Generating Discourse Structures for Written Texts
Huong LeThanh, Geetha Abeysinghe, and Christian Huyck
School of Computing Science, Middlesex University
The Burroughs, London, NW4 4BT, United Kingdom
{H.Le, G.Abeysinghe, C.Huyck}@mdx.ac.uk
Abstract
This paper presents a system for automati-
cally generating discourse structures from
written text. The system is divided into two
levels: sentence-level and text-level. The
sentence-level discourse parser uses syntactic
information and cue phrases to segment sen-
tences into elementary discourse units and to
generate discourse structures of sentences. At
the text-level, constraints about textual adja-
cency and textual organization are integrated
in a beam search in order to generate best dis-
course structures. The experiments were done
with documents from the RST Discourse
Treebank. It shows promising results in a rea-
sonable search space compared to the dis-
course trees generated by human analysts.
1 Introduction
Many recent studies in Natural Language Proc-
essing have paid attention to Rhetorical Structure
Theory (RST) (Mann and Thompson 1988; Hovy
1993; Marcu 2000; Forbes et al. 2003), a method
of structured description of text. Although rhe-
torical structure has been found to be useful in
many fields of text processing (Rutledge et al.
2000; Torrance and Bouayad-Agha 2001), only a
few algorithms for implementing discourse ana-
lyzers have been proposed so far. Most research
in this field concentrates on specific discourse
phenomena (Schiffrin 1987; Litman and Hirsch-
berg 1990). The amount of research available in
discourse segmentation is considered small; in
discourse parsing it is even smaller.
The difficulties in developing a discourse
parser are (i) recognizing discourse relations
between text spans and (ii) deriving discourse
structures from these relations. Marcu (2000)’s
parser is based on cue phrases, and therefore
faces problems when cue phrases are not present
in the text. This system can apply to unrestricted
texts, but faces combinatorial explosion. The dis-
advantage of Marcu’s approach is that it pro-
duces a great number of trees during its process,
which is the essential redundancy in computa-
tion. As the number of relations increases, the
number of possible discourse trees increases ex-
ponentially.
Forbes et al. (2003) have a different approach
of implementing a discourse parser for a Lexi-
calized Tree Adjoining Grammar (LTAG). They
simplify discourse analysis by developing a
grammar that uses cue phrases as anchors to con-
nect discourse trees. Despite the potential of this
approach for discourse analysis, the case of no
cue phrase present in the text has not been fully
investigated in their research. Polanyi et al.
(2004) propose a far more complicated discourse
system than that of Forbes et al. (2003) , which
uses syntactic, semantic and lexical rules. Polanyi
et al. have proved that their approach can provide
promising results, especially in text summariza-
tion.
In this paper, different factors were investi-
gated to achieve a better discourse parser, in-
cluding syntactic information, constraints about
textual adjacency and textual organization. With
a given text and its syntactic information, the
search space in which well-structured discourse
trees of a text are produced is minimized.
The rest of this paper is organized as follows.
The discourse analyzer at the sentence-level is
presented in Section 2. A detailed description of
our text-level discourse parser is given in Section
3. In Section 4, we describe our experiments and
discuss the results we have achieved so far. Sec-
tion 5 concludes the paper and proposes possible
future work on this approach.
2 Sentence-level Discourse Analyzing
The sentence-level discourse analyzer constructs
discourse trees for each sentence. In doing so,
two main tasks need to be accomplished: dis-
course segmentation and discourse parsing,
which will be presented in Section 2.1 and Sec-
tion 2.2.
2.1 Discourse Segmentation
The purpose of discourse segmentation is to split
a text into elementary discourse units (edus)1.
This task is done using syntactic information and
cue phrases, as discussed in Section 2.1.1 and
Section 2.1.2 below.
2.1.1 Segmentation by Syntax – Step 1
Since an edu can be a clause or a simple sen-
tence, syntactic information is useful for the
segmentation process. One may argue that using
syntactic information is complicated since a syn-
tactic parser is needed to generate this informa-
tion. Since there are many advanced syntactic
parsers currently available, the above problem
can be solved. Some studies in this area were
based on regular expressions of cue phrases to
identify edus (e.g., Marcu 2000). However, Re-
deker (1990) found that only 50% of clauses
contain cue phrases. Segmentation based on cue
phrases alone is, therefore, insufficient by itself.
In this study, the segmenter’s input is a sen-
tence and its syntactic structure; documents from
the Penn Treebank were used to get the syntactic
information. A syntactic parser is going to be
integrated into our system (see future work).
Based on the sentential syntactic structure, the
discourse segmenter checks segmentation rules to
split sentences into edus. These rules were cre-
ated based on previous research in discourse
segmentation (Carlson et al. 2002). The segmen-
tation process also provides initial information
about the discourse relation between edus. For
example, the sentence “Mr. Silas Cathcart built a
shopping mall on some land he owns” maps with
the segmentation rule
 ( NP|NP-SBJ  <text1> ( SBAR|RRC <text2> ) )
In which, NP, SBJ, SBAR, and RRC stand for
noun phrase, subject, subordinate clause, and re-
duce relative clause respectively. This rule can be
stated as, “The clause attached to a noun phrase
can be recognized as an embedded unit.”
The system searches for the rule that maps
with the syntactic structure of the sentence, and
                                                                
1 For further information on “edus”, see (Marcu 2000).
then generates edus. After that, a post process is
called to check the correctness of discourse
boundaries. In the above example, the system
derives an edu “he owns” from the noun phrase
“some land he owns”. The post process detects
that “Mr. Silas Cathcart built a shopping mall
on” is not a complete clause without the noun
phrase “some land”. Therefore, these two text
spans are combined into one. The sentence is
now split into two edus “Mr. Silas Cathcart built
a shopping mall on some land” and “he owns.” A
discourse relation between these two edus is then
initiated. Its relation’s name and the nuclearity
roles of its text spans are determined later on in a
relation recognition-process (see Section 2.2).
2.1.2 Segmentation by Cue Phrase–Step 2
Several NPs are considered as edus when they
are accompanied by a strong cue phrase. These
cases cannot be recognized by syntactic informa-
tion; another segmentation process is, therefore,
integrated into the system. This process seeks
strong cue phrases from the output of Step 1.
When a strong cue phrase is found, this process
detects the end boundary of the NP. This end
boundary can be punctuation such as a semic o-
lon, or a full stop. Normally, a new edu is created
from the begin position of the cue phrase to the
end boundary of the NP. However, this procedure
may create incorrect results as shown in the ex-
ample below:
(1) [In 1988, Kidder eked out a $46 million
profit, mainly][ because of severe cost
cutting.]
The correct segmentation boundary for the sen-
tence given in Example (1) should be the position
between the comma (‘,’) and the adverb
“mainly ”. Such a situation happens when an ad-
verb stands before a strong cue phrase. The post
process deals with this case by first detecting the
position of the NP. After that, it searches for the
appearance of adverbs before the position of the
strong cue phrase. If an adverb is found, the new
edu is segmented from the start position of the
adverb to the end boundary of the NP. Otherwise,
the new edu is split from the start position of the
cue phrase to the end boundary of the NP. This is
shown in the following example:
(2) [According to a Kidder World story about
Mr. Megargel,] [all the firm has to do is
"position ourselves more in the deal flow."]
Similar to Step 1, Step 2 also initiates discourse
relations between edus that it derives. The rela-
tion name and the nuclearity role of edus are
posited later in a relation recognition-process.
2.2 Sentence-level Discourse Parsing
This module takes edus from the segmenter as
the input and generates discourse trees for each
sentence. As mentioned in Section 2.1, many
edus have already been connected in an initial
relation. The sentence-level discourse parser
finds a relation name for the existing relations,
and then connects all sub-discourse-trees within
one sentence into one tree. All leaves that corre-
spond to another sub-tree are replaced by the cor-
responding sub-trees, as shown in Example (3)
below:
(3) [She knows3.1] [what time you will come3.2][
because I told her yesterday.3.3]
The discourse segmenter in Step 1 outputs two
sub-trees, one with two leaves “She knows” and
“what time you will come”; another with two
leaves “She knows what time you will come” and
“because I told her yesterday”. The system com-
bines these two sub-trees into one tree. This pro-
cess is illustrated in Figure 1.
Figure 1. The discourse structure of text (3)
Syntactic information is used to figure out which
discourse relation holds between text spans as
well as their nuclearity roles. For example, the
discourse relation between a reporting clause and
a reported clause in a sentence is an Elaboration
relation. The reporting clause is the nucleus; the
reported clause is the satellite in this relation.
Cue phrases are also used to detect the con-
nection between edus, as shown in (4):
(4) [He came late] [because of the traffic.]
The cue phrase “because of” signals a Cause re-
lation between the clause containing this cue
phrase and its adjacent clause. The clause con-
taining “because of” is the satellite in a relation
between this clause and its adjacent clause.
To posit relation names, we combine several
factors, including syntactic information, cue
phrases, NP-cues, VP-cues2, and cohesive de-
vices (e.g., synonyms and hyponyms derived
from WordNet) (Le and Abeysinghe 2003). With
the presented method of constructing sentential
discourse trees based on syntactic information
and cue phrases, combinatorial explosions can be
prevented and still get accurate analyses.
3 Text-level Discourse Analyzing
3.1 Search Space
The original search space of a discourse parser is
enormous (Marcu 2000). Therefore, a crucial
problem in discourse parsing is search-space re-
duction. In this study, this problem was solved by
using constraints about textual organization and
textual adjacency.
Normally, each text has an organizational
framework, which consists of sections, para-
graphs, etc., to express a communicative goal.
Each textual unit completes an argument or a
topic that the writer intends to convey. Thus, a
text span should have semantic links to text spans
in the same textual unit before connecting with
text spans in a different one. Marcu (2000) ap-
plied this constraint by generating discourse
structures at each level of granularity (e.g., para-
graph, section). The discourse trees at one level
are used to build the discourse trees at the higher
level, until the discourse tree for the entire text is
generated. Although this approach is good for
deriving all valid discourse structures that repre-
sent the text, it is not optimal when only some
discourse trees are required. This is because the
parser cannot determine how many discourse
trees should be generated for each paragraph or
section. In this research, we apply a different ap-
proach to control the levels of granularity. In-
stead of processing one textual unit at a time, we
use a block-level-score to connect the text spans
                                                                
2 An NP-cue (VP-cue) is a special noun (verb) in the NP
(VP) that signals discourse relations.
3.1-3.2
Elaboration
3.1-3.2
3.1-3.3
Cause
3.33.1 3.2
3.33.1-3.2
Elaboration
3.1-3.3
Cause
3.1 3.2
that are in the same textual unit. A detailed de-
scription of the block-level-score is presented in
Section 3.2. The parser completes its task when
the required number of discourse trees that cover
the entire text is achieved.
The second factor that is used to reduce the
search space is the textual adjacency constraint.
This is one of the four main constraints in con-
structing a valid discourse structure (Mann and
Thompson 1988). Based on this constraint, we
only consider adjacent text spans in generating
new discourse relations. This approach reduces
the search space remarkably, since most of the
text spans corresponding to sub-trees in the
search space are not adjacent. This search space
is much smaller than the one in Marcu’s (2000)
because Marcu’s system generates all possible
trees, and then uses this constraint to filter the
inappropriate ones.
3.2 Algorithm
To generate discourse structures at the text-level,
the constraints of textual organization and textual
adjacency are used to initiate all possible con-
nections among text spans. Then, all possible
discourse relations between text spans are posited
based on cue phrases, NP-cues, VP-cues and
other cohesive devices (Le and Abeysinghe
2003). Based on this relation set, the system
should generate the best discourse trees, each of
which covers the entire text. This problem can be
considered as searching for the best solution of
combining discourse relations. An algorithm that
minimizes the search space and maximizes the
tree’s quality needs to be found. We apply a
beam search, which is the optimization of the
best-first search where only a predetermined
number of paths are kept as candidates. This al-
gorithm is described in detail below.
A set called Subtrees is used to store sub-trees
that have been created during the constructing
process. This set starts with sentential discourse
trees. As sub-trees corresponding to contiguous
text spans are grouped together to form bigger
trees, Subtrees contains fewer and fewer mem-
bers. When Subtrees contains only one tree, this
tree will represent the discourse structure of the
input text. All possible relations that can be used
to construct bigger trees at a time t form a hy-
pothesis set PotentialH. Each relation in this set,
which is called a hypothesis, is assigned a score
called a heuristic-score, which is equal to the
total score of all discourse cues contributing to
this relation. A cue’s score is between 0 and 100,
depending on its certainty in signaling a specific
relation. This score can be optimized by a train-
ing process, which evaluates the correctness of
the parser’s output with the discourse trees from
an existing discourse corpus. At present, these
scores are assigned by our empirical research.
In order to control the textual block level, each
sub-tree is assigned a block-level-score, depend-
ing on the block levels of their children. This
block-level-score is added to the heuristic-score,
aiming at choosing the best combination of sub-
trees to be applied in the next round. The value of
a block-level-score is set in a different value-
scale, so that the combination of sub-trees in the
same textual block always has a higher priority
than that in a different block.
• If two sub-trees are in the same paragraph,
the tree that connects these sub-trees will
have the block-level-score = 0.
• If two sub-trees are in different paragraphs,
the block-level-score of their parent tree is
equal to -1000 * (Li-L0), in which L0 is the
paragraph level, Li is the lowest block level
that two sub-trees are in the same unit. For
example, if two sub-trees are in the same
section but in different paragraphs; and there
is no subsection in this section; then Li-L0 is
equal to 1. The negative value (-1000) means
the higher distance between two text spans,
the lower combinatorial priority they get.
When selecting a discourse relation, the relation
corresponding to the node with a higher block-
level-score has a higher priority than the node
with a lower one. If relations have the same
block-level-score, the one with higher heuristic-
score is chosen.
To simplify the searching process, an accu-
mulated-score is used to store the value of the
search path. The accumulated-score of a path at
one step is the highest predicted-score of this
path at the previous step. The predicted-score of
one step is equal to the sum of the accumulated-
score, the heuristic-score and the block-level-
score of this step. The searching process now
becomes the process of searching for the hy-
pothesis with highest predicted-score.
At each step of the beam search, we select the
most promising nodes from PotentialH that have
been generated so far. If a hypothesis involving
two text spans <Ti> and <Tj> is used, the new
sub-tree created by joining the two sub-trees cor-
responding to these text spans is added to Sub-
trees. Subtrees is now updated so that it does not
contain overlapping sub-trees. PotentialH is also
updated according to the change in Subtrees. The
relations between the new sub-tree and its adja-
cent sub-trees in Subtrees are created and added
to PotentialH.
All hypotheses computed by the discourse
parser are stored in a hypothesis set called
StoredH. This set is used to guarantee that a dis-
course sub-tree will not be created twice. When
detecting a relation between two text spans, the
parser first looks for this relation in StoredH to
check whether it has already been created or not.
If it is not found, it will be generated by a dis-
course relation recognizer.
The most promising node from PotentialH is
again selected and the process continues. A bit of
depth-first searching occurs as the most promis-
ing branch is explored. If a solution is not found,
the system will start looking for a less promising
node in one of the higher-level branches that had
been ignored. The last node of the old branch is
stored in the system. The searching process re-
turns to this node when all the others get bad
enough that it is again the most promising path.
In our algorithm, we limit the branches that the
search algorithm can switch to by a number M.
This number is chosen to be 10, as in experi-
ments we found that it is large enough to derive
good discourse trees. If Subtrees contains only
one tree, this tree will be added to the tree’s set.3
The searching algorithm finishes when the num-
ber of discourse trees is equal to the number of
trees required by the user. Since the parser
searches for combinations of discourse relations
that maximize the accumulated-score, which rep-
resents the tree’s quality, the trees being gener-
ated are often the best descriptions of the text.
4 Evaluation
The experiments were done by testing 20 docu-
ments from the RST Discourse Treebank (RST-
DT 2002), including ten short documents and ten
                                                                
3 If no relation is found between two discourse sub-trees, a
Joint relation is assigned. Thus, a discourse tree that covers
the entire text can always be found.
long ones. The length of the documents varies
from 30 words to 1284 words. The syntactic in-
formation of these documents was taken from
Penn Treebank, which was used as the input of
the discourse segmenter. In order to evaluate the
system, a set of 22 discourse relations (list, se-
quence, condition, otherwise, hypothetical, an-
tithesis, contrast, concession, cause, result, cause-
result, purpose, solutionhood, circumstance,
manner, means, interpretation, evaluation, sum-
mary, elaboration, explanation, and joint) was
used.4 The difference among cause, result and
cause-result is the nuclearity role of text spans.
We also carried out another evaluation with the
set of 14 relations, which was created by group-
ing similar relations in the set of 22 relations. The
RST corpus, which was created by humans, was
used as the standard discourse trees for our
evaluation. We computed the output’s accuracy
on seven levels shown below:
• Level 1 - The accuracy of discourse seg-
ments. It was calculated by comparing the
segment boundaries assigned by the dis-
course segmenter with the boundaries as-
signed in the corpus.
• Level 2 - The accuracy of text spans’ combi-
nation at the sentence-level. The system gen-
erates a correct combination if it connects the
same text spans as the corpus.
• Level 3 - The accuracy of the nuclearity role
of text spans at the sentence-level.
• Level 4 - The accuracy of discourse relations
at the sentence-level, using the set of 22 rela-
tions (level 4a), and the set of 14 relations
(level 4b).
• Level 5 - The accuracy of text spans’ combi-
nation for the entire text.
• Level 6 - The accuracy of the nuclearity role
of text spans for the entire text.
• Level 7 - The accuracy of discourse relations
for the entire text, using the set of 22 rela-
tions (level 7a), and the set of 14 relations
(level 7b).
The system performance when the output of a
syntactic parser is used as the input of our dis-
course segmenter will be evaluated in the future,
when a syntactic parser is integrated with our
system. It is also interesting to evaluate the per-
                                                                
4 See (Le and Abeysinghe 2003) for a detailed description of
this discourse relation set.
formance of the discourse parser when the cor-
rect discourse segments generated by an analyst
are used as the input, so that we can calculate the
accuracy of our system in determining discourse
relations. This evaluation will be done in our fu-
ture work.
In our experiment, the output of the previous
process was used as the input of the process fol-
lowing it. Therefore, the accuracy of one level is
affected by the accuracies of the previous levels.
The human performance was considered as the
upper bound for our discourse parser’s perform-
ance. This value was obtained by evaluating the
agreement between human annotators using 53
double-annotated documents from the RST cor-
pus. The performance of our system and human
agreement are represented by precision, recall,
and F-score5, which are shown in Table 1.
The F-score of our discourse segmenter is
86.9%, while the F-score of human agreement is
98.7%. The level 2’s F-score of our system is
66.3%, which means the error in this case is
28.7%. This error is the accumulation of errors
made by the discourse segmenter and errors in
discourse combination, given correct discourse
segments. With the set of 14 discourse relations,
the F-score of discourse relations at the sentence-
level using 14 relations (53.0%) is higher than
the case of using 22 relations (52.2%).
The most recent sentence-level discourse
parser providing good results is SPADE, which is
reported in (Soricut and Marcu 2003). SPADE
includes two probabilistic models that can be
used to identify edus and build sentence-level
discourse parse trees. The RST corpus was also
used in Soricut and Marcu (S&M)’s experiment,
in which 347 articles were used as the training set
                                                                
5 The F-score is a measure combining into a single figure.
We use the F-score version in which precision (P) and recall
(R) are weighted equally, defined as 2*P*R/(P+R).
and 38 ones were used as the test set. S&M
evaluated their system using slightly different
criteria than those used in this research. They
computed the accuracy of the discourse seg-
ments, and the accuracy of the sentence-level
discourse trees without labels, with 18 labels and
with 110 labels. It is not clear how the sentence-
level discourse trees are considered as correct.
The performance given by the human annotation
agreement reported by S&M is, therefore, differ-
ent than the one used in this paper. To compare
the performance between our system and SPADE
at the sentence-level, we calculated the difference
of F-score between the system and the analyst.
Table 2 presents the performance of SPADE
when syntactic trees from the Penn Treebank
were used as the input.
Discourse
segments
Un-
labe lled
110
labels
18 la-
bels
SPADE 84.7 73.0 52.6 56.4
Human 98.3 92.8 71.9 77.0
F-score(H)
- F-score(S)
13.6 19.8 19.3 20.6
Table 2. SPADE performance vs. human per-
formance
Table 1 and Table 2 show that the discourse
segmenter in our study has a better performance
than SPADE. We considered the evaluation of
the “Unlabelled” case in S&M’s experiment as
the evaluation of Level 2 in our experiment. The
values shown in Table 1 and Table 2 imply that
the error generated by our system is considered
similar to the one in SPADE.
To our knowledge, there is only one report
about a discourse parser at the text-level that
measures accuracy (Marcu 2000). When using
WSJ documents from the Penn Treebank,
Marcu’s decision-tree-based discourse parser re-
ceived 21.6% recall and 54.0% precision for the
Level 1 2 3 4a 4b 5 6 7a 7b
Precision 88.2 68.4 61.9 53.9 54.6 54.5 47.8 39.6 40.5
Recall 85.6 64.4 58.3 50.7 51.4 52.9 46.4 38.5 39.3System
F-score 86.9 66.3 60.0 52.2 53.0 53.7 47.1 39.1 39.9
Precision 98.7 88.4 82.6 69.2 74.7 73.0 65.9 53.0 57.1
Recall 98.8 88.1 82.3 68.9 74.4 72.4 65.3 52.5 56.6Human
F-score 98.7 88.3 82.4 69.0 74.5 72.7 65.6 52.7 56.9
F-score(Human) –
F-score(System)
11.8 22 22.4 16.8 21.5 19.0 18.5 13.7 17.0
Table 1. Our system performance vs. human performance
span nuclearity; 13.0% recall and 34.3% preci-
sion for discourse relations. The recall is more
important than the precision since we want dis-
course relations that are as correct as possible.
Therefore, the discourse parser presented in this
paper shows a better performance. However,
more work needs to be done to improve the sys-
tem’s reliability.
As shown in Table 1, the accuracy of the dis-
course trees given by human agreement is not
high, 52.7% in case of 22 relations and 56.9% in
case of 14 relations. This is because discourse is
too complex and ill defined to easily generate
rules that can automatically derive discourse
structures. Different people may create different
discourse trees for the same text (Mann and
Thompson 1988). Because of the multiplicity of
RST analyses, the discourse parser should be
used as an assistant rather than a stand-alone
system.
5 Conclusions
We have presented a discourse parser and evalu-
ated it using the RST corpus. The presented dis-
course parser is divided into two levels: sentence-
level and text-level. The experiment showed that
syntactic information and cue phrases are quite
effective in constructing discourse structures at
the sentence-level, especially in discourse seg-
mentation (86.9% F-score). The discourse trees at
the text-level were generated by combining the
hypothesized discourse relations among non-
overlapped text spans. We concentrated on solv-
ing the combinatorial explosion in searching for
discourse trees. The constraints of textual adja-
cency and textual organization, and a beam
search were applied to find the best-quality trees
in a search space that is much smaller than the
one given by Marcu (2000). The experiment on
documents from the RST corpus showed that the
proposed approach could produce reasonable re-
sults compared to human annotator agreements.
To improve the system performance, future work
includes refining the segmentation rules and im-
proving criteria to select optimal paths in the
beam search. We would also like to integrate a
syntactic parser to this system. We hope this re-
search will aid the development of text process-
ing such as text summarization and text
generation.

References

Lynn Carlson, Daniel Marcu, and Mary Ellen Oku-
rowski 2002. Building a Discourse-Tagged Corpus
in the Framework of Rhetorical Structure Theory.
In Current Directions in Discourse and Dialogue,
Kluwer Academic Publishers.

Katherine Forbes, Eleni Miltsakaki, Rashmi Prasad,
Anoop Sarkar, Aravind Joshi and Bonnie Webber
2003. D-LTAG System: Discourse Parsing with a
Lexicalized Tree-Adjoining Grammar. Journal of
Logic, Language and Information, 12(3), 261-279.

Edward Hovy 1993. Automated Discourse Generation
Using Discourse Structure Relations. Artificial Intelligence, 63: 341-386.

Huong T. Le and Geetha Abeysinghe 2003. A Study to
Improve the Efficiency of a Discourse Parsing
System. In Proc of CICLing-03, 104-117.

Diane Litman and Julia Hirschberg 1990. Disambiguating cue phrases in text and speech. In Proc. of
COLING-90. Vol 2: 251-256.

William Mann and Sandra Thompson 1988. Rhetori-
cal Structure Theory: Toward a Functional Theory
of Text Organization. Text, 8(3): 243-281.

Daniel Marcu 2000. The theory and practice of dis-
course parsing and summarization. MIT Press,
Cambridge, Massachusetts, London, England.

Livia Polanyi, Chris Culy, Gian Lorenzo Thione and
David Ahn 2004. A Rule Based Approach to Discourse Parsing. In Proc of SigDial 2004.

Gisela Redeker 1990. Ideational and pragmatic mark-
ers of discourse structure. Journal of Pragmatics,
367-381.

RST-DT 2002. RST Discourse Treebank. Linguistic
Data Consortium. http://www.ldc.upenn.edu/Cata-
log/CatalogEntry.jsp?catalogId=LDC2002T07.

Lloyd Rutledge, Brian Bailey, Jacco van Ossenbruggen, Lynda Hardman, and Joost Geurts 2000. Generating Presentation Constraints from Rhetorical
Structure. In Proc of HYPERTEXT 2000.

Deborah Schiffrin 1987. Discourse markers. Ca m-
bridge: Cambridge University Press.

Radu Soricut and Daniel Marcu 2003. Sentence Level
Discourse Parsing using Syntactic and Lexical Information. In Proc of HLT-NAACL 2003.

Mark Torrance, and Nadjet Bouayad-Agha 2001.
Rhetorical structure analysis as a method for understanding writing processes. In Proc of MAD 2001.  
