M REPRESENTATION OF SEMANTIC INFORMATION IN GMAN 
MACHINE READABLE DICTIONARIES 
Winfried Lenders 
lnstltut fur Komm~nikatlonsforschung u. PhonetAk, Univ. Bonn 
Poppelsdorfer Allee 47, D-5300 Bonn, 
The basis of this paper is a comparative stud~ on German 
machine readable dictionaries which has been conducted during 
the last two years In connection with a research project 
financed by the German minister for science and technology. 
The aim of this project was to compare and to describe about 
12 dictionaries, used in computer systems, in more or less 
conventional language analysis and synthesis systems as well 
as in understanding systems. The comparison and description 
was oriented to the data basis, the theoretical background, 
the field of application, and especially the microstructura 
of these dictionaries. In this paper we want to discuss a 
specific problem of lexAcal mlcrostructure which has arisen 
in the context of thls project, the problem of semantic In- 
formation. This problem Is most relevant since on the one 
side all of the exAstlng machine readable dictionaries are 
relatively restricted in their semantic dimension, and on the 
other side more detailed information about the semantics of 
words is extremely necessary for language analysis and synthes- 
Is programs as well as for understanding systems. 
The first point of my discussion is to explain how the 
meaning&of German words are actually represented in the 
mlcrostructure of the Intended dictionaries. My examples are 
taken from the dictionary components of the systems SUSY, 
LIMAS, HAM-RPM, PLIDIS, BEAST/BACONp from the German word data 
- 176 - 
bank "WahrAg" and from several other more theoretical 
approaches. 
My second point Is the question how far in these approach- 
es the general IAnguIatic problem of representation of word 
meanings has been solved and which elements of the semantic 
level are normally not to be found in these dictionaries. In 
the third point I want to advocate the thesis that in a mach- 
ine readable dActlonar~j the meaning of a word can be repres- 
ented by a set of semantic relations. In connection with this 
thesis recent developments An word semantics especially An the 
fields of frame theory and semantic network theory have to 
be discussed. 
The general assumption of this paper is that "dictionary" 
- this concept Is used synonymous with "lexicon" - desJ~nat- 
es an~ collection of word elements of a specific language; 
every element of such a collection should be connected with a 
certain set of linguistic Information. This set of linguistic 
Anforuatlon Is called the m/croatructure of the dictionary. 
The mlc~ostmcture should be considered as an n-tupel of 
descriptive elements, each of which consists of an Information 
field. Examples are the field of morphological Information, 
the field of mox*phological information, the field of phonolog- 
Ical Anfomation and the field of semantic Information. The 
field of semantic information again may be regarded as a 
n-tupel of Information. The elements of this n-tupal may be 
specific relations which combine the concerned lexical element 
with other lexical elements of the Intended dictionary. - The 
object of this paper Is not to discuss how the semantic di- 
mension of an Individuals dictionary or lexicon may be organ- 
Iced but to show in what way machine readable dictionaries 
which are determined for use in different computer systems 
could be constructed In their semantic level with respect to 
the results of recent discussion in Computational Linguistics 
and A~tifAcial Intelligence Research. 
- 177- 
