digest Sunday, 14 Aug 1988 Volume 8 : Issue 49 Query Responses: How to compile a psychologists' email directory? Ornithology as an AI domain AISB Proceedings Church's Y-operator PCES Sigmoid transfer function Feigenbaum's Citation English grammar (open/closed classes) ---------------------------------------------------------------------- Date: 6 Aug 88 21:39:24 GMT From: mind!harnad@princeton.edu (Stevan Harnad) Subject: How to compile a psychologists' email directory? This is a copy of letter to Bob Morecock, Editor of Psychnet: Bob, Here is a thought I had: You could perform a double service, to Psychnet as well as the psychological community if you systematically put together a psychologists' email address directory. I heard (perhaps from you) that APA will be publishing members' email addresses, but that's in hard copy and a while away. If you could get the addresses in an electronic file/listserver it would be a great service to the field AND would give Psychnet an automatic broad subscribership. I don't know what official policy and rules are on this. I suspect that they're only now being improvised on the fly. But it seems to me that a newsletter and email directory are sufficiently non-invasive so you can probably treat email address information as public-domain -- like (listed) phone numbers. People who want to be "unlisted" could easily put up software that returned unwanted messages unread, and they could even distribute passwords to the only ones they want to hear from; but most psychologists, I suspect, would like to see email used more widely and imaginatively, at least for the time being. Once we reach the junk mail threshold we can start putting in safeguards. A method for compiling such a directory might be this: Besides requesting APA's cooperation (i.e., asking them to give you all the email lists they've gotten as they go along) you could send email queries to all the major universities and research institutions, either requesting their directories of psychologists email addresses, if possible, or else requesting that your appeal for psychologists' email addresses be posted on the local electronic bboards and msgs to ask psychologists to send in their email addresses for the directory and newsletter directly to you. In exchange you could promise to provide email addresses to those who inquire -- this would not have to be done by you personally, but by software, if the directory were set up properly. This is EXACTLY the right time to set up such a psycholgists' email directory; it will get already-emailing psychologists more actively involved and it will encourage others to get email addresses. You might even be able to get a grant to help you do this from APA, NSF or NIMH. What do you think? [You may want to post this on Psychnet to get readers' reactions, but really the Psychnet readership is still far too small and unrepresentative, so in talking to ourselves now we are just preaching to the converted. This also needs to be posted to a much larger population. I'm going to put it on some of the USENET groups to see whether there is other information on compiling such a directory, perhaps from experience in other fields, and also to beat the bushes to see whether this has already been begun or done by anyone else for psychology or related fields.] Stevan -- Stevan Harnad ARPANET: harnad@mind.princeton.edu harnad@princeton.edu harnad@confidence.princeton.edu srh@flash.bellcore.com harnad@mind.uucp BITNET: harnad%mind.princeton.edu@pucc.bitnet UUCP: princeton!mind!harnad CSNET: harnad%mind.princeton.edu@relay.cs.net ------------------------------ Date: 7 Aug 88 00:20:58 GMT From: sunybcs!dmark@rutgers.edu (David Mark) Subject: Re: How to compile a psychologists' email directory? In article <2721@mind.UUCP> harnad@mind.UUCP (Stevan Harnad) writes: > >This is a copy of letter to Bob Morecock, Editor of Psychnet: > >Bob, > >Here is a thought I had: You could perform a double service, to Psychnet >as well as the psychological community if you systematically put together >a psychologists' email address directory. I heard (perhaps from you) that >APA will be publishing members' email addresses, but that's in hard copy and >a while away. If you could get the addresses in an electronic file/listserver >it would be a great service to the field AND would give Psychnet an automatic >broad subscribership. I have had good success in compiling such a directory for geographers and other spatial scientists. Three of us began the project by merging our own lists about 3 years ago. Then, we ran a workshop on e-mail at the Association of American Geographers' national meeting in May 1986. Periodically, I send the file to all in the file, asking them to confirm their entries and suggest colleagues to add. It just grows and grows. I did a mass e-mailing in June to about 240 users, and got about 60-70 new ones back. I even have a few spatial psychologists! We have suggested that a field for email address be added to the AAG's membership form. David Mark, Chair, AAG Gegraphic Information Systems Specialty Group geodmm@ubvmsc.cc.buffalo.edu geodmm@ubvms.BITNET ------------------------------ Date: 8 Aug 88 17:01:21 GMT From: jbn@glacier.stanford.edu (John B. Nagle) Reply-to: glacier!jbn@labrea.stanford.edu (John B. Nagle) Subject: Ornithology as an AI domain Rod Brooks at MIT has been addressing this problem with his "artificial insects". He is referring to the size of the brain, though, rather than an attempt to emulate real insect behavior. His most advanced "insect" to date is supposed to wander around the AI lab searching for empty aluminum cans. I suspect that the time has come to make more detailed studies of low-level animal behavior than have usually been made in the past. It might be useful, for example, to study grasping behavior in squirrels by videotaping their activities as they are presented with food made up in specific shapes, and reducing the videotape data into kinematic models, then trying to find control equations that produce similar behavior. Studies of animal locomotion, from Muybridge to Raibert, have used similar techniques, and the most recent work has resulted in just such control equations. Raibert now has machines that walk, run, and most recently, turn flips. The state of the art in grasping is much worse; most of the work is based on very elaborate computational geometry and still doesn't work too well with complex hands. I have a conjecture that animals do grasping by moving the hand into a relatively standard configuration for the type of grasp and then turning control over to a feedback process that can be modelled by energetic means along the lines of Khatib or Witkin. One could validate or refute a conjecture of this type with properly analyzed photographic studies. Trying to actually build nests or emulate other sorts of low-level animal manipulative behavior will be very difficult until the simpler tasks of basic manipulation are achievable routinely under varied conditions. John Nagle ------------------------------ Date: Tue, 9 Aug 88 14:03:53 +0100 From: Tony Cohn Subject: AISB Proceedings Further to the posting of the call for papers for AISB89 (due by November 1 1988), I have been asked about the availability of past AISB conference proceedings. The first AISB conference was in 1974 and thereafter biennially until 1982 which was the first ECAI. AISB conferences restarted bienially from 1985. The availability of past proceedings is as follows: 1974 (Sussex): not available. 1976 (Edinburgh): not available. 1978 (Hamburg, joint with GI): small numbers available from AISB office 1980 (Amsterdam): small numbers available from AISB office 1982 (Paris, retrospectively became the first ECAI): small numbers available from the AISB office; selected and revised papers available as "Progress in Artificial Intelligence, Steels and Campbell (eds), Ellis Horwood, 1985". 1985 (Warwick): selected papers published as "Artificial Intelligence and its Applications, Cohn and Thomas (eds) Wiley, 1986". 1987 (Edinburgh): proceedings published as "Advances in Artificial Intelligence, Hallam and Mellish (eds) Wiley, 1987". The address of the AISB office is Judith Dennison, AISB Executive Officer, School of Cognitive Sciences, University of Sussex, Brighton, BN1 9QN, UK (email: judithd@uk.ac.sussex.cvaxa). Please contact the AISB office for details of how to join the society. Benefits currently include the AISB quarterly, AI Communications (European members only), and reduced entry to AISB events. AISB is currently considering publication of selected papers from the unpublished proceedings (1974 to 1980). Any comments on this project including suggestions for papers to be included please contact me. _______________________________________________________________________________ |UUCP: ...!ukc!warwick!agc | Tony Cohn | |JANET: agc@uk.ac.warwick.cs | Dept. of Computer Science | |ARPA: agc%uk.ac.warwick.cs@nss.cs.ucl.ac.uk | University of Warwick | |BITNET: agc%uk.ac.warwick.cs@UK.AC | Coventry, CV4 7AL | |PHONE: +44 203 523088/(secretary: 523193) | ENGLAND | ------------------------------ Date: 10 Aug 88 07:38:56 GMT From: munnari!banana.cs.uq.oz.au!farrell@uunet.UU.NET (Friendless Farrell) Reply-to: farrell%banana.OZ@uunet.UU.NET (Friendless Farrell) Subject: Church's Y-operator In a previous article, GODDEN@gmr.COM writes: >Subject: Church's Y-operator Maybe not Church's Y-operator, but Curry's. I know for certain (since I have it in front of me) that it is defined on p178 of Combinatory Logic Volume 1 Curry, Feys and Craig (QA9.C84 v1 1958 in our library system) Y is commonly called the paradoxical or fixpoint combinator. Its important property is that Y f = f (Y f) Look up the ACM Guide to Computing Literature under combinators, lambda calculus or possibly applicative languages if you're really interested. Friendless farrell@banana.cs.uq.oz - mail me if you can ! ------------------------------ Date: Thu, 11 Aug 88 11:57:39 GMT From: IT21%SYSB.SALFORD.AC.UK@MITVMA.MIT.EDU Subject: PCES Subject: expert systems on PCs From: Andrew Basden In July -- parvis@gitpyr.gatech.edu asked for info on the usability of ESs on PCs. He said: >I'm doing research on the usability and feasibility of expert systems on >personal computers such as the Apple Macintosh and the IBM PC. > >There are certainly limitations due to memory size and time efficiency. >What are typical problems when developing and/or using a PC based expert >system? What do users (not only developers) think about expert systems >on PCs? What domain solutions are successfully realized on a PC? Are >the users satisfied with the features and efficiency or are such >systems 'just expensive toys'? We developed ELSIE 1986-6 as one of the Alvey Community Club projects. It consists of four ESs, linked into one system via a common database, to give Quantity Surveyors advice when acting in a Lead Consultant role (hence its name! - work it out if you don't get it; finding the name took 6 months of intensive research!). In this role they help clients who want to build, say, offices, at the initial stages of planning. At this stage the client wants to know, among other things how much the building will cost, so as to set a budget, how long it will take, what the development appraisal over the life of the building will be (taking into account interest, inflation, maintenance, etc.) and how they should go about organising the building project. ELSIE has 4 modules to cover these: Budget module, Time module, Development Appraisal module, Procurement module. We started the project in Jan 86, with the aim of creating awareness of ES technology capability, but in fact found that by June 87 we had produced four truly usable systems. These have now been packaged for sale, and are selling at a rate of over two per week. Some companies are coming back for second copies. Therefore we feel there is evidence that ELSIE is NOT just an expensive toy, and that users (not just developers) think it a good thing. We developed it in Savoir, which is a mature and flexible shell. SAvoir is better than some because it is designed to be fast, and to performs good checking of the KB, such as for loops. The Budget module, the biggest, for instance, has around 2500 rule equivalents. (Savoir is an inference net system rather than rule based.) At this size it is getting near the limits of Savoir on PCs, but we did not hit any such limits during the whole project. We developed on PCs. This actually gave an advantage during validation, in that we could send out copies of the ES for testers to run on their own PCs. The knowledge acquisition and other aspect of building the four modules was based on a methodology mentioned (in an early version) in Attarwala and Basden (1985) References: Savoir, from ISI Ltd., 11 Oakdene Road, Redhill, Surrey, UK. 1000 on PC For description of ELSIE, see Brandon P.S., Basden A., Hamilton I., Stockley J. (1988) 'Expert Systems - the strategic planning of construction projects', The Royal Institution of Chartered Surveyors, London, UK. Attarwala, F.T., Basden. A. (1985) 'A methodology for building Expert Systems', R&D Management. Trust this info helps. I can give more if you ask me specific questions. Andrew Basden, I.T. Institute, University of Salford, UK. ------------------------------ Date: 12 Aug 88 17:48:58 GMT From: pacbell!eeg!marcus@ames.arpa (Mark Levin) Subject: Sigmoid transfer function > >Try this one : f(x) = x / (1 + |x|) I followed up before by claiming that this function is used in physiology and psychophysics to describe neural properties. However the function I gave showed only one half of the function and had an obvious discontinuity, which you may find distastful. If the function is graphed with the x-axis in log coordinates the function becomes your favorite Sigmoid Function. FOOTNOTE: I used this on my thesis which was on the psychophysics of light adaptation. This is a convenient form for displaying it in the area of adaptation since the changing of sigma (the constant 1 above) will shift the function along the axis without changing the shape of the function (change the *threshold*). And this is what we want adaptation to do. But, remember that the real function has the discontinuity at 0 (or Threshold). For those who are interested, a better model of adaptation is by scaling the inputs to the function and keeping sigma constant. This looks the same, but hypothesizes that pre-processing accounts for adaptation rather than changes in neural properties. marcus@eeg.com Mark Levin RA at the EEG Systems Lab. 1855 Folsom St., San Francisco, CA 949103 {pacbell,lll-winken,ucsfcgl}!eeg!marcus ------------------------------ Date: Fri 12 Aug 88 22:02:23-CDT From: Charles Petrie Subject: Feigenbaum's Citation No help for the citation. But I offer the suggestion that Feigenbaum's suggestion has been bumped up a level (and more, recursively) by research into the explicit control of reasoning. It isn't enough simply to have a lot of rules executed by a "dumb" interpreter than it is to have a clever, domain-independent theorem proving strategy. You can get pretty far, but you soon bump up against the complexity barrier. In the former case, it's because rules interact in complicated ways that need to be controled to produce useful behavior. In the case of theorem proving, the community is awaiting the explicit representation of mathematicians' expertise rather than depending upon clever encodings and syntactic search strategies. "Dumb" interpreters have built-in strategies: for instance, OPS. Users take advantage of built-in strategies by hiding control strategies in domain data, e.g., ad hoc control predicates and extra rule antecedents. Genesereth's, de Kleer's, and others' work in making reasoned control explicit has the potential to make systems much smarter. That's where I'd place bets on rule-based success these days: research in the representation of reasoned but explicit control knowledge. ------------------------------ Date: Sat, 13 Aug 88 00:50:25 -0700 From: mcguire@aerospace.aero.org Subject: English grammar (open/closed classes) John B. Nagle writes: > I understand that there is an approach to English grammar based on >the following assumptions. > 1. There are four main categories of words, essentially nouns, > verbs, adjectives, and adverbs. These categories are > extensible; new words can be added. > 2. There are about 125 "special" words, not in one of the four > main categories. This list is essentially fixed. (New > nouns appear all the time, but new conjunctions and articles > never.) >Does anyone have a reference to this, one that lists all the "special" >words? The proper technical term for what I think you are referring to is the distinction between "open class" v.s. "closed class" words. Certain classes of words (where a class is defined by its members in some way behaving the same) contain a finite number of members while other classes contain a potentially infinite number. If you want to construct a list of all closed class words in English you might start with the prepositions, determiners, articles, auxiliary verbs, conjunctions, numerals, verb features, etc. - though your ultimate list depends upon how you define your classes, what "behave the same" means, and what counts as a words. While I'm familiar with this distinction, and think that it may have been around in linguistics for quite some while (Bernard Bloch maybe?), I don't remember it being used much. The only references that spring to mind are some studies in speech production and slips of the tongue done in the 70s by Anne Cunningham (she's a Brit though I'm not sure of her last name) and maybe Victoria Fromkin claiming that less errors are associated with closed class words and that they play some privileged role in speech_production/syntax/lexical_access/the_archetecture_of_the_mind. I can't think of any explicit influence the "open/closed" distinction has had on generative grammer. I feel however that implicit awareness of this distinction has lead people to construct and prefer theories where closed classes correspond to atomic linguistic categories. Coupled with the generativist bias on how classes are defined, this preference has left most most current theories analyzing the examples: "John loved Mary" "John has loved Mary" "John might love Mary" "John seems to love Mary" as having practically nothing in common. ------------------------------ End of AIList Digest ********************