Transportable Natural Language Interfaces f or
T
axonomic Knowl edge Representati on System
s
Lutz Prechelt, Finn Dag Bu, Rol f Adams
(prechel t jnndagjadams@i ra.uka. de)
Insti tut f ur Programmstrukturen und Datenorgani sati on
Uni versi tat Karl sruhe, W{7500 Karl sruhe, Germany
Appeare d in the pr oce e di ngs of t he
Ni nt h IEEE Conf e r e nc e on
Ar t i c i al I nt e l l i ge nc e f or Appl i c at i ons
Or l ando, Fl or i da, Mar ch 1-5, 1993
1
Introduction
The r e have be e n s e ve r al at t e mpt s t o c r e at e nat ur al
l anguage i nt e r f ac e s f or dat abas e s [15, 10, 5]
of t he s e nat ur al l anguage i nt e r f ac e s t
t r ans por t abl e , t hat i s , t he y
domai ns by changi ng
de pe nde nt knowl e
A general approach is presented for bui l di ng transpornal dat ab
t abl e nat ural l anguage i nterf aces f or questi on answee
ri ng syst ems based on a KL-ONE-l i ke knowl edge represent at i on. An exampl e system, YAKR, i s descri bed: The
YAKS knowl edge representati on of concepts and rel at i ons i s annot ated wi th mi ni mal syntacti c i nf ormati on
t o generat e a semanti c case f rame grammar wi th i nheri tance of cases. The generated grammar di rects a case
f rame parser, whi ch processes wri tten i nput i nto i nst ant i at ed case f rames. These i nstanti ati ati ons are easi l y t ransl at ed i nto knowl edge base queri es. The same
met hod i s appl i cabl e to other object-ori ented knowl edge
bases and ot her parsi ng techni ques.
The ori gi nal
cont ri but i on of thi s work i s to show an approach wi th
whi ch nat ural l anguage i nterf aces can wi th l oweort be
adapt ed t o work wi th any new knowl edge base: Whi l e
most ot her syst ems requi re a compl ete model of the domai n f or t he nat ural l anguage i nterf ace knowl edge represent at i on, we deri ve most of thi s i nf ormati on f rom
t he appl i cat i on's knowl edge base. Thi s techni que reduces t he amount of work needed to create the i nterf ace
t o about addi t i onal 15 percent af ter bui l di ng the knowl edge base f or the appl i cati on kernel .
AI topic: knowledge acquisition, knowledge representation, natural language interface
Language/Tool: Unix, C++
Status: implementation complete, evaluation in progress
Eort: about 4 person years
Impact: quick development of restricted natural language
interface for certain classes of knowledge-based applications (15 % additional knowledge acquisition work).
e xpr e s s f ac t s ) and as s e r t i onal knowl e dge ( f ac t s about
i ndi vi dual s (i nstances) i n t he appl i c at i on domai n) .
The t e r mi nol ogi c al knowl e dge c ons i s t s of c onc e pt de ni t i ons and r ol e de ni t i ons . A concept c an be t hought
of as an abs t r ac t s e t of i ndi vi dual s . The c onc r e t e i ndi vi dual s t hat be l ong t o a c onc e pt ar e c al l e d t he i nstances of t hat c onc e pt . A rol e i s a bi nar y r e l at i on
f r oma c onc e pt A t o a c onc e pt B , i . e . , a s e t of
of i ns t anc e s . A i s c al l e d t he domai n of t h
i s c al l e d t he range of t he r ol e . C
de ne d wi t h c ons t r uc t or
of t he s e t of al l
r es
and s mal l , but s uc e t o c ons t r uc t al l i nf or mat i on t he
nat ur al l anguage i nt e r f ac e ne e ds . I n t hi s s e c t i on we
de s c r i be t he t ype of i nf or mat i on t he annot at i ons c ont ai n and how i t i s us e d i n YAKR t o ge ne r at e t he c as e
f r ame s . Se e [ 6, 14] f or a de t ai l e d de s c r i pt i on.
The r e ar e t wo mai n t ype s of i nf or mat i on pr e s e nt i n t he
annot at i ons : ( 1) i nf or mat i on about i ndi vi dual wor ds
and ( 2) i nf or mat i on about gr ammat i c al c ons t r u
The wor d i nf or mat i on as s oc i at e s e ach
wi t h i t s nat ur al l anguage s yno
s i mpl e phr as e s t hat r e pr
var i e t y of t he
c onc e p
Onl y t he wor ds Zu g r i and L e s e n ne e d t o be i n t he di c t i onar y, t he c ompound i s al gor i t hmi c al l y br oke n i nt o
t he s e c ompone nt s .
4. 2
Gr a mma t i ca l Cons t r uc t i on Annot a t i ons
Al l t he above annot at i ons me r e l y de s c r i be phr as e s t hat
r e pr e s e nt i ndi vi dual c onc e pt s ; no c as e f r ame s ar e bui l t
f r omt he m. The s our c e of c as e s f or t he c as e f r ame s ar e
t he r ol e s . Thi s i s whe r e t he i nf or mat i on about gr a
mat i c al c ons t r uc t i ons i s us e d, whi ch t e l l s
t o i ns e r t whe r e . Si mi l ar Annot at
t o ge ne r at e t he i nf or mat i o
ot he r nat ur al l ang
As an e xampl
an ac t
t hat has t o be us e d i n a que r y i f t he c as e c ont ai ni ng i t
has be e n l l e d. Mos t of t he i ns t ant i at i ons c an s i mpl y
be pr oc e s s e d i n a t op- down manne r , onl y t he i ns t ant i at i ons f or c e r t ai n gr ammat i c al c ons t r uc t i ons r e qui r e
s ome mor e c ompl i c at e d pr oc e s s i ng. I n any c as e , t hi s
pr oc e s s i ng i s pur e l y me chani c al . No f ur t he r s e mant i c
pr oc e s s i ng i s ne c e s s ar y. Mos t ambi gui t i e s t hat r e mai
af t e r par s i ng ne e d not be e xpl i c i t l y r e s ol ve
s i nc e t he wr ong i nt e r pr e t at i ons r e
r e t ur n no ans we r at al l .
Whe n a par s i ng t e chni
s i ng s hal l be u
ge t s
pr oach, t he l e xi c on c an be r e duc e d t o a di c t i onar y ( wi t hout s e mant i c i nf or mat i on) ; t he s e mant i c i nf or mat i on
c an be de r i ve d vi a t he annot at i ons . The c onc e pt ual
s che ma c ons i s t s of s or t i nf or mat i on and c ons t r ai nt s on
t he ar gume nt s of nons or t pr e di c at e s . Wi t h our appr oach, no s uch s che ma i s ne c e s s ar y at al l ; t he i nf or mat i on c an c ompl e t e l y be de duc e d f r omt he knowl e dge
bas e i t s e l f . The dat abas e s che ma c ons i s t s of i nf o
t i on t hat e nabl e s t he mappi ng of t he i nt e
pr e s e nt at i on t o a que r y e xpr e s s e d
l anguage . Wi t h our appr
mappi ng c an be d
de r i ve d f
on
T
[ 9] Ke i t h E. Gor l e n, Sanf or d M. Or l ow, and Pe r r y S.
Pl e xi c o. Data Abstracti on and Obj ect-Ori ented
Programmi ng i n C++. Wi l e y, Chi che s t e r , 1991.
[ 10] B. Gr os z , D. Appe l t , P. Mar t i n, and F. Pe r e i r a.
TEAM: An e xpe r i me nt i n t he de s i gn of t r ans por t abl e nat ur al l anguage i nt e r f ac e s . Arti ci al Intel l i gence, 32( 1987) : 173{243, 1987.
[ 11] I EEE Comput e r Soc i e t y. The Seventh Conf erence
on Arti ci al Intel l i gence Appl i cati ons , Mi ami Be ach, Fl or i da, Fe br uar y 1991. I EEE Comput e r Soc i e t y Pr e s s .
[ 12] Mat t hi as Ot t .
Mode l l i e r ung von UNI XKommandos mi t YAKR. St udi e nar be i t , Uni ve r s i t a t Kar l s r uhe , I ns t i t ut f ur Pr ogr amms t r ukt ur
und Dat e nor gani s at i on, D- 7500 Kar l s r u
many, Nove mbe r 1992.
[ 13] Lut z Pr e che l t . Ei n Fal l s chab
De ut s ch. Mas t e r ' s t he
s t r ukt ur e n u
Kar l s r
[ 14]