| FRENCH | GERMAN | DANISH | ITALIAN | ENGLISH | SPANISH | 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
arbre Baum albero tree árbol 
trae timber madera 
Holz legno leña 
bois 
monte 
bosco wood 
Wald skov bosque 
forêt foresta forest 
selva 
  
  
  
  
  
  
Table 1: Comparison of equivalents in five European languages 
an object have no meaning and make no sense. In times of 
analogue and analytical image processing this was not partic- 
ularly noticed due to the fact that the semantic processing of 
the image content fell entirely to the human operator. 
In a fully digital environment, the human operator in charge 
is assisted by the computer and will in the long run be largely 
substituted by the machine. The effort of modelling the con- 
gnitive process with the support of digital procedures leads to 
a paradigm shift. Among the many problems related to this 
entirely new approach we shall devote our attention to only 
one: the linguistic aspect. 
The intimate relationship between language and the real 
world, between concept and object, between significante and 
significado has been first analysed by Ferdinand de Saus- 
sure (1916) and very sharply defined by Ludwig Wittgenstein 
(1921): 
" The sentence is an image of reality" 
Formulated the other way round, we obtain: 
" The reality of an image is verbalisation" 
An image, be it mental or digital, abstract or concrete, only 
acquires sense when the semantic information contained in it 
can be communicated. 
Analysis and modelling of semantic image information de- 
pends largely on language. The question is to find out how 
far modern linguistic and language theory may be of help 
in structuring semantic problems at image processing (Rapp, 
1995). 
2 LINGUISTICS, SEMANTICS AND FIELDS OF 
KNOWLEDGE 
The digital image with its scale of gray values is a Carrier 
of semantic information which must be "interpreted". So is 
human language. 
Human language uses sounds and signs instead of gray values 
to convey meaning. A linguistic message is a string of mental 
images conveniently coded in sounds and signs according to 
certain rules, that will be decoded (understood) by those who 
master the code and are able to attach the correct meaning to 
such strings of sounds or signs i.e. to extract the semantic in- 
formation, reconstruct the images the emitter of the message 
had in mind. 
Differing sensibilities and capabilities as well as language and 
culture dependent constraints, make of human beings imper- 
fect interpreters of the "real world”: some cultures see - and 
8 
speak of - only three colours in the rainbow whereas we have 
learnt that there are seven; some understand mountains + 
valley as only one mental unit whereas we dissect them as 
separate entities with some misterious limit in between. If 
we add to the culturally imposed fragmentation of reality our 
individual perception influenced by personal background and 
experience, discrepancies are still deeper. 
An example of the way in which different languages segment 
part of the real world is shown in Table 1 where Umberto 
Eco's example has been expanded to include English and 
Spanish " equivalents". 
We see here that even languages pertaining to the "occiden- 
tal" world look upon Nature in different ways. What mental 
image does the concept "Wald" awake in a Dane and in a 
Spaniard when they see the word or the corresponding symbol 
on a map? Does it correspond to the semantic field of the 
German term? Is it fair to translate the Spanish " bosque" into 
the Danish "skov" knowing that both terms cover conceptual 
fields which are so different? How to know which of the Spa- 
nish words does apply when translating "skov" into Spanish? 
There is hardly any exact equivalent between the terms of 
two languages even when traditionally they are treated as 
synonymous in bilingual dictionaries. 
A great deal of attention has been devoted in terminology 
to the definition of the field of knowledge so as to estab- 
lish the structure of conceptual systems (the semantic space 
corresponding to individual terms or concepts). In this area 
in particular, thinking is still largely dominated by the tradi- 
tional pen and paper processing techniques and by the rela- 
tively simple relationships that have sufficed until now (Sager, 
1990). Here we have another paradigm shift which has not 
been fully recognised and assimilated. 
Communication sciences and computation have come to the 
help of linguists opening new horizons in the study of lan- 
guage: modelling has entered linguistics. A model of knowl- 
edge is conceived in linguistics as a multidimensional space 
in which intersecting axes represent some kind of conceptual 
primitives or characteristics. They may also be seen as fea- 
tures or components. A concept, i.e. a unit of knowledge, 
can thus be represented and identified by reference to its co- 
ordinates along each axis. The sum of all the values with 
respect to each axis is then equivalent to defining the unique 
position of the concept in the knowledge space. 
There is assumed to be a certain degree of dependency be- 
tween dimensions - the characteristic "square" may limit the 
possibility of attribution of other primitive characteristics by 
rejecting a simultaneous characteristic of "round". Some 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B6. Vienna 1996 
struc 
may 
ted, 
ship: 
cept 
the 
strai 
tion. 
conc 
a rai 
ther: 
set c 
We : 
natu 
of th 
be v 
1: tl 
The 
mati 
In Li 
" Cox 
The 
age 
Othe 
grap 
i.e. 
Both 
nizal 
Sem: 
trans 
resul 
the € 
ter 
the i 
with 
for ir 
Tran 
they 
ships 
prod 
quire 
(cod 
into 
mati 
A ve 
Buag 
tion, 
and : 
spea 
modi 
enou 
port: 
acco 
ative 
whet 
(Sag 
Tran 
to q