Closing the Gap Between Discourse Structure and 
Communicative Intention 
Paul Wu Horng Jyh 
National University of Singapore 
Singapore, 0511 
The Republic of Singapore 
E-mail: paulwu@iss.nus.sg 
1 Introduction 
Ill tile past, quite a few phenomena related to discourse structures have been studied, such as lexica\] 
cohesion (HMliday and Hasan, 1976), coherence relation (Hobbs, 1985; Wu and Lytinen, 1989), and 
rhetorical relation (Mann and Thompson, 1988). On the other hand, there Mso exists al)uadaut 
research on communicative intention such as speech act theory (Searle, 1968) and relevance maxim 
(Wilson and Sperber, 1986). One logical question to ask is then: 
What is the relationship between discourse structures and communicative intention? 
In tile following, we shall l, ake a.n objective approach to bridging tile gap between these two s(~ts o\[' 
I)henomena. More specifically, certain discourse structures and communicative intentio,,s are tak('~,. 
as literally as possible, from their "native" theories; Then, without any i)redefined fram(,wol'k, w(' 
observe the emergence of an integral discourse model encompassing these structures and iJltellt.i()lls 
via necessary "bridging" knowledge and inferences. Thus, on the one hand, our integrative vi(,ws 
toward discourse structures and intentions are similar to (Grosz and Sidner, 1986) and (Moore a ll(I 
Pollack, 1992) in the sense that we hold 
• The separation of attentional and intentional "tracks" of discourse \] . 
• The stratification of coherence (informational) and rhetorical (intentional) rela.ti(Jlls. 
On the other hand, unlike previous work, we formulate knowledge sources which expl:,i.l, how the, 
emergence of integrated discourse model is possible, rather than just identifying such integrati()=~. 
hi particular, the following knowledge sources are identified: 
KN-LC which associate concepts denoted by words (anchoring Lexi(:a.l Cohesion) 
KN-SA which interpret utterances in terms of the speaker's mental models whi(:h consist or I)r(,I)~,- 
sitions embedded in attitude contexts (anchoring Speech Acts) 
KN-CR which specify a. domain theory of physical world (anchoring Coherence Relal.ious) 
KN-RR which specify a domain theory of mental dynamics (anchoring Rhetorical Relatiom~s) 
The above four knowledge sources in turns are governed by the ultimate principles of R.elevall('(' 
Maxim (KN-RM). In short, by explicating the knowledge behind discourse structures and illt('tlti(,Hs. 
we a.(lvoc~tte that the stratification of discourse structures and intentions is motivatod/stll~l)~='t(,~l 
by modularized, but related, knowledge sources. 
1This, howcver, will not bc discussed here; interested readers are referred to (Wu, 1993) for some' (I dis~'u.~siol~. 
152 
2 A concrete example 
In the following, we propose a sequence of specific questions concerning lexica.I cohesions (I,('). 
coherence relations (CR), rhetorical relations (RIR), speech acts (SA), and conversa.tioual maxiu~s 
(C,M); more specifically, we intend to demonstrate the following strata 
• X coherence relation encompasses Y lexical cohesion (CR > LC). 
• X rhetorical relation encompasses Y coherence relation (RR > CR). 
• X speech act is achieved by Y rhetorical relation (SA > RR). 
• X relevance maxim is obeyed by Y speech act (RM > SA). 
Let's look at an example which is composed for an independent purpose; It is taken froln (Man. 
and Thompson, 1986): 
($1) I love to collect classic automobiles. 
($2) My favorite car is my 1887 Duryea. 
First, let's observe that the semantic/propositional interpretation of SI and $2 yields th(, lid 
lowing (main) propositions: 
LOVE(l, { a, I OLD(a) & AUTOMOBILE(a) }) 
FAVORITE(I, b), CAR(b), b = 1887-Duryea 
That is, I love antique automobiles and my favorite car is the 1887 Duryea. By detecting the 
following lexical cohesions (between the words denoting the concepts): 
KN-LC-I: 
'CAR' is a synonym of 'AUTOMOBILE' 
'FAVORITE' is similar to 'LOVE' 
We conclude the following coherence relation, 
Exemplification(S2, S1) - (F1) 
since as (Hobbs, 1985) defines it, we can 
KN-CR-I: 
Infer p(A) from S1 and p(a.) from $2, where a is a member or subset of A 
in our case, 'p' is 'LOVE' (or 'FAVORITE') and 'A' is the set of automobiles a.nd 'a' is th(' ('a.,' (i,y 
1887 Duryea). 
Now, we have demonstrated the connection between coherence relations a.nd lexical cohesi(,lls. 
Let's look at how this is connected to rhetorical relations and speech acts. First, we need to (,xl.eml 
the interpretation of $1 and $2 from purely propositional to including prol)ositional attitudes (( 'f. 
(Cohen and Levesque, 1990), (Wu and Lytinen, 1991) and (Wu and Lytiuen, 1992)), 
KN-SA-1: 
$1 = BEL(Spkr, LOVE(Spkr, { a \[ OLD(a) & AUTOMOBILE(a) })) (1,'2) 
$2 = BEL(Spkr, (FAVORITE(Spkr, b), CAR(b), b = 1887-Duryea)) 
153 
That is, the speaker (spkr) 'believes,' as well as 'stated,' that S1 and $2. In order to conceive th,' 
intention behind the speaker's making $2, it is reasonable to assume that 
MBEL(Hr, Spkr, ?BEL(Hr, LOVE(Spkr, { a\] OLD(a) & AUTOMOBILE(a) }))) - (F3) 
That is, the speaker and the hearer (Hr) mutually believe (MBEL) that the hearer weakly believes 
(?BEL) S1. Thus, in order to facilitate the "acceptance" of S1, the speaker uses $2 as an 'evidence' 
to support S1. That is, the following speech act and rhetorical relation are achieved simultaneously. 
ASSERT(Spkr, S1) and Evidence(S2, S1) 
The Evidence rhetorical relation is achieved through associating the following knowledge concerning 
persuasiveness and the Exemplification coherence relation (an instance of which is recognized in 
F1): 
(KN-RR-1) 
IF the speaker provides more detailed information about a fact P (such as stating an 
example of P) 
THEN the hearer is more obliged to believe P (because the speaker really wants him to 
believe P) 
Once the Evidence rhetorical relation is recognized, the hearer recognizes that the speaker really 
wants him to believe P from the THEN-part of KN-RR-i: 
WANT(Spkr, Hr, BEL(Hr, LOVE(Spkr, ( a\[ OLD(a) & AUTOMOBILE(a) })))- (F4) 
Based on F2, F3 and F4, the hearer has recognized, according to Searle, the sincerity, the prep¢tra- 
tory, and the essential conditions of the Assertion speech act, respectively. Together with the 
Evidence rhetorical relation, it is clear that the speaker is making an assertion. 
Now, the last question remains as which relevance maxim has been obeyed during the a bov,~ 
process. According to (Sperber and Wilson, 1986), relevance is obeyed by producing the following 
contextual effects (KN-RMs) (1)enhancing a statement (2) canceling a statement (3)introducing ;t 
statement. Thus, in $2, the speaker has addressed the relevance by producing a definite contextual 
effects to enhance the certainty of a proposition. 
3 Discussion and conclusion 
In summary, we have demonstrated through tracing detailed examples that an integrated model of 
discourse does encompass discourse structures and intentions. We also take an objective approach 
in arguing so, since we do not predefine any framework to demonstrate our thesis; Rather. we 
follow existing theories and by identifying the necessary bridging knowledge and int~erences, we 
just observe the integration emerging out of the individual discourse structures and intentions ;is 
defined in their native theories. 
This integrative view is similar to others, e.g., (Grosz and Sidner, 1986) and (Moore and Pollack. 
1992). Due to the lack of space, we did not deal with the full ranges of issues, but many of our 
positions are identical to others, in particular: 
1. Discourse structure and intention form a strata 
154 
2. Many to many mapping exists between adjacent layers of the strata, e.g.. I)etween coh('reHC,' 
and rhetorical relations for both text generation and interpretation 
3. The nucleus and satellite, if exist, may not coincide between two adjacent layers. 
What we have demonstrated in this paper is a particular (one-to-one) instance of a strata o\[' 
discourse structure and intention. To be more specific, we demonstrate how a lexical cohesion gives 
rise to coherence relation; a coherence relation to a rhetorical relation; then to a speech act and a 
relevance maxim. In doing so, we feel we have extended the research in the following ways 
• Address a wider range of strata; not only between coherence and rhetorical relations, but a.ll 
the way from lexical cohesions till relevance maxims; 
• Formulate knowledge which explain the emergence of such connections between layers o\[' 
discourse 
The second point has been kept in the background in many previous research. For example, now 
it can be readily understood that an Exemplification coherence relation can give rise to an Evi- 
dence rhetorical relation. However, knowledge such as (KN-RR-1) is kept implicit; but upon closer 
inspection, it becomes obvious without such knowledge the connection between Exemplification 
and Evidence can not be established. Thus, flushing out the explanatory theory for the mapping 
between discourse strata has been identified as our most urgent research topic. In particular, th(, 
knowledge in the maxim level (KN-RM) is much in need of further study, especially with respe(l 
to other layers of the strata. 

REFERENCES 
(Cohen and Levesque, 1990) Cohen, P. and Levesque, H. "Rational Interaction as the Basis for C:ommunica- 
tion," In Cohen, P., Morgan, J. and Pollack, M. (Editors) Intentions in Communication, MIT Press, 1990. 
(Grosz and Sidner, 1986) Grosz, B. and Sidner, C. "Attention, Intentions, and the Structure of Discourse." 
Computational Linguistics 12, 1986. 
(Halliday and Hasan, 1976) Halliday, M. and Hasan, R. Cohesion in English. Lognman, 1976. 
(Hobbs, 1985) Hobbs, J. "On the Coherence and Structure of Discourse" (Draft) In Polanyi, L. (Edil.or). 
The Structure of Discourse. Ablex. 
(Mann and Thompson, 1986) Mann, W. and Thompson, S. "Relational Propositions in Discourse." Dis~,~,',, 
Processes, 1986. 
(Mann and Thompson, 1988) Mann, W. and Thompson, S. "Rhetorical Structure Theory: Toward a fun,- 
tional theory of text organization," Text 8, 1988. 
(Moore and Pollack, 1992) "A Problem for RST: The Need for Multi-Level Discourse Analysis," Coml~tda- 
tional Linguistics, vol.18, no.4, pp.537-544. 
(Searle, 1969) Searle, J. Speech Acts: An Essay in the Philosophy of Language. Cambridge Uniw~rsity Press. 
1969. 
(Sperber and Wilson, 1986) Sperber, D. and Wilson, D. Relevance: Communication and Cognition. Harvar,\[ 
University Press, 1986. 
(Wu and Lytinen, 1990) Wu, P. and Lytinen, S. "Coherence Reasoning in Persuasive Discourse." Pro,.. ,'1 
The Twelfth Conference of the Cognitive Science Society. Cambridge, 1990. 
(Wu and Lytinen, 1991) Wu, P. and Lytinen, S. "Attitude and Coherence Reasoning in Persuasiv," I)i.-- 
,:ourse," Proc. The 1991 AAAI Spring Symposium on Argumentation and Belief. Stantbrd, 1991. 
(Wu and Lytinen, 1992) Wu, P. and Lytinen, S. "Attitude Emergence- An Effective \[nterpret;ttion S,'b,'lll,' 
for Persuasive Discourse," Proc. of The Fifteenth International Conference on Computational Linglti.~/i, ~. 
Nantes, France, 1992. 
(Wu, 1993) Wu, P. H-J. "What's Going on in these Advertisements? - A Case Study of Indirect Sp,',','b. 
Proc. of The First Meeting of Pacific Association for Computational Linguistics. Vancouver, 1993. 
