Can text structure be incompatible with rhetorical structure? 
Nadjet Bouayad-Agha, Richard Power and Donia Scott 
Information Technology Research Institute 
University of Brighton 
Lewes Road 
Brighton BN2 4G J, UK 
first name. lastname@it ri.bton.ac.uk 
Abstract 
Scott and Souza (1990) have posed the problem 
of how a rhetorical structure (in which proposi- 
tions are linked by rhetorical relations, but not 
yet arranged in a linear order) can be realized 
by a text structure (in which propositions are 
ordered and linked up by appropriate discourse 
connectives). Almost all work on this problem 
assumes, implicitly or explicitly, that this map- 
ping is governed by a constraint on compatibil- 
ity of structure. We show how this constraint 
can be stated precisely, and present some coun- 
terexamples which seem acceptable even though 
they violate compatibility. The examples are 
based on a phenomenon we call extraposition, 
in which complex embedded constituents of a 
rhetorical structure are extracted and realized 
separately. 
1 Introduction 
Text planning (or more broadly, document plan- 
ning) can be divided into two stages. In the 
first stage, material is selected, perhaps from a 
knowledge base, and organized rhetorically. In 
the second stage, the rhetorical structure is re- 
alized by a text structure (or document struc- 
ture), through which the material is distributed 
among sentences, paragraphs, vertical lists, and 
perhaps even diagrams. The RAtS (1999) pro- 
posal for a standard NLG architecture distin- 
guishes tile outputs of these two phases by the 
data types l:l.hetRep (rhetorical representation) 
and DocRep (document representation). 
We focus in this paper on the second stage 
of text planning -- the passage from RhetRep 
to DocRep. NLG researchers have addressed 
this issue in various ways, but everyone as- 
sumes some kind of structural compatibility be- 
tween rhetorical structure and text structure. 
The most popular discourse framework in NLG 
is R ST (Mann a.nd Thompson. 1988). which 
makes the crucial distinction between nucleus, 
which is the most important part of a message, 
and satellite, which is the peripheral part of the 
message. Scott and Souza (1990) provide guide- 
lines for the realisation of RST trees into a co- 
herent text. One of them is to avoid dangling 
sentences, that is, to avoid putting "information 
that is only weakly relevant to the message" in 
a separate sentence because it will feel as if it 
has been introduced as an afterthought or as 
introducing a new topic which is then abruptly 
abandoned, disrupting the comprehensibility of 
the text. As an example, the authors provide 
the attributive satellite of an elaboration rela- 
tion. 
Marcu (1996), in order to build a valid text 
plan from a set of rhetorical assertions, uses 
the "nuclearity principle", that is the observa- 
tion in Mann and Thompson's framework that 
"whenever two large text spans are connected 
through a rhetorical relation, that rhetorical re- 
lation holds also between the most important 
parts of the constituent spans". Therefore, the 
resulting text plans are valid in the sense that 
they are isomorphic with one of the rhetorical 
structures that can be built from the rhetorical 
assertions using this nuclearity principle. 
Our aim in this paper is to formulate more 
precisely a notion of structural compatibility 
which is necessary in order to describe the real- 
isation of a RhetRep into various DocReps, and 
then .to discuss some examples (mostly taken 
from the domain of patient information leaflets) 
of apparently acceptable texts in which this no- 
.tion of compatibility is violated..:To discuss 
this issue clearly, an assmnption must be made 
about the kinds of information represented by 
rhetorical and text structure; we outline in sec- 
tion 2 the common assumption that these rep- 
resentations are trees, labelled respectively with 
rhetorical and textual categories, the rhetorical 
structure being unordered and the text struc- 
194 
ture ordered. Section 3 then defines a notion 
of.structural compatibility that:is weaker than 
isomorphism; section 4 shows that we can find 
plausible counterexamples even to this weaker 
formulation, and discusses why these passages 
occur. Section 5 discusses some implications for 
NLG, and finally, section 6 raises further impor- 
tant issues. 
2 Rhetorical structure and 
text structure 
To distinguish clearly between FthetRep and 
DocRep, we need to define the kinds of infor- 
mation that should be included in the two rep- 
resentations. Bateman and Rondhius (1997) 
compare several approaches to rhetorical rep- 
resentation, citing in particular RST (Mann 
and Thompson, 1988) and Segmented Discourse 
Representation Theory (Asher, 1993). These 
approaches share the idea that rhetorical repre- 
sentations are composed of propositions linked 
by rhetorical relations; SDRT includes as well 
the logical apparatus of DRT, thus covering 
notions like necessity and logical scope which 
are missing from RST. For the most part, 
NLG applications have used the RST frame- 
work, adapted in various ways; the most com- 
mon representation, proposed also as the RAGS 
standard, is that of a tree in which terminal 
nodes represent elementary propositions, while 
non-terminal nodes represent rhetorical rela- 
tionships. This representation, proposed for ex- 
ample by Scott and Souza (1990), is illustrated 
by figure 1, which might be realized by the fol- 
lowing passage: 
(1) Elixir occasionally provokes a mild allergic 
reaction B, because it contains gestodene C. 
However, Elixir has no serious side- 
effects A . 
Assuming an RST-based framework, an im- 
portant issue is whether the rhetorical represen- 
. tation should already.imply a linear order. Most 
researchers have followed Scott and Souza in as- 
suming that linear order should be left unspeci- 
fied; it is during the transition to the document 
representation that the material is distributed 
among linguistic units (or perhaps diagrams, in 
a multimedia document) arranged in a specific 
order. Thus the cause relation in figure 1. for 
example, could be realized with nucleus first, or 
satellite first, or satellite embedded within nu- 
cleus: 
not(serious-side -e ffe~ts(elixir)) 
NU~EUS S,~LLITE 
B C possible(allergic-reaction(elixir)) contain(elixir, gestodene) 
Figure 1: Rhetorical representation 
(2a) Elixir occasionally provokes a mild allergic 
reaction B, because it contains gestodene c. 
(2b) Because it contains gestodene C, Elixir 
occasionally provokes a mild allergic 
reaction B. 
(2c) Elixir, because it contains gestodene C, 
occasionally provokes a mild allergic 
reaction B . 
In the RAGS proposal, which aims to extract 
a useful common approach from current work 
in NLG, the DocRep comprises an ordered tree 
corresponding roughly to the 'logical markup' 
in notations like HTML and LaTeX. More pre- 
cisely, a distinction is made between abstract 
and concrete levels of representation, where the 
abstract representation corresponds to logical 
markup (e.g., concepts like 'paragraph' and :em- 
phasis'), while the concrete representation also 
covers graphical markup (concepts like ~vertical 
space' and 'bold face'). In terms of this dis- 
tinction, it is the AbsDocRep that is specified 
during text planning; graphical markup can be 
deferred to a later formatting stage. 
Figure 2 shows two alternative document rep- 
resentations expressing the rhetorical content in 
figure 1. Following Power (2000), the nodes of 
the tree are labelled with 'text-categories' us- 
ing a system that extends the 'text grammar' 
proposed by Nunberg (1990). 1 These document 
1Nunberg's terms 'text-phrase', 'text-clatise',and 
'text-sentence' refer to textual categories, which 
should not be confused with their syntactic coun- 
terparts. They are defined not by syntactic forma- 
tion rules but by their role in text-structure, which 
is typically marked as follows: tezt-sentences begin 
with a capital letter and end in a full stop; text- 
clauses are separated by semicolons; tezt-phrases are 
195 
PARAGRAPH 
TEXT-SENTENCE 
TEXT-PHRASE " TEXT-PHRASE 
B 
possible(allergic-reaction(elixir)) / 
TEXT-SENTENCE 
TEXT-PHRASE TEXT-PHRASE 
concession A 
"however" not(serious-side-effects(elixir)) 
TEXT-PHRASE TEXT-PHRASE 
cause C 
"because" contain(elixir, gestodene) 
TEXT-SENTENCE 
(b) ~~ / 
TEXT-CLAUSE TEXT-CLAUSE TEXT-CLAUSE 
C 
TEXT-PHRASE TEXT-PHRASE TEXT-PHRASE TEXT-PHRASE 
cause B concession A 
"consequently"possible(allergic-reaction(elixir)) "however" not(serious-side-effects(elixir)) 
Figure 2: Document representations 
representations can now be passed to the tac- 
tical generator for the syntactic realization of 
the elementary propositions; the resulting texts 
might be as follows: 
(3a) Elixir occasionally provokes a mild allergic 
reaction B, because it contains gestodene C. 
However, Elixir has no serious side- 
effects A . 
(3b) Elixir contains gestodeneC; consequently, 
it occasionally provokes a mild allergic 
reactionS; however, Elixir has no serious 
side-effects A . 
3 Structural compatibility 
Summarising the argument so far. we have made 
three main points: 
o Rhetorical structure has typically been 
represented by unordered RST trees such 
as figure 1. 
o Document structure, which conveys in- 
formation similar to logical markup in 
HTML~ can be represented by ordered 
trees in which nodes are labelled with text- 
categories (figure 2). 
constituents of text-clauses, sometimes separated by 
commas, although within text-clauses the hierarchi- 
cal- structture is expressed mainly through syntax. 
A given rhetorical representation can be 
expressed by a variety of different docu- 
ment representations, in which the propo- 
sitions occur in different orders, and in 
different text-category configurations, and 
the rhetorical relations are expressed by 
different connectives. 
This formulation of the problem raises an obvi- 
ous question: how can we characterize the set of 
document representations that adequately real- 
ize a given rhetorical representation? Elsewhere 
(Power, 2000), we have argued that an adequate 
realization must meet three conditions: 
Correct content: 
All propositions and 
nmst be expressed. 
rhetorical relations 
Well-formed structure: 
General formation rules for document 
structure must be respected (e.g. a text- 
sentence cannot contain a paragraph, un- 
less tile paragraph is indented). 
Structural compatibility: 
The docmnent representation nmst orga- 
nize the propositions in a way that is com- 
patible with their organization in rhetori- 
cal structure. 
196 
The first two conditions are relatively straight- 
forward, but what is meant,exactly .by 'struc- 
tural compatibility'? 
Assuming that we are comparing two trees, 
the strongest notion of compatibility is isomor- 
phism, which can be defined for our purposes as 
follows: 
DocRep is isomorphic with RhetRep 
if they group the elementary propo- 
sitions in exactly the same way. 
More formally, every set of proposi- 
tions that is dominated by a node in 
DocRep should be dominated by a node 
in RhetRep, and vice-versa. 
Under this definition, the rhetorical representa- 
tion in figure 1 is isomorphic with the document 
representation in figure 2a, but not with that in 
figure 2b: 
* Proceeding top-down and left-to-right, the 
five nodes in figure 1 dominate the proposi- 
tion sets {A,B, C}, {A}, {S,C}, {B}, and {c}. 
o Ignoring nodes that express discourse con- 
nectives, the nodes in figure 2a dominate 
the proposition sets {A,B,C}, {B,C}, 
{B}, {C} (twice), and {A} (twice). These 
are exactly the same sets that were ob- 
tained for figure 1. 
* Tile corresponding sets for figure 2b are 
{A,B,C}, {C}, {B} (twice), and {A} 
(twice). Since the set {B,C} is missing 
from this list, there is a grouping in figure 
1 that is not realized in figure 2b, so these 
representations are not isomorphic. 
Since structures like figure 2b are common, iso- 
tnorphism seems too strong a constraint; we 
have therefore proposed (Power, 2000) the fol- 
lowing weaker notion of compatibility: 
DocRep is compatible with RhetRep 
if every grouping of the elementary 
propositions in Docgep is also found 
in R.hetRep. 
Formally, every set of propositionS 
that is dominated by a node in DocRep 
sh.ould be dominated by a node in 
RhetRep -- bat the converse is not re- 
quired. 
Under this constraint, we allow tim document 
representation t.o omit rhetorical groupings, but 
"you forfA~ T~ITE to take C~ 
your tablet" SUS_ 1 NUC~USD 2 
"Go on as before" 
~,~J~ L E U S _ I N U C~.E~ S _ 2 
B C 
"take another as soon "wait until it is time as you remember" to take your next dose" 
Figure 3: Rhetorical representation of in- 
struction 
not to introduce new ones. The resulting struc- 
tures may be ambiguous, but this will not mat- 
ter if the unexpressed rhetorical relationships 
can .be inferred from the content. 
4 Extraposition 
The compatibility rule may be a useful text- 
planning heuristic, but as a constraint on ade- 
quacy it still seems too strong. Looking through 
our corpus of patient information leaflets, we 
have noticed some exceptions, especially in pas- 
sages giving conditional instructions: 
(4) If you forget to take your tablet A, take an- 
other as soon as you remember B or wait 
until it is time to take your next dose C. 
Then go on as before D. 
From the point of view of document structure, 
this passage is a paragraph comprising two text- 
sentences: thns the proposition D is separated 
from the other three propositions, which are 
grouped in tile .first sentence. However, rhetor- 
ically speaking, D belongs to the consequent of 
the conditional: it is the final step of the plan 
that should be activated .if_the patient forgets 
to take a dose (figure 3). Compatibility is vio- 
lated because tile DocRep contains a node (the 
first text-sentence) dominating the proposition 
set {A, B, C}. which is not dominated by any 
node in figure 3. 
Such examples might be explained as the re- 
sult of loose punctuation or layout, perhal)S 
197 
through imitation of the patterns of conversa- 
tion, in which extra:.materi~! is_often ~tagged. on- 
as an afterthought. Thus proposition D remains 
grouped with B and C -- they occur consecu- 
tively -- but through a minor relaxation of nor- 
mal punctuation it has been separated by a full- 
stop rather than a comma. However, this expla- 
nation fails to cover variations of the example 
in which the propositions in the consequent are 
not realized consecutively in the DocRep: 
(5) Consult your doctor immediately A if a 
rash develops B. It might become seriously 
infected C. 
In this example, A must be grouped rhetori- 
cally with C rather than with B, unless we take 
the radical step of allowing rhetorical structure 
to contradict logical structure. The proposition 
C cannot be logically conjoined with the con- 
ditional because it contains a hypothetical dis- 
course referent (the rash) that is bound to the 
antecedent, and is therefore inaccessible outside 
the conditional. 
If passages of this kind are not artifacts of 
loose punctuation, why do they occur? A plau- 
sible reason, we suggest, is that some com- 
plex rhetorical patterns cannot easily be real- 
ized in a way that maintains structural com- 
patibility, usually because text-clauses are over- 
loaded. Conditionals are especially prone to this 
problem because the only common discourse 
connective ('if') is a subordinating conjunction 
which can only link spans within a syntactic 
sentence (and thus within a text-clause). If ei- 
ther the antecedent or the consequent is com- 
plex, the author is faced with a tricky problem. 
We have found examples in patient informa- 
tion leaflets of conditional sentences so long that 
they are ahnost incomprehensible. More skilled 
authors, however, succeed in presenting the ma- 
terial clearly either by using layout (e.g., a com- 
plex antecedent is presented as an indented list), 
or by a trick of rhetorical reorganization that we 
will call eztraposition. It is this trick that intro- 
duces an incompatibility between RhetRep and 
DocRep. 
Extraposition typically occurs when a rhetor- 
ical representation R contains a complex em- 
bedded constituent C. To respect structural 
compatibility, R should be realized by a doc- 
ument unit that contains the realization of C: 
instead, in extraposition, a document unit real- 
ising/?. - C is coordinated with one realizing C. 
so that the extraposed material C is raised in 
the DocRep to the same level as R. To recon- 
...... struct ~:the:.: meanings.of .the:-.whole:. passage, .the 
reader has to plug C back into R. In most 
cases, the author facilitates this task through 
an explicit deictic reference to the extraposed 
material (Bouayad-Agha et al., 2000): 
(6) If you have any of the following, tell your 
doctor: 
difficulty in breathing 
........... al)dominal..Dains 
nausea or vomiting 
Occasionally, however, the author leaves the 
extraposition implicit, assuming that the reader 
can infer the correct location of C within R from 
the propositional content. In such cases, the ex- 
traposition looks like an afterthought, because 
the unit realizing R - C contains no signal that 
a gap in its content will be filled in later. 
We have also come across rare examples 
of another kind of incompatibility in which 
Marcu's (1996) principle of nuclearity is vio- 
lated by grouping together two satellites which 
have the same nucleus. Suppose that the rhetor- 
ical representation in figure 1 is realized by the 
following passage, in a context in which the 
reader knows nothing about gestodene: 
(7) Although Elixir has no serious side- 
effects A, it contains gestodene c. Conse- 
quently, it occasionally provokes a mild al- 
lergic reaction 8. 
The apparent concession relation between A 
and C here is paradoxical, since in rhetorical 
structure they are unrelated. Of course a con- 
trast between A and C nfight be perceived by 
a medical expert; however, one can construct 
similar examples in which the apparent relation 
is even less plausible: 
(8a) Although we usually work fl'om nine' to 
five A, today is Friday C. Consequently, we 
can go home early B. 
This may be rather loose, but many people find 
it acceptable. It could be explained as a rhetor- 
ical trick in which the sheer paradox of the con- 
cession serves as a signal that it is incomplete. 
The device might be spelled out as follows: 
Although Elixir has no serious side- 
effects A, there exists a contrasting 
state of a~hirs resulting fl'om the flzct 
that it contains gestodene c. This 
state of affairs is that it occasionally 
provokes a nfild allergic reaction t3. 
198 
Unlike the conditional examples above, this de- 
vice works only.when t he-.rhetorically grouped 
propositions B and C are consecutive in the 
DocRep. Thus whatever view is taken of exam- 
ple (Sa) , everyone finds its variant (Sb) much 
worse: 
(Sb) # Today is Friday C although we usually 
work from nine to five A. Consequently, we 
can go home early s. 
5 Implications for NLG 
For many NLG applications, the notion of com- 
patibility defined above is a useful hard con- 
straint; even if violations of this constraint are 
sometimes acceptable, they are not essential. 
However, for some kinds of material (e.g., com- 
plex instructions), extraposition is a convenient 
rhetorical device which might improve the read- 
ability of the generated texts, so it is worth con- 
sidering how a text planner might be configured 
so as to allow solutions that violate compatibil- 
ity. 
In terms of the RAGS framework, there 
are broadly two possible approaches. First, 
we could introduce incompatibility by defin- 
ing transformations on the RhetRep; alterna- 
tively, we could relax the constraints govern- 
ing the transition from RhetRep to DocRep. 
The RAGS proposal (1999) allows for rhetorical 
transformations through a distinction between 
abstract and concrete rhetorical representa- 
tions. The abstract representation AbsRhetRep 
expresses the rhetorical content of the under- 
lying message, while the concrete RhetRep ex- 
presses the rhetorical structure directly realized 
in the text and corresponds to the representa- 
tion used by Scott and Souza (1990) to discuss 
textual realisation. If KhetRep is incompati- 
ble with AbsRhetRep, the text structure DocRep 
will also be incompatible with AbsRhetRep, 
even though the rules for realizing rhetorical 
structure by document structure are themselves 
compatibility-preserving, qYaalsformation oper- 
ations are also used by Marcu (2000) to map 
Japanese rhetorical structures onto English-like 
rhetorical structures, but these are mappings 
between two PdaetReps rather than from an 
AbsRhetRep to a RhetRep. 
If transformations are allowed, there are obvi- 
ous dangers that the message will be expressed 
in such a distorted way that the reader cannot 
recover the original intention. For this reason, 
rhetori(:al transformations must be defined with 
care. A fairly safe option would appear to be 
.... -the ..extraposition-iof.:a ,proposition. ~lab6rai~ing 
the antecedent of a conditional --.even though 
such a transformation would violate Marcu's 
(1996) 'nuclearity principle' (assuming that the 
antecedent is regarded as the satellite). The fop - 
lowing examples show that this transformation 
leads to acceptable texts regardless of the order 
of nucleus and satellite within the conditional: 
(9a)~ Dcr.uot" use :Elixirif you :have had' an al: " 
lergic reaction to Elixir. An allergic reac- 
tion may be recognised as a rash, itching 
or shortness of breath. 
(9b) If you have had an allergic reaction to 
Elixir, do not use Elixir. An allergic re- 
action may be recognised as a rash, itching 
or shortness of'breath. 
However, the approach based on rhetorical 
transformations leads to difficulties when the 
acceptability of the resulting text depends on 
linear order as well as grouping. For instance, 
suppose that we try extraposing the elabora- 
tion of a satellite when the main relation is not 
a conditional, but a concession. The following 
passages show two texts that might result, but 
in this case the second version sounds anoma- 
lous: even if they are not grouped together in 
the DocRep, the satellite and its elaboration at 
least need to be consecutive. 
(10a) You should not stop taking Elixir, even 
though you might experience some mild 
effects. For example, feelings of dizziness 
and nausea are very common at the begin- 
ning of treatment. 
(lOb) # Even though you might experience some 
mild effects at tile beginning of tile treat- 
ment, you should not stop taking Elixir. 
For example, feelings of dizziness and nau- 
sea are very common at the beginning of 
treatment. 
A transformation from AbsKhetRep to 
RhetRep cannot distinguish these cases, so that 
10a is,allowed while 10b is protfibited; unless the 
l:l.hetRep is at least partially specified for linear 
order. Adhering strictly to the RAGS frame- 
work, where linear order is specified only in 
tbsDocRep, one would have to adopt the alter- 
native of building an incompatible /~bsDocRep 
from RhetRep. constraining the linear order at, 
this stage. 
199 
6 Conclusion 
We have discussed various examples Of extra- 
position. This phenomenon is due to various 
factors: the complexity of the material (exam- 
ple 4), the presence of logical information (5), 
the use of referring expressions to access infor- 
mation at various degrees of accessibility in the 
text structure (5,6,9), and the use of particular 
rhetorical strategies (7,8). This last group of ex- 
amples concerns.a concession constr.uction sim- 
ilar to the one discussed by Grote et al. (1997), 
namely the substitution concession. This type 
of concession groups together the conceded part 
A and the explanation C but leaves the conclu- 
sion B unverbalised. The difference in the case 
of examples 7 and 8 is that A and C are grouped 
together but B is required to follow them be- 
cause there is not enough information for the 
reader to infer B from A and C. 
The extraposition phenomenon shows that 
the nucleus-satellite distinction is not the only 
factor influencing the segmentation of the mes- 
sage. In example 10, the injunction you should 
not stop taking Elixir obviously expresses the 
main intention of the author. However, the 
fact that the subordinated concession is placed 
after its main clause makes it available for 
further expansion. The sometimes compet- 
ing informational and intentional roles of dis- 
course segments have been at the centre of 
the debate over the nucleus-satellite distinction 
(Moore and Pollack, 1992; Moser and Moore, 
1996; Bateman and Rondhius, 1997); the acces- 
sibility of discourse segments on the right fron- 
tier of a discourse structure is a phenomenon 
that has already been discussed by several re- 
searchers (Webber, 1991; Asher, 1993). Extra- 
position provides a useful and sometimes im- 
portant means of rearranging complex material 
in an abstract discourse representation in order 
to satisfy the constraints posed by linearisation 
into text. 

References 
N. Asher. 1993. Reference to Abstract Objects 
in Discourse. Kluwer Academic Publishers, 
Netherlands. 
J. Bateman and K. Rondhius. 1997. Coher- 
ence relations: Towards a general specifica- 
tion. Discourse Processes, 24(1):3-50. 
N. Bouayad-Agha, D. Scott, and R. Power. 
2000. Integrating content and style in doc- 
mnents: a case study of patient informa- 
tion leaflets. Information Design Journal, 
-9(2):161~176. -- • :-'.::- 
B. Grote, N. Lenke, and Stede M. 1997. 
Ma(r)king concessions in english and german. 
Discourse Processes, 24( 1):87-117. 
W. Mann and S. Thompson. 1988. Rhetorical 
structure theory: towards a functional theory 
of text organization. Text, 8(3):243-281. 
D. Marcu, L. Carlson, and M. Watanabe. 2000. 
The automatic translation of discourse struc- 
tures. In Proceedings of the North Ameri- 
can Chapter of the Association for Comptu- 
ational Linguistics (NAACL'2000), Seattle, 
Washington. 
D. Marcu. 1996. Building up rhetorical struc- 
ture trees. In Proceedings of AAAI-96. Amer- 
ican Association for Artificial Intelligence. 
D.J Moore and M.E. Pollack. 1992. A prob- 
lem for rst: The need for multi-level dis- 
course analysis. Computational Linguistics, 
18(4):537-544. 
M. Moser and J.D. Moore. 1996. Towards a 
synthesis of two accounts of discourse struc- 
ture. Computational Linguistics, 22(3):409- 
419. 
G. Nunberg. 1990. The Linguistics of Punctu- 
ation. CSLI, Stanford, USA. 
R. Power. 2000. Mapping rhetorical struc- 
tures to text structures by constraint satis- 
faction. Technical report, ITRI, University of 
Brighton. 
RAGS. 1999. The RAGS project: towards 
a reference architecture for natural lan- 
guage generation systems. Technical report, 
Information Technology Research Institute, 
Brighton, UK. 
D. Scott and C. de Souza. 1990. Getting the 
message across in RST-based text generation. 
In R. Dale, C. Mellish, and M. Zock, editors, 
Current Research in Natural Language Gen- 
eration. Cognitive Science Series, Academic 
Press. 
B.L Webber. 1991. Structure and ostension in 
the interpretation of discourse deixis. Lan- 
guage and Cognitive Processes, 6(2):107-135. 
