Conceptual Lexicon Using an Object-Oriented Language 
Shoiehi YOKOYAMA Kenji HANAKATA 
Electrotechnical Laboratory Universitat Stuttgart 
Tsukuba, Ibaraki, Japan Stuttgart, F. R. Germany 
Abstract 
This paper describes the construction of 
a lexicon representing abstract concepts. 
This lexicon is written by an 
object-oriented language, CTALK, and forms a 
dynamic network system controlled by 
object-oriented mechanisms. The content of 
the lexicon is constructed using a Japanese 
dictionary. First, entry words and their 
definition parts are derived from the 
dictionary. Second, syntactic and semantic 
information is analyzed from these parts. 
Finally, superconcepts are assigned in the 
superconcept part in an object, static parts 
to the slot values, and dynamic operations 
to the message parts, respectively. One word 
has one object in a world, but through the 
superconcept part and slot part this 
connects to the subconcept of other words 
and worlds. When relative concepts are 
accumulated, the result will be a model of 
human thoughts which have conscious and 
unconscious parts. 
1.Introduction 
Semantic relations among words or 
concepts have been represented as a 
slot filler definition in the lexicon 
\[Bobrow 77\]. These relations are usually 
represented as frame lists in LISP or 
Prolog. When the representation of a 
large-scale lexicon is needed, it is 
difficult to make and maintain the lexicon 
because the relation of super/subconcepts 
dramatically increases. 
An object-oriented language has a strong 
inference mechanism. It has already been 
used for parsing sentences \[Hayes 84, 
Nishida 84\]. It can also describe the 
representation of semantic relations more 
simply than LISP. CTALK is such an 
object-oriented language and can dynamically 
handle the network using the concept of a 
world a group of objects -- and an 
object. 
Now the systematization of semantic 
relations is made for the words which 
represent concrete objects. However, the 
organization of abstract words is difficult 
because it is not so clear how to obtain and 
deal with abstract concepts. On the other 
hand, research is being conducted to 
computerize or categorize the common 
dictionary \[Yokoyama 77, Tsurumaru 84\]. 
This paper describes a trial of 
systematization and categorization of 
abstract words. The procedure is as follows: 
first, the definition parts of abstract 
words in the dictionary are semantically 
analyzed, and the semantic relations are 
derived. Second, an object is established 
for an abstract word. Third, the 
super/subeoncepts of these words are 
assigned to a father-son relation among the 
objects corresponding to the semantic 
network. Then the slot definitions are 
assigned to the object. From these slot 
values, new objects are made under the 
relative objects. Finally, the network is 
constructed from these abstract words. 
The network formed by this method is 
considered to be a model of human 
consciousness. That is, when a human is 
reminded of a concept, other concepts 
relative to it are simultaneously extracted 
from the memory and stay beneath the main 
memory. This approach is considered to be an 
implementation of interactive activation 
model \[Rumelhart 82\] in abstract concept 
level. 
2.Object-oriented language CTALK 
The object oriented language, CTALK, was 
developed at Universitat Stuttgart \[Hanakata 
84\]. In this section, a simple summary of 
CTALK is described. 
The main elements of CTALK are worlds and 
objects. A world is a set of objects, and 
some active worlds stay in the main memory. 
When the number of worlds exceeds the 
constant defined by the system, the least 
recent worlds are gradually swapped into the 
secondary memory. The hierarchy or the 
network among the objects, that is, the 
father son relation of the objects is simply 
set to every object, independent of the 
world structure. 
An object consists of four parts, SUPERC, 
SONS, SLOTS, and METHODS. SUPERC means the 
superconcepts, that is, the fathers of the 
object. SONS, literally, means the sons of 
the object. Actually, in the input of the 
object, the names of sons are not specified 
explicitly. The control of these father-son 
relations are performed on the bit table. 
The use of this bit table enables the fast 
operation of creating and removing these 
relations, and also of treating inheritance 
mechanisms. The inheritance usually works 
inside a world, and can also be spread to 
other worlds. The direction of inheritance 
can be specified by VIEWED AS operator 
\[Selbmann 85\], with which an object inherits 
the features of specified father(s). 
SLOTS is a static property of an object. 
The form of SLOTS is as follows: 
\[slotname (restriction) : default-value\] 
\[slotname slot-value\] 
The restriction and default value are 
optional, and slot value must fill the 
restriction in the same slotname in the 
antecedent object. If a conflict occurs at 
value assignment, no value can be assigned 
when the restriction is violated. When there 
is no restriction or the restriction is 
filled, another default-value can be 
assigned in a descendant object. 
METHODS is the message part of the 
object. Sending the message invokes the 
operation written in the METHODS part. Of 
course, logical functions such as ~if ° , 
226 
"then", and "else", and numerical functions 
such as "equal", "greq" (greater than or 
equal) can be written in the METHODS part. 
User functions can be defined using system 
functions and other user functions. 
3.Construction of the conceptual network 
3.1.Procedure 
The procedure to make the conceptual 
network is as follows: 
I. The definition part for an abstract 
word is derived from the corresponding entry 
word in a dictionary \[Kindaichi 74\], and 
semantically analyzed. 
2. A new world is made for an entry word. 
3. The top level object named TOP is 
made. This is the control object for message 
passing and slot handling. The father object 
of TOP is now CTALK_IN_CTALK (default 
highest object in the whole world). All 
objects in the same world are the sons of 
TOP. 
4. An object corresponding to the entry 
word is made as a son of TOP. 
5. If the evident superconcept is 
derived, it is explicitly written in SUPERC 
part. In the case of an isa relation (a 
change to another word), the slot value 
"isa" is filled for the corresponding word. 
6. Objects of the words written in the 
definition part of the entry words are made 
in the world of the entry word. These 
objects have an object with the same name as 
the father object in another world, that is, 
a superconcept. 
7. The arc relations relative to the son 
object are assigned to the slot name, and 
the slot value is the corresponding word at 
the other side of the arc. 
Finally, one world is formed for one 
entry word as a "lump" of the concepts. In 
the following sections, the details of this 
procedure is described. 
3.2.Making the objects 
Fig.l (a) shows the top-level object TOP 
in the world "mokuteki" (object or aim). Its 
superconcept is now CTALK_IN CTALK, but if 
two or more conceptual words can be sublated 
("aufheben°), the integrated concept becomes 
a new superconcept. The sons are the entry 
word itself and the objects (words) which 
are derived from the definition part of the 
entry word. These form the components in the 
semantic network. In the slot "mean", the 
literal meaning cited from the dictionary is 
written, and used as a normal search (i.e. 
consultation) of the dictionary. In order to 
consult the dictionary, the following 
message should be sent to CTALK IN CTALK: 
(CTALK mean mokuteki\]. 
The operation written in METHODS part in 
CTALK..IN_ CTALK: 
\[mean ?x -> (* do: (x chw:) 
(TOP gv: mean))\] 
is invoked. The variable x is connected to 
"mokuteki", and the literal meaning is 
output. "* do:" is a PROGN-Iike function in 
LISP, and the return value is the evaluation 
of the last part of the following functions. 
"Chw:" is a system function changing a 
world, and "gv:" takes a slot-value. 
TOP IN mokuteki 
SUPERC 
\[ C'rALK IN CTALK ) 
SONS 
\[ koto IN mokuteki \] 
\[ mono_IN mokuteki \] 
\[ mokuteki IN mokuteki 1 
s s' 
( mean "kodosuru mokuhyotosite kangaerarela 
sositai nanigotoka, sonaritai nanimonoka" ) 
/* something considered as m behavioral aim */ 
METHODS 
Fig. 1(a) Object TOP IN mokuteki. 
mokuteki IN mokuteki 
SUPERC 
\[ TOP IN mokuteki ) 
( koto IN mokuteki \] 
( mono IN mokuteki \] 
SONS 
( mokuteki_IN_syusi \] 
\[ mokuteki IN meate \] 
Fig. I (b) Object mokuteki IN mokuteki• 
koto IN mokuteki 
SUPERC 
( TOP IN m~uteki 1 
\[ koto IN.koto ) 
SONS 
\[ mekuteki IN mokuteki \] 
S~/)TS 
\[ JBO isi \] 
( OR mono \] 
\[ YNA nani 
METHODS 
Fig. 1(c) Object koto IN mokuteki• 
Fig.1(b) shows the object of the entry 
word "mokuteki" (object or aim). Many of the 
entry words in the dictionary are explained 
as the subcategory of "mono" or "koto" , 
"Mono" represents a concrete thing or object 
in many cases, but also represents the 
abstract things in some cases like in 
"mokuteki " . On the other hand, " koto" 
represents abstract things or matters in 
almost a\].l cases. According to the 
dictionary, both ~mono" and °koto" are 
assigned as SUPERC's here, but it could be 
possible to make only °koto" a SUPERC. For 
the problems concerning the definition power 
of the dictionary, see Conclusion. The entry 
object can be connected to its relative 
words as SUPERC relations. In SONS part, the 
words referring to "mokuteki" are described, 
Here the \].ink from ~mokuteki" iN_'syusi" 
(purpose) and "mokut eki " ...IN _" meate" (guide 
or aim) is connected. The more complex the 
semantic network is, the greater the number 
of SONS objects is. In SLOTS part, there is 
no slot name nor va\].ue because no ~isa" 
relation exists in the definition part in 
the dictionary. 
Fig.l (c shows a part of objects which 
should be made on the world "mokuteki'. 
"Koto ° is one of the superconeepts of 
~mokuteki ° and by itself a subconcept of 
"TOP ~ and "koto._IN_koto ° . 3 slot names are 
227 
assigned at SLOTS part, that is JBO, OR, and 
YNA. The strange names such as JBO and YNA 
indicate the reverse direction of arcs. This 
guarantees the two way relation of objects. 
Slot value is, of course, the corresponding 
word name. In the case of "kodo" (behavior) 
the structure is similar to "koto". 
3.3.World "mokuteki" 
Fig.2 shows a part of a conceptual 
diagram for the world °mokuteki ° after all 
relative objects have been made. --:> 
indicates the arc from father objects in 
other worlds, --> the arc to son objects in 
other worlds, and --~ the connection inside 
the world. When new terms or example 
sentences for the word are added to this 
world, the structure inside the world will 
change. On the other hand, when new entry 
words are added to the system, the number of 
worlds increases. If these words are 
relative to the world °mokuteki" or refer to 
it, connection arcs are newly formed, and 
the network becomes more complex. 
CTALK IN CTALK l 
\ \ I ~isi IN isi  TOP - 
/ :, <o o 
"~mokuteki~" is i / 
/ mokuteki 11 \\ ./ 
Fig. 2 Part of the structure of 
world "mokuteki". 
4.Conclusion 
The world shown in Fig.2 is considered to 
be a model of the human conscious process. 
When the human considers a certain concept, 
the central concept is fetched to the main 
memory. That concept corresponds to a world 
in this system. Relative concepts are 
connected to this concept and exist near (or 
beneath) the main concept. Some of these 
concepts often go to the surface of 
consciousness, but usually stay beneath the 
consciousness. These concepts correspond to 
other worlds connected to this world. 
In the current system, relatively static 
properties are defined, and only a few 
dynamic connections are used. Dynamic 
functions in CTALK including restriction 
conditions in . SLOTS part should be 
considered in the next version. 
If a similar structure is found between 
two or more worlds, integration of these 
worlds can be considered. On the other hand, 
when one world is too big to be treated, the 
separation of the world would occur. 
Separation and integration processes of 
worlds are also considered as a model of 
human thinking. However, the automatization 
of the process is very difficult. 
Knowledge for the world is now derived 
from a Japanese dictionary. One reason for 
using the dictionary is that it is difficult 
to deal with the words having abstract 
concepts, and the dictionary is one of the 
228 
most convenient clues for treatment. The 
other reason is that this dictionary has 
been computerized \[Yokoyama 77\] and will be 
used for automatic semantic analysis. 
However, definitions of entry words in a 
dictionary have various problems such as 
definition, semantic elements, and 
co reference \[Nakano 85\]. Description of 
other dictionaries should be discussed, and 
the efforts to find suitable representation 
and connection will be continued. 
Acknowledgements 
This research started at Universitat 
Stuttgart when Yokoyama stayed there as a 
visiting researcher sent by the Japanese 
Government. The authors express their 
appreciation to researchers in Universitat 
Stuttgart and Electrotechnical Laboratory, 
as well as staffs in both governments. 
References 

\[Bobrow 771 Bobrow, Daniel G. and 
Winegrad, Terry: "An Overview of KRL, a 
Knowledge Representation Language ° , 
Cognitive Science, Vol.1, No.I, pp.3-46 
(1977). 

\[Hanakata 84\] Hanakata, Kenji: °A 
Powerful Agent for the Control of 
Inheritance and Message Passing in the 
Object-Oriented Language CTALK", GWAI-84 
(1984). 

\[Hayes 84\] Hayes, Philip J.: 
"Entity-Oriented Parsing', COLING-84, 
pp.212-217 (1984). 

\[Ishiwata 84\] Ishiwata, Toshio: "Features 
in Japanese Representation" (in Japanese) in 
Tanaka, Akio et al: °Performance II", 
Asakura Publishing Co. (1984). 

\[Kindaichi 74\] Kindaichi, Kyosuke et al. 
(eds.): "Shin-Meikai Kokugo Jiten ° (Second 
Edition) (Concise Japanese Dictionary) (in 
Japanese), Sansei-.-do (1974.). 

\[Nakano 85\] Nakano, Hiroshi: "Problems 
about the Description of Semantic 
Definitions" (in Japanese) in Kusanagi et 
al.: "Grammar and Semantics II", Asakura 
Publishing Co. (1985). 

\[Nishida 8411 Nishida, Toyoaki and 
Doshita, Shuji: "Combining Functionality and 
Object-Orientedness for Natural Language 
Processing ° , COLING-84, pp.2\]8 221 (1984). 

\[Rumelhart 82\] Rumelhart, David E. and 
McClelland, James L.: "An Interactive 
Activation Model. of Context Effects in 
Letter Perception: Part 2 ° , Psychological 
Review, Vol.89, No.i, pp.60 94 (1982), 

\[Selbmann 85\] Selbmann, Frank: "Entwurf 
und Implementierung eines kontrollierbaren 
Backtrack Mechanismusses so wie einer 
universell verwendbaren undo-Funktion fur 
die objekt- und botschaftsorientierte 
Sprache CTALK" (in German), Diplomarbeit der 
Universitat Stuttgart (1985). 

\[Tsurumaru 84\] Tsurumaru, Hiroaki et al.: 
"A Consideration about the Development of a 
Semantic Dictionary" (in Japanese), 50th 
Conference of IPSJ, 5K-7 (1985.3). 

\[Yokoyama 77\] Yokoyama, Shoichi: 
"Preparation for the Database Management of 
a Japanese Dictionary" (in Japanese), Bul. 
Electrotechnical Laboratory, Vol.41, No.t1, 
pp.855-868 (1977). 
