Implementing an Integration of 
the Systemic Flowchart Model of Dialogue and 
Rhetorical Structure Theory 
Yuen Q. Lin and Robin P. Fawcett 
Computational Linguistics Unit, Aberconway Building, University of Wales, Cardiff CFI 3XB, UK 
e-mail: liny@cardiff.ac.uk, fawcett@cardiff.ac.uk, fax: 01222-874806 
1. Introduction 
There are two major types of language 
generation.l The first of these is monologue 
generation, which focuses on generating mon- 
ologue text, typically of paragraph length. 
The second is dialogue generation, the aim of 
which is to produce co-operative, interactive 
discourse. Typically projects in natural lang- 
uage generation focus on one or the other, but 
some researchers are now considering how the 
two fit together. This paper is one of a series 
that discuss how this may. be done, at the 
second stage of generation, i.e. the generation 
of the structure of discourse. 
The most influential approach to planning 
discourse monologues in recent years has 
undoubtedly been Rhetorical Structure Theory 
(RST). While there is much to discuss about 
this framework, for our present purposes we 
shall adopt and accept the 'standard version, as 
found in Mann and Thompson (1987) and the 
many papers that build on it. For dialogue we 
use what we take to be the most holistic and 
most explicit development of the Birmingham 
School of discourse analysis (originating with 
Sinclair and Coulthard 1975), namely the 
Systemic Flowchart Model (SFM). This was 
first described in Fawcett, van der Mije and van 
Wissen (1988). 
The present paper follows on from two 
earlier papers in particular. The first describes 
the implementation of the SFM as part of the 
COMMUNAL Project (Lin, Fawcett and Davies 
1993).2 The second is Fawcett and Davies' 
~rogrammatic paper (1993), which describes 
ow an integration of RST and the SFM might 
be attempted, The purpose of the present 
paper is to give an account of an 
implementation of the point where the two 
models are integrated - including a new system 
network through which this is achieved. We 
see no reason why other contributions that 
draw on a broadly systemic functional 
approach to modelling RST relations, such as 
Vander Linden, Cumming and Martin (1993) 
and Maier and Hovy (1993), may not be 
integrated with the present proposals. 
2. Some problems in relating RST to the SFM 
Fawcett and Davies (1993) provide summ- 
ary descriptions of the two approaches, so we 
shall not replicate those descriptions here. That 
paper also discuases some of the possible 
difficulties in integrating the SFM and RST. 
Fawcett and Davies concluded: 
We do not expect a 'seamless join' between 
the two models to occur effortlessly; there are 
bound to be difficulties of many kinds as we 
explore possible ways of bringing the two 
models together. But, given the richness and 
flexibility of the descriptive and implement- 
ational tools available to us, it is an enter- 
prise on which we can enter with an expect- 
ation of some measure of success. 
Here we shall report both the way in which we 
have successfully modelled the point in the 
overall discourse grammar at which RST and 
the SFM meet, and the surprising ease with 
which we were able to do it. 
For the Birmingham School of Discourse 
Analysis, a dialogue consists of a series of 
transactions, each of which is made up of a 
number of exchanges. These in turn have 
constituents called moves, and each move con- 
sists of one or more acts. Thus their model 
assumes that there is a rank scale relationship 
between these units (Halliday 196l). This 
concept is found in the SFM too. 
In RST, on the other hand, there is the pot- 
ential for the unlimited recursion of structures 
consisting of a nucleus with a satellite in the 
relation of elaboration or reason, etc. In the 
integration of the two to be described here we 
model this as the recursion of acts within acts. 
The SFM model of discourse structure is 
both richer and at the same time more trad- 
itional than the types of structure built in RST. 
The main difference in 'richness' is that each 
node is labelled twice. It is shown as both an 
element of structure in the unit above, and by 
the name of the unit that fills it. Thus the 
concepts of 'nucleus' and 'satellite' are treated 
here as elements of structure. The SFM is 
more traditional in that its structure is based on 
'constituency' rather than the concept of 'sister 
dependency', as in RST. This provides a richer 
and so more informative labelling, with the 
rhetorical relation shown as the class of act. 
3 A typical structure 
We will now look at a typical example of 
the sort of structure that occurs at the point 
where the rank-based structure of the SFM 
meets the potential recursion of RST relations 
- using for both a 'constituency' approach that 
has at each node both elements and units. 
41 
Ivy/R 
move: give_information 
N S 
act: inform act: elaboration 
N S N S 
act: act: act: act: 
inform background inform reason 
I II II II I 
I went to He'd agreed to It took us as we dis- 
see Fred. discuss my a long agreed on 
paper with me. time, several points. 
Figure 1: A fragment of a discourse structure 
at the point where RST and the SFM meet 
How is this structure generated? To see this, 
we must look at the new system network shown 
in Figure 2 - and also at Its associated realiz- 
ation rules, because it is through these that the 
chosen features are converted into structures. 
4 How the structure is generated 
The discourse generator that we are about 
to describe is called GENEDIS, because it 
GENErates DIScourse (For details see Lin et 
al. 1993.) Many of its choices are guided by 
the goals set in the higher planning 
component, but much of the 'discourse 
potential' is captured here. As it enters the 
network in Figure 2, it knows that it has the 
goal of filling the Respond element (hence 'R') 
of an exchange, and that it is to be uttered by 
Ivy (i.e. the system). The network operates in 
the following manner. We enter it at \[discourse 
unit\], and find that the probabilities in the 
initial system are set 100% to \[move\]. (On 
(re-entry to the network to fill out the lower 
part of the tree, however, this choice will be 
reversed.) As you will see if you look at the 
realization rules, the selection of \[move\] inserts 
first the unit 'move' into the structure, and then 
locates the element nucleus (N) at Place 2 in 
that move's structure. (We shall shortly see 
why it is at Place 2 rather than Place l.) 
We now encounter two 'simultaneous' 
stems. In the MOVE CLASS system, 
NEDIS typically choos~-s one of the five 
major classes of move. The realization rule on 
each of these inserts this into the structure (the 
relationship of 'class' being shown by a colon). 
Notice that the categories that get inserted into 
the growinu structure may have similar or even 
identical labels to the features in the network. 
In principle the two are separate, but there is 
little point in multiplying terms unnecessarily. 
In the lower system, i.e. MOVE_COM- 
PLEXITY, our choice is \[with_satellite_m\]. 
(Lack of space prevents us from describing the 
effects of choosing \[simplem\] or \[co_ordin- 
atedm\], but the rules given cover the former.) 
However the typical choice - and the most 
interesting one - is \[with satellite m\]. The 
choice of this feature aff~ts the structure in 
four ways. First, its realization rule (0.4) 
rovides for re-entry to the network to fill N. 
is this fact that provides for the recursion 
typical of RST relations in the present 
framework, as we shall shortly see. 
But this feature also leads on to two further 
simultaneous systems: one for RHETORICAL 
RELATIONS and a second for SATELLITE 
THEMATIZATION. We will take the latter 
first. The concept of 'thematizing' an element 
in the structure of a unit comes from systemic 
functional grammar, where it is mainly assoc- 
iated with the clause. We use the same term 
here because the motivation is similar in both. 
Typically, the satellite follows the nucleus, 
but the different types of rhetorical relation 
have different typicalpatterns in this respect. 
So how does GENEDIS know which to 
choose? The answer lies in the upper system, 
and the entry to this is the third effect of 
choosing \[with_satellitem\]. As you will see, 
the features here are precisely those of the 
rhetorical relations of RST. But notice that 
these features have attached to them same pass 
(sp) preference resetting rules (i.e. these rules 
are not realization rules). It is these that 
express the likelihood that the satellite will 
precede or follow the nucleus - as the sp rules 
below the network show. One advantage of 
using probabilities on features in systems is 
that they can be varied, e.g. for different types 
of genre {cp. the. prevalence of thematized 
'purposes' m certain types of instructions). In 
some cases the probabilities are absolute, as in 
our case, where the sp rule on \[elaboration\] 
makes it 100% certain that the S will follow. 
In any given instance, of course, the Performer 
may have a reason to thematize a satellite that 
overrides the probabilities, but in a sophisticat- 
ed model he/she should be able to set this 
against the knowledge of the general probab- 
iffties for a ,ziven type of rhetorical relation. 
The finaVeffect of selecting \[withsatellite 
_m\] is to provide that the network will be re- 
entered to fill S, as in Rules 0.5 and 0.51. 
We have now generated all of the structure 
of the move. This part of the network is non- 
recursive, but as we now re-enter the network 
to generate the acts that will fill the N and S of 
the move, we meet a recursive system network. 
The recursion is modelled through the realiz- 
ation rules; a new layer of structure is added 
by each re-entry to the network. And the pot- 
ential for recurslon is, in principle, infinite. 
On re-entry to the network to fill N, the 
preferences have been reset to \[act\] and 
\[nucleus _act\]. Similarly, on re-entry to fill S 
they are reset to \[act! and \[satellite_act\]. Rule 
0.7 inserts the unit act' in both cases. The 
class of act to fill N comes from the network 
tor ACT_CLASS, and in our case Rule 0.8 
inserts 'inform' after 'act'. It is Rule 0.71 on 
\[satellite act\] that determines what the class ef 
the act filling S will be. Thus, since in our case 
\[elaboration\] has been chosen on the 'mother' 
42 
discourse 
_unit 
100% 
- move < 
(o.1) 
0% 
- act-- 
(0.7) 
MOVE 
CLASS 
i give_information (0.2) 
solicit_information (0.21) 
t propose_own action (0.22) 
others 
satellite 
_act - 
(0.71) 
MOVE F simple_m (0.3) 
N COMPLEX-\[- with_satellite_m~" 
ITY \[ (0.4) " 
co_ordinated_m ... \ 
\ \[-intbrm~ 
\[-" assert'--\[ (0.08) \ 
-nucleusf AC~T L'°thers~ 
act~i CLASS I \[---exclamation 
k" 1- express~\] (0.087) 
I-others ~ others 
ACT_ F simple-a ( J 
COMPLEX-~ with satellite a t 
ITY L co_~dinat~d0~a~.: 
i elaboration) 
(sp0.41) 
purpose 
(sp0.42) 
RHETORICAL reason 
RELATIONS S (sp0.43) 
background 
(0.44) 
others 
t I satellite_ SATELLITE thrematized THEMATIZ- (0.5) ATION satellite unthematized 
(0.51) 
SELECTED SAME-PASS PREFERENCE RE-SETTING RULES 
sp0.41 : elaboration : for same_pass prefer \[satellite unthematized\]. sp0.42 purpose : for same_pass prefer \[40% satellite thematized / 60% satellite_unthematized\]. 
sp0.43 : reason : for same_pass prefer \[20% satellite_thematized / 80% satellite_unthematized\]. 
sp0.44 : background : for same_pass prefer \[ 1'% satellite_thematized / 99% satellite_unthematized\]. 
SELECTED REALIZATION RULES 0.1: move : N @ 2. 
0.2 : give_information : give information. 
0.21 :solicit_information:solicit information. 
0.3 : simple_m : for N prefer \[act, nucleus_act, simple\], for N re_enter_at discourse_unit. 0.4 : with satellite m or with satellite_a : 
if ~,ith_satdilite_a then-N @ 2, 
for N prefer \[act, nucleus_act, for N re_enter_at discourse_unit. 0.5 : satellite thematized : 
S @ 1, for S prefer \[act, satellite_act, for S re_enter_at discourse_unit. 0.5t : satellite_unthematized : 
S @ 3, for S prefer \[act, satellite_act, for S re_enter_at discourse_unit. 0.7: act:act. 
0.71 : satellite_act : 
if on_mother_pass elaboration then act_class elaboration, 
if on_mother_pass purpose then act_class purpose, if on_mother_pass reason then act_class reason, 
if on_mother pass background then act class background. 
0.8 : inform: act_class inform. 0.87 : exclamation: act_class exclamation. 
Figure 2: A system network (and its realization rules) that integrates the Systemic Flowchart 
Model of dialogue structure with the relations of Rhetorical Structure Theory 
43 
pass through the netw, ork the class of the act 
will be an elaboration. 
The final two systems are ACT_COM- 
PLEXITY and ' SATELLITE THEMATIZ- 
ATION - and it is these that ~ovide for the 
recursion. As you will see, the first is entered 
from both \[nucleus_act\] and \[satellite_act\]. 
And one of its features is \[with_satellite_a\], 
whose Rule 0.4 provides for re-entry to fill N. 
• . , • . O t Finally, it leads on to choices m themauzm_ 
the S, and to the realization rules that proviffe 
for re-entry - like those for the S in a move. 
Thus the network and realization rules 
together provide for the recursive embedding 
of acts with further acts at either N or S or at 
both, In, this way,,then, we do indeed move 
virtually seamlessly from the SFM structure to 
RST relations. 
5. Conclusions 
The generation of the discourse structure of 
monologues and dialogues are usually treated 
as separate research paradigms, but in this 
paper we have shown how the gap between the 
two can be bridged. As we have seen, RST 
relations can be fully integrated with the richer 
SFM framework by restating the 'sister-dep- 
endency' relations of RST as occurring via 
their mother unit, i.e. in an 'element and unit' 
model. In this way we have arrived at a 
unified framework for generating both 
dialogues and monologues. 
How was this integration accomplished so 
relatively easily? The answer seems to be that 
GENEDIS, which is itself adapted from the 
sentence generator GENESYS, provides an 
appropriately rich and relevant array of o.per- 
ators. It can therefore be extended, m a 
principled manner, to incorporate the simpler 
set of relations of RST. This suggests in turn 
the value of the particular set of concepts 
incorporated into the Cardiff Grammar, and of 
the grammar-writing tool DEFREL, in which 
both GENESYS and GENEDIS are written. 
Essentially, we have here incorporated RST 
within the SFM rather than the other way 
about. There is a good reason for this. It is 
that dialogues regularly contain monologues, 
but,monologues (w, hile they can be interrupted 
by mini-dialogues to clear up misunderstand- 
ings and to challenge, etc.) occur - either 
explicitly or implicitly - within a dialogue. 
The fundamental principle underlying all 
discourse generation is that there is always an 
Addressee - so that there is always at least an 
implicit potential for dialogue. 

References 
Dale, R., Hovy, E.H., Roesner, D., and Stock, 0., 
(eds.) 1992. Aspects of Automated Natural Language 
Generation. Berlin: Springer. 
Fawcett, R.P., and Davies, B.D., 1992. 'Monologue as a turn in dialogue: towards an integration of 
exchange structure and rhetorical structure theory'. In 
Dale et al., (eds.) 1992, pp. 151-66. 
Fawcett, R.P., Tucker, G.H., and Lin, Y.Q, 1993. 
'How a systemic functional grammar works: the role 
of realization in realization'. In Horacek and Zock 1993, pp. 114-86. 
Fawcett, R.P., van der Mije, A., and van Wissen, C., 
1988. 'Towards a systemic flowchart model for local 
discourse structure'. In Fawcett, R.P., and Young, 
D.J., (eds.) 1988. New Developments in Systemic 
Linguistics, Vol 2." Theory and Application. London: 
Pinter, pp. 116-43. 
Halliday, M.A.K., 1961. 'Categories of the theory of 
grammar'. In Word 17, pp. 241-92. 
Horacek, H., and Zock, M., (eds.), 1993. New 
Concepts in Natural Language Generation. London: 
Pinter. 
Lin, Y. Q, Fawcett, R. P. and Davies, B. L. 'GEN- 
EDIS: the discourse generator in COMMUNAL'. In 
A. Sloman, D. Hoggs, G. Humpherys, A. Ramsay, 
and D. Partridge (eds.) Frontiers in Artificial Intell- 
igence and Applications. Amsterdam: IOS Press. pp. 
148-57. 
Maier, E., and Hovy, E., 1993. 'Organizing dis- 
course structure relations usin,2 metafunctions'. In 
Horacek and Zock 1993, pp. 69Y86. 
Mann, W.C., and Thompson, S.A., 1987, 
'Rhetorical structure theory'• In Polyani, L., (ed.) 
Discourse structure. Norwood, N.J., Ablex. 
Sinclair, J. and Coulthard, M. (1975) Towards an 
Analysis of Discourse. Oxford: Oxford University 
Press. 
Matthiessen, C.M.I.M., and Bateman, J.A., 1991. 
Text Generation and Systemic Functional Ling- 
uistics. London: Pinter. 
Vander Linden, K., Cumming, S., and Martin, J., 
1992. 'Using system networks to build rhetorical 
structures'. In Dale et al. 1992, pp. 183-98. 
