Rhetorical Relations, Action a,nd Intentionality in Com:ersation* 
Da,vid R,. Tra,um 
Computer Science Department 
University of Rochester 
Rochester, New York 14627-0226 USA 
tramn(~)cs.rochester.edu 
Abstract 
This paper contains an investigation of the relationship between rhetorical relations and 
intentions. Rhetorical relations are claimed to I)e actions, and thus the prolmr objects of inten- 
tions, although some relations may occur \])e independent of intentions. Explicit identification of 
particular relations is shown to I)e not always m~cessary when l.his information~ can be cal)tnred 
in other ways, nevertheless, relations are often useful I)oth in plamfing aml r~cognil,ion. 
Rhetorical Relations and Intentions 
There are a number of different types of relationships between l)roxima.te segments of language, as 
well as a number of terms used to dif|erentiate groups of these types. For the present l)Url)oses 
I make no distinction (as do, e.g. \[Sidner; Moser a.nd Moore\]) 1)etw(,en afflictive and other types 
of discourse coherence relations. Since my primary interest is lit conversation (rather than, say, 
single producer text), I also include relations between utterances by (lilli:rent sl)eakers. Another 
dimension along which a distinction couhl be made is what exactly i.s being related'/ Are relations 
between explicit spa.ns of language text or between elements of the expressed contents of texts or 
perhN)s mixes between these categories? Aga.in, I will not make any such distinctions here, and 
simply use the term rhetorical relations to refer to any of these relations. 
As with other types of linguistic meaning, there are both Semantic and Pragmatic views of 
rhetorical relations. Semantic views concentrate on whether or not a relationship can be inferred 
fl'om features of the discourse and whal: other in|brmation can 1)e inferred when the relation hol(Is 
or does not. Pragmatic views are more concerned with how a rela.ti(mshil) is established and what 
effects this relationship (or lack of relationshil) ) has on the context) For the purposes of engaging 
in conversation and relating rhetorical relations to intelltiorts, I take the pragmatic viewpoint as 
more central, although both are iml)ortant. 
From the pragmatic viewpoint, rhetorica.\] relations are the same general kind of thing as speech 
acts or actions in generM: al)stractions of agent-directed change in the worhl, in this case change in 
the conversational and metM state of the col.,versallts. The only difl'(qence between speech acts and 
rhetoricM relations is that the latter a.re exl~lMtly concerned with the linka.g¢, of separate segments 
of language. \['I~'aum and Hinkehnan, 1992\] presents a nmltistra.tal tlteory of Conversation Acts, 
*This material is ha.ted upon work supported in part by the NSF under research gr;tnt ltO. IRI-9(1\[13841, by ONR 
under research grant no. N0{I014-9(I-J-1811, alld I)y DARPA/ONR under contract Nlllllll4-.q2-J-1512. l wouhl like to 
thank James Allen and Peter Heeman for helpfid discussions. 
1This distinction iv also discussed by \[Maylmry\] a.s the distinction between rcl, tion.~ and act.i, although here I will 
use these terms more or less synollylnously. 
132 
which includes both traditional sl)eech acts and several kinds of rhetorica.l arts. There are low level 
acts which link together utterances into turns a.nd content into units which axe acknowledge(i a, nd 
understood together, as well as a level of a~yumt:nlation acts which link together 1)roposition level 
and larger illocutionary acts. Argumentation acts include act types such a,s ans'wcriug a questiou, 
as well as most traditional rhetorical relations. 
Actions play a useflfl role as ,% concel)tual intermediary between bundles o\[" <)l)serve(l features on 
the one hand, and stereotyl)ica,1 causes a.nd efl'ects. They are ;Llso the \])rol)(~r ()l)jects of intentions, 
and play a central role in planning, l)lan execution an(I l)lan recognition. PI;i.nning is the process 
of selecting a set of actions which, when 1)erforme(l under the prol)er conditions will lea,(l to a. 
desired state of affairs, or goal. Plan execution is the process of pertbrming the designated actions, 
monitoring them for desired effect ~uld repairing or replanning in case of a, l)rol)lem. Plan recognitiou 
is the process of reasoning ahout an agent's menta,1 state (including intentions, beliet~, an(1 goals) 
based on observed actions in context. 
Intentions are commitments towards a course of action. \[Bratnlan, 1990\] discusses three roles 
that intention plays in deliberative behavior: serving a,s a motivation for l)lanning, a "filter of 
admissil)ility" on plans and 51rther intentions, a,nd a controller of (:on(luct, motivating execution 
monitoring and tel)air and rel)lanning when necessa.ry. 
Rhetorical relations are thus actions in the worhl (listinguishe(l 1)y conditions on their occurrence 
and effects, which will generally l)e changes to the conversational state all(I the I)eliefs of the 
conversants. Relations can also be 1)lanned, intended, performed, and recognized. As with other 
actions, relations can be performed intentionally or incidentally. Of intentiona.l a.ctions, it is also 
possible to draw the distinction made in sl)eech a.ct theory between illocutionary acts, those in which 
part of the intended effect includes a.n awa.reness on the t)art of the hea, rer ()1" this intention, and 
pcrlocutiouary acts, in which it is only the efi~ct that ma.tters a.n(l not recognition of the intention 
\[Austin, 1962\]. 
For non-illocutionary acts, the intention of the speaker is not relevant - these actions can be 
produced as side-effects of the speaker's intention, so that a determination of the intention is not 
necessary to determining whether the act was 1)eribrmed. For exa,ml)le, the evidence relation of 
\[Mann and Thompson, 1987\] may hohl \])etween two text spans even if the speaker di(l not inten(I 
such a relation, all that is required is that tlm hea.rer's belief in the nucleus is increase(I though 
the understanding of the satellite. For an illocutionary act, on the other hand, the recognition of 
communicative intention is crucial to understanding. For exa.mple, the answer relation can only be 
recognized by attributing to the speaker the intention to answer the question. 
Is Identification of Rhetorical Relations Necessary? 
It is on the following point that tile main criticisnl of bounded sets of sl)e('ch acts or rhetori('al 
relations (e.g. \[Cohen a,nd Levesque, 1990; G rosz a, nd Sidner, 1986; Lochba.u m; I! ughes a,nd Mc(',oy\]) 
is centered: intuitively, all that is needed tbr successful communica,tion is tha.t the hearer understand 
the speaker's end intentions, not that the a.ct types themselves be rec()gniz(,d. This intuition, along 
with the lack of general agreement on the precise set of acts or rela.tions lead some to reject the 
utility of relations aJtogether and concentra.te only on intentions. While I ha.re some sympathy fi~r 
this view, relations are often convenient for inh~re,tia.l purl)oses. It is not so iml)ortant that the 
particular set of relations use(I 1)e the "right" set, or even tha.t the s(,t be sha.r(~d by tim discourse 
participants, a.s long as both pa.rticipa.nts can r~a.ch a. ha.sic a.gre(~nmnt t)n t.h(~ inl.~md~d ~fli~rts. Still, 
relations prove to be a. conveni(~nt intcrm(~(lia.ry b(q.w(~cLi rea.soning a.bout high leve| int, entions a.nd 
actual surf~tce forms. 
In light of the preceding discussion, the question of whether rhetori(:al rela,tions must be identi- 
: 133 
fled can be broken down into two subquestious. On the one ha.nd, if we are talking about whether 
the semantic relationship is inferable from the resulting discourse representa,tion, the answer must 
he yes for illocutionary relations. On the other hand, it may not 1)e necessary to explicitly rec- 
ognize the act itself in the interpretation process - this knowledge may be iml)licit within the 
representation. 
As an example, \[Moser and Moore\] discuss a ro-specificaliou relation which hohls between two 
propositions which refer to the same entity. Thry only consider relati~us which "must be recognized 
in order to understand the discourse", yet it seems l)erfectly pla.usil)l(, to claim that an agent couhl 
recognize that the propositions pl and 1)2 refer to the same entity E without necessarily representing 
or noticing any relationship 1)etween pl and p2 (though this relationship would be deducible from 
the individual designations of pl and p2). 
Rhetorical Relations in the TRAINS System 
In the TRAINS Conversation System implenmntation \[Allen and Schubert, 1991\], we take a fairly 
pragmatic approach towards rhetorica.1 relations. Those relations that are conventiona.lly signalled 
by surface features (e.g. by clue words such a.s "so", "no", "okay", purpose clauses) are hypothesized 
by the Speech Act Interpreter \[Heeman, 1993\] and used by the DiMogue Manager \[Tra.um, 1993\] 
to guide further interl)retation. In the case of more implicit relatiol,ships wr often (Io not identify 
the precise relation, merely operating on the sl)ee('h act level tbrms. ()f (:ourse, r(,la.tions (:ould be 
id(mtilied based on how the content is treated with rcspect to l)revi(~us conteut, but that doesn't 
seem helpful presently. This is particularly true tot subject matter relations. For example, a 
purpose clause is nseflfl ior the domain plan recogMzer \[Ferguson anti Allen, 1993\] in incorporating 
new content into an existing (partial) plan, but in the absence of such a cue, the recognizer will 
still try to connect the new content to 1)revious content. It wouhl then 1)e possible to deduce the 
relations that this item hohls with 1)revious items, but we currently see no need to do this. As an 
example, consider the following possible pair of utterances in a situation in which the expressed 
content is not already in the current plan~ but is easily incorl)orated: 
(1) Move Engine E1 to Dansville to pick Ul} the boxca,r there, ... 
(2a) Move Engine E1 to Dansville, 
(2b) pick up the boxcar there .... 
In (1), the "to" clause would lead the Sl)eech a.ct interpreter to pr(q)ose a PURI)OSE relation, 
while no Such relation would be proposed fi)r (2). (1) might have wider applicability (e.g. in a 
case in which it is already known tha.t the boxcar a.t Dansville will be moved) and may be easier 
to incorporate into a current plan structure (e.g. for (2a) the planner might choose a different 
reason for moving El), but in the current case these utterances would end u l) with identical plan 
structures. Now if relation identification were an explicit task of the system, we could say that a 
purpose relation was deduced (there a.re of course other possibilities, such as SEQUENCE), though 
currently we see no need to assign a particular rela.tion or set of relatiolLs. 
Conclusions 
Smuming up, rhetorical relations, like speech acts, are abstract actions and thus tile prol)er object 
of intentions. They may realize intentions, Mthough, Like other actions, some may be performed 
without having been intended (e.g. as side effects of actual intentions). Rhetorical acts have 
"134 
as their direct effect a change in hearer's belief (and speaker's belief about hearer's belief, etc.). 
The conversational and mental state will also be cruciM in determining whether relations actually 
hold hetween segments of language. Identification of relations can often be usefltl in discourse 
understanding, but is not Mways strictly necessary, as several sets of relations Inight lead to the 
same results, and often it is not possible to identit), the particular relation uutil after calculating 
effects, in which case there is no flu'ther need tbr labelling the relation. Which set of relations should 
be part of the working ontology is still an open (luestion, Mthough I wouhl suggest that an approach 
toward this question wouhl be to use only those relations which seem useful in interpreting and 
producing text. 

References 
\[Allen and Schubert, 1991\] James F. Allen and Lenha.rt K. Schubert, "The TRAINS Project," TRAINS 
Technical Note 91-1, Computer Science Dept. Uniw~rsity of Rochester, 1991. 
\[Austin, 1962\] J. A. Austin, How f.o Do Th.iug.~ with Word.~', Harvard University Pr~ss, 1962. 
\[Bratman, 1990\] Michael E. Bratman, "What Is hltention?," In P. R. Cohen, J. Morgan, and M. E. Pollack, 
editors, Intentions in Communieaiiou. MIT Press, 1990. 
\[Cohen and Levesque, 1990\] Phillip 1(.. C, ohell aud llector .\]. Levesque, "Ra.tional hlteraction ms tile Basis for 
Communication," In P. R. Cohen, J. Morgan, and M. E. Pollack, editors, lul~Tl.tiou..s in Communicatiou. 
MIT Press, 1990. 
\[Ferguson and Allen, 1993\] George Ferguson and James Allen, ~.~ .nerlc Plan Recognition for Dialogue 
Systems," In DARPA Human Languagf:s Technology Workshop, March 1993. 
\[Grosz and Sidner, 1986\] Barl)ara Grosz and Candice Sidner, "Attention, Intention, and the Structure of 
Discourse," CL, 12(3):175-204, 1986. 
\[Heeman, 1993\] Peter Heeman, "Speech Acts and Mental States in Task-Oriented Dialogues," In Working 
Notes AAAI Spring Symposium on Rcasouiug about M~:nlal Stal~:s: Formal Thcor'ics and Applications., 
pages 68-73, March 1993. 
\[Hughes and McCoy\] John Hughes and l(athleen McCoy, "ObserwLtions and Direct, ious ill Text Structur~," 
This volume. 
\[Lochbaum\] Karen Lochl)aum, "The Need for hltentioaally-Based Apl)roaches to Language," This volume. 
\[Mann and Thompson, 1987\] William C,. Mann and Samlra A. Thompson, "Rhetorical Structure Theory: 
A Theory of Text Organization," Teclmical Report ISI/RS-87-190, USC, hlforma|.ion Sciences Institute, 
June 1987. 
\[Maybury\] Maybury, "On Structure ,g:. hltelltioll," This volume. 
\[Moser and Moore\] Megan Moser aml Johamm Moore, "Au Invest, igatioll of Discourse Relations Expressed 
by Lexical Markers," This volume. 
\[Sidner\] Candace Sidner, "On Discourse Relatiolls, Rhetorical Relations and Rhetoric," This volume. 
\[Traum, 1993\] David R. Traum, "Mental State ill the TRAINS-92 Dialogue Manager," In Working Not¢:s 
AAAI Spring Symposium ou R~:asoniug abo,ul, M~'ulal States: Formal 7'h¢:orics and Applications., pages 
143-149, March 1993. 
\[Traunl and tliukehnan, 1992\] David R. Traum and Elizalmth A. Hillkellnau, "Conw~rsation Acts in 'r~mk- 
oriented Spoken Dialogue," Computational Iu.tdlig~.ucc, 8(3):575-599, 1992, Special Issue on Non-lit~;ral 
language. 
