TOPIC-COMMENT STRUCTURE OF TEXTS 
(AND ITS CONTRIBUTION TO THE AUTOMATIC PROCESSING OF TEXTS) 
Ferenc Kiefe r 
KVAL, Stockholm, 
116 45, S6dermalmstorg 8 
Abstract 
In general, topic is informally defined as that 
part of the sentence which the sentence is about 
and comment as what is said in the sentence. 
This is, of course, not a formal definition but 
so far noone has succeeded to provide a viable 
formal definition for these notions. In the 
analysis of sentences and texts these notions 
can nevertheless be used with considerable 
Success. 
Typically, topics are noun phrases and com- 
ments are open sentences (that is, sentences 
containing a variable). For example, the 
sentence John is writing a letter is about John 
and it says that he is writing a letter. Formal- 
ly, 
(i) Johnx: (x is writing a letter) 
where John is a variable binding expression and 
is writing a letter is an open sentence in 
which the variable x is bound by John. 
A sentence may have more than one topic. The 
sentence John saw Mary in the garden may be a 
statement about John and Mary and what is said 
about them is that he saw her in the garden. 
Formally, 
(2) \[John x ~ Marry\]: (x saw y in the garden) 
If a sentence has only one topic this topic is 
very often (but not always) the subject of the 
sentence. Consider the sentence John wants to 
see Mary with heavy stress on John (the sen- 
tence can, for example, be considered as an 
answer to the question Who wants to see Mary?) 
This sentence is evidently about Mary and says 
that John wants to see her. 
Texts are built up from sentences and the topic- 
comment structure of a text will consist of the 
topic-comment structures of its constituent 
sentences. The topic-comment structure of 
texts is, however, not simply a cumulative 
function of the topic-comment structures of the 
constituent sentences. I shall present in my 
paper a series of observation concerning the 
topic-comment structure of texts on the basis 
of a lega, text (Rules of the Arbitration Insti- 
tute of the Stockholm Chamber of Commerce). 
(i) Each text has a set of terms which are not 
defined (not introduced) in the text: they are 
taken for granted. These terms belong to the 
stock of common knowledge with respect to the 
given text (SCK). SCK defines the frame of the 
text (or its scenario). Thus, for example, in 
the text in question the following terms belong 
to SCK: Stockholm Chamber of Commerce, 
The Swedish Law of Arbitration, the parties, 
the arbitrator(s), the (arbitral) tribunal, the 
claimant, the respondent, the request for 
arbitration. It is a characteristic feature of 
texts that topics belonging to SCK are empty 
(non-informative) topics. If such a topic is 
found it should in general be replaced by a 
more informative topics. The latter can be 
reconstructed on the basis of the preceding 
text. For example, 
(3) Each party shall be given an opportunity to 
state his views. 
Taken in isolation, this sentence could be 
analyzed as having the topic each party and the 
comment it shall be given . . . In the present 
context, however, the sentence is about the 
interpretation and correction of an award. 
Sentence (3) must thus be reconstructed as (4). 
(4) Each party shall be given an opportunity to 
state his views on the award (or the inter- 
pretation of the award). 
The topic of (6) is, then, the views on the 
award and it is stated about this topic that each 
party shall be given an opportunity to express 
it. 
(ii) In several cases the topic of the sentence 
is given by the heading of the whole paragraph. 
On the surface the topic of sentence (5) is 
the Institute. 
(5) The Institute may request a party to 
amplify any submission to the Institute. 
However, as indicated by the heading of the 
paragraph, what is at stake here is the amplifi- 
cation. The question which one would as about 
(5) is (6): 
(6) What about the amplification of a submis- 
sion? 
240 
(iii) Typically, the topic of a sentence is a 
definite noun phrase and a term introduced in 
the comment is an indefinite noun phrase. 
Taken in isolation this observation may lead to 
incorrect analyses. Consider, for example, 
(7) The claimant shall pay a registration fee 
and both parties shall deposit with the 
Institute a sum to cover the costs of the 
proceedings. 
In the given context, this sentence is not about 
the claimant and the parties but rather about 
the registration fee and the deposit. 
(iv) Conditional sentences raise special 
problems. Take 
(8) If an arbitrator appointed by a party dies, 
Such a sentence may be an answer to the 
question (9). 
(9) What happens if an arbitrator appointed by 
a party dies ? 
Formally, conditional sentences can often be 
analyzed as implications: A D B. In such 
sentences, as shown by (8) and (9), the if- 
clause (=A) is the topic and the comment is that 
this clause (i.e. the underlying proposition) 
entails the consequent clause (i.e. the underly- 
ing proposition), that is, we get the structure 
(lO): 
(10) Ax: (x ~ B) 
In connection with if-topics several problems 
must be solved: the function of if-topics in 
texts, the domain of if-topics, the internal 
structure of if-topics. Each of these problems 
shall be given ample consideration in my paper. 
On the basis of the above observations I am 
going to stipulate a general strategy for the 
determination of the topic-comment structure 
of texts. In addition, I shall briefly report on 
an experiment, made at KVAL, concerning the 
implementation of a model based on topic- 
comment structure. In this model questions 
are answered by means of topics. Finally, I 
shall present some arguments in favor of the 
approach chosen. 
241 - 
