Formal redundancy and consistency checking rules 
for the lexical database WordNet TM 1.5 
Dietrich H. Fischer, GMD-IPSI 
Abstract 
In a manually built-up semantic net in which not 
the concept definitions automatically determine 
the position of the concepts in the net, but rather 
the links coded by the lexicographers, the for- 
mal properties of the encoded attributes and re- 
lations provide necessary but not sufficient con- 
ditions to support maintenance of internal con- 
sistency and avoidance of redundancy. Accord- 
ing to our experience the potential of this meth- 
odology has not yet been fully exploited due to 
lack of understanding of applicable formal 
rules, or due to inflexibility of available soft- 
ware tools. Based on a more comprehensive in- 
quiry performed on the lexical database Word- 
Net TM 1.5, this paper presents a selection of 
pertinent checking rules and the results of their 
application to WordNet 1.5. Transferable in- 
sights are: 1. Semantic relations which are 
closely related but differing in a checkable 
property, should be differentiated. 2. Inferable 
relations - such as the transitive closure of a hi- 
erarchical relation or semantic relations induced 
by lexical ones - need to be taken into account 
when checking real relations, i.e. directly stored 
relations. 3. A semantic net needs proper repre- 
sentation of lexical gaps. A disjunctive hy- 
pernym, implemented as a set of hypernyms, is 
considered harmful. 
I Introduction 
When building large-scale lexical/semantic resources, 
subsequent - or better, simultaneous - validation of con- 
tent is essential. Basic validation includes formal redun- 
dancy and consistency checks. Taking WordNet 1.5 as 
an example, we illustrate the development and applica- 
tion of such rules. The computational environment for 
this enquiry is TerminologyFramework, an object- 
oriented generic tool developed to represent and consis- 
tently maintain concept-oriented dictionaries of different 
types, including WordNet \[Fischer et al., 1996\]. Into this 
system, the content of WordNet 1.5 was downloaded 
from the "dict/data.*"-files; some checks were per- 
formed during download as part of the operational se- 
mantics of our relation definitions, but most of the 
checks were simulated a-posteriori by database queries. 
The main aim of the enquiry is not to produce an error 
report on WordNet 1.5, but to develop a methodology of 
redundancy and consistency checks for re-use. There- 
fore, not only WordNet 1.5 has been checked, but our 
ideas have been developed and checked for validity and 
relevance using WordNet 1.5. Compared to the thesauri 
we had previously modeled and downloaded, WordNet 
1.5 offers a richer set of semantic and lexical relations 
which give rise to new questions of redundancy or con- 
sistency (cf. \[Fischer, 1993\]). The more relations intro- 
duced into a manually built up net, the more dependen- 
cies are created which hold each other in check; once 
they are explicated, formulated as guidelines and imple- 
mented, they can be used to support internal consistency 
- a necessary, but not sufficient condition for the cor- 
rectness of a semantic net. 
In the following - after having characterized in gen- 
eral the status of formal checks in semantic nets of the 
WordNet-type - we present and comment a series of 
constellations which shall give exemplary insight into 
the topic. 
How to read the pictures in this paper 
All pictures in this paper are snapshots from the screen, 
and clippings from a window of TerminologyFrame- 
work's graphical browser \[M6hr and Rostek, 1993\]. The 
original, and hence uncorrected WordNet 1.5 data is 
shown as a graph where a node may represent a concept 
(i.e. synset, named by its first synset-element) or occa- 
sionally a term, i.e. synset element linked to its concept 
by the designation relation which is shown by its inverse 
from the concept side and is thus labeled term. Terms 
are represented in TerminologyFramework as unique 
objects having the synset element word/phrase as their 
possibly homographic name, and a system-generated and 
maintained homograph counter number which is stored 
separated from the name string as another term attribute. 
A term node label is distinguished from a concept label 
by a lowercase prefix indicating the part of speech 
(hence N for noun concepts / n for noun terms, V for 
verb concepts / v for verb terms). A concept's label body 
takes its content from its first synset element which is 
also transformed on download into a term with this label 
22 
body. We do not present examples from the adjectives 
and adverbs which are also contained in the database. 
2 Formal checks versus checks based 
on concept definitions 
We suggest that Woods and Schmolze \[1992\] implicitly 
also critize semantic nets of the WordNet-type when 
they characterize "typical frame-based representation 
systems" by stating that in these systems the semantics 
of manually linking concepts by is-a or a-kmd-of rela- 
tions is "strictly operational - defined by how they work 
and what they cause to happen." Then they continue: 
"There is no external criterion of correctness to which 
these decisions should adhere" (p. 135). 
We wonder what would be an external criterion of 
correctness even in a system of terminological logic, or 
is there a criterion which checks the adequacy of a 
manually supplied logic concept definition? In any case, 
we would call the criteria which are based on the rea- 
soning of the system internal ones. However, in the 
framework of a KL-ONE-type system they can be neces- 
sary and sufficient - relying on a set of undefined con- 
cepts, axioms, and a logic calculus - while in a manually 
built-up thesaurus or semantic net of the WordNet-type 
only necessary conditions of correctness can be checked. 
Note that Woods and Schmolze \[1992\] assume that 
something "happens" when the lexicographer draws a 
semantic link; they do not criticize - as we do - that 
sometimes nothing happens or not enough happens 
which differentiates this type of link from some other, 
except that they have different labels. Then linking 
would be nothing but connecting nodes. Our aim is to 
dig out and apply rules which approximate the intended 
semantics of the links, or to populate the inventory of 
those checks which are based on formal properties of the 
used relations and attributes and their logical dependen- 
cies, thus constituting their operational semantics. They 
are not based - as in a terminological logic system - on 
an "understanding" of the individual concept definitions. 
Therefore we call them formal. 
WordNet concept definitions are given by glosses in 
natural language - not referring explicitly to the net it- 
self, thus not making use of its disambiguated polysems. 
These glosses are attached to the concepts, but some- 
times they are not available and sometimes they are 
intermixed with or replaced by usage examples. Checks 
may be conceived which are based on an understanding 
of these definitions whether the reader of the glosses is a 
human reader or a program, based on linguistics. How- 
ever, sometimes the glosses are faulty and the link 
structure is correct, and sometimes the contrary is true; 
thus they could hold each other in check, as long as they 
are independent of each other. 
We continue these notes on formal versus content 
based checks at the end of this paper (see section 5) by 
presenting three examples, contrasting both types of 
checks when we already have explicated the formal ones 
we then can refer to. 
3 Checks to avoid redundancy 
From the results presented in the following, one can 
infer that redundancy-free data was the aim of the 
WordNet lexicographers, but apparently if by insertions 
redundant data was generated this occasionally was 
missed, if the redundant data was out of sight of the 
human checker or the checking program, i.e. distant by 
some hierarchical levels from the point of update. 
3.1 Short cuts in the generic hierarchy 
Figure 1 shows a short cut, spanning four levels of the 
hyponymy hierarchy from noun concept hptd to noun 
concept triglycerzde. Because of the transitivity of the 
generic hierarchy relation this link is redundantly 
stored in WordNet 1.5. Note however, that the short cut, 
so obvious in this picture, will probably be hidden to the 
eye, even in a hierarchical line print-out, when all hypo- 
nyms of hpid, od \[2\], animal oil, and glycerlde would 
be displayed. The numbers of hyponyms from lipid 
down to glycertde are 4, 32, 29 and 1 respectively. 
Note that '2' in the label N. od \[2\] is the homograph 
counter number, generated by TerminologyFramework. 
This number need not be identical with the sense number 
shown by WordNet's read-only browser. 
The database contained only 11 short cuts with respect 
to the hyponym / hypernym hierarchy for noun concepts 
and 2 short cuts with respect to the troponym / tropony- 
mOf hierarchy for verb concepts. 
Note however, that the result of the redundancy check 
is valid if and only if the premises are valid: If one of 
the non-redundant hyponym links is wrong, and has to 
be removed the link diagnosed as redundant may be 
correct and non-redundant. For the cases of redundancy 
mentioned here the premises apparently were correct. 
N: II 
HYPON'YlVl 
'N: oll \[2\] 
1 H~ON'CM 
N: animal od 
I ~-rCPONVM 
N: glycende 
HYPON"YIvl 
N: tngt 
31d 
IHYPONYM 
tcerlde 
Figure 1 A short cut in the generic noun hierarchy 
23 
3.2 Are short cuts in the meronymic 
hierarchy redundancies? 
Short cuts in the meronymic hierarchy - as represented 
in Wordnet 1.5 by the semantic relations part/partOf, 
member/group, and ingredient/substance - have also 
been supervised by the system on download. 73 short 
cuts were detected, all in the part/partOf hierarchy 
However these can be judged a-priori to be redundancies 
if and only if transitivity holds for the part/partOf rela- 
tion, and this depends on definitions (not given by 
WordNet) and actual usage. 
A preliminary result of our analysis is: 1. The lexicog- 
raphers implicitly made heavy use of transitivity, other- 
wise the data would be highly incomplete; in other 
words, if transitivity of the partOf-relation had not been 
presupposed there should be many more short cuts. 2. 
By an occasional distorted use of the partOf-relation 
applied to individual concepts transitivity was invali- 
dated; this pertains to cases of the type: "The Alps are 
part of Yugoslavia; the Alps are part of France, ere .... " 
With respect to generic concepts (not individual ones, as 
the Alps) there is also an acceptable example of an im- 
plicit exclusive disjunctive partOf-value-set. These dis- 
junctive partOf-value-sets invalidate transitivity and 
inheritance of meronymic relationships through generic 
relationships. These cases have to be singled out, sepa- 
rated into a special relationship or simply corrected by 
introducing e.g. the concept French Jlps etc.; then for 
the rest transitivity holds and short cut checking and 
checking for redundancy by inheritance is meaningful. 
Meronymy in WordNet needs the space of another 
paper, see also \[Bloksma et al., 1996\], and \[Prlss, 1996\]. 
3.3 Redundancy by implication of 
different semantic relations 
On the class of verb concepts two semantic relations are 
defined which are not logically independent: Troponymy 
(i.e. hypernymy on verbs) and entailment. Example: hmp 
is a troponym of (or a special way to perform) walk, and 
snore is an entailment of sleep, if simulated snoring is 
not snoring. It is stated in Fellbaum's WordNet paper 
(\[Miller et al., 1993\]) that "Troponymy is a particular 
kind of entailment, in that every troponym V 1 of a more 
general verb V2 also entails V2" (p. 47). 
Accordingly, the database does not include a verb V 1 
which is a direct troponym of a verb V2 and directly 
entails V2, because the latter link would be redundant. 
However, because of the assumed transitivity of tro- 
ponymy, the database stores only direct troponym links 
(2 exceptions, see above subsection 3.1), and taking this 
into account, the rule should be formulated more exphc- 
itly: If a verb concept V1 is a direct or indtrect tropo- 
nym of V2, then V1 entails V2. As a formula: (V1 tro- 
ponymOf* V2) => (V1 entailment* V2). (Here the star 
operator designates the transitive closures of the rela- 
tions it is applied to.) Hence the check must look for 
non-empty overlaps of these two virtual relations. Re- 
suit: The database included 15 redundant direct entail- 
ments; one is shown in Figure 2. 
Figure 2 also includes an overlap with redirect entail- 
ment (from massage \[2\] via rub \[1\] to touch \[7\]); this 
however vanishes if the redundant direct link (from rub 
\[1\] to touch \[7J) is removed, but what is to do, if the 
troponym chain from manipulate \[2\] to rub \[1\] would 
not exist in the database? The constellation then would 
not be tractable by automatic update. The database in- 
eludes another three indirect redundant entailments of 
this type. 
With hindsight and foresight we are emphasizing the 
following: The logic of the generic subsumption de- 
mands that every instance of a subeoncept is also an 
instance of its superconcepts, otherwise the logic, sup- 
posed to be started from, is changed. From this follows 
that poly-hypernymy has to be treated with care: Multi- 
ple superconcepts to be implicitly combined by or 
(disjunction) and not by and (conjunction) invalidate 
"that every troponym V 1 of a more general verb V2 also 
entails V2" (see below Figure 12 and Figure 13). 
+ ,:V touch \[7\] 'rake physical 
- / ~ "~ : contnctwlth, come 
, ,~" \ Omss'-,.~Incontactwm: + 
" "rl~:tOPOltPa'~ \ "Touda.,he stone for 
, / + ~ ." goodltek+'She 
' , ' ~ never toudled her 
£V handle \[2\] ~ husband'; "The two 
' l , \ Ixildlnosalrnost 
mOP+NYM '+ " . \toum- 
,, ." , ' r ENTAILMENT 
\ t = 
V mampulete \[2\] : \ " 
gloss I "rROPON~,lvl , 
'hold ,omathl.g I "~'~,~ , .., - ~ rno~/e ~/er 
In one's handy i v gu=oe \[lJ ~ something 
imdmc'.~a' \[ gloss ~ ~ with 
I . ." ~. ~ pressure' TROPONYM wr example, pllsS TROPONVM \ ¢ 
l- °~'=" ~ \/ I V rub \[1\] 
I ENTAILMENT / 
~ , V 
massage \[2\] 
""'gloss..... ' 
'usuaJly for medicinal or relmmtlon purposes" 
Figure 2 The entailment hnk from rub i7\] to touch \[7\] is 
redundant However the entailment hnk from massage \[2\] to 
rub \[1\] is not redundant, - or could massage \[2\] be a tropo- 
nym of rub \[1\].9 
May a troponym of x also be 
an entailment of x? 
It has been pointed out in this subsection that a tropo- 
nym t of x also entails x. Therefore, if in addition x en- 
tails t, then x and t are equivalent, logically. It follows 
that - if such a constellation occurs, and the premises are 
valid - the concepts t and x should be merged. 
The database contains exactly three constellations of 
the type in question. They are shown in Figure 3 and in 
Figure 4 which contains two cases (the pair exhale \[1\] / 
breathe \[1\] and the pair inhale \[1\] / breathe l"1\]). 
24 
, ', r * 
;' ,,,, ...... ' _,___ :.._,.~...~ v: care for \[1\] " 
, ,. 1 =,,, . . ", :, V. t reat\[ \]~_term.~:..:L..,,,::vtreat;q ~ ., 
', ',. + \[ , ', ' , rpr0dde treatrnerrt for" ' 
~,-.~I"ROPONVM : - '-" ,:~.a ,L~,rr~.rL~,L';~. - " ' " " , i = ~1 l, |~ t,a -re. ~2 e*ltJ .r Ii.jLIi~. ~ - 
¢ . 
. '.:~ ,: - , ..... . . ,~ .",~L.t:v. dispense \[1\] , , 
, ', ........... ..,..,_.,...mrro-TS:-T.,~.. ~ , ~,,;. : - 
V: administer \[1\] -.--..r.term.:j..:.. j "~ , ': ..... " 
' "-'-~tmgr~ t %.'' "-----'~ "~ .,~'~v: aammlstert/j "~ 
, ~ ' 0 S = J i1" I ,*- - 
'-',','t~';,i'> '' :~ .~, -'-.~,-~'",'ofhledlcafJons" , ~ ~' • =.,.Hi, H?. \[ ,,-~-;r-\], ~'t.0., ,,.~?'~- -, , r,". . 
.,"-~-'- t. ",-'J l ' -~:; ".~" ",.'.;~.'~, ~,-',, ;. . , 
,' :',' ENTAILMENT :r1~,..OPO.:~:;~,l~. ',~ ,?!'!:" ; o ~. 
J,,," , ,' - , ~' ~'~=3,.,.~ ~.: ~,',- - -, 
: ' \] ~ : I- ' ' ~, '~.%~:~:dr~.JL~ '~ : ,' -"J" -," '. , ~, ";,.'," ~ I /' "." ~$'~:~-~.~,~'~"='"'-~".-, " " 
.t :: ', " r' ~ . ,~ ,I'-~.,,.;;~" ;~," ,.'~."~- ',,,, 
, . V: give \[22\] ,..~.,......_,lerm.~.L:,~.,.~,ff, v ' hive r221 = .-~- ~ " 
.... ° .... - - gloss. ,,-!,:" ~.~..~, ,,,.'~ ,'~ ~'~ r' ." :, 
' ,', ,, ' 'as of t,.re~itr0ents or proi~esses:,: , ,..,,, , .' , . "= ,~,~. ~..,.. 
, ' ,~, , '' ShagaveNmRrst,,~.ki. I '. - 
:. ,, :, .,, ' . gavehl.rh~dI.u~ ,~ ,,,'. :-', ~ ,- , ;,': , ',,,' , , , , , ,"u "- ""-;~7-~: ~',: >-'~' " " 
Figure 3 Are admzn~ster \[I\] and gtve \[22\] equtvalent? 
According to our view the premises in Figure 4 are not 
valid: exhale \[1\] and mhale \[1\] are not troponyms of 
breathe \[1\], because breathing needs exhaling and in- 
haling; both troponym links should be removed. On the 
other hand, for the constellation in Figure 3 we would 
assume that the synsets {administer, dispense} and 
{give, apply} should be merged. 
? ,, , , 
',' ' ~ ~o.~;'(~\] .... o .... , ,., , 
"~ "'" -- "" ~ ' ~ tMl~t" ' v take a breath , , , '1~ olr inla; term - ~ .... , , 
' ' ' ' "I'Tt~P(~NYM ENTAIl,R" EI'/'I'NI~F ~ I * '" ' "~ breathe ,n 
''' ~" " " Iomt ~ . ..... hede \[1\] 
..;, , ~., ~ ~,.--., ., %. ./~T~-'~" ~. , 
,~ Iv exhnle\[1 "~,.-,tllrfrr~,~ "' , ' " ~:, , %'-~ ~'..~.~n~._~,,v draw n mr 
. .~, , . _j~rrfl~,_,~V exhale \[1\] ,, . ' ~V inhale 1\]-,...i~,.•L 
v exp,re \[11"7- tm~--/'/~ Irk ' " ~ I ~ "----Inh~e ~eep~ 
...... .,=4"Lm m /-/ - . ~ ' 
. /t ' 
youlltthl ENTAILMEWr ' "~ ' • ... . I, t' " =--"e " ' /' 'IROPONYM%~ 
, wmglt", / 1ROPON'~M ........ 7ROPOI~,1:1'1:~O1~ "'" '':' 
, / ', I ~. it ,~, t -~~\v.~,,=o\[~\] ,' ,:, '/, I ,°"ff '', \,,,/, , \\N 
, ' ENT~IJvEN'r f 
, ,V.sneeze ~" ~ ' ~/.~ ' V snort\[3\] 
, , - ~. t 'V snort \[1\] It V puff \[7/\] V hLIf \[2\] 
tlwheh m t, tate a sn~'tl~9 " ~ , 's~:k In er udm, u ol alr~ "draw' 
kdtant enlmled sound ~ V blmv\[l) a deep brmh'~ drew on a 
o~e's nose' e0duJ~g hmf 'e~ale herd: 
I 
Figure 4 exhale \[1\], tnhale\[/\], breathe \[1\], and their enw- 
ronment 
3.4 Redundancy by inheritance 
via troponymy 
We did not exemplify what is meant by redundancy 
caused by inheritance via the generic and meronymic 
relations for nouns (of. subsection 3.2). Instead we show 
an example for inheritance via troponymy, see Figure 5. 
- . . r- .. ," 2" ',- ,;' veil Bd, lravelt0ward :,: : - .', "lzt,~or~tnl;'o=m\]now., r:,.. ~7,,:,~ ~r~oh~6~:,~,. -- . 
-',~-'c~~;=;~+~'--,,~+i"" ~:, ;J.~':.,7.~:.:+:'+ +.' , 
, ¢ 2.- a ~-aj.~,. 3,,. p -~- . - "k-~ - i k- ~" s - .¢- - , - 
=" ..... " "'~- , "= -;'~'.'-~, " ,~' .... '-" " ~come\[6\] ", 
; ..... '- -" ~'~V Ixng~\]~.Z~r~NI:.~ILMEI~ ~.-l~Vcome~"52~'~. - ~:!.~ :~ 
" v.fe~Pl 't~,m~; | ' " - :~' :; '~ r '/~L~=..-~-,~: -'-,,v=mo~t3} 
:vbnng121 .... " .... , ..,d .,=,. ...... ~ 
_" ..... .~:-- :ENTAItMEN '~'. '=~-'"l " ; -" 
i ::  ii: - ; 
- '" : : "q;V~ngaong"r'-%~.; ", ,'-:1"-~':~'"-'" "- -:Z "~ -~_ 
,, .... t~\[tb~, - . - . ..... , , :.: , 
V 10ring around \[. v txlng ~lt~pg " :. ' - " : 
' ' -'~.. ~.'.r ~tv bnng with - " 
Figure 5 A redundant entazlment link from brmg along to 
come \[6\] 
Nine concepts qualify with redundant inherited entad- 
merit links, however five of them also qualify as redun- 
dant by implication. (see above subsection 3.3). One 
example of the rest (4 concepts) is shown in Figure 5. 
The database contains some further redundancies by 
inheritance via troponymy. 
3.5 Redundancy by synecdoche versus 
auto-relationships 
Our trainee students, in their search for synonyms of a 
concept c, tended to take designations of some super- 
concept of c -whether the superconcept already exists in 
the thesaurus or not - and added them as synonyms of c 
because these words are also used to denote the concept 
c. The examples given then may be related to the phe- 
nomenon of synecdoche: According to the OED synec- 
doche is a rhetorical "figure by which a more compre- 
hensive term is used for a less comprehensive term and 
vice versa, a whole for a part or a part for a whole, genus 
for species or species for genus". While such a synecdo- 
chical use of designations, carefully applied in discourse 
does not lead to polysemy, it would lead to an absurd 
overload of polysems in a dictionary, if the principle 
would be transferred to it, even if reduced to inherttance 
of designations, i.e. top-down, not bottom up, and along 
the generic relation only. 
25 
' 'a tool with a sharp, point and euttJng edges 
* ' for making holes In hard materials (usually ' 'a motor-driven tool". , " 
,. , ~ rotating ~apicfly or by repe~ed blows)" 
' gloss " ~. ,' gloss .. -- 
' " N', powertool ~- :: .... iN', drill \[3\], ' "~rm~ 
. .,-- ,N.,;,,-','/-,-:.:- ---- 'I' ' 
; : .: . I -i 
. . T . ~ " , r" "Z ~j," .-~ - "= 
-.' " , -' ~N'p~erdnl*:.-:-t,~'~n:po~erdnJI -' I .. : 
' " - ' *-" ,:, enlesaxe 
'-: "-'- : I '-=G~oss~?: o' .:' ', , . z- , : 
I I I . I ~ . ~1 ¢ fv" -11-1 . I ~ ,, , .' I . - apowerto~forrnaIdng I . 
• " " J..h.'POl~C, tl~! "holes In her'¢l'mitl~rlills; , I , i 
- , 'i ,, :',r, O~(-.~ J 3' ' -"1 ", 
_ ' , ,'-:: ,N. drlll\[,:l\]::~.~LTr~rn~7.~n, drdl\[4 l r-:.~ 
... :?. , : 
"~ " -.. , '.- ,,.~ ~ ,' ,. gloss :" n: electrlc drill _',.'~ , ',.' 
', a . _r- ¢ ' ~ ," - 
L. . : ", ",.>..~.: ;:?'a rotatllng ddli pofvered by an~elearlc rnot fir ~ . - ~ ".i 
Figure 6 Redundancy by synecdoche or an auto-relationship9 
An interactive or batch check may look for direct or 
indirect superconcept-subeoncept pairs which have as- 
signed terms with identical names, thus giving rise to 
homographs. However, the interactive check's action 
part should be a warning and not a roll back, otherwise 
the system would exclude that polysems can be arranged 
in a generic hierarchy as superconcept and subconcept. 
Fellbaum \[1996\] calls this an "Autorelation" (in Ger- 
man). A good example from WordNet is the noun drink 
in the sense of beverage (drmk\[2J),, and in the sense of 
alcol~ohc beverage (drink \[3~. However, what about the 
example shown in Figure 6? 
Is the assignment of term drill as a synonym of elec- 
tric drdl avoidable redundancy leading to avoidable 
homography, or not? If not, why was drill not also a 
synonym of power drdl? If yes or no, and even if we 
confine ourselves to a synchronic view of language, 
what would be a working guideline which delineates the 
assumed good example drmk from the alleged not so 
good one (Figure 6)? 
In the bilingual DUDEN OXFORD we find as a trans- 
lation of drill into German not only the general meaning 
Bohrer (WordNet: drdl \[3\]), but also Bohrmaschme 
(WordNet: power drdl) in the context (Carpentry, 
Butldmg). Given that the concept or meaning unit is 
context independent, this may bring to mind that not 
only translation needs context, but also synset member- 
ship. 
The database contains only 23 noun homographs se- 
lected by the mentioned check. These homographs be- 
long to 21 pairs of noun concepts. Five of them are sin- 
gled out due to the type of mistake which is shown be- 
low in Figure 15: A hyponym link has been coded where 
a meronymic one would have been correct. The corre- 
sponding number of verb homographs is 283, and they 
belong to 336 pairs of verb concepts, among them five 
cases which are synecdochical or auto-related triplets. 
A similar type of questionable synecdoche pertains to 
cases such as {drumhead, head}, a synset which is a 
hyponym of membrane, but also creates a new sense of 
head (total number of senses: 24). 
4 Consistency checks obtained for 
semantic relations which are induced 
by lexical relations 
When asking for possible checks not yet performed for 
WordNet's term-term relations (lexical relations) we 
became aware of their interference with the concept- 
concept relations (semantic relations). This enquiry is 
well supported in TerminologyFramework by its easy-to- 
handle facility to define virtual (i.e. inferable or comput- 
able) relations on top of real or virtual relations. 
4.1 Exclusivity of hypernymy and 
antosemy 
The WordNet antonymy relation is a lexical relation, i.e. 
a relation between synset elements of different synsets 
\[Miller et aL, 1993\]. Therefore it was modeled in Ter- 
minologyFramework as a term-term.relation. With re- 
gards to concepts, two concepts can be defined to be 
opposed if at least two of their terms are antonyms. 
TerminologyFramework implements the opposed rela- 
tion (synonymously: antosemy) as a vtrtual, i.e. comput- 
able concept-concept relation (see Fischer et al. 1996). 
We say that the semantic relation antosemy is mdueedby 
the lexical relation antonymy. 
Antosemy relationships and hypernymy or hyponymy 
relationships are exclusive to each other, i.e. both rela- 
tionships cannot hold in conjunction between any pair of 
concepts. This rule may be based on the feature model of 
concepts: If we assume a concept representation by fea- 
tures, then hypernymy entails inclusion of all features of 
the superconcept, and this cannot be compatible with an 
antosemy between superconcept and subconcept, which 
may, for example, be based on a meta-antosemy relation 
between features. 
For a cheek of this rule the overlap of two virtual re- 
lations has to be determined: antosemy and the transitive 
closure of hypernymy or troponymy. The number of 
antosemy relationships is about 800 for nouns, and about 
500 for verbs; there was no overlap with direct or indi- 
rect hypernymy, and in only two cases there was a non- 
empty overlap with indirect troponymy. These two vio- 
lations of the exclusion rule are shown in Figure 7. 
Note that there is no overlap with direct troponymy, 
but with indirect or inferable troponymy. What is wrong 
with this counter-example? The troponym link from 
prove \[1\] to negate l"1\] is an error, and it may have its 
origin in a fallacy: To prove by negation is a troponym 
of to prove, but this is different from to negate in the 
sense of to show to be false. In other words, one may 
prove A by showing that the negatton of A is false, but 
the point is, that the negation of A is another object than 
A, i.e. the object to be proved has changed, and indeed, 
it cannot reasonably be maintained that to negate A is a 
special way to prove A 
26 
The two violations of the exclusion rule Figure 7 
4.2 About binary and n-ary (n > 2) 
antonymy and antosemy in WordNet 
WordNet 1.5 permits cardinality > 1 for antonymy, how- 
ever a cardinality check of antonymy is blind with re- 
spect to fundamental semantic differences of which the 
cardinality check for the induced relation, the antosemy 
relation, is more sensitive. We explain this by the fol- 
lowing examples from,WordNet 1.5: 
The verb term trust \[1\]is an antonym of the verb terra 
mistrust and of the verb term distrust \[1\], and both - as 
variants - are synonyms of each other. Thus, the cardi- 
nality of the antonym set of trust \[1\] is 2. However, 
because these two antonyms are synonyms, the value set 
of the induced antosemy relation of the concept belong- 
ing to trust \[1\] has cardinality I. Therefore we say that 
trust \[I\] has no genuine multi-value antonymy or, syn- 
onymously, it has binary antonymy - although its anto- 
nym set has cardinality 2. 
On the other hand, there are cases (one is presented in 
Figure 8 below) where the induced relationship has car- 
dinality 2 and the basic antonym relationship has cardi- 
nality 2: In Figure 8 the term ~,. artse \[3\] has genuine 
multi-value antonymy or (in this case) ternary anton- 
ymy. However, not all cases of cardinality 2 of antosemy 
are caused by deliberate ternary antonymy. Cardinality 2 
may also be caused by an error of a type we try to de- 
scribe in the next subsection. WordNet does not differ- 
entiate between binary and n-ary opposition (n > 2), or, 
basically, between binary and n-ary antonymy (n > 2), 
and due to this implicit merging of relations, the lexi- 
cographers intentions cannot be automatically checked 
by a simple cardinality test for antosemy; i.e. we do not 
know which of the cases of cardinality 2 of antosemy are 
inconsistent or not. The comparatively low number of 
these cases (about 200, but only 40 for nouns and verbs) 
allows for intellectual perusal. 
~,,,~',..~lJ~slt\[4\] ~'-~.~"l'~ ~'~'.7,="- ~1 ~,"~'-'" ," ~v:.ller41 ~: 
Fngure 8 An interesting vmlation oftransitiwty 
However, if n-ary antonymy is admitted the law of tran- 
sitivity is at stake. We singled out all cases where the 
transitivity rule for antosemy was not fulfilled, resulting 
in 29 connection components or 85 involved concepts. 
All these constellations seem to need correction. The 
corrections we would suggest are: Cutting of an antonym 
link, or a merging or a splitting of concepts, and all 
these cases belong to the type characterized in the next 
subsection. There was just one case, shown in Figure 8, 
where we first thought that an antonym link was forgot- 
ten (between v: sit down \[2\] and v' he down ), and this 
was caused by the fact that the database contains the 
following transitive complete group: he \[2\], stt \[1\], 
stand \[3\] are antonyms of each other (see Figure 13). 
Coming across other perhaps more harmful cases (see 
below Figure 12 and Figure 13) of a concept which is to 
be represented as a disjunction of concepts and itself a 
lexical gap, we now suggest that the constellation of 
Figure 8 was intended to say: arzse \[3\] is an antosem of 
the one concept sit down \[2\] or he down. However, if 
this single-valued binary relationship is split here into 
two binary ones linked to the components of the dis- 
junction, then all antonym value sets must be interpreted 
as disjunctmns, and transitivity need not hold for n-ary 
antosemy with n > 2. On the other hand, the antonym 
value set of trust \[1\] is to be interpreted as a conjunc- 
tion. If the example of Figure 8 is interpreted in this 
27 
way, we also loose simple symmetry of antosemy: From 
the mere fact that the antosem of stt down \[2\] is artse 
\[3\] one can no longer infer that the antosem of arise \[3\] 
is sit down \[2\] 
4.3 Commutativity of synonymy and 
antonymy is equivalent to maximal 
cardinality 1 for antosemy 
t 
Fzgure 9 A case which qualifies as a wolatmn of transitivity 
of antosemy. 
Was the constellation shown in Figure 9 intended as 
ternary antosemy? Is the case equivalent to Figure 8? We 
do not think so, and this may be backed also by the fact 
(not shown in this figure) that V. sedate is a top concept 
while V" de-energize is not. The constellation shown in 
Figure 9 may be characterized by another law which is 
pertinent here if we may assume that binary antosemy 
was intended. It is an abstraction from one possible cor- 
rection of the constellation: The antonyms of the syno- 
nyms v: energue and v: smnulate \[1\] should be syno- 
nyms, i.e.v, de-energize and v: sedate should be syno- 
nyms, and in order to achieve that, the concepts V. de- 
energtze and V sedate need to be merged. 
Abstracting, the rule postulated for binary antonymy 
is: For each set of synonyms S the set of antonyms of the 
elements of S must again be a set of synonyms or it is 
empty. By "a set of synonyms" we mean a set of ele- 
ments which are defined as synonyms. If we would in- 
terpret "a set of synonyms" as a synset then the rule 
reads: For each synset S the set of the antonyms of the 
elements of S must be a Subset of another synset, the 
subset may even be empty or be the whole synset. Trans- 
forming the rule into a formula, would help to better see 
that the rule is a kind of law of commutativity between 
antonymy and synonymy. Another paraphrase of this 
rule is that binary antonymy is a structure-preserving or 
homomorphic mapping with respect to synonymy. 
There is a simple checking rule for this because the 
rule is equivalent to maximal cardinality 1 of antosemy. 
However, this would help for future checking only if 
WordNet is modified and updated so that binary and n- 
ary (n >2) antonymy would be different lexical relations. 
Note that the value sets of binary antonymy need not 
have maximal cardinality 1, although this is true for 
binary antosemy. 
4.4 Commutativity of antosemy and 
hypernymy / troponymy 
The commutativity rule for antosemy and hypernymy or 
troponymy - as a heuristw rule - has already been sug- 
gested by Fischer et al. It may be justified by feature 
inheritance. Formally the heuristic rule of commutativity 
states: For each concept c: If antosem(c) is not empty, 
then the equation hypernyrn (antosem(c)) = antosem 
(hypernym(c)) or set inclusion in one direction or the 
other should hold. Applying the rule on the WordNet 
data has resulted in 91 noun concept and 260 verb con- 
cepts fulfilling neither strong (equality) nor weak (set 
inclusion) commutativity. These cases have not yet been 
evaluated by a native speaker, but we suggest that they 
deserve revision. An example is shown in Figure 10. 
' ;1- : ' ' i ': 1- : 
: ~,: t,~" ' 
N nghteousness ~ , . ~ppo~ld, , ~N unnghteousness 
v 
N honoral~leness.~.-..opposetl,--J.-N d=shonorab/eness, . ,, N dJshonor\[1\] , 
l '~ : _ "'aLOS~' - -"GLOSSl : 
" I " , ~-. .... ,',-,~ -I-: Hwo~ .~equ=~,,~ ,, ~.~" Vta,OmM- " 
" t ~ des~oro¢ ,;,,, honor£r + L .-~ ., 
'l ' " r~,='" , ' +, +:' I,~t,,~,r~,,, +: I • 
N fidel*ty\[2\] 4, ;ppoN¢l . ,;, ~N' infidelrty ~ 
tccPO~ - ' .'~,. ~ t.ta,oNvfl 
"N loyalty\[2\] ~ ~posed ~ N dtsloyalty 
-=. , 
Figure 10 Four of 91 noun concepts fulfilhng nezther weak 
nor strong commutatwity of antosemy and hypernymy" The 
four concepts are honorableness /dtshonorableness, and fi- 
dehty \[2\] / mfidehty. For morahty \[1\] and immorahty \[2\] 
weak commutativity holds 
28 
4.5 Disjunctive hypernyms - 
or may antosems share hyponyms? 
If we rely on monotonic feature inheritance the above 
question needs a negative answer. All the more the em- 
piricist may be interested in convincing counter- 
examples. These are the results of the respective data- 
base search: 
For noun concepts (after we have corrected that irre- 
sponsibility was an antonym of itself) there is exactly 
one hit, shown in Figure 11. 
., ..% 
' a',,, - . % " ~r~ r ~ ~, ' I~ lit~ml y- -, - 
, :%.:N assetiZ\] ~" .., ";~'~. ~.. "'~...':I~II Iilli01hty\[~\] ".-." " "j . 
-'~i~t' ;' ~S~ve . -: A-.': , ~ , ~ "- ,~ : .... -=~ r. ""n'p*~Vo~,o~on . , 
-"- .... # ~I01~,~; cred'lt~ ~1-" "~- -'¢' " ¢":",7-ff~.; ".~.'.. "- " "n flnencmlobltgatlon- -,ll.i~lle.ll~,--_ l'i !./J . ~'.,-. ~,' 2,,~ N oeOt\[l : j,~ .......... ~ , =, 
.~.~.'~.lTt~.-~'{,,~-~ ....-,I,-j.r.. I'. '-~: " .F '.=--" ~r'r,~- -2"'n ,nde~edness \[1\] ; ~ " 
. -~:. ,<~. ='.. . .,~--~, . :~ - , .-.,. ~ . ., ~, ..- -.~_ ." 
f n11~ll~, Z ,:'" N ctedlt line \[2\] ~."',: if" ". ,o HYPON~. "~ llOWld" " : ' : 
,~. "P~.-':"--' I T".; ..'-;-, " " LI '? ",-'~-...: ,:.. 
"~'==d~d"~- " v"~... ~-'" -. '-~ 'Y,';'- '. 'thepm,,,~ " , 
: " 
h ." :"- ',' ,~,~.~,~y,f ,',. :, --,r.,;,,y# -,., >. _~,',, ~'-'~'7"-'-~ ""~a~, )' " " ~, " 
,:. . !:;. 
~_~,~ ',,, ,-'-,IN home eqmy cfedlt \[ - I.~. ~ .N installment wedlt = ,, ' ', , 
IUIt~ ., - J- , ~ .I '~ - .22, I - - ~ . ,. e -.. 
',-.'-~', -t2 ",:e~. .... '." ."' -'-.~ ~ ,," -~-: =, " :" , , ~ , ~,,,'*~ "., -I-, ,' ~, ,-7,':" ~,,', t ' ,' "" , - .. 
. :.-~-.:.,, ,~.~' ,.= .~.~.~,=~.,t ..... .' 
::,,:, -':-=,3' .I, =\,',': .... ~ :' ,: " " . , : ':' "" r: " - ' - ~ " 
Figure I l Special credits represented as oxymora (Webster' "a 
figure of speech in which opposite or contradictory ideas or 
terms are combmed")9 Or" Can a credit which has become a 
real debt be accepted as an asset entry? If that meaning of 
credit exists, where credit \[7\] (an offer) has been transformed 
into a debt by drawmg on the offer, should it not be differenti- 
ated from credtt \[7\]? 
For verb concepts - if we subtract the two cases pro- 
duced by the error treated in subsection 4.1 - the check 
detects exactly six pairs or two constellations shown in 
Figure 12 and Figure 13. 
In Figure 12 the verb concept smuggle is a troponym, 
directly shared by export and by import \[2\]. Although 
the gloss says that smuggle is a troponym of transport, 
we may perhaps redefine: "export or import illegally". 
However, if someone is smuggling it cannot be inferred 
that he is exporting illegally, nor that he is importing 
illegally, only the disjunction can be inferred. The se- 
mantics of the generic subsumption is that every instance 
of a subconcept is also an instance of its superconcept, 
otherwise the subsumption is not justified. Therefore 
smuggle is a troponym of a concept export or import (or 
yet transport?). The stored troponym links have to be 
removed. Note also that otherwise transitivity of tro- 
ponymy would be invalidated and short cuts could no 
longer be deemed a priori to be redundant. 
, r 
',,vseliabroad, t " " "- "~ ..... , t.t-,'l'~: ~ .... 
, .: ~i~. .term .,'r ,-.:' ,~, ". . Aerrrl :--.-.. ~' " 
, =" --;'-~\,"1 "-"~r :",~r""'~ " ~1.,I .~,1 ~ "~.~'- 
" ~-~'7"~" ;'%'~V' export :'" '~p'poseil '~' V, IrrlportlZ\] ~'.~ ~" :~ "~ "., 4..--., .... ~. ~ ~ ~'~-, o-.: -~ ,;~ -, .,: ~ ' ',4.. ~--~ 
:v transfer abroad ,,- ~ '~ - ' %, +,.' , ~- ~'l~'ln" ir~fr~l~'abrbed; " 
- '-~'~- "', c "~-o E'~.-.-,~.~X--2. - -.;-^-' ~" " " " "~'"':' i; Y ,';', ' 
-, 41 '- :I ,,:'. r-:.,~ ~'.~'~'..\~ :i~::.,', ~°.~-~ .. o .~\].~!,:~ ,- ~. :. 
-~,r'---" :,-, .- :a~-"~lOIl'-" tarrn, ;j : '-'- -tz-~ ~T - " 
, : .,,, ¢:,:. ::...~ .~:. transport l)legally"~ ..~-~ smuggle \[2\] ,.., ,,,, = "T4,d~-". ,'\]~ 
Figure 12 The "disjunctive hypernym", Implemented in a way 
which is harmful: smuggle \[2\] should be a troponym of a 
concept export or tmport \[2\] 
From this diagnosis we are led to the question what 
was the general practice in WordNet with respect to 
multiple direct superconcepts? The database contains 
558 noun and only 25 verb concepts with more than one 
direct superconcept. Among them, of course, also those 
found by the short cut check (see subsection 3.1), and 
for them the value set of superconcepts implicitly is a 
conjunction. Apparently this was aimed at for all nouns, 
but we saw a case where the hypernym relation implic- 
itly was substituted by ts-used-for and this made up an 
or. For the verbs we see far more ands than ors, among 
the latter also undifferentiated verbs which are tropo- 
nyms, e.g. of both change \[1\] ("undergo a change") and 
of change \[3\] ("cause to change"); el. also below section 
5. 
",L~"Wlhatt/tti' \[, .', ..... ~ ~,t'- "' " - --. "'"" " . _.-_..' ~ ~.- ., '~-,-. .-,:-.. -, ,: ,-i~e,ylng, be v 
~uma~a~,m~: --'~ • - - , " ' - - . : .- ,, ..-,.' . -. .... ".':j" : ". :'. '- ~ro~ian; be - 
• ~', , ., , "'~ ~posea ; , d "" "' .... ' "" 
, p~rrdrman~l-,~ .>-I"~.~- '" . '-, ~ :-".,,k'- ," /e "' ....; , , 
', " " gloel " " ', .I;- -' ,~ " ~ ,~ - "..' ,-, " ..... ,.~. ' 
• .'%r' ,p. --. .... .,-,m.i. ,. _- ,.r ~.,,, ,-. "?~a-"~ " 
'-V: stand \[3\] ~l'='oppoiad "71,,~V: Sit \[1\] :,,,,,~,,,,:op'p.~.ed ~,~,,~v: ,,; tel ' 
- ~ : "TFIOPONY M TROPONVM TROPOI~I ~': ' Y'., 
•, ,., - , sprawl \[1\] ~ oss " 'sit or Irffwlt!i -, 
• . ,..,,..v ,r:- --' :-~one'sllml~ ~ " 
. :- , , .: .S, TF~OPON'YM ' - r spread, q~' 
- ,. ", V: spread-eagle \[3\] .-- -. '=7., 
"stahd with arms and legs Ipread out' ' ;' " ' " ' 
Figure 13 Beyond details to correct we want the reader to see 
that again subsumption is of the type of the harmfully imple- 
mented disjunctwe hypernym. 
29 
5 Continued: Formal versus content 
based checking 
At the end of section 2 we promised to present three 
examples which illustrate the difference and the limits of 
formal checks in contrast to checks based on concept 
definitions; here they are: 
Figure 14 shows an example which was retrieved by 
the check described in subsection 4.4: Neither weak nor 
strong commutativity holds for the concept pair N. em- 
ployee and N: employer, and the formal rule suggests 
that their direct superconcept should be opposed. How- 
ever, reading the gloss of N. employer the human reader 
(or the program analysing the gloss) may infer that the 
superconcepts need revision: The gloss logically de- 
mands the existence of a concept person or firm which 
does not exist in the database, although Fellbaum \[1996\] 
argues to consider lexical gaps; however, if one would 
propose that an add+tional hypernym link to N. firm 
might help, that would be an error of the type discussed 
in subsection 4.5. Another unfortunate decision would 
be to create two polysems: 1. An employer who is per- 
son, 2. an employer who is a firm. Bloksma et al criti- 
cized that WordNet practice. 
The second example, shown in Figure 15, was among 
the 21 noun pairs which were retrieved by the check for 
generic synecdoche or generic auto-relationship. This 
topic was treated above, see subsection 3.5. The concept 
pair we present here is satmwood \[1\] and satinwood \[2\], 
linked by a hyponym link. Because supereoncept and 
subconcept are both designated by the same word, thus 
creating homography, they were detected by the check 
which relates to generic synecdoche. However, this ex- 
ample is neither a case of generic auto-relationship nor 
is it a case of avoidable generic synecdoche: The hypo- 
nym link between satinwood \[1\] and satmwood \[2\] must 
be replaced by a substance link, also available in Word- 
Net. No formal check could prevent the lexicographer to 
select the wrong link type unless the synecdoche checker 
would lie in wait and catch this special case because of 
name equality. Asking for a check which would be ade- 
quate in more generality, we draw the reader's attentmn 
to the two superconcept chains which - before they end 
up in a common ancestor - are headed by hfe form and 
object \[1\], in the sense of the negation of life form (see 
glosses). Assuming that there exist entities which have 
aspects of a life form as well as aspects of not being a 
life form, we may be interested whether WordNet re- 
fleets this non-Boolean logic view. In fact, in Wordnet 
1.5 we find 15,087 subconcepts of object \[1\], and 
13,806 subconcepts of ltfeform, and only 6 concepts in 
the overlap, among them, of course satmwood \[2\]. In 4 
of these 6 cases the same error occurred, and in the other 
2 cases the hyponym link was mistaken for a member 
link, also available in WordNet. WordNet does not en- 
code plain logical incompatibility, to express disjoint- 
ness of hierarchies, but what about antosemy? Was it by 
chance that life form did not get the term ammate object 
which then might associate the used term mammate 
object as an antonym? Then we would be led back to the 
check of subsection 4.5: May antosems share hyponyms? 
. ' + '" " ; N: person \[1\] " , " /I 
" 
, . ~ ",.-~ ..... ,..~., 
:..- °,- :" . ~ I.i~POPlf~4 " ~ ,' I..I'~: + ' .:~. ~ ' 
+ \ ..... :" "~"':++ "'+~.,',." 5,",, '+- ~ .... ' 
" .N,,nonworker+..oppo~ed_~N:womr~ \] ~" -,.~" ,N,,leader\[2\].,~-.,r~r_t.Jt,7.yN,,follower\[1\]+: 
'- " "-: - : '": 
': .:,:' .' +m,. ..-:/,' ' -:+ :.--'.- , 
- . " " " . , " +2 ~ P -*L "' : ~,t" t~ .1 - - + ' ~ • 2 
'-; +',,~+i+e .,- ", '.~ :-, ,~N:empkr/ee+.~,~,N em,d~r, .~ +, :'.'. " I ~" +", "! 
'.',,, " + ++ +--:+':C:,;,,g'+'~';-PP'":~-+r+'~'~' +~.~,.,:,+v::~,+;.,++ 
'.,'-:';+."- : ++~-: ~-~t0SS,~::~.+:;+~.o:, .+~: k- Ot0s~,+'~ - ' :..;+ ~. ; - " , ~+ * r + - +.,, .-.t,. ¢i;i .-r a 
-+-k:, %+-;) ++:',;l--',~-':+'°J:. :..,:'++:' ?~-++ -, : .., = -,:- + 
.. - ,.., ~-.~,mpe~og.~ <...,; ,~-++. ~,... ~t~r+l~ .~', •, .~+ 
, - -"+ = - ~.~.. n -"+" - - +- : - ' +. + *+_ -... 
Figure 14 The posmon of employer in the net does not reflect 
tts gloss. 
t ' ~' " ' " , -+-- - -'7+-'._ ,~ -- , , - ,. 
.~* ; +- " '-.: +-+'. "~'~'"NentJty 7 " . ., . - . 
-n tnanlrrlateo~eet." '~ - ~" - ,f'- ~ ' +. ,, . 
', .~-',' ',+""+-t~:_." IP¢I~ON~'MP ~ I'IWON~M " " te(In~ "n Iangthtng ~ '" 
"+-n.obje=\[1\] ~...+ :" ~ + :=7 ° j.%k- '+fo"~.~.,.n organ,sm\[1\] . " 
- ~ "" " " .+.tllrll~r'l oqectLlJ ~ - . " 9N ifemiim . . 
" n physcalo~ect ;',,..+'+++ L '' , +' -' "" r" ~"tllll'll% . ' - 
- I + :., -.;, ,+"+z-,I- =.~ ." ..... N plunt\[ll ":' ' *lnyP~4ngently' " 
,,.. r ,- .Nsubstance\[l\]-/; +~ , , . . ~ - '- - . - -" 
" "-:-+"":"':i, ',I - -+,,,. ,,t.m,6m~+ _, ........ 
' 11" + 1+ +~ 1 + " + IN vasm+ plant ' " 
, . -, :,+N mmen~ " . .t: ' L "' -" +, 
..... , h, I - I-lYPOI, Md' , . '. ...-, ;: ~Po~ ;" ,." + . . , 
.... "-"" T . ,+' + "N, woody plant " " 
.. ;. -"I~ ptant material .... ,, I -' 
, , HvPOnI~R 
+'~.' ". '' .~'" ;'" ~ "" "Ntree\[l\] +' ~','-' , + 
"~" "+':' 2 + J~N, wood\[4\] G '., I ' ~r~Chtoroxylonswmtenm " 
+o.,\ . I , . , ,. , I iN satmwood\[1\].rlmra~n satmwoodtree . 
-/i "Z \'-" " - - 'mel~nlm:,m= " ?.~ n saur~ood\[11 
HYPOP~M ¢oss ,. 
tlohllleil , ~ 
" ~ Indlm~l wdh , 
lhlbBJ'kofl~H~" +' ' ;N s~r~.,lood\[2 \] X99~II31111~IdkJSU'DUS + - 
.. +. 
V -tom '+ 
'hmd Wllowtd~ wooa of e snllnwomt ~e ht~'~llny ', ' 
lustlm'~tlledforlklec~lma~,~podtlmdtol31s' n'sat~muood\[2\] r . 
L 
Figure 15 How could this type of false hnk selection between 
sattnwood \[1\] and satmwood \[2\] be prevented9 
We would term it a formal check when for a given ge- 
neric relation of a concept class the system is asked to 
retrieve all top concepts (or the direct subconcepts of a 
possible unifying root), and especially select those 
which are isolated, i.e. which of these do not have sub- 
concepts themselves. With respect to this check the tro- 
ponymy hierarchy of WordNet's verb concepts has 573 
30 
top verbs, however it is surprising to find that 236 of 
them are isolated. Now, we already have reached the 
limits of formal checking because formal checking can- 
not tell us whether all this was intended or is acceptable, 
and if not, what to do. A human reader of the glosses 
may infer that this or that top concept should be sub- 
sumed under a more general one, or may have already 
existing or not existing subconeepts. 
6 Conclusion 
To our knowledge, the WordNet lexicographers were not 
supported by dynamic checking on update, or by an 
easy-to-use database query language for batch-checking, 
nor by a graphical browser/editor for visual feedback. 
WordNet's database set-up program, the "grinder", ob- 
viously controls consistency, however we are not in- 
formed about this. According to TerminologyFrame- 
work's error report on download, the "grinder" did not 
faultlessly perform range-checking because it allowed a 
few semantic (synset-to-synset) pointers where they 
should be lexical ones (synset-element to synset- 
element), and in very few cases it missed that a synset- 
element should not be an antonym of itself (irreflexivity 
of antonymy), and that a synset should not be a hypo- 
nym of itself (irreflexivity, as an entailment of aeyely- 
city of the generic relation). Beyond these few slips we 
found more interesting examples of errors or of redun- 
dancies which were not detected by chance, but by trig- 
gers (created by TerminologyFramework from the speci- 
fication of the operational semantics of the relations) or 
mostly by queries to the database, guided by our meth- 
odological interest. 
A practical lesson is that the design of dictionary rela- 
tions should be such that they are tractable by formal 
checking, and this is severely impeded if different rela- 
tions are merged of which one has the checkable prop- 
erty p and another lacks it; the merged relation has then 
lost the checkable property p. Examples from WordNet 
1.5 are meronymy (which is treated in this paper only by 
a short note) and antonymy. Another point we had to 
struggle with was WordNet's treatment of disjunctive 
hypernyms, especially when they are lexical gaps. The 
topic of implicit logical junctors in value sets of Word- 
Net's generic and meronymic relations was also treated 
by Bloksma et al., but the therapy they propose we could 
not get to like. 
This paper does not contain a complete list of check- 
ing rules for WordNet 1.5: Whenever we tried to evalu- 
ate a rule we got hints for another rule, and we have not 
yet taken into account all WordNet relations and attrib- 
utes. Of course, there are important and less important 
relations, but note, if one takes only the important ones, 
or the most important relation, the generic relation, then 
formal checking in this type of semantic net is very lim- 
ited. In any case, formal checking is only a kind of syn- 
tax checking, the next step after spelling checking, but 
100 new pennies will make up an Euro. Some of our 
concrete diagnoses may be wrong, or fall short, or be- 
come obsolete by a new release, but the questions to be 
posed for this type of semantic net remain. 
Acknowledgments 
I am indebted to Melina Alexa and John Bateman for 
encouraging this work, and to them both and Wiebke 
Mt~hr, Renato Reinau, Lothar Rostek, and Ingrid 
Schmidt for valuable help to improve this paper. 

References 
\[Bloksma et al., 1996\] Laura Bloksma, Pedro Luis Dlez- 
Orzas, and Pick Vossen. User Requirements and Func- 
tional Specification of the EuroWordNet project. Deliv- 
erable D001, WP1, LE2-4003, Version 4, Draft, May 14, 
1996. 
\[Fellbaum, 1996\] Christiane Fellbaum. WordNet: Ein 
semantisches Netz als Bedeutungstheorie. In 
Bedeutung, Konzepte, Bedeutungskonzepte. Theorle und 
Anwendung in Lmgutstik und Psychologte, pages 211- 
230. Joachim Grabowski, Gisela Harras and Theo 
Herrmann, editors. Westdeutscher Verlag, Opladen, 
1996. 
\[Fischer, 1993\] Dietrich H. Fischer. Consistency Rules 
and Triggers for Multilingual Terminology. In TKE'93 
Terminology and Knowledge Engineering, pages 333- 
342. Klaus-Dirk Schmitz, editor. INDEKS-Verlag, 
Frankfurt/M., 1993. 
\[Fischer et aL, 1996\] Dietrich H. Fischer, Wiebke M6hr, 
and Lothar Rostek. A Modular, Object-oriented and 
Generic Approach for Building Terminology Mainte- 
nance Systems. In TKE'96. Terminology and Knowledge 
Engmeermg, pages 245-258. Christian Galinski and 
Klaus-Dirk Schmitz, editors. INDEKS-Verlag, Frank- 
furt/M., 1996. 
\[Miller et al, 1993\] George A. Miller,. Richard Beck- 
with, Christiane Fellbaum, Derek Gross, Katherine 
Miller, and Randee Tengi. Five Papers on WordNet TM. 
CSL Report 43, Cognitive Science Laboratory, Princeton 
University, July 1990, Revised August 1993. 
\[M6hr and Rostek, 1993\] Wiebke M6hr and Lothar 
Rostek. TEDI: An Object-oriented Terminology Editor. 
In TKE'93 Terminology and Knowledge Engmeermg, 
pages 363-374. Klaus-Dirk Schmitz, editor. INDEKS- 
Verlag, Frankfurt/M., 1993. 
\[Priss, 1996\] Uta Priss. Relational Concept Analysis: 
Semantic Structures in Dictionaries and Lexical Data- 
bases. Thesis, TH Darmstadt, October 1996 
http://php.indiana.edu/-upriss/papers/cdissi.ps 
\[Woods and Schmolze, 1992\] William A. Woods and 
James G. Schmolze. The KL-ONE family. In Semantw 
Networks m Artoqcial Intelhgence, Part 1 Fritz Lehman, 
editor. Computer&Mathematics, 23(2-5): 133-177, 1992. 
