Interlingual Lexical Organisation for Multilingual Lexical Databases 
in NADIA 
Gilles Sdrasset 
Gi\]le~.Seraeset@imag. fr 
GETA, IMAG-campus (UJF & CNRS) 
BP 53, 38041 Grenoble Codex 9, France 
Abstract 
We propose a lexical organisation fi)r mullilingual 
lcxical databases (MLDB). This organisation is based on 
acceptions (word-senses). We detail this lexical 
organisation m~d show a mock-up built to experiment with 
it. We also present our current work in defining and 
prototyping a specialised system for the manage,ncnt of 
acception-based M LDB. 
Keywords: mnltilingual lexical database, acccplion, 
linguistic structure. 
Introduction 
Needs for large scale lexical resources for Natural 
Language Processing (NLP) in general and for Machine 
Translation (MT) in p:uticular, increase every day. These 
resources ,are considered to ::epresent the most expensive 
part of ahnost any NLP system. Ilcnce, an increasing 
interest in the development of reusable dictionaries can be 
observed. 
q'o develop a Multilingual Lexical Database (MLDB), 
we think of two main approaches. First, the tran.~'r 
approach where the links between the languages are 
realized via unidirectional bilingual dictionaries. This 
approach is used by many MT systems and by some 
lexical database projects (notably Acquilex or Multilex). 
Second, the interlingttal approach where the links 
between the languages arc realized via an nnique 
interlingual dictionary. The KBMT-89 project 
(Knowledge Based Machine Translation) at Carnegie 
Mellon University in the US and the EI)R. (Electronic 
Dictionary Research) project in Japan use tiffs approach. 
In the context of multilingnal MT systems, we arc 
interested in the problems posed when constructing "rod 
using an application and theory independent MLDB. Wc 
are developing a Lexical l)atabase Management SyslcIn, 
NADIA, based on an inlerlingual approach. Wc chose 
acceptions as inlerlingual milts. NADIA provides re:my 
tools for the management of MLI)Bs. Moreover, this 
system gives the linguist a great liberty in tbc choice of the 
linguistic structures. 
We first give ,an overview of the project, beginning with 
its lexical organization. Then, we give the results of our 
experimentations on this lexic,'d organization. Finaly, we 
present our current work: the definition and prolotyping of 
a specialized system for the management of acception- 
based MLDBs. 
NADIA is the continuation of a work done for the 
Multilex ESPRIT project. The coherence checkcr and 
software architecture have been defined for Multilex and 
adapted to our lexical orga,fizatinn. 
I. Acception-basod lexical organization 
After studying and comparing different projecls of 
lexic:d dat;dmses, including I"J)R (E\[31~. 1993), KBMT-89 
(Nirenburg and 1)cfrise 1990; Good,nan and Nirenburg 
1991) Multilex and of Mullilingual MT syslems, such as 
CICC (Uchida and Zhu 1991) and ULTRA (Farwcll, 
,guthrie ctal. 1992), wc lmve concluded in favor of an 
iutcrlingual lexical organization for our MLDBs. 
Some of the inter,rational projects of lexical databases 
are based on a multbbilingual approach (e.g. Multilex) 
while others use knowledge representation as an 
interlingua (e.g. KBMT-89 or EI)P,). Much like ULTRA, 
our approach is interlingual and linguislic rather than 
knowledge-based. 
1. The dictionaries 
A MLI)B consists of two kinds of diclionarics: lhc 
monolingual dictionaries and the acception dictionary. 
1,1, Monoli~ dielionarics 
The monolingual dictionaries arc accessible by entries. 
These entries are le,mnas ("normal form" of words, e.g. in 
Snglish, infinilivc lk~r verbs, singt, lar for nouns, etc.). 
Items of the monolingual dictionaries (monolingual 
acceptions) are generally accepted meanings of words or 
expressions, as wc can find them in standard printed 
dictiona,'ics. These monolingnal acccplions arc combined 
with their linguistic inlbrnmtion. 
Monolingual acceptious of a language L arc acccptions 
that are connected to a word or an expression of L. Such 
an acccption can be accessed from one (or mo,'e) entries. 
1.2, Ace.option dicdonm2~ 
The interlingual diclio,mry, called acc(v~tion ¢lictionaty, 
contains interlingual acctT)timzs. Some inlkwnnalion c\[in be 
linked to these intcrlingu:fl acceptions. 
In a MLI)B composed of n monolingual dictiouaries, the 
set of intcrlingual acccplions is equal Io the uuinn of the 
sets of monolingual acceptions of the n dictionaries, with 
an equality relation bound to the semantic identity. 
Some contrastivc problems may appear when two 
monolingual acceptious of two different languages are 
semantically slightly different. This appears when there is 
a non-direct translation of a word (e.g. 'river' can be 
translated in French by 'rivibre' or by Tleuvc 'l ). This kind 
of problem is solved by a relation from acceplion to sub- 
acception which is prc-defincd in all NADIA lcxical 
databases: the contrastivc relation. It is intended to code 
contras|ive problems induced by a non-direct translation, it 
1 A 'rivibre' is a ralher small river fl(~wlng into annlher river. A 'fleuve' 
is a large river flowing into the ,see. 
278 
/'7"'\ / ....... .':,,,o\ 
Fig. |: illustration of the acception-based lexical organization 
iS not inlended lo code any kind of ontological 
itffonnalion. 
2. I,exical organizati(m 
In the acception-based lexical organization, the 
monolingual acceptions and the interlingual acccplions 
must satisfy the lbllowing criteria: 
2,1. Well-formcdness crileria 
, Each interIingual acctTtion correspomls to at &ast 
one monoIingual acception. This criterion slates 
that an interlingual acccption must correspond to at 
least one entry of one language (as ntonolingual 
acceptions). 
• An interliagual acception corresponds to at most 
one monolingual acception of the same language. 
An interlingual aeception is not necess'~rily 
connected to a mcmolingual acccption of each 
language of the MLDB. 
• A monolingual acception corresponds to one and 
only one interlingual acceptiom A monolingual 
acception is always related to an intcrliugual 
acception and (as stated by the preceding criterion) 
is one-one. 
2.2. Translation criteria 
• Two monolingual acceptions ofdiffi~rent languages 
correspond to a unique interlingual acception if, 
and only if they have the same meaning. This 
criterion sta~es the semantic identity of two 
monolingttal acceplions of different langtmgcs 
(provided that they correspond to the same 
interlingual acception) allowing the use of the 
interlingual dictionary for lexieal translation 
purposes. 
• If entry el of language TA is translated by entry e2 
in language L2 via a non-dh'ect equivalence, the 
corresponding interlingual acception must be linked 
by the conm~stive relation or by a relalion of quasi- 
synonymy. This criterion allows the use of the 
acception dictionary for lexical translalion purposes 
even when there is no direct translation. 
II. Experimentation 
l. The Parax mock-up 
In order to experiment this lexical organizatio,t, 1;:. Blanc 
has built the Parax mock-up (Sdrasset and Blanc 1993). 
This mock-np is a small acceplion-based lexical database 
of 5 languages (Fmglish, French, German, I~.ttssian, 
Chinese). 
Parax, produced on Macinlosh with IlypcrCard TM, was 
designed to experiment prol~letns inherent to the 
acception-based lexieal organization, llence, items of the 
monoliugual diclionaries are combined with rather simple 
linguistic information. 
An enlry of a monolingual dictiotmry is linked Io several 
acceplions. These acceptions arc provided with their 
linguistic information (lcfl c(flumn in fig. 2). l:.ach of these 
monolingual acceptions is related to an interlingual 
acccption along with its (lefinition (in French) and s0me 
scmanlic infonnation (right eoltnnn in fig. 2). 
....... ft~n~ai~ ~- 
~ I) \[iu.~P.r 
SENS: I, 
~l~¢r ~ ~pou~erJor rf~e 
St NS:2. 
SENS:~ 
~po~er-~¢mlrier$ prer~ir~ pour 
~pou×, ~pou#e, #e marler eveo t'2# 
"1~ pOU~It r_fornle $'ad6ptef 
exactemeld ~ Juice ferric, un 
mouvemer~t\] I'}'o~ qu/ ~¢ ?es 
/orals du corp,~,1 
~.pou=er_ld~l ~'~tt~cher d~ propo# 
d~lll~rk et ~vec erdeur h qqeh 
qvu~3 
Fig. 2: Monolingual entry: "dlXmSer" (to marry, to fit, to espouse) 
To accede to the acceplion dictionary, Ihe user selects an 
acception in the middle column. The acception is 
displayed ahmg with its sul~-aceeptions (middle column of 
fig. 3). From Ihis point, it is possible Io gel a translation by 
selecting a target langtmge for one of the acceptions. The 
lranslation appears in Ihe right eohnnn of fig. 3 (which 
shows the German translation of the acception). In Ihe 
acception~ 
SOURCE: franCois #~pouser_semnrier$ CIBLE: nllemnnd I affbmenus 
~pou~er 1 "*~p0u3erJemurier$ 
SENS:I. 
~.,~.t~o.q.~.~,r.~ ~..m.~.r.i.e.r.$...L&c.,..~.&N..LE g ............ 
prendre pour 6p.0ux, ~p0uae~ 3e marier avec 
,.(Ze e~:~.~..~.~e2...,'..~ ........................ 
............... ~.~. ~.u..~.~.r.. ~ ~..,. ~,.~.r:i. ~.r.. l~.h.~..m.....m..~..* R.u.... 
................ ~...m..~.r.i~.r....~.~.r...~.¢.b.o.mm~.L..L': ................. 
............... ~.~. ~.u.~.e.r ~.e..!~.r.i.~r.l~.m..m. ~..~.RU...,. 
.~r...~.~.d.~.L.{ ~.~,~,.,.q..qe,..f~.~.m.~Z...t.~ ................................ 
................ ~.Et~ .u ~.~.r~ ~.m.<r. j...e,.r.l..r...e.t~.¢ ~ A..X.. ........ 
• IF -i" 
hei raien a"E pouae r.,Je m~rie r $ 
SENS:I. 
Fig. 3: Acception: #dpouser scmaricr (to roan'y) and it's sub-acceptions. 
279 
example given, there is no direct equiv:dence from l:rench 
to Russian as Russian introduces a distinction on the 
gender of the subject. To get the Russian translation, we 
have to select one of the sub-acceptions. Then, we can get 
'>getlrrrrr,¢>l ' for fl man or '3aMyx'-: ', 'aaMym (gblflTI| - as)' 
or ~3,t~MyIK(~M '* for a woman. 
2. Indexing methodology 
2.1. Indexing in Parax 
As the platform we used for this mock-up was not 
specialized for such a task, we have used an indexing 
methodology lbr the construction of this MLI)B. 
The starting point of onr work was a smaU French corpus 
we wanted to index, llence, we begau to index French 
words and for each created acceptions, we gave a 
translation in the other languages. 
After creating an entry, the lexicographer gives its 
different word-senses ,and their linked linguistic informa- 
tion (the kind of information depends on the language of 
the entry). 
Then, the lexicographer links the word senses to an 
interlingual acception. As lhe number of acceplions is still 
small, it is possible to select an ,already existing aeception 
by browsing directly in the acception dictionary. If the 
searched aceeption does not exist, it is created along with a 
definition in French and some semantic inlk)rmation. 
2.2. General c, ~'ls.~ 
When developing a lmge scale MLI)B, it is no longer 
possible to select existing interlingual acceptions by 
directly browsing through the acceptiou dictionary. 
Moreover, the different dictionaries will have to be 
iudexed by different lexicographers. 11encc, it is necessary 
to define another methodology. 
The process of creation of an entry and its monolingual 
acceptions does not change. AftEr creating an enlry, Ihe 
lexicographer selects a possible translation for the 
considered acception in a language of the database. If this 
translation is already indexed in the target language, he 
selects the corresponding acception in the target 
dictionary. The source and target monolingual aeceptions 
are automatically linked to the same inlerlingual 
acception. If the translation is not ah'eady indexed in the 
target language, the lexicographer indexes it (partially) 
and asks the person in charge of the target dictionary to 
complete the new entry. 
The acception dictionary is thus constructed and 
managed by the system and the lexicographers work in 
more or less the s,-une way as when indexing bilingual 
dictionaries. This automatic management of the 
interlingual dictionary involves the automatic verification 
of the criteria defined abxwe. 
When a problem is detected the system attaches a 
w,-u-ning for the lexicographer in charge of the acception 
dictionary, m~d proposes a default solution. 
3. Some results 
The corpus we wanted to index in the Parax mock-up 
consisted of 135 entries in French corresponding to a 
representative set of verbs, nouns, adjectives, and adverbs 
of gener:d vocabulary. We have indexed these entries and 
the related aeceptions. As we sutrted the indexation with q 
French corlms, only some of the entries in the other 
languages have been given all their acceptions. 
The distribution of the entries and aeceplions of the 
different languages is the lollowiag: 
l~ntries Acccptions 
\[ Frencl~ 135 484 
~ 304 484 
l German 388 509 
\[~m 394 545 
This represents a total of 589 interlingual acccptions. 
Among these intcrlingual aeceptions, 58 are sub- 
acceptious introduced by contr,'t~tive problen~s. The size of 
this mock-up is of the same order as that of Multilcx. 
III. Current work 
Our current work consists in the dcfiniti(m and 
prototyping of a specialized management system for 
acceplion based MLI)FIs. 
1. Related projecls 
Some internalional projects have already started the 
development of a system for MLl)lis. We have studied 
and we nse some of their reStllis. 
In Fmropc, we have participated in the Mullilcx project 
(CFC - I)G XIII - EsPRrI" project) which aims at the 
definition of standards for lexieal databases systems. We 
use some o1' its rcstzlts (e.g. the software architecture, some 
of the tools). 
Multilex's software architecture, based on three layers 
(presentation level, internal level and database level), 
clearly separates the presentation from the coding and the 
coding from the storage of the information. This 
organization allows to change Ihe presentation of the 
structures (giving the possibility to define user interfaces 
hiding the internal structure). 
We have also studied the Jap:mesc I:I)R project which 
has developed large dictionaries of about 300,000 words in 
bofll English and Japanese (200,000 of general vocalml:u-y, 
100,000 of terminological vocabulary). FDR has also 
developed diction:uies of 4(X),000 concepts, dictionaries of 
300,000 co-occun'cnces (bolh in F.nglish and Japanese) 
and dictionaries of 300,000 bilingual culries (both for 
J,'qmucsc-l:.nglish and l';nglish-Japauesc) (I';I)R 1993). 
In EDIt., illdividtml concepts arc introduced in the word 
dictionary and correspond to the word senses, llence, our 
acceptions are really close 1o their concepts, l lowever, 
they do not use a contrastivc relation t() code problems 
between the languages. 
The CICC (Center of International Cooperation for 
Computerization, Japan) has also used a very close 
organization to construct a MLI)B (Japanese, Chincse, 
Thai, Malay, and Indonesian) for its Multilingu:d Machine 
Translation system. This lexical database coulains 50,000 
wonls or terms (Uchida and Zhu 1991). 
2. Toward a specialized management system 
A specialized management system for acception-based 
lexical databases must offer ways to automatisc the 
management of the aceeption dictionary. It must also offer 
tools to define, index and manage the monolingual 
dictionaries. 
280 
The NAI)IA system has to detect potential errHrs in the 
acceptiHn database. Fach pHtenlial Error is giVEn IH a 
lexicographer who is in charge of the correctio,. This 
detection is independent of the linguistic structure of the 
monolingual dictkmary. It consists in tile detection of 
geHmclrie inconsistencies in the relations between the 
elements Hf the database (entries, nlonolingtlal acceptions, 
inleflingtull acceplions). 
"llm NAI)IA system also provides tHOIS to help the users 
define, index, and manage a MI,I)B. 
These tHHIS depend on the linguistic slructure Hf Ihe 
different dictionaries, Ilence, a lingnist has to declare the 
slructure of the mlicles of the dictkmaries via a specialized 
la,tguage. To encode the linguistic informatkm, the 
linguist can use p,'edefined basic data structures (strings. 
lists, sets, trues, graphs, autHmata or Typed Feature 
S It'llcttll'eS). 
Several tools have been defined to help tlm users: 
. l,',ditor: lhis tool provides a default interface to edit 
items Hf a dictionary. It is alSH ixmsible to customize 
the interface -- this tHol is at sh'uctured Edilor h la 
GRIF (Andr6, Furttla et ill. 1989). 
Browser: this tool gives ways to browse through Ihe 
database. 
• Colterence checker: the linguist may defi,e some 
coherence all(l integrity rules that apply on an article, 
on a dictionary Hr on the whole lexical database. 
These rules are checked alld file result lfepet~ds on 
the strength of the nile. 
l)iq'aulter: the linguist may also define rules IH 
default entries HI a dictionary. These rules can be 
applied in batch mode (in order to expand an existing 
dictionmy) Hr in interactive mode (to hel l} the 
lexicographer in the indexing proCESS). 
• lnq)ort/eaport: the linguist m:ty write importing and 
exporting prHcedures from the intermtl structure to an 
external Rmnat based on the SGMI, langLUlgc and 
TEl guidelines. 
3. I)cfinition of the lixngtnistic slructures anti 
cohe,'ence checking 
As an example of the use tlf tile NA1)IA matmgenlcnt 
system for acceptiHn-based MI,I)B, WE give the definition 
of the linguistic struclu,'e used ixl Ihe Parax mock-.up (sue 
above). Then, we give sonic constraints that can be 
defined on this database. 
3,1, l)¢\[htition of l,ing!!i~'~ I~I,S~ 
The linguistic structttre used in Pa,ax is inspired by the 
structures of the dictionaries of GI'TI'A's ARiANE system. 
It is at flat list Hf attrilmte-value pairs. 
3.1.1. An example: l'arax "I)I.S" 
We give here a LISP form Hf the defixfition of the 
slructures. Oflter dialects will be defined in order to hide 
lhis LISP Rmn to the linguist (see below). 
lAJinition of the database 
BefHre defining the structures of :t dictionary, the 
linguist has to define lhe database. This definition consists 
Hf a dechtratiHn of the diction:wies cHmainEd in il (hErE, a 
database called "Parax" with 4 monHlingual diclionaries). 
FHr each dictionary, the li.guist cnlcrs its namE, its 
language, its owner, all optional comment and tile classes 
used to CH(Io its Entries and acccptiHns. 
(de f ine-da taba.~,a Parax 
: owtier "GETA" 
:comment "rfh\[:3 datab+lse Js the same as t;hc~ Parax 
illo¢.:k-ilp defined by Eti(?llt\]o 1{\]~111(2 wi_Lh \]lyper<:;lI:{\]." 
:dictionaries 
( do t \[ rlo- di ct i onal y l?Kollch 
: lalitglaOe "FLatl¢&\[!;" 
: o'dIlei "GETA" 
: (hiLly ' Frellch-etlt l;y 
: acception ' l?1 c~{Ich ~t(?ttop\[; toil) 
(define--di ctJ onary Vng I ish 
: I~lIlOuafJe ~I.\]nlll \[sh" 
I OWtle V "G}",TA" 
:entry 'l,~tl(i\]ish etltvy 
:ilcception 'l'\]nol ish-acCelJtJon) 
+,.) 
Structures of the Frem:h dictiom, y 
The linguist defines lhe linguistic structures of the 
dictionaries with an Hl~ieCt-HrientEd hmguage. This task is 
analogous to the definition Hf classes ill all 0bieCl-Hricnted 
hlngttage, or to the clEf trillion of tile structttrE of a 
structarcd dHcumcnt (h Ill GRIF, l ,aTEX or FranleMaker). 
Two "classes" are ah'eady dcfixled by the system: enlry 
"rod accEption, The linguist determines the structures to he 
associated with these HbiECtS, l lerc, we give the dcfinilion 
of the structure of the French diclkmary. 
The predefined class entry implemEnls a tree with 
accEptions on ils leaves. In the following example, an 
Entry consists in a feature structure with two fcalures (a 
graphic-lbmi and \[i categHry). 
((lef-lin(itllutl_c clas:; lrorlch--~!litl-y {ol/tlry) 
( t oat;Ill o-. st" £ tlct ur e 
(glaphic-folu/ str inq) 
(cat ogory (:it t.(!g(~ y ) 
)) 
(dot linquistic--c\]a:;s cate, lo~y () 
(ont.' (,\[ 'n(e 'rip +vb 'vhimp 
'vblTell 'adj 'Cald 
' d/tic:t; ' lTolll ' f:~lb ' cool7(\]) ) 
The predefinEd class ttcception prHvides a way to code 
its rehltiHn with an interlingual aeception. In the Example, 
wE dcl\]ne an acceptiHn as a fealttre structure with fEalures 
represenling derivation infHrm'ltion (with Ihe kind and Ihe 
source Hfa derivation), information on valencies, etc. 
(oh!f-linguistic clam; fronch-/toception (ac:c(q~t:ion) 
{ t{~/ILufo ,t;t;I IICI;III (2 
(cat cat:~flol y ) 
; ; I\[l\[Ollllill io\[I o\[I \[ }14~ d(.i Ivitl~\[oi\], 
(d~vv ( |¢~/lt,ul o t;(;\[ 11(?lTut o 
( ~lt.'~ iv ..k i nd 
{oI1(~ ol 'll~tctioll 'tltTO,'¢ulh '111 i(u 'IlilGI/}\[\]\[~ 
'\[li\[\]£;ILr '~l(Ijec\[; '/tdjpil~lf; 'adjpotpas 
'~tdiyC.ll~lct 'v(~llu~) ) 
(der l v I I om :;ymbol ) ) ) 
(£\]FVII { f(~ll;/ll Q -t;LEIICI;III (~ 
(de~ iv kind 
(oIlo.of 'iicol\]d 'rl\]\[otl 'llitlstl7 'nc(~tlo.ct; 
'verhe) ) 
(doriv \[l-o\[tl symbol ) ) ) 
((live3 (f(,~tttlf(~ ~i\[~lll<Tt\[tll() 
(dt~l iv kind 
(oll(~ of 'llilb:;I; 'l\]IJal \[l()\[I 'v<!\[\])(~) ) 
( (le t i V- f l/c~ltt :;ymbol ) ) ) 
; ; iIlfOIRI}ltiOll Oil t-lie vah!tlcios 
(v/ll0 vii 1/!llcty ) 
(villi valency) 
(val2 va\] oIlcy} 
(val3 w~ Icnlcy ) 
(val4 valency) 
; ; el h(}I" i tl I ormat iorl 
(t/TIE (gltly-o~ 'ltl/lflC ' IOltl) ) 
(\[Ibl (/iny o£ ',¢ifl 'p\] ) ) 
(a/IX {olte-of ' <!t;l(~ ' aVO\[ 17) ) 
(rectproque (one of 'algO atoll 'argl-algl}) 
(/la;poct (()lie Of 'achovt5 ' \[n&chov(! '(\]*'.,\]I(IL ' t ill 
' i\]lli &t \[I ' frtSquent ' i t\]\[~Lilll\[;/lll6) ) 
)) 
281 
(def-linguistlc-class valency ( ) 
(any-of 'nom ~+nom 'aveclnom 'comme+nom 
'conhre+nom 'dans+nom 'de+nom 'en+nom 
'erltre+nom 'par~nom 'parmi+nom 'pour+hum 
'sur+nom 'inf '~+inf 'de+inf 'adj 'que+ind 
'que+subj 'se-moy 'se-pass 'lieu-slat 'lieu-dyn 
'mani&re 'z6ro) ) 
3.2. Coherence checkinff 
When the definition of the structure is done, the linguist 
can define coherence rules that will be applied on the 
entries. 
3.2.1. Three kinds of rule 
The linguist cml define three kind of rules: 
• Integrity rules apply to an article of a dictionary. 
They ensure that none of the ,article of the lexical 
database has an ill-funned configuration. 
• Local coherence rules apply to different articlcs of 
the stone dictionary. They ensure that the dictionary 
is coherent. 
• Global coherence rules applyto different articles of 
different dictionaries of tile lexical database. They 
ellSUl'e some coherence between dictiouaries. 
3.2.2. Three levels or cohere,ce rules 
"llle rules are ,associated wilh a strength: 
• Warning: when the constraint is overridden, a 
message is passed to tile lexicographer, but all 
treatments ,arc ,allowed. The warning disappears as 
soon as lhe lexicographer validates tile entry. These 
constraints are nsed Io detect potential emirs. 
• Delay: when the eonsmtint is overridden, tile 
lexicographer receives a mess:,ge and sonic 
treatments are forbidden on tile concerned entries. 
Incorrect entries will not be accessible by extraction 
requests. Interactive treatments such as browsing and 
editing are allowed. These constraints are used to 
handle temporarily incomplete entries. 
• Criticah these constraints can't be overridden. If a 
transaction overrides such a constraint, it will be 
cmlccled (rollback). 
3.2.3. Exmnple of coherence rule declarathm 
A coherence rule declaration is a method (ill the sense of 
I.ISP/Common Lisp Object System) which is applied on 
all objects of the class defined in tile parameter lisl. The 
body of tile rule is a lisp expression that must t'CltJrn T or 
nil. If the result is nil, the exception nmchanism 
corresponding to tim strength of tile rule is inw~kcd. 
Ilere is an example of an integrity rule for the French 
dictionary. This rule verifies that the derivation 
information is coherent with the category of tile aceeption. 
(def-lntegrity dry-cat-coherence 
( (acception french-accept;ion) 
(dlctJonary french) } 
critical 
(cond ( (is-one-of (cat acceptlon) 
'vb 'vbimp 'vbrefl) 
(and (empty-p (drvrl accept\[on) ) 
(empty-p (drva acceDtion) )) ) 
( (eq~lal (cat acception) 
'no) 
(and {empty-p (drvv acception)) 
(empty-p (drva acception) ) ) ) 
( (equal {cat accept;ion) 
' adj ) 
(and (empty-p (drvv accept.ion)) 
(empty-p (drvn acception) ) ) ) 
(t t))) 
Conclusion 
In this paper, we have presented our work on MI.DBs. 
After a study of existing international projects and the 
definition and testing of the proposed lcxical organizatkm, 
we ,are currently defiuing and prototyping a specialized 
system fer the management of acception-based MLI)Bs: 
the NADIA system. 
This system introduces new interesting points. First, the 
acception-based lexical organization seems to offer the 
advantages of an interlingual approach while avoiding 
some of the theoretical and methodological problems of 
the kuowledge-bascd approach (Sdrasset and Blanc 1993). 
Second, it gives the linguist the possibility to freely dciine 
a collection of linguistic structures with a raflmr complete 
set of predefined data structures. 
Our objective now is to integrate in this prototype 
features coming from research in the field of structured 
documents and a multidialectal facility ill all tools, ill 
order to provide lexicographers and other users with an 
interface iu their mother tongue. 
Acknowledgments 
A part of this work was conducted ill tile Multilex 
project. 1 wish to tha,lk all Multilex parmers and GETA 
members for flmir supptu~ and feed-back. 
References 
An&d, J., R. Furuta, et al. (1989). Structured 
l)ocuments. C,'unbridge series on Electronic Publishing, 
Camb,idge Universily Press. p. 220. 
I~I)R (1993). I~I)R Electronic I)ictionary Technical 
Guide. Japan Eleclronic l)ictionary P, esearch Institute lad, 
Project report n ° TR-042, August 16, 1993. p. 144. 
Farwell, 1)., I.. Guthrie, et al. (1992). The Automatic 
Creation of Lcxical Entries for a Multilingual MT system. 
COI,ING-92, Nantes, 23-28 July 1992, wfl. 2/4. 
pp. 532-538. 
Goc, dman, K. and S. Nirenburg, Ed. (1991). Tile KBMT 
project: a case study in Knowledge-Based Machine 
Translation. San Marco, Calilornia, Morgan Kaufmann 
Publishers. p. 330. 
Nirenburg, S. and C. 1)efrise (1990). l.exical and 
Conceptual Structure for Knowledge-Based Machine 
Translatkm. ROCI,ING lit, Taipeh, 20-22 August 1990. 
pp. 105-130. 
Sdrasset, G. and I". Blanc (1993). Une approche par 
acceptions pour les bases lexicalcs multilingues. T-TA- 
TAO 93, Monlr6al, 30/9-2/10/93, to be published, p. 15. 
Uchida, II. and M. 7ku (1991). Interlingua. International 
Symposium on Multilingu~d Machine Translation (MMT- 
91), Beijing, 19-21 August 1991. pp. 22-30. 
282 
