SUBORDINATE CLAUSES AND BELIEF - DOMAINS IN VERBAL 
INFORMATION PROCESSING 
Istv~n B~tor~ 
EWH Koblenz, Sen~nar fur Informatih, Sehwerpunkt Lin6~tettk 
Rheinau 3-4, D-5400 Koblenz 
There is agreement among ltn6~/stB , logicians and 
computer linguists on the treatment of belief-statements as 
embedded, hierarchical structures. On the syntactic level of 
analysis lin6ulsts treat belief-statements as subordinate 
clauses (S-over-S-structures, LYONS 1977, KAPLAN and BRESNAN 
1981, etc.). 
On the semantic level of analysis, following the lo~o- 
81 tradition, they are manifested as predicate formulae, in 
which full propositions are used as arguments in higher pre- 
dicates (e.go CRESSWELL 1973). In computational treatments 
acoordin~ly the recovery of the hierarchical propositional 
structures is considered as primary objective: The informat- 
Ion conveyed by a belief-statement is presumed to be stored 
(or retrieved) in a data base as some kind of nested (S-over- 
-S) structure (e.g. RUSTIN 1973, EISENEERG 1977). Consider 
the sentence (borrowed from Lyons, slightly modified): 
(1) Hr° Smith bel'ieves that professor Brown Is the Dean. 
The syntactic analysis reveals the followin 6 structure: 
(2) S (Nr(N(Sn~Ith) ) VP(V(believe) 
S (NP(N(professor)N(Brown))VP(V(be)NP(Det (the)N(Dean)) ) ) )) 
The semantic structure is very sim~larz 
- 25 - 
(3) Pl(bslieve (Smith, P2(be (nrofeseor Brown, Dean)))) 
The CL-aneZysts comprises acoordLngly the followLng lo~Lcal 
etepss 
(4) a) recover the structure of Pl o) process the structure Pl 
b) recover the structure of P2 d) process the structure P2 
This scheme will be referred in the following as the 
standard model. 
Due to unsolved problems in evaluation of composite 
predicates and to the extremely high computational costs 
embedded structures are avoided in application systems (such 
as USL, ~DIS, HAM-RPU, of. BOLC 1980). Moreover the super- 
imposed predicate appears from the point of view of the 
(embedded) lower predicate as purely accidente~. The dome~n- 
-concept of beliefs avoids some of the difficulties of the 
8tande4rd model. Considering the-basic con~nunicative function 
of human language the information conveyed by a verbal state- 
ment should be decomposed into two components: 1. semantic 
(material) information and 2. modal information conoer~ng 
belief-statue of the statement (UNG~ 1972). The fundam- 
ental difference between the two kinds of information should 
be reflected in the linguistic design and should be taken 
into consideration in models for man-machine con~nunioation. 
Accordingly the comunication partners dispose of belief- 
-registers in addition to the information registers (as it is 
the case in current systems). The belief-register provides a 
belief-~ent and a belief-value (belief-oertitute) for each 
statement received. It is natura~ to i~itialize at the beginn- 
ing of a discourse a number of belief-dome£ne for the EGO, 
the PARTNER and the PERSONS MENTIONED. Notice that these 
processing frames (in the technical sense of the word, like 
in ~TZING 1980) are necessary in any system aiming at the 
same degree of sophistication. 
- 26 - 
Zn case of overtly marked belief-agents the objective 
of analysts is the proper asstg~aent to a belief-don~tn and 
not the recover7 of a structure. This means that expensive 
reoover~j procedures can be dispensed with, since the surface 
structure usually provides tnnedAate keys for the proper 
assignment of belief-agents. Each belief-s~ent has his own 
domain of beliefs. In actual oom~unAcation the appearenoe of 
a partner A initializes a do,wAn of A'. As soon as in the 
conversation further actants B, C, D etc. occur there will be 
ocr~esponding new belief-do~ains B°, C°, D" etc. initialis- 
ed. Statements overtly ~rked in view of a belief-agent will 
be transferred to the corresponding donntin. 
For u~arked statements a general strategy of assi~ent 
of default values can be developed along the lines of Gri0e 
and of MEGGLE (1981). Havi~ initialized the appropriate be- 
lief-frwne with the appropriate belief-a~ent, there is a 
particular domain of each belief-agent. In the case of (1) 
there is a belief-do.An called "world-of-M~-Smith" within 
the ttuiverse of discourse. The prooesaing involvess 
(5) a)lo0ate belief-domain by belief-s~ent key (- Mr Smith) 
b) enter/retrieve p in the activated belief-don~in and 
o) determine ourx~nt de~'ee of certitude (~ to believe) 
The present concept of belief-donuiins should be regarded in 
a more general procedural view of language, such as presented 
in B~TORI (1981). The 8u~ested treatment of belief-domains 
as pointer based areas in DB is another instance of prooedux~- 
al solutions, in which d~nsmic language structures turn out 
to be simpler than their static, purely representionally 
oriented description. On the linguistic level of analysis the 
domain concept of beliefs is motivated by the observation 
that the superimposition of belief-structures on statements 
does not make the comprehension of these sentences more diff- 
icult, at least not in the measure as the processing of the 
embedded structures in the standard model would let this to 
- 27 - 
expect. Even if storing S-over-S structures cannot be dispens- 
ed with entirely, the number of their occurrences in actual 
analysis-procedures can be substantially reduced. Notice that 
the concept of belief-domains provides a framework to treat 
opaque references i~ a natural way. 
In the final version of the presentation it will be 
attempted to substantiate the proposed model on the treatment 
of belief-agents and esp. of their derivation from agents and 
subjects across clause boundaries. 

References

B~TORI. lStv~n: Die Grau~natik aus der Sicht kognitiver Pro- 
zesse. ~bingen, Gunter Narr 1981 

BOLC, Leonard (ed.): Natural Language Based Computer Systems. 
Munohen, Hauser 1980 

CRESSWELL, M.I.: Lo@ics and Languages. London, Methue~ and Cos 
1973 

EISENBERG, Peter (ed.): Semantlk und K~mstliche Intelligenz. 
Berlin. Nalter de Gruyter 1976 

KAPIAN, Ronald und BRESNAN, Joan W. (in press): Lexical- 
Functional Gran.nar: A formsl system for grammatical re- 
presentation. I.~N: BRESNAN, Joan W. (e~) The mental re- 
presentation of g~smmatical relations, Cambridge, ~IT 
Press 

LYONS, John: Semantics. Cembridge University Press 1977 • 

MEGGLE, Georg: Grundbegriffe der Ko---unikation. Berlin, Wal- 
ter de Gruyter 1981 

tr~TZING, Dieter (ed.): Frame of Conceptions and Text Under- 
standing. Berlin, Walter de Gruyter 1980 

RUSTIN, Randall (ed.)" Natural Language Processing. New York, 
Algorithmics Press Inc. 1973 

UNGEHEUER, Gerold: Sprache und Kommunikation. Hamburg, Helmut 
Buske 1972 
