Assertions from Discourse Structure 
William C. Mann 
USC Information Sciences Institute 
and 
Sandra A. Thompson 
UCLA Linguistics Department 
I nt rod u ction 
As part of an ongoing study of discourse structure of natural texts, we have 
identified a particular class of propositions that affect the bearer's perception of the 
coherence and communicated content of texts. As an example, if the text (spoken in 
a suitable situation) is: 
"I'm hungry. Let's go to the Fuji Gardens." 
then the most natural interpretation is that the Fuji Gardens is a restaurant at which 
the speaker would like to eat with the hearer. The text is heard as exhibiting a 
problem-and-solution structure. Consequently, we can say that there is a proposition 
which says that there is a "solutionhood" relation between the two sentences. In this 
case, going to the Fuji Gardens (partially) solves the hunger problem. 
The solutionhood construct is one type of re/ational propositions. Note that 
the proposition about solutionhood is not stated explicitly in the text. 
Although phenomena resembling relational propositions have been 
recognized, there is no widely accepted explanation of how they arise from text. This 
paper characterizes relational propositions and presents a theory of discourse 
structure to explain them. In this Rhetorical Structure Theory (RST), relational 
propositions arise in direct correspondence to particular elements of the structure of 
a discourse. 
We present Rhetorical Structure Theory progressively during analysis of a 
published, two-paragraph political advocacy text. The text involves substructures for 
informing, giving evidence, conceding, requesting an action, justifying a presentation, 
asserting conditionally, and others. 
The two elements that form the basis for this paper, relational propositions and 
Rhetorical Structure Theory, have both been described in more detail elsewhere. The 
explanatory relation between them, however, has not \[Mann & Thompson 83, Mann 
84\]. 
257 
\[Mann & Thompson 83\] examines various other theoretical constructs, 
including implicature, presupposition and indirect speech acts, to see whether they 
account for the textual properties of relational propositions. It concludes that these 
constructs do not account for them. The paper also discusses relationships between 
relational propositions and the work of Grime& Hobbs, van Dijk, Martin, Longacre, 
Beekman and Callow, and it includes analyses of several texts. 
1 The Phenomenon of Relational Propositions 
The Fuji Gardens statement and the solutionhood relation have already 
illustrated that relational propositions need not be signalled explicitly in order to be 
recognized. 1 Extending the example, we now describe properties relational 
propositions hold more generally, giving special attention to those properties that can 
be accounted for by Rhetorical Structure Theory. 
1.1 Relational Propositions Assert 
In our informal presentations of relational propositions, virtually everyone 
recognizes that texts such as the political letter analyzed in this paper convey the 
particular relational propositions that we attribute to them, even though it does not 
represent them explicitly. If the text author were to deny a particular relational 
proposition, most readers would be surprised--and puzzled about the status of the 
part of the text containing that proposition. 
This general consensus testifies that the relational proposition is conveyed. 
Further evidence lies in the apparent redundancy or somewhat bizarre markedness 
that occurs when the relational proposition is asserted explicitly by adding a clause to 
the text: 
"I'm hungry. Let's go to Fuji Gardens. Our going to Fuji Gardens would 
contribute significantly to solving the problem of my hunger." 
1.2 Relational Propositions are Coherence Producing 
One way to demonstrate that a relational proposition is coherence-producing 
is to insert a denial of the relational proposition into the text. Doing so makes some 
portion of the text a non-sequitur: 
"I'm hungry. Let's go to the Fuji Gardens. Of course, going to the Fuji 
Gardens won't do anything about my hunger." 
1We use constructed examples in this section only, tor expository reasons. RST is a theory of natural 
text; it was developed entirely on natural texts, such as the political advocacy text analyzed below. 
258 
The second sentence above has become a non-sequitur, and as a result the 
text as a whole is incoherent. Relational propositions are always 
coherence-producing in this way. We will see later that this is a consequence 
predictable from RST, particularly from the structural forms that RST posits. Also, 
relational propositions are always present in coherent multisentence texts. 
1.3 Other Kinds of Relational Propositions 
The list below names several kinds of relational propositions besides 
solutionhood, and gives an example of an asserted, coherence-producing proposition 
for each. These are drawn form the larger collection of \[Mann & Thompson 83\]; we 
believe that still more kinds of relational propositions could be discovered or perhaps 
created .2 
EVIDENCE: They're having a party again next door. I couldn't find a parking 
space. 
ELABORATION: I love to collect classic automobiles. My favorite car is my 1899 
Duryea. 
MOTIVATION: Take Bufferin. The buffering component prevents excess 
stomach acid. 
THESIS/ANTITHESIS:Players want the referee to balance a bad call benefiting one team 
with a bad call benefiting the other. As a referee, I just want to call 
each play as I see it. 
CONCESSION: I know you have great credentials. I'm looking for someone with 
great experience. 
CONDITION: Give her a subscription to Science magazine. She'll be in seventh 
heaven. 
REASON: I'm going to the corner. We're all out of milk. 
JUSTIFICATION: Let me make one thing perfectly clear. I am not a crook. 
We desire a theory that will answer two questions about relational 
propositions: 
1. What relational propositions are possible? 
2We have abundant natural correlates for these constructed examples. They have been designed to 
illustrate the fact that the relations and relational propositions are identifiable, without any explicit 
signalling, such as a dause, conjunction, or lexical selection. 
259 
2. What relational propositions does a particular text assert? 
The answers come from studying discourse. 
2 Rhetorical Structure Theory 
RST has not been developed simply to account for relational propositions; it 
arose from a much broader desire to understand text and communication and to learn 
how texts may be created. We identified and began studying relational propositions 
only after RST had largely assumed its present shape. 
We wanted a theory of text organization--a way to describe what kinds of parts 
a text can have, how they can be arranged, and how parts can be connected to form a 
whole text. We especially valued the following attributes. 
1. Comprehensiveness: The theory should apply to many kinds of text. 
2. Functionality: The theory should be informative in terms of how text 
achieves its effects for the writer. 
. Scale insensitivity: The theory should apply to a wide range of sizes of 
text and should be capable of describing all of thevarious sized units of 
text organization that occur in a large text. 
4. Definiteness: The theory should lend itself to formalization and 
computer programming; 
5. Constructive potential: The theory should be usable in text 
construction as well as text description. 
We developed this theory primarily in response to small written texts, although 
it has also been applied to larger texts. We have constructed RST descriptions for a 
variety of texts, including: 
- Administrative memos 
- Personal letters 
- Advertisements 
- Editorial letters in magazines 
- Complete Scientific American articles 
- Newspaper articles 
- Public notices in magazines 
- Research technical reports 
- Travel brochures 
- Cookbook recipes 
260 
To introduce the theory, let us consider the analysis of a text that appeared in 
a political newsletter, The Insider, Vol. 2.1, July 1982. 3 The Insider is the California 
Common Cause state newsletter. This text was the "con" part of a "pro" and "con" 
pair of letters on the issue of California Common Cause's endorsement of the nuclear 
freeze initiative, which was then on the California state ballot. 
The text has been reformatted, numbered, and divided the text into units. 
Units are roughly equivalent to clauses, except that that relative clauses and 
complement clauses are considered to be part of the unit in which their governing 
item appears, rather than as independent units. As long as the whole text is analyzed, 
the size of the minimal units can vary without affecting the larger analysis. 
1. I don't believe that endorsing the Nuclear Freeze Initiative is the right step 
for California CC. 
2. Tempting as it may be, 
3. we shouldn't embrace every popular issue that comes along. 
4. When we do so 
5. we use precious, limited resources where other players with superior 
resources are already doing an adequate job. 
6. Rather, I think we will be stronger and more effective 
7. if we stick to those issues of governmental structure and process, broadly 
defined, that have formed the core of our agenda for years. 
8. Open government, campaign finance reform, and fighting the influence of 
special interests and big money, these are our kinds of issues. 
3Le~er used by permission 
261 
9. (New paragraph) Let's be clear: 
10. I personally favor the initiative and ardently support disarmament 
negotiations to reduce the risk of war. 
11. But I don't think endorsing a specific nuclear freeze proposal is 
appropriate for CCC. 
12. We should limit our involvement in defense and weaponry to matters of 
process, such as exposing the weapons industry's influence on the 
political process. 
13. Therefore, I urge you to vote against a CCC endorsement of the nuclear 
freeze initiative. 
(signed) Michael Asimow, California Common Cause Vice-Chair and 
UCLA Law Professor 
How is this text organized? At the most general level, the text as a whole 
functions as a request to vote in a certain way. At its coarsest level of decomposition, 
it has two parts. One part presents the request, presented in segment 13, and the 
remainder supports that presentation. 
The theory has a number of patterns, called rhetorical schemas, that represent 
organizational information about text. To represent the particulars of two-part 
decomposition of the text, we use one of these rhetorical schemas, the Request 
Schema, Figure 1. 
Request 
I .._T.~.~a b lem en t 
Figure 1: Request Schema 
A text that instantiates the Request Schema has a nuclear part, called the 
nucleus, that presents a request. It also has one or more supplementary parts, called 
satellites, that are functionally related to the nucleus. Satellites are related to the 
nucleus by a named relation. Here we have relations named motivation and 
enablement. 
262 
Let us illustrate the parts of a Request Schema in a short example: 
"Call me. I have a surprise for you. My extension number is 110." 
The nucleus is "Call me," the motivation satellite is "1 have a surprise for you," and 
the enablement satellite is "My extension number is 110." These elements can be 
arranged in any order and still be an instance of the Request Schema. Schemas do 
not encode the order of segments; in this case, the segments can be rearranged freely 
without disturbing their meaning or structural relation. 
Satellites are all optional, so we can delete either one in this example and still 
instantiate the Request Schema-but there must be at least one satellite. The political 
text has a motivation satellite, segments 1 through 12, but no enablement satellite. 
We analyze each of the two top-level segments of the political text in the same 
way. The final segment is a single unit, so we don't try to divide it. The first segment, 
12 units long, consists of a claim (unit 1) and two arguments that give evidence for the 
claim. We analyze this arrangement with the Evidence Schema (Figure 2), in which 
the claim is the nucleus and an evidence relation connects the nucleus to the ,satellite. 
Figure 3 shows the resulting structure. 
Evidence 
Figure 2: Evidence Schema Request 
Evidence 
~idence // 
2 3 4 5 6 
Figure 3: 
7 8 9 10 11 12 
The Upper Structure of the Political Text 
13 
263 
Both of the nuclei obey what we call the Most Favorable Audience Rule: F.or 
the most knowledqeable and D0sitivelv Dredisoosed hearer, the nucleus al0n@ would 
be sufficient to perform the function of the structure; the ~atellit~ function to inCreasE 
the likelihood that the nuCleus will succeed. This rule is a summary of many 
observations about the rhetorical structures of texts. It does not always hold, but 
there is a strong, unexplained tendency for it to hold. 
In this case of the Request Schema, presenting the request (to vote in a certain 
way) to a favorably predisposed hearer would be enough to get that reader to vote as 
desired; the supporting argument makes the desired vote more likely for most readers. 
This application of the Evidence Schema contains two arguments: One says 
that the proposal is wasteful, and the other says that better alternatives exist. They 
make the reader more likely to accept the claim that endorsement is not right. 4 
The analysis goes on, down to single units. 
schemas used. They are drawn from 
through use of about 30 relations. 
Thesix /Antithesis J 
~sis~tithesis 
Figure 4 shows the additional 
a larger set of about 25 schemas, defined 
Concessive 
concession~ / 
lnform 
backg . /rou~ 
Just05 justification/"~"~"~ \[ Co.ditional 
Figu re 4: Other Schemas Used to Analyze the Political Text 
4Although the unit begins "1 don't believe that...," the claim here is really about whether this step is right 
for CCC. The evidence that follows in units 2 through 11 is about what benefits C, CC, not about whether the 
author believes this claim or not. RST does not represent the indirectness of the form of the claim. 
264 
To illustrate the relations used here, we turn back to the text. 
The thesis~antithesis relation connects units 11 and 12. 11: "But I don't think 
endorsing a specific nuclear freeze proposal is appropriate for CCC." 12: "We 
should limit our involvement..." 
The concessive relation connects units 2 and 3: 2: "Tempting as it may be," 
3: "we shouldn't embrace every popular issue that comes along." 
Unit 8 is an elaboration for the Inform Schema; it lists instances, such as open 
government and campaign finance reform. 
Unit 9 ,says "Let's be clear." This is in a justification relation to the argument 
that follows, in 10 through 12. It obtains permission to present a second argument, 
defending the same conclusion. 
Finally, units 4 and 5 are in a condition relation. 4: "When we do so..." 5: "we 
use precious resources..." 
Figure 5 shows the structure of the whole text. 
2.1 Definition Mechanisms of RST 
How is RST defined? How do claims about particular texts arise from an RST 
analysis of it? The theory is defined in terms of just three mechanisms: schemas, 
schema application conventions, and relation definitions. 
Schemas are simply sets of relations. There is no schema-specific information 
beyond the identities of the relations that comprise the schema. 
Schema application conventions are descriptions of what it means for a 
particular span of text to instantiate a schema. Its conventions are easy to state: 
1. One schema is instantiated to describe the entire text. 
2. Schemas are instantiated to describe the text spans produced in 
instantiating other schemas. 
3. The schemas do not constrain the order of nucleus or satellites in the text 
span in which the schema is instantiated. 
4. All satellites are optional. 
5. At least one satellite must exist. 
6. A relation that is part of a schema may be instantiated indefinitely many 
times in the instantiation of that schema. 
265 
e¢ c! 
o 
c 
o 
/ 
ill 
C, 
/ 
=. 
- i C <1o 
I o 
0 r 
el >~ 
Figu re 5: Full Rhetorical Structure of the Political Text 
266 
7. The nucleus and satellites do not necessarily correspond to a single 
uninterrupted text span. 
8. The relation definition must be consistent with the spans of text related by 
the instantiation of the schema containing the relation. 
It is possible for the conventions to apply to a text in more than one way, so 
that the text is rhetorically ambiguous. 
A relation definition specifies three kinds of information: 
1. A characterization of the nucleus. 
2. A characterization of the satellite. 
3. A characterization of what sorts of interactions between the conceptual 
span of the nucleus and the conceptual span of the satellite must be 
plausible. 
To define, for example, the motivation relation, we would include at least the 
following. 
1. The nucleus describes an action performable, but not yet performed, by 
the reader. 
2. The satellite describes the action, the situation in which the action takes 
place, or the result of the action in ways that help the reader associate 
value assessments with the action. 
3. The expected value assessments are positive, to lead the reader to want 
to perform the action. 
The relational propositions do not arise independently of the relation 
definitions. Rather, finding that a relation definition holds is sufficient to establish the 
corresponding relational proposition. As readers recognize the functional relations of 
the parts, they are recognizing that the relation definitions hold. The content of the 
relational proposition is identified in this process. As a consequence, the definition 
scheme for RST requires no additional definitions in order to specify the relational 
proposition. In any particular case, the proposition can be derived from the way the 
relation definition fits. 
We have found the relation definitions, useful in predicting other facts about 
the text, such as the kinds of conjunctions that will appear in certain places. We have 
analyzed a large number of texts, including thousands of clauses, in this way. We are 
confident that we can perform this analysis, with fairly high reliability, for virtually any 
small, written, multisentence monologue in general American culture, using only 
about 25 schemas. 
Note that these rhetorical schemas are defined in terms of the functions of 
segments of text. The evidence relation applies when one segment supports another 
as evidence, Solutionhood applies when we can see one segment as identifying a 
267 
problem and another as a partial solution to that problem. These are not criteria of 
form; as one might expect, the relationship of these function categories to form 
categories is quite loose. The rhetorical structure of text, in these terms, is composed 
of function-specific units. The structure does not express categories of knowledge or 
form as much as it expresses the roles of specific parts in relation to the whole text, 
especially the role of each satellite relative to one particular, immediate portion of the 
text, the corresponding nucleus. 
3 RST as an Account of Relational Propositions 
The key observation for the purposes of this paper is: 
For every relation of the rhetorical structure of a text, a 
corresponding relational proposition is asserted. 
For solutionhood relations, the discourse structure asserts a solutionhood 
proposition. For evidence relations it asserts an evidence proposition, and so forth. 
Readers attribute the assertions to the text because they recognize the functional 
relations of the parts. 
Now we can explain why relational pi'opositions are coherence-producing. 
First, RST structures always have the connectivity of trees. The schema 
application conventions guarantee this, because when a span is decomposed, each 
of the parts is further decomposed separately. 
If a portion of the text is to be connected to the whole without being a 
non-sequitur, some relation must be established. A relation is established through 
implicit assertion of a relational proposition. Since the relations form a tree, denial of 
any one relational proposition is sufficient to separate the structure into two parts, 
thus destroying connectedness, a key attribute of coherence. 
Now we can also explain why relational propositions are always present in 
coherent multisentence texts. In regarding the text as a single whole, readers impute 
rhetorical structure to it, necessarily positing relations between the parts; the relations 
give rise to assertion of relational propositions. 
We can also see how to create more precise specifications of relational 
propositions. They can be developed from the relation definitions of RST. RST tells 
what sorts of propositions can be relational, gives the conditions under which 
relational propositions arise, and tells how to alter a text or a situation so that the 
asserted relational proposition is changed. Rather than simply searching texts for 
potential relational propositions, we can search rhetorical structures for the 
necessary relational propositions. 
268 
4 Uses of Rhetorical Structure Theory 
Rhetorical Structure Theory provides an attractive basis for explaining 
relational propositions, although some details need development. 
In addition, RST satisfies some of the attributes identified in section 2, above, 
as desirable for a descriptive theory of text organization. It is comprehensive enough 
to apply to many different kinds of text; it is functional, in that it explains what various 
portions of a text do for the writer. And it is scale insensitive, applying to a wide range 
of sizes of units, from simple clauses up to whole magazine articles. 
However, RST still lacks two desired attributes: It needs for more detailed 
expression of each part, and it would be useful to develop a constructive counterpart 
to the descriptions, a way to select schemas and plan texts. 
In addition to these attributes, 
benefits of RST. 
we recognize other opportunities for and 
1. It gives a partial account of the distribution of interclausal and 
intersentential conjunctions. 
2. It leads to new observations of text phenomena, including some related to 
nuclei, such as the most-favorable-reader hypothesis. 
3. The advantages of a recursive theory are obtained for text structure. 
Beyond the phenomena identified above, RST appears to be useful in 
accounting for other kinds of discourse phenomena. We have found no boundary for 
its uses; it is like trying to delimit the uses of a grammar. We have identified the 
following as particularly attractive applications: 
• Thematization and text development 
. Distributions of tense selections in text 
- Lexical selection 
- Patterned shifts of hypotheticality, identifiability, or conditionality 
- Patterns of use of conjunctions 
- Purposeful clause combining 
. Distribution of topicalization markers 
- Text ordering (under way) 
- Relating coherence to cohesive devices 
5 Summary 
The assertion of relational propositions is a hitherto unexplained 
phenomenon. Rhetorical Structure Theory provides a way to explain such assertions 
in terms of discourse structure. In addition to explaining relational propositions, 
269 
Rhetorical Structure Theory can be used to explain other text characteristics as well, 
and it provides a way to address a wide range of discourse phenomena. 
References 
\[Mann 84\] Mann, W. C., Discourse Structures tor Text Generation, USC/Information 
Sciences Institute, Technical Report RR-84-127, February 1984. Also appeared 
in the proceedings of the 1984 Coling/ACL conference, July 1984. 
\[Mann & Thompson 83\] Mann, W. C., and S. A. Thompson, Relational Propositions in 
Discourse, USC/Information Sciences Institute, Technical Report RR-83-115, 
July 1983. To appear in Discourse Processes. 
270 
