Providing Help and Advice in Task Oriented Systems
Proceedings of the Eighth International Joint Conference on Artificial Intelligence. Expert Systems. Pages 176-177.
© 1983 International Joint Conferences on Artificial Intelligence
Abstract
This pap( ·r d esc rib es c urr e nt w o rk at th e U niv ersity o f Pr· nn sy l v:-t ni a ce nt e r ed a r oun d p r ov idin g ! nt cll i g~ nt h r lp :1: 1d advire to user s of iHt .e r :.c ti ve task oriented ;yste m s . Thi s wor< foc u ses on tt r ce ge ner a! t hem es: (I) H e lp syst e ms s hould be ac tiv e r:<thd t!1a n passiv e; ( 2) he lp syste m s s h ou ld conta in exp lic it mo d c!s of t:1e u se r , the to ' k a nd the s ystem util ity being U'C d :md (3 ) the he lp syste m s ho ul d e n gage in an int e ra c tiv e di a log ue with the user in o rd e r to id enti fy th e info rma tio n he r:·:-t ll y n ee d s . An expr rim f' n ta l hclJ• sysL em , W!Z .- \RD , ha s bee n impl eme nt e d fo r the VA.:Y..jVMS opr ra lin g sy s t e m to exp lo re so me o f t h ese iss u es.
- References l. 13all , E., a nd P hil I bFs, 'Re prese ntation of Task-Specific Knowledge in a G rac efu lly Int eracting Use r l nte rf:~ce" , i:~ Proc . AAA I. 1980.
- Brach m an, R., A St rudu ral Paradigm for l1eprc, e nt in?, 1\nowledge , BBi\1 Hcport no. 3605, 1978.
- Drac hm an, R., E. Cicc:~rdli , N. G ree nfeld and ;\I. Yonk e , "!\:LONE Hefcrence Manu al• . 13BN repo rt no. 38,18 , 1978.
- Bu rt o n, R. and J . Brown , 'An inv c's tigation of C<>mp ut.er Coac hing fo r Infor mal Learning Activ it ies " , DON [kp ort :l\) 14 , 1978.
- Fi n in , T., 'Help :<nd Advice in Task O rient ed Sys tems" . 1 1cport MS-CJS-1982-22, C<1 mputer a nd Info. Science. U. of Penn. 1982.
- Ge nese rcth , M. , "Autom ated Co nsu ltat ion for Complex Computer Sys tems ' , Ph .D. T hes is, Dc p:~r tmcnt of Computer Sc ience, Ha rv a rd Univ ersity, 1978.
- Gencscr;' th , M., ' The Ho le of Plans in Automated Co nsultat ion" , P roc. IJ C.\I-7i, 1977.
- Howe, A., "HOW? A C ust o mi zable, Assoc i :~tiv e Ne tw ork B~se d Help f'acility • , Hepo rt MS-CIS-1983-14, Co mputer :1nd Info. Sc ience, U. of Penn. 1983.
- Go ldstein, 1. , "T he Co mput e r as Coac h : A n Ath letic Paradig m fo r Intellect ual Ed ucatio n' , A IM 389. M.I. T. 1979.
- Hi<'sbf'ck , C. and H. Schank , Compnhension by Comp uter: Ex pec t :~t i o n Based Analy sis of Se ntences in Co nt ext. Yale Uuiv ersity CS report no. 84.
- Sc h:• :;k , R. and R. Abelson , Scr ipts, P lan s. Go als a nd Llnd ersta nding , Law erence Erba um P ress , H illsda le NJ, 1977.
- Sc hu ster, E. a nd T. f'inin, "Understanding Misund erstanding -Recognizing and 11csponding to User Misund erstand ings ' , repo rt MS-C IS-83-12, Co mputer and Info. Sc ience, U. of Penn. , 1983.
- Sc hu ste r, E ., "Custom-Made Responses: M:~intaining ar.d Updat ing the Use r Mode l' , re po rt MS-C JS-83-13, C IS , U. of Pe nn ., 1983.
- Shr ager , J . and T . Finin , 'An Expe rt System that Volunteers Advice ' , Proc. AAAI-82, 1982.
- Shr ager, J., 'Invokin g a Beginn er 's Aid Processo r by Recog ni zing J CL Goals ', re port MS-C IS-81-7 , Co mputer and Info . Scie nce , U. of Penn., 1981.
- Wilensky, R ., ' Understanding Goal-based Sto ries • , Yale Univ ersity Research Re po rt No. 140, 1978.
- Wilensky , R. , 'Talking to U nix in E nglish • , tecnical report, Un iv . of Ca. -Berkeley , 1982.
References were extracted automatically from the PDF and may contain errors.
Full text extracted automatically from the PDF; figures, tables and equations may differ. The PDF is the version of record.
1. Introduction
The s tate o f t.he art in "hel p · sys tems h:-ts no t changed much in the h st dcc·:-tde (·H two! ). Cu rrent hdp systems typic:dly bll into one o f three gene ra l c:t t c·gories: key word systems , hierarehic:-t l key word sys tems a nd network systems. (dcs!'ribcd mo r~ fully in !F'inin 8:2J) . Present approachs to prov idi ng help to inexperienced o r infrequent users bils for t wo impor t:lnt reasons. First , c urr~nt iwlp syst<·ms have no real unders tanding of who they are tryi ng to help (e.g. what a particular user docs a nd does not know, what goals users in gene ral arc like ly to have), the topic they are providing inform::ttion :tbout, o r the information they can provide. Second , cu rrent help systems a re not inLc rac ti vc. The process of suc-:c·ssfully requesting and getting the informal ion the user needs is one in which both the system and the UH :r 'hould be :11't.ivc agents. At a ny po int. the party should be able to take the initiative.
:\s a n example o f a kind of he lp that cu rrent technology can not prov ide, cous ider the prohkrn of the user who doesn't know he n ~cds J,elp . It is quite common for a complex system such as a t ext editor o r opera ting sys t.em to be designed so that a new user can learn a few b:1sic comma nds which are suffic ient to accomplish mos t t a&ks. Add itional commands which g reat ly ext end the conven ience a nd p ractical power o f the syslem :>re provided for the user to ·g row into · . It is common, however , fo r some use rs to become trapped by th e s imple complete se t o f basic commands a nd never progress to learning the full power o f the system.
A huma n adv isor , wa tching such a user , might notice that he is not making full use o f t he system a nd volunteer advice abou t us ing mo re adv anced comma nds. For examp le, if you see a n inexperienced user issuing the following commands to your op~ r ating system :
DELETE PARSE.PAS. l
DELETE P ARSE. P AS .2
DELETE PARSE.PAS.3
and leav ing the fil e PARSE.PAS.4, you might o ffe r t he following adv ice:
It looks like you're trying to delet e all of the old vers ions of the file PARSE.PAS. You can do this directly with the PURGE command. For example, you could have said:
PURGE PARSE.PAS
and it would have had the same effect.
th <· r<·lev an t top ics.
- H'hat infor·mation ~an t.'te system provide? The uS<·r :llld the system n··ed t rChave ,;om~ morkl dcscribi :1g the infu r m:~ tion t hat the help sys tem caB dc·livN. The user cou ld d i,covrr this himself if the help sy ~ t <· m pr ~>v ided a "meta-hdp' facili ty tho t could pru·· irk info r111 a tioH on t he help sys tem itself. The sys tem cou ld t ry to inform t.h c usr· r about the range of information !t cou!J provide by presenting summaries ,, f thr available he-l p or by r nun. erat ing only wh :1 t it est ima t es to be the rclcv:10t topics.
- Which offering· m.atch the nee ::Ia? Once we have a ric,rription of t.b c information the user nerds and d<'oniptions uf the inform :1t ion that the system can provi rl<-, we nerd to irl enl ify thr bc3t matches. In general t here will not br a orll'- lo-onr n.:>tch. Th r user can usc his own fac ilities to co rn parr· the desc riptions or the syd<'m euuld IJ>C some sort of description-matching process to , ug!;<''t ma tches.
- How does one ask for a particular piece· of information? Once the user has dec ided what inform:l tiun he needs and which of thr offered informa l ion ma t cho'S his need , he has to know how to spr·rify tli :lt to the· help syst r< n . A wide range of mock s ::trr possible which a llow the u~c r to describe the in format ion he wo nt" topic n:>mes, key words. key words with qualifir·rs, s imple na tural language-like phr:>scs a nd unrest ri cted natura l language . T he system , on t he other hanri . could t:. ke the in itiat ive at this step by present ing a menu of likely cho ices or by selec ting the rnost likely candid a te :wd 1- imply presenti n!', it to the user .
- Does the user understand the help? Once the user rPceivcs tlw inform :~ tion he has to unders t and it. It shc>tlid be t he sys tems goal to tailor , when poss ible, the inform:1tion it. presents the user to tha t user 's level of cx p<'r tise :>nd previous rx pcricnrr. The sys tem should IH' :1rt ively t ry ing to verify th:1t the user h :~s correctly uudN stood the informat ion he was given . The use r should he :~hie to as k for cbrification or examples . Thr system might be prepared to offer background inform :~t ion (e.g. defini t ions of t erms ) and alterna tive presentat ion of t he sam(' inform:l tion.
3. WIZARD
WI X. \ /?/) is an ex Jwrimcntal l!elp sys tem whose dom :~ i n is a ~ubsc t of t h<' VAX/ VMS opera ting system ([Shrage r 1031j, jShrager and F inin , J VS~j f. It attempts to provide inr xperien<·ed or infrequent " " '" with " Cf.·rta in kind o f advice to help them learn to usc the ope rati ng system cfficiel!tly.
In dcH·Ioping ll'fl,, \1?0 we a t.tempted to focus on t he problem of rcco~n i z in g when t he user required help and volun teering advice. Ou r a pproach is to recognizes correct yet ineffic ient command "'CJUcnces and help the bel> inner become more proficient by indicat ing huw t hese tasks rn a.v be done more effic iently.
T here are severa l dimens ions to inefficiency in operating syst~rn in l<'rac tions: operating sys tems provide many features (such as wilrl car<b in file names, lists of verb t argets, etc) which are meant to
minim ize the user's work (e.g. , typ in~ ) . Consider: $!'!tiNT PfWGRESS.MEI\1 $!'!11NT l\1EETING.MEM <· t. c .. r:1 t.h<·r t h:IB : 3
$!'fliNT *.l\ lEl\1
Un another dimension . we measu re inefficiency in terms of system r<·sourccs. The system typ ically provides spc<'ial fa nctions wh ich Jwrfurm opNat.ions much more sp:.ringly than more general means wou ld permit . Cont rast:
with : $COPY NOTES* OLDNOTES* $DELETE NOT r-;s**
$11ENAME NOTES. * OLDNUTES.*
1\'JZAUD recognizes such "less dficient' sequences and constructs help :11cssag··s t h:~ t pruvid ~ either immed i:tte advice or a poi~tcr to a manua l or online HELP cnt.ry . In t he following example, WIZ:I!lD's advice is shown indent("(!:
$C:OPY TEST l . TEST:2. $DELETE T EST!..*
I f you arc trying to change the name of the f ile TJ;;!:>TI. to TI\ST I.!. you cou ld have d011c this m ore directly with the RIINAAfE' cmntnnnd. e.g.
$fi£NAME' TJ;;STI. TEST2.
1~ypf f!EIJ' HE'NAME for more informat ion 011 the J?ENA1\1E command.
In ou r c urr~n t IVIZAUD implcment a t.ion we a ttc·rnpt to prov ide a non- interruptive inte rface through t he usc of mul t iple windows. IV!7AUIJ '~ comments a re presented in a spec ia l pop··U[J advice window which t.h e use r is fr e<' to change the size of. elim inate or simply ignore.
The difficulty of WfZAI?D's task is, of eo1m:e, to rccop,ai'c wh r a Sf, tr. <' sequence of comm:tnds const it ut es a plan th a t a pe rson is us ing to ar hievc a go:1 !. We h:tvt· approached tltis problem by constructing a c:> t. :do~ of "lnd plans ' which novice uc-c rs often use to ac hieve w mnw n goals. T he probl<·m thus reduces to rn a tch.ing command SNJ·t cnr: <·s to d··scriptinns of generic plans from the cat :~ logu e. In I his appJ; ,,a tion , the matching process is complicated by the following t~sues:
- o Non-contiguity - T he ind ividual commands which make up a sequence might be spread out over a session. Each of the intervening commands may or may not affect tlw goal which the overa ll sequence is meant to ac hieve.
- Non-lineari t y - A single event may play a part in severa l pbns.
- Ambiguity- The m:~pp i ngs from sequences of events to plans and from plans to goals are both many to many. A given sequence may ins t:>nti a te sev e r:~! plans and a certa in plan can be ins t ant iated wit.h a number of sequences. Similarly , a pi:ln might be used to satisfy sever:~! goals and a particular goal can be realized by several phns.
- Eztenaional Knowledge- In general we may need to have det ailed knowledge of the user's environment since a given sequence may have side effects or use information
not directly expressed in the sy nt~x of the commands. In o:d(·r to recogn ize which uf ,·,eve r :~ l possible go:lls is being :~ttempt:·d one may. for example, h:we to expand "wildca rd" pattNned fil rn:.mb and select n:tmes from the current directory which :~rc refer red to by the command at hand.
The goal recognition process is driven by an expectat ion-hosed parser [r:irsbcr k and Schank[ which uses a knowledge base represent ed in a network i>ased representation language based on Kl~ ONE. This knowledge base represen ts many of the concepts for objects and processes which underly the task domain , generic and spec ific commands that the sys tem provides, and the actual history of tlw int eract ion between the user and the task system. Thus, the model of the l:tsk and of the system have been merged by having a single taxonomy which contains both generic t ask concepts (e.g. commands which create files), generic system comm:~nds (.e.g. COPY) and individu:tl instantiations of commands issued by :<particular user (e.g. th<· COPY command that user TIM issued a t 3:02pm on 1\.hy :kd) . T he only information that WIZARD keeps about an individual usrr is which comm:~nds he h:ls been observed using and which piec~s of :~dvice he has already been giH n.
When the user issues a operati n sys tem command , individual concept s representing the comm:1nd are added to the knowledge base. If this event matches the initial event in a sequence of a plan , a demo n is created to monitor the knowledge base for the instantiation of a concept matching the next event in t he plan's sequence. If a concept matching that event is subsequently inst antiated , then the next event in the sequence is monitored for . When all of the events in the sequence have occured , the plan is checked to ensure that it s:~tis fics any loc:tl and global constraints placed on it (e.g. what kind of intervening events, if any, can occur) .
If all of the constuints are satisfied , t hen WIZARD takes the goal assoc iated with that. plan as the use rs go:~ll WIZARD's advice is generated from the advice template assoc ia ted with with the pbn . The advice template may contain references to individual objects (e .g. file names used :IS command arguments) in the even ts matching the plan sequence. Before offering t he advice, WIZARD checks to see if this user has already been shown it.
4. Conclusions
The W'J ZARD Help system has add ressed only the problem of recognizing ccr t:~in s itu :~t. ions in which the user may need help or adv ice. Ou r cur rent work is direc ted toward developing better apprn:~chcs to modeling t.he individual user (i.e. his interests, level of expe rtise, etc. ), the generic user (i .e. his a priori goals, his planning stategics, etc) and toward making the interact ion between the user :tnd help system more of a dialog.
In particular, we have been working on recogni zing and responding to a user's misunderstandings [Schuster 83[, using a model of the user as a has is for customizing the help offered [Schuster 83) and various spread ing activation techniques for identifying relevant help informat ion from a set of weighted keywords [Howe 83[.
1 The current implementation does not handle multiple goals associated with a ftiVPO nJ~. n.
5. Acknowledgements
Thi' work IS partially supported by the N:.ti·Jnal Science Foundation under grant number IST81-I :2139. .Jeff Sh r:<ger collaborated on much of this n'~ea rc h and is re,poiJsible for i he current WIZARD implementation.
6. References
- 13all , E., and Phil I bFs, 'Representation of Task-Specific Knowledge in a Gracefu lly Int eracting User l nte rf:~ce" , i:~ Proc . AAA I. 1980.
- Brach man, R., A St rudural Paradigm for l1eprc, ent in?, 1\nowledge, BBi\1 Hcport no. 3605, 1978.
- Drachman, R., E. Cicc:~rdli , N. Greenfeld and ;\I. Yonke, "!\:LONE Hefcrence Manual· . 13BN report no. 38,18, 1978.
- '1. Bu rton, R. and J . Brown , 'An inv c'stigation of C<>mput.er Coaching for Informal Learning Activ it ies" , DON [kport :l\) 14 , 1978.
- Fi n in , T., 'Help :<nd Advice in Task Oriented Systems" . 11cport MS-CJS-1982-22, C<1mputer and Info. Science. U. of Penn. 1982.
- Genesercth, M. , "Automated Consu ltat ion for Complex Computer Systems ' , Ph .D. T hesis, Dcp:~rtmcnt of Computer Sc ience, Harv ard University, 1978.
- Gencscr;'th , M., 'The Hole of Plans in Automated Consultat ion" , P roc. IJC.\I-7i, 1977.
- Howe, A., "HOW? A Customi zable, Assoc i :~tiv e Network B~sed Help f'acility · , Heport MS-CIS-1983-14, Computer :1nd Info. Sc ience, U. of Penn. 1983.
- Goldstein, 1. , "T he Computer as Coach: An Ath letic Paradigm for Intellectual Education', AIM 389. M.I.T. 1979.
- Hi<'sbf'ck, C. and H. Schank , Compnhension by Computer: Expec t :~t i on Based Analysis of Sentences in Cont ext. Yale Uuiversity CS report no. 84.
- Sch:· :;k, R. and R. Abelson , Scripts, P lans. Goals and Llnderstanding , Lawerence Erbaum P ress, Hillsdale NJ, 1977.
- Schuster, E. and T. f'inin, "Understanding Misunderstanding - Recognizing and 11csponding to User Misunderstand ings' , report MS-CIS-83-12, Computer and Info. Sc ience, U. of Penn. , 1983.
- Schuster, E., "Custom-Made Responses: M:~intaining ar.d Updat ing the User Model' , report MS-CJS-83-13, CIS , U. of Penn ., 1983.
- 14 . Shrager , J . and T . Finin , 'An Expert System that Volunteers Advice ' , Proc. AAAI-82, 1982.
- Shrager, J., 'Invoking a Beginner 's Aid Processor by Recognizing JCL Goals', report MS-CIS-81-7 , Computer and Info. Science, U. of Penn., 1981.
- Wilensky, R., 'Understanding Goal-based Stories· , Yale University Research Report No. 140, 1978.
- Wilensky , R. , 'Talking to Unix in English · , tecnical report, Univ . of Ca. - Berkeley , 1982.
@inproceedings{ijcai1983p41,
title = {Providing Help and Advice in Task Oriented Systems},
author = {Finin, Timothy W.},
booktitle = {Proceedings of the Eighth International Joint Conference on
Artificial Intelligence, {IJCAI-83}},
publisher = {International Joint Conferences on Artificial Intelligence},
editor = {Alan Bundy},
pages = {176--177},
year = {1983},
note = {Expert Systems},
url = {https://ijcai.org/proceedings/1983/41},
}
TY - CPAPER TI - Providing Help and Advice in Task Oriented Systems AU - Finin, Timothy W. T2 - Proceedings of the Eighth International Joint Conference on Artificial Intelligence, IJCAI-83 PB - International Joint Conferences on Artificial Intelligence PY - 1983 SP - 176 EP - 177 UR - https://ijcai.org/proceedings/1983/41 ER -Copied
Machine-readable versions of this paper, for tools and AI agents: