KBMS Requirements of K Matthias Jarke  Bernd Neumann Yannis Vassiliou  and Wolfgang Wahlster STRA CT This overview paper pr ovid es a cust erspecti ve of the require ments fo base managemen systems
127K - views

KBMS Requirements of K Matthias Jarke Bernd Neumann Yannis Vassiliou and Wolfgang Wahlster STRA CT This overview paper pr ovid es a cust erspecti ve of the require ments fo base managemen systems

The cu sto er is ta ken to b eveloper of k nowledge based applic ation systems such as rul based expe t sy stem s natural language interface vision systems and design support c t t a a k management functions that if pro b a K c s s t d plications

Tags : The sto
Download Pdf

KBMS Requirements of K Matthias Jarke Bernd Neumann Yannis Vassiliou and Wolfgang Wahlster STRA CT This overview paper pr ovid es a cust erspecti ve of the require ments fo base managemen systems

Download Pdf - The PPT/PDF document "KBMS Requirements of K Matthias Jarke B..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.

Presentation on theme: "KBMS Requirements of K Matthias Jarke Bernd Neumann Yannis Vassiliou and Wolfgang Wahlster STRA CT This overview paper pr ovid es a cust erspecti ve of the require ments fo base managemen systems"— Presentation transcript:

Page 1
KBMS Requirements of K Matthias Jarke , Bernd Neumann Yannis Vassiliou , and Wolfgang Wahlster STRA CT This overview paper pr ovid es a cust erspecti ve of the require ments fo base managemen systems. The cu sto er is ta ken to b eveloper of k nowledge- based applic ation systems such as rul -based expe t sy stem s, natural language interface , vision systems, and design support c t t a a k management functions that, if pro b a K c s s t d plications . H wever, the r ange of re quireme ts a wide to permit the development of a comp g K i t s s a e g generalize DBM . 1. A

Customer View of KBMS One way to introdu ce a Knowledg e Base Mana (KBMS) is t im pose its a u o t u " i a g purpose sy ste - xture of the best eas fro m artificial intelligence da t w s p i he i a n do main s." Th is de lib er at el exagger view h s be follow m ny of us who are attempting to ex tend ing ex istin database or kn owled e-directed techno lo gies. This p r and the sub equent on es in this pa rt o th e book take a d fferent view. We ask: "Who a e th e pot enti al cus s o K MS s nd wha a the se rvi th ey woul d like to buy Following [MY O86 , a e tha t e dir ct cu sto s for KB MS s

w be th developers of knowledg e-b sed a pplication , rather than the en d u ers o those application . Generally speaking, we expect d velopers of knowled e- based ap licatio to n eed th e fo llo win knowle dge re prese ntation la ngua ges that press the struc ure of the gi ven ap plicati on effectivel y (i.e., form all , conc isel naturall y, etc.) know led e or ga niz on t a f t s a e h of lar e am ounts of lex kn owledge s ructures; meth odolo ies a nd en viro nmen ts through w ic h the c stom er ca n cre te , m knowledge bases efficiently and effectively Additionally , end users often need

to have multiple know ledge-based ap plications and other app tion int ra ct wi th ach oth r ffi ci ently . This l ead s to a four th gen stom er nee : support for the re-use of existing hardwa re and softwa re facilities. In gen ral, knowledge ba ses can b ch aracterized by the gree to which they offer the above facilities. Fo r ex ample, Brach man and Le vesqu [BL86 defin (the content of) a knowledge base as th e set of non-ruled-ou t possibilities. If the knowledge b se is empty eve hing i pos sib An op er atio n c ll ed TE LL info s the kno wl ed ge b abo t c rt ain Un iv ersitt Passau

FRG iv ersi t t H mb F G Foundation of Research a nd Technology in Hellas, Greece Un iv ersitt d s Saarlan s, FRG In: J. W. Schmidt and C. Thanos (eds.): Foundations of Knowledge Base Management: Contributions from Logic, Databases, and Artificial Intelligence. Berlin: Springer, pp. 391-195.
Page 2
re str tion on the se t of pos sibil . Th e op er ation AS K a llo ws th e us er to q ery the knowledge base abou t its kno wled ge. It is sh own that data bases are special kn owledge bases in wh ich only very si T ELL op erati ons are possibl e. In th e rel onal dat bas mod l, for

example, only function and variab le fre n- ary predi at es of pred efin ed ruc ur can b ed. e of this r ri ct ed for of the TE oper at ion. the AS K o er atio of DBMS can be ficiently im pl emented by dire ct re tr ie val. As more co mplex TELL eration (e.g ., in cluding rules, in co mpl fo co mp ect s, p e ( w furt her expressiveness, even decida bilit y) is reduced. The onl y now n o o d not to start with an em knowledge base but t mu dom ain-specific knowledg e as possible into a KBMS, i In the past, m st AI researchers have therefore b ilt their own tailored knowledge base nagement

faciliti es. The onl y products available on the m rket, commercial databases and, m re recently, expert sy stem s s d t The se ge ne ra l t appear insufficient for serious knowledge-b ased applications; however, r KBM fo r each app cati sev erely rest ri ct s th e ec feasibilit y of AI. Therefore, t e question arises: Are there classes of (knowledge-ba ed ) applica tions large enough to ma ke th e co tru ti o special KBMS economically feasible, yet foc d ough to pr ov ide e ffec iv and e ffic ie nt suppor t t d veloper? This ov erv ew ch apter, based on four p sition p rs deliv ered at the

Xan a wo rkshop, attempts an affirmativ e answer to this question by sketch ng knowledge b se men requirements for fou po rt ant cl asses o AI appl ication : rule-based expert s stem s, natural lan ge un dersta ndi ng and ge neratio n s st em s, visio s ems, and design enviro ents. We con lud that th e K requi re me s of su ch ap plication are rath er diverse. With present-day technology , we do not exp ct a gen rali zed KBMS to be ab le to h le ll of these app cations, nor will a s mpl tool -kit appro ach suf ic e. Ins ead , w advo ca te a " evo utiona ry " app oa ch [MY O86 th at develops

specialized rep esen tation sch mes for classes of knowle dg e-b sed application like those ske ch ed abo e. o mo il ed exa pl of su s em are prese te d in subsequ nt chapters by Neu nn (g eo me tri al s ene d pti ons) and by Borgid a et al. (d at ab as soft wa re d velop environ ent ). 2. KBMS Requirements of Rule-based Expert Systems Expert Sy st ems (ESs) are co puter programs tackling p le w ere, e en no straigh forward an aly tic methods exist, hu mans are able to achiev e results. The first exp rt sy stems were d loped more th an a d cad ago for a nu mber of well-bound ed stand- alon e

applications: the assistan ce of o ganic ch emis ts (e.g., DENDRAL), similar sy stems for d agnosis o pulmon ry diseases, internal med cine, and in fe ctions (e.g., MY CIN, INTERNIST/CADU CEUS, PUF ). Sin e then, ES s were appli d to several application do mains in cluding eng neering, mathematics, and geology . More recently , attempts to solve bu sin ss prob le ms with Exp S ste s in ar s such as insuran ce and banking [PAU86, REIT84 h cau ed great int rest. S ral E surv ey s ex t in t e literature [HWL83, NAU83, SZ OL86 Abs ting f th e d ve lo ent of ES s, thre e co mmo funct ona l c ar ac

te ri sti s can b ident ed - arac teristic s tha ha ve becom de fa ci e guidelin es fo r building ESs. These are: a. Knowledge b se. A representation of heuristic nd factu l information, t f o f assertions and deduction rules b. Infer nc e eng ne. A mechanism play ing the rol of an inte applies the knowledge as represented in a suitable way to ac hie e res
Page 3
c. Man-machine interface. A mechanis that trans ers queries from and answer s to the user, so ti mes seeking additi onal inform ation for t Inference En gine. Th is in cludes ex pla atio facilities for t e user. While th e

latter two co mpo ents are of p ramount im rtan ce fo r th e efficien t op eration and usability o an ES, th e Knowledg e Base is g rally consid er ed th e "heart and soul" of the sy stem, and, cons equ ntly , the on e th at h a ttract d most research i terest. It i of ten mi ak enly as su med th at rul s the for o IF it uati on> TH EN < action> , are the ba sis for represen ting knowledge in all ESs . Even though m ny othe r w ys to re pres t knowledge exist (e.g., frames ), rules h ve indeed been the do mi nant representation, esp cially in co mmercial applications Logic-base d sy ste s can

al be consider ed rule-ba ed , where rul and the factu l part o the knowled e cannot be distinguished. Our focus th en i ased ESs. The co mme li tion of rule -ba ed E s h s b en t e sou of mor r se ar inte re st, bu t ne requi rement s ar e al so su rf acing. Infor lly co mmer li zation o E s i s the abil ity to dev lop th in a MIS d rtment, and to have end-u er dep rtmen s as well as th e MIS d part ment use th em in th eir daily work . The first requ irement fo r ESs that was id entified is th at they can be built and used by MIS personnel. ES dev lop ent too s and appropriate MMI d si gns t

le is r quir em ent ( she ls "). The second major requirement, the importan e of which was only recen tly recognized in indu stry and conc ern thi chap te r, i th e n eed to t t e E with o her MIS app ic ation [V 83 ]. For ex pl e, the XCON sy stem has cu rren tly a bout 4000 rules but work s on an underly ng database of 9000 VAX co mponen d scription with 20-125 attributes each [FM86 ; co mmerci al d tabases, used, e.g., in connection with ESs for planning applic ations, are of co urse much larger but ten used with smaller rule bases. Sev ral examples of larg e organi zation att mpting to ext

nd the scope of ES s fro smal l sp eci li zed application to main stream applic ation have en reported. The b ackbon e of such exten io ns is th e abili ty to tie t e ES with t e organ zat ons' l rg e ope ration l d bases. As an indicativ e exampl e, we me ntion th e e ffort at a maj r ch arg c co any to cr eat e a ver tion sy ste w ith th e he lp of an ES that requ ir es equ nt rev of th e cu st er cr edi histo Try ng to et th e abov e requirement, and parallel to research effo rts, th e indu stry is following three sep pa th s. The fi rs t path is a cus -ba ed ap proa ch to co uple sp ec ifi

ES s and co mme ci al da ta base em s (e .g., I fe re nce rp. . The s econd th a l- pu rpose me chan ism, with in the E software d velop ent sy stem, th at translates ES co mmands to databa se qu eries (e.g., Kee Connection ty ing directly into an S d tabase). Fin lly , the third th is th e int grat ion and st orag e o ES rules with the d tab se itself (e.g., Postgres by Relation l Technology , Inc.). Th ere is little ind catio n that, in curr ent co mmer l ef fort s, the ma ge ment o v ry la rge of rul pl ay s an i por tan p rt ; but th may co me with increasing sophistication of th e ru le

ses, fo r example, wh en ry large sp ecification have to b managed Even though the philosoph methodolo and time horizon for their effective use d ffer fo r the thre e app oa es , th e ob je tive and dir tion ar e co mmon: sh ifting the em ph asi in an ES to wards th e ty pe of sy stem that we cal a KBMS , wh ile meeting th e requi rement to ti e an ES with oth r ap plication We ca n mm arize the KBMS re quirem en of rul -based ES s as follo ws: Knowled e r presentati on : Sim ilar to databases, kn owledge r present relati vel sim le , consist ng of a re pre sentat on of underlying facts a d of

a represe ntation of du ction rules. Often, the se rules c pond t H rn cla se form such th at gen ralized reasoni ng mechanisms can be rather efficient. Know led e or ga niz on: If fact bases be com large or external fac bases m accesse d, DBMS echanism m st be m de a aila ble to the ES . ca ll y, t is i ude s s m takn ow led e about the da ta base t i c a t a i aces through wh facts can be stored and re trieved. Sophisticated de ductive query
Page 4
keep re spon se ti me s cept bl e. Th es e b c w o s t p c a plication. Environments: Basi cally , these ar e the t ools available in

th e re spe tive e xpe rt sys em s ls . In one hodol ogy ght c oupl ng) da bas ac cess will b dd en fro m th e u er as fa r as p ssib e; in th e coupling , th e us er wi ll l ad r quir ext rn al d exp itly befor consulting dialog [ V84 . Coup lin g: Be sides accessi ng the external databases mentione d above, e nn ect on t ot he r i fo rm ati stem s too s, in cluding nu merical co mpu ation [KOWA86 o graphics, as, e.g., b F The ch apters in part II of this book present so me pro posals th at ad dr ess th e qu estion of integrated fact and rule manag ment require d to support KBMS for

rule-based sy stems. Howev r, the follo wing sections will demo trate that oth r knowledge-based app ic ations may req ire more substantial extensions to DBMS. 3. KBMS Requirements of Natur al Language Access Systems There are sev ral way of viewing the relation hi between natural langu age int rf s ( an KBMS. First, one can try to build NLI KBMS. Second , one may need KBMS to pl ement NL I. Finally a can build NLI (suppo rted by KBMS) to knowledg e-base d sy stems (suppo rted by KBMS), cf. Fig. 1. In this section, we con entrat e on th e sect* and th ird pro lems. A pi ce o sof i ca ll a NLI

if (a) s th i e o th e sy i c in a ra language, and (b the processing of the i nput (the generation of t e out put) is based kn dg e abo sy nt act , ma a ma a a natural languag Fi g. 1. This d finitio im ies on the on e h nd that NL sy st s e kno wl edge -ba ed appli ation whi ould be ne fit from KBMS. O the othe r ha nd, the by de finition not e xpe rt em becaus ery hu man knows at least p rt o one n tu ral languag without necessarily be ing a lang uage exp rt. ge ne ra pr em of ge tt c ut ers t u and natural lan guage is far fro b ing solved. Current NLI are a long way fro bro d-based univ

rs al languag cap bilities co arab le to hu man dialog partners. Mo st co mmercial NLI provide only da ta base ac ces but are insuffic t for cce ss to more co mplex knowledg e bases; for a su rv ey of su ch sy stems, see [WAHL86 .
Page 5
Though man datab se features (e.g., concurren and secu rity ) appear less critical in NLI than in trad ition l datab se app ation due to the sing le -u ser wo rkst at on approach usually taken by such sy stems, th e avail bili ty of KBMS cou d signi fi can tl extend the practi cal scop e for NL I. T e st im po rtant issue is th e integ ation of

ltiple div rse knowledge sou ces for wh ic h it seems infeasib le to find a co mmo knowledge representation For ex am e, KBMS for NLI h ve to integrate th e following kinds of kno wledge: Ling uistic kn owledge co ncerns a wide variet of aspects, rangin f om t unders tanding of s eech act s for c opera tive q Concept ual k nowled e t e o a o t s o reality relev nt to the dialog One of th e r asons why NLIs for d tab se ac ce ss h be en rel tiv ely succ es sful is that the conceptual knowledge d under yin a databa se is rather cl ear and well- delim ited. Inferenti l k owledge refer to a r le

base sim ilar to t at of ru le -ba expert stem s which works on the factual and c nceptual wle deri ve i plicit facts or ru les from st ored ones A use m del describ s the sy stem's knowledge about stereo ty pes of c u t N in a of backgr ou nd kn owledge, xperien ce, dialo stor y, or goals. As evid en ced by experime ntal sy stems such as HAM-ANS [HMM8 , th e knowledge b se structure of NL I can b eco me extr ely co mpl x. Figur e 2 su mma es so me of the port nt kno wl edg sou requi red for NLI sy st ems. In ter s of th e asp ct s in oduced in th e introdu ction, t e KBM req irement can be su

mmari d as follo ws : Knowledge representatio n: The required knowled e representation s hould be a hy brid o ny individual representations, such as fr es, d du O kn le units exp essed in o e of t ese lan guages m st be ag grega ed t t e ybrid k now ledge representatio n of one kno wledge sour ce (such as lin guistic, conceptual, e fur he r ag gre ate t a g oba ll y co nsis tent now led e base . It is im porta nt ha ve varie y i expression while a h d o c v t hybrid representation echa ism ca l Know le dge ord lexic n -morpholog ic al data -spel ling co rre tion data Sy ac tic Kno ledg e -p

hras al l exi co n na si g amm -ge er at io n gr am mar Sema nt ic K e -ca se- fra le xico n -te rmi olo cal / c onc ep al t -re ere al / as serti al ne Di sc ou rse Kno le dg e -us r mo de l -ju ti fi cat io n t / in fer ce me mor -te xt / dial m In feren ia K -infere ce r les -meta-inference rules -d ial g r les Fi g. 2. Knowledge organizatio n: There ar e two in knowledge bases i nvolved if NLI for knowledge- based sy stems are concern d. The NL KB can be organized by the scope of validity of knowledge,
Page 6
using a distinction of genera l backgroun d knowledge, task-speci fic

knowledge, and dialog-sp ecific knowledge (cf. Fig. 1). Additionall the KBMS ould also have to nage the background KB o the knowledge-based sy stem to which the NLI provi des access. Finally , th er e is a standarized dialog pr ocedure consisting of the anal is of a user request, of the evaluation of this reque st with respect to the knowledge-based application sy st , and of the generation of a, hopefully cooperative, answer. Environment and cou lin First, one n eeds an environm ent through w ba ses be is t as pec of knowle dge acquisition or learning which can be p rtially done through th e

NLI itsel f (o r an r NLI), an wh ich mu st b p rtially do ne b ling istic and domain experts in cooperation with co mputer scientists through a form al knowledge representation language [HAHN87 . Th e degree of coupling with ot her software is basically determ in ed by th background sy stem to whi h the NLI pr ovides acce ss . In particular, all of the syste s desc ribed in o s o t c t possib e ex cep tion of design KBMSs - hav been l NLI: da ta bases , expert stem s, a d ge om etrical sce ne de scriptions 4. KBMS Requirements for Visual Data i n mb o -b a , i in th ose in (robot) vision,

natural language understanding an d ge neration, p a s r easoning require a repre sentation vary ing scenes of spatial objects. Such a representation, called a et ri cal sc en e de sc rip (GSD), calls for sub tantial KBMS suppo rt not b d m s GSDs ar e i tan fo r sev ral rea . Th ey serv th e o ectiv e o com puter vision, "to know is e by looking," nd a sa tisf various ne eds in robotic s research. F rtherm ore, GSDs can serv as an rm ia te- el q at sc des b tw ee v n tural language understandin . This representati on is high-level enou gh to s esentation for a natu ral la ng uage te rface ,

ye p e t a voi d loss of information with respect to inf rm ation obtai ned t e vis al cha nnel. GS Ds ar e a q antit at ive d sc ript ion of v su al p ope rties of time-v arying scen es, including 4D object shap es, 4D o ject locat on s, illu min asp s, an t t a cep al k b Na ral language provides interesti ng space-ti me concepts ("events") which one m want to r N l m c l t i pri itives which express one of a li mited set of inte re ting qua lita tive p opositions bout a lim ite d set of obser vable scene p operties; for inst obse va bles i cl ude p si ti on , orie nta n, dis ance , nd vel

ty with resp ect to referen e po sition s, whe eas qua lita tive prope ties b s a " " easing," "decreasing," larger/s maller than reference," etc. Hence the requi re nts for event re trie va l may be defined largely without reference to a particular applica esented in the cha ter by Neum ann in this vo lum A final observation is that event pr im itives are "dur ative" pr opos itions bou a mp e ma c o s terrelated prim itiv es, the correspondin time interv al may be constrain d m a no n-t ial wa Hence ent recognition involves te mpor al c ons tr ai nt s Th e constrai nt sat sfaction para digm

is as fu ndam ntal to event recogni tion as the instantiati on pa radigm In term s of the four as pects ntione d in the intr oduct n, t e KBMS r su mmarized as follows: Knowledge representation: There is a need for a specialized knowledge representation of four- dimensional scenes. Additionally highe r-level concep ts such as collision, occlusion, m tion ty pes, and expectations m st be available. Reasoning capabilities cannot be just based on de duction rules as in rule-based expert s ste s but frequently involve the concept of constraint satisfaction, e.g., finding any a swer that satisfies

ill the tem poral constraints i posed by the sc ene description and the quer Know led e o ga niz on: Visual data require extre ely large, s are d storag e and retr ieva l erations sin ce ther e i o t i a r en vi ro en t. e n eed quantitati ve and qualitati ve descripti ns im poses different le vels of stract io a ong w ic h KBMS s ou ser ve t m diate vi ron ent and coup ling Th e GSD KBMS is o ten coupled with p
Page 7
video cam ra) on the low end, and with natural la nguage query a d comm and interface on the high end. In summ ary GSDs dem onstrate that the e are basic requirem nts

that are of ge ne ra l im portance for a w de class of a plic ations , are not met by current datab ase knowled e rep esentation techno logy will not b met by KBMSs if a unifo rm approac to all a cat io ns ta ke n. 5. KBMS Requirements of Design Environments A growin g cl ass of kn owl dge-based a pplicat io ns is no l er co ncerned wit just descri bin and classifyin g th e world but with help in g changi ng it . The activit of creating a nd m intaini ng artifacts is calle d desi gn [ IMO8 1] . Com puter-a ided desig ( AD) spans a wide ran of a ppl icatio ns, fr om ship bui ldi to VLSI an d

software d sign, t co puter aided plann models i busi ess. From early on, AI researchers have been inte rested in this area; m re recently data base researchers have also given it a great deal of attention. More t an in ot her kn owl dge-based applicat io ns, t e term "kn ledge" a ppe ars questi ona ble i design su pp t. The desig "kn wled base is a co llectio n of ev olvi ng partial stored belief about a "go s ste and its e nvir onm ent. These belie fs are represented as functional s ecifications designs , or n t e case of VLSI or software design s stem s) implem entations w ich may be contri

bu ted b differen peop le with possib e in onsistent views of the problem to be solved. Knowled e b se consistency an d d yna c belief ac quisition are therefore of critical im portance in design su pp t env ro nm ents. There are several differen t views of what a KBMS for design support s hould offer. Database resear her [KL84 tend to stress the aspect of com le x objects which evolve over tim under the cooperative influence of multiple designers. Sev ral KBMS req irem ents can be derived fro t is view: It st be possible to co mpose co le x configurati ons from si mpler objects in a flexible

manner which cannot be easily for eseen in a rigid database schem . Multiple equivalent or at least overlapping viewpoint and transform tions among t em st be supported. Version management st allow the tem poral development of partial to full specifications, the concurrent exploration o alternative designs , and the m int nance of designs, i.e., error corrections, adaptation to changing envir onm ents, and enhancem ents. Nested transaction concept are required to co ordinate th e cooperative and concurrent development of the design objects. These concepts define ranges of visibility for the

versions of a design object, for instance: public releas e, project, w rk group, in dividual desi gner. The also form the u its of recovery , in order to avoid unnecessary repe tition of design work in case of error . AI researchers view design as a sear ch process in a very large space of (partially spe ified) alternatives. To cont rol t is sear ch pro cess effectivel a KBMS for desig su pp ort sho not onl manage kn ledge ab ou t the desi gn states , as described by the m lti-ver ion com lex objects mentione d ab ove. In ad dit on , a process perspective becom s necessary whic h in vol ves

knowled e abou t [ OST85] : the goal structure of the design process. Knowledge about the desired functionality , perform ance, and critical factors reduces the searc space for "good designs." More importantly , explicit knowledge about t e design goals allo ws auto tic choice am ong design alternatives, especially i the case of maintenance where certai design deci sions have to be retracted and subsequently "replay d." the design de cisions and their rationales. Knowledge about the re asoning behi nd design decisions facilitates th e co mmunication between designers and inten nce personnel,

as well as the checking of consistency b tween existing and ne w design decisions. This knowledge reveals the
Page 8
assu tions and co mm it ments made by a designer; it al so deter ines proof obligations for im plem entations with respect to the corresponding sp ecifications. the control mechanisms that determ ine how the k ledge about design o jects, design g als, and the docum entation of design decisions are actually b ought to bear in designin well with li mited search effort. This include knowledge about a specific desi gn methodol ogy which enforces "procedural rationalit [

IM081] in the design proce ss. Thes e conc ts al so i pl a need for working wit incomplet specifi cations, possibly with an exception-h ndling m chanism to deal with overabstr action. If possible, design rules and structure should be organized in g neralization hierarchies which facilitate auto matic rule acquis on e p ( b o because design tas are so divers e that not all r les appl cable to a spe ific c be de ne d bef rehan . Ad di tiona ll y, desi KBMS are an expert suppo rt to ol wh ere th e users are o ten the b st ex perts. Fina lly, resea rchers th a comm unica tions pers pec tive

require design to m dia a a communica tio ns medium ong differe nt desi gners, s t s a a f to look at different asp ects of a desi gn i a co nsi tent m nner [ F86] . This perspective stresses the im portance of developi ng tually under r a i b i o words, the KBMS is vie ed as a knowledge sharing syste [J ARK86b t d d t p o d s differ widely. Ne vertheless, can be identified. I general, a design KBMS can be a a d entation knowle dge base attache to a problem solving e vironm ent ( st be able to m nage consistently large sets of design objects, m st of the incom te and developed fr om particular

viewp rm or e, transformatio n and back tracking operations st contr ed in this com x objec envi ronm ent, o i i d often over extende d a to ol to enh ce docu en tation, und erstan dab lity o for designers and end users ( overing issu es such as requirem nts validation, explanation, and verification of transforma tion correc ness), and de bugging and odification. In term s of the four areas define d in the introduct ion, the requirem ents a K c b summarized as follows: Knowledge representation: Design KB MS need a r presentation not only for co lex design objects but also for knowl dge about

the design pr ocess and its underl ng rati onales (e.g., in the form of rules and justifications). Knowledge representation m st be visualized at lea t exte rnally such that the design knowledge base serves a a tually understandable documentation sy stem within and be yond the dev lopment gro up. Knowledg e organization Ty pical ruc ur es in clud e con gurat ions, views d versions which can so serve as a sta ting point for learning. Usually , there are only fe w ba objec ts (e.g., one larg e software sy stem, or one aircraft wing to be developed) wh ich, howev er, have a ry co lex organiza n.

t c mp x relationsh ps betw ) , kn owle sho als b orga nize d i a teleol ical way, should be justified in terms of the sig goal stru ctu e and constraints. ve u ef knowledge organization structure is that of a b lief ma int Envi ron en Que ng fa cil ti es mu st in clud e inf at ion abo t the d gn history a w ll as about different views (a bstraction , representatio s) th e d tabase. Con istency ch ecking mu st b po ssible acro ss d ffer ent rep esen tation , and st in cl ude th e po ssibility that d sign portion s mad by differen t design ers can b tem porarily inconsistent. Th is in consisten

mu st be gradually resolved using nested tran sact io n concepts which or nize formal group cooperation. Coupling Ideally it should be possible to couple a design KBMS with th e usag e env ro ent. In this case, the end u er can qu ery the design knowledge b se to get an und erstanding of wh at the d gned sy ste do and why it d es it lik e th at. Ho wev r, this i curr tly only done w th expe rt sy st she ( d gn obj ec ts be ing xpert sy st s) wh er e the sh ell r ma s as a usag e envi ro ent as a velo ent enviro ent. The chapter by Borgida et al. provides an exam ple of this kind of K MS.

Page 9
Fi g. 3. 6. Conclusion The purpose of this chapter was to provide a cust er vi ew o KBMS: what KB MS tools do knowledge-b ased applications really need? Our revi ew of four i port nt KBMS applica tion areas h s shown so me co mmonalities but also significant differences. Common to all applications is the need fo r handling large amounts o factual data as provided by databases. In so me cases, su ch as GSD and design KBMS, the stru cture of the data appears so co mplex that sp eci aliz ed DBMS with cap bilities for h ndling co lex object s are need ed. In the case of design

KBMS, transacti on concepts st also be extended. A second comm on requirement appears to be that metaknowledge about the av ailable information is needed as a dictionary to integrate knowledge from different sources. This feature was particularly visible in the case of KBMS for NLI but also in desig KBMS. The differences lie in the kinds and specific representati ons of knowledge dictated by the need of effi cien cy in different do mains. As can also be seen in earl er sectio ns of this boo k, ny capabilities o si rule-b as ed expert sy ste s can be e dded into extended DBMS which provide lim

ited deductive capabilities in an effi ci en t, set-orient ed manner. On the other hand, the re advanced knowledge-b ased applications require either very dedicated data struct ures not provided by DBMS (geo metrical scen e descriptions are a good exam ple), or the special purpose integration of a large num r different sources of knowledge in diffe rent special-purpose represen tations. One feature o growing importance in th is cont ext is t e representat on of time . I GSD, w d ribe ti -vary ng spatial objects and relationships. In design KBMS, we describe a history of our knowledge and

conceptu al-work on an artifact which in itself may contai n a tem oral model. Tim also play s an im rtant role in natural langu age. Thus, th e different uses of time, ea ch requiring s eci aliz ed knowledge representation nd (expen ive) reasoning approach es su ch as constraint satisfaction dependent: tracing, fu zzy reasoning, etc. may be on e of th e main fa ctors that prevent a mo re gen KBMS appro ach. In su mmary generalized KBMS for advanced know ledge-b ased app Ei a K h t i to supporting efficiently very sim le knowledge re pres ta tions (a t m ductive databases), or it st e d

application-area-spe ific k now I e in it s a chit tur to prov ide ef fi ci ent re ason ng wi th li mit d re 7. Ref rences [BJ86b Bolc, L., and M. Jark e (ed .), Coopera tive Interfaces to In forma tion S stems , Spring er -Verl g, Berlin, Heid elberg, 1986 [BL86 B , R J. , an d H. J. L sq ue, Wh t Makes a Knowledge Base Kn owl dge abl A Vi of Datab ses fro m th e Kno ledge Le vel, in KERS86 , 1986, pp. 69 -78.
Page 10
[BM86 Bro ie, M.L., and J. My lopoulo (ed .), On Knowledge Ba se M anag em en t Systems, In tegra ting Artificial Intelligen ce and Database Tech nologies , Spri

nger- Ver ag, Berlin , Heid elberg, May 19 86. [FM86 Fox, M.S., and J. McDermott, Th Role of Databases in Kno le e B sed Sy st , i [BM86 198 6, pp. 407-43 0. [HAHN87 Hahn , U., Modelling Text Und rstanding: The Methodo logical Asp cts o Auto matic Acqu isition o Knowledg e Through Tex Analy is, Proc. 1st international Symposiu m on Artificial Intellig en ce a nd Exp rt S stems , Berlin, 1987 [HMM86 Ho eppner, W., K. Morik, an d H. Marbu r, Talking It Ov er: Th e Natural Language Dialo Sy stem HAM-ANS, in [BJ86b 1986, p . 189-258. [HW 83 Hay s-Ro th, F., D. A. terman, and D. B. Len t (ed

.), Building Exp rt Systems , Addison- Wesley , Read ing, MA, 198 3 [JARK86b Jark, M., Kn wledge Sharing and Nego tiation Suppo rt in Multip erson Decision Support Sy stems, Decision Suppo rt Systems , Vol. 2, No . 1, 19 86, pp 93-102 . [JV84 Jark e, M., and Y. Vassiliou, Cou pling Exp rt Sy stems with Datab se Management Sy stems, in [REIT84b 1 984, pp. 65-8 . [KE S86 Ker chb rg, L. (ed ), per D se Sys , see [KERS84 , t Intern at ional Workshop on Expert Database Sy stems, Ben amin /Cu mmings Publish r Co mpany Inc., Menlo Park , CA, 1986 [KERS84 Kerschberg, L., (ed.), Pr oc. 1st Intern

ational Workshop on E xpert Datab se Sy stems, Kiawah Island, South Carolin a, October 1984. [KL84 R.H., and T. Leh an , at aba e Supp ort for Versio ns and Alte rn ativ es of Larg e Design Fil s, IEEE Tran saction on Softwa e Engineering , Vol. SE-10, No . 2, pp. 191-2 00. [K OW A86 Kow k, J.S (ed. ), Couplin g Symbolic an d Nu merical Compu ting in Exp rt Systems , North Holland, Ams erd m, 1986 OS 5] ow J. ow ar r Models of the Desig Process, Art ifi cial Int llig en aga zin , Vo l. 6, No. 1, 19 85, pp. 44-57 [MYLO86 My lopoulos, J., On Kno ledg e Base Ma nag ment Sy stems, in [BM86a]

pp. 3-8. [NAU83 Nau, D.S., Exp ert Co mput er Sy ste IEEE Co mputer , Vol. 16, No. 2, 1983, pp. 63-85. [PAU86 Pau, L.F. (ed.), Artificial Intelligen ce in Econo mics and anag ement , North Hol and, Amsterdam, 1 986 [REIT84b Reitman, W. (ed.), tif icia l Int ll igen ce Appli ation for Bu sin ss , Ablex, Norwood, NJ, 1984 [SIM O81 Si mon, H. A., e S nc e of the A ifi cial , 2nd ed., MIT Press, Camb ridge, MA, 198 1 [SMW B85 ith , R. G., T M. Mi tch ll , P.H. Win ton and B.G. Buchan an, Represen ation of Use of Expli it Justification for K nowledge Base Refinement, Proc. 9th International

Joint Conferen ce o Art ifi cial Int ligen ce , Lo s Angeles, CA, 1985, pp. 673 -680. [SZOL86 Szolovits, P., Knowledge Based Sy stems: A Su rvey , in M86 19 86, pp.339-35 2. [VCJ83 Vassilou, Y., J. Cl ifford , and M. Jark e, ow D es a Expert Sys em G t its Da ta ? , Proc. 9th Conferen ce on Very La rge Databa ses , Floren ce, Italy 1983, pp. 70-72. [WAHL84 Wahlster, W., C ooperativ e Access Sy stems, Futu re Gen ration Computing S st ems , V l. 1, No. 2 198 4, pp. 103-11 1. [WAHL86] Wahlste , W., he Role of Natura l La nguage in A dvanced K now ledge -Based System s, Repo rt No . 6, erf

rsc un gs bereic h 3 . des Saarla ndes Sa arbr c ken FR G, 19 . [WF86] Winograd, T., and F. Flores, Unders tanding Com uters and Cogn ition: A New Foundation of Design Ablex, N rw ood, N , 1986. 10