Get Constraints in Discourse 172nd Edition Anton Benz (Ed.) free all chapters

Download the full version of the ebook now at ebookultra.com
Constraints in Discourse 172nd Edition Anton
Benz (Ed.)
https://ebookultra.com/download/constraints-in-
discourse-172nd-edition-anton-benz-ed/
Explore and download more ebook at https://ebookultra.com

Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.
Wilhelm Reich in Hell Robert Anton Wilson
https://ebookultra.com/download/wilhelm-reich-in-hell-robert-anton-
wilson/
ebookultra.com
Conversion to Islam in the Balkans Anton Minkov
https://ebookultra.com/download/conversion-to-islam-in-the-balkans-
anton-minkov/
ebookultra.com
Uncle Vanya Anton Chekhov
https://ebookultra.com/download/uncle-vanya-anton-chekhov/
ebookultra.com
Advances in Insect Physiology Vol 29 1st Edition Peter
Evans
https://ebookultra.com/download/advances-in-insect-physiology-
vol-29-1st-edition-peter-evans/
ebookultra.com

Calculus 10th Edition Howard Anton
https://ebookultra.com/download/calculus-10th-edition-howard-anton/
ebookultra.com
Lenin as Philosopher Anton Pannekoek
https://ebookultra.com/download/lenin-as-philosopher-anton-pannekoek/
ebookultra.com
Immunogenetics 1st Edition Anton W. Langerak
https://ebookultra.com/download/immunogenetics-1st-edition-anton-w-
langerak/
ebookultra.com
Anton Chekhov 1st Edition Rose Whyman
https://ebookultra.com/download/anton-chekhov-1st-edition-rose-whyman/
ebookultra.com
Adobe InDesign Classroom in a Book 2022 release 1st
Edition Kelly Anton
https://ebookultra.com/download/adobe-indesign-classroom-in-a-
book-2022-release-1st-edition-kelly-anton/
ebookultra.com

Constraints in Discourse 172nd Edition Anton Benz (Ed.)
Digital Instant Download
Author(s): Anton Benz (ed.), Peter KÃ¼hnlein (ed.)
ISBN(s): 9789027291431, 9027291438
Edition: 172
File Details: PDF, 4.06 MB
Year: 2008
Language: english

Volume 172
Constraints in Discourse
Edited by Anton Benz and Peter Kühnlein
Editor
Andreas H. Jucker
University of Zurich, English Department
Plattenstrasse 47, CH-8032 Zurich, Switzerland
e-mail: ahjucker@es.uzh.ch
Editorial Board
Shoshana Blum-Kulka
Hebrew University of
Jerusalem
Jean Caron
Université de Poitiers
Robyn Carston
University College London
Bruce Fraser
Boston University
Thorstein Fretheim
University of Trondheim
John C. Heritage
University of California at Los
Angeles
Susan C. Herring
Indiana University
Masako K. Hiraga
St.Paul’s (Rikkyo) University
David Holdcroft
University of Leeds
Sachiko Ide
Japan Women’s University
Catherine Kerbrat-
Orecchioni
University of Lyon 2
Claudia de Lemos
University of Campinas, Brazil
Marina Sbisà
University of Trieste
Associate Editors
Jacob L. Mey
University of Southern
Denmark
Herman Parret
Belgian National Science
Foundation, Universities of
Louvain and Antwerp
Jef Verschueren
Belgian National Science
Foundation,
University of Antwerp
Emanuel A. Schegloff
University of California at Los
Angeles
Deborah Schiffrin
Georgetown University
Paul Osamu Takahara
Kobe City University of
Foreign Studies
Sandra A. Thompson
University of California at
Santa Barbara
Teun A. van Dijk
Pompeu Fabra, Barcelona
Richard J.Watts
University of Berne
Pragmatics & Beyond New Series is a continuation of Pragmatics & Beyond and
its Companion Series. The New Series offers a selection of high quality work
covering the full richness of Pragmatics as an interdisciplinary field, within
language sciences.
Pragmatics & Beyond New Series (P&BNS)

Constraints in Discourse
Edited by
Anton Benz
Zentrum für Allgemeine Sprachwissenschaften
Peter Kühnlein
Rijksuniversiteit Groningen
John Benjamins Publishing Company
Amsterdam/Philadelphia

Library of Congress Cataloging-in-Publication Data
Constraints in discourse / edited by Anton Benz, Peter Kuhnlein.
p. cm. (Pragmatics & Beyond New Series, issn 0922-842X ; v. 172)
Includes bibliographical references and index.
1. Discourse analysis. 2. Constraints (Linguistics) I. Benz, Anton, 1965- II. Kühnlein,
Peter.
P302.28.C66 2008
401'.41--dc22 2007048314
isbn 978 90 272 5416 0 (Hb; alk. paper)
© 2008 – John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands
John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa
The paper used in this publication meets the minimum requirements of
American National Standard for Information Sciences – Permanence of
Paper for Printed Library Materials, ansi z39.48-1984.
8
TM

Table of contents
Acknowledgements vii
1. Constraints in discourse: An Introduction 1
part i
The Right Frontier 27
2. Troubles on the right frontier 29
Nicholas Asher
3. The moving right frontier 53
Laurent Prévot and Laure Vieu
part ii
Comparing Frameworks 67

4. Strong generative capacity of rst, sdrt and discourse dependency dags 69
Laurence Danlos
5. Rhetorical distance revisited: A parameterized approach 97
Christian Chiarcos and Olga Krasavina
6. Underspecified discourse representation 117
Markus Egg and Gisela Redeker
part iii
The Cognitive Perspective 139
7.
Dependency precedes independence: Online evidence from discourse
processing 141
Petra Burkhardt
8.
Accessing discourse referents introduced in negated phrases: Evidence for
accommodation? 159
Barbara Kaup and Jana Lüdtke

 Table of contents
part iv
Language Specific Phenomena 179
9. Complex anaphors in discourse 181
Manfred Consten and Mareile Knees
10. The discourse functions of the present perfect 201
Atsuko Nishiyama and Jean-Pierre Koenig
11. German right dislocation and afterthought in discourse 225
Maria Averintseva-Klisch
12. A discourse-relational approach to continuation 249
Anke Holler
13. German Vorfeld-filling as constraint interaction 267
Augustin Speyer
Index 291

Acknowledgements
The contributions collected in this volume are based on the proceedings of the first
conference on Constraints in Discourse held at the University of Dortmund. All con-
tributions have been reviewed again and thoroughly revised before publication. The
conference was organised by the two editors Anton Benz and Peter Kühnlein together
with Claudia Sassen. Both editors regret that Claudia Sassen, who did a great job at
organising the conference, had to leave the editorial board.
We thank Angelika Storrer from the Institute for German Language at the Univer-
sity of Dortmund as well as the Deutsche Forschungsgemeinschaft for their financial
support. Furthermore, we have to thank our employers, the IFKI at the University of
Southern Denmark, the University of Bielefeld, the ZAS in Berlin and the University
of Groningen for their help and encouragement.
John Tammena has helped reduce the unreadability of our introductory chapter.
We want to thank him as well as Paul David Doherty who helped setting up the index.
Our special thanks, however, go to Andreas Jucker, the series editor of PBns, and
of course to Isja Conen from John Benjamins’ publishing company, for their untiring
help and patience.

Constraints in discourse
An introduction
1. General remarks
For a long time the development of precise frameworks of discourse interpretation has
been hampered by the lack of a deeper understanding of the dependencies between
different discourse units. The last 20 years have seen a considerable advance in this
field. A number of strong constraints have been proposed that restrict the sequencing
and attaching of segments at various descriptive levels, as well as the interpretation of
their interrelations. An early and very influential work on the sequencing and acces-
sibility of expressions across sentence boundaries was concerned with the rfc (Right
Frontier Constraint), often associated with a paper by Polanyi (1988). The rfc formu-
lates a restriction on the possible discourse positions of pronominal expressions. Another
much discussed constraint governing pronominal reference is the centering principle
formulated by Grosz and Sidner (1986). In addition to the proposal of new discourse
constraints, recent years saw the development of competing formal frameworks for
discourse generation and interpretation, most importantly, Rhetorical Structure Theo-
ry (rst, Mann and Thompson 1987) and Segmented Discourse Representation Theory
(sdrt). Especially the recent publication of Asher and Lascarides (2003), which sum-
marises more than ten years of joint research in sdrt, gave a strong impulse to the field
of discourse semantics and led to the publication of an increasing number of papers.
Constraints play a role not only in diverse fields of linguistics, but in a wide variety
of fields of research in general, such as computer science, especially artificial intelli-
gence (cf., e.g., (Blache 2000)). What the use of constraints has in common in all these
fields is that they describe properties of objects in order to specify whether certain
objects are well-formed from the point of view of the background theory. As soon as
an object carries the property or properties specified by all of the constraints defined
by the theory, it counts as well-formed and is accepted as (part of) a model of the
theory. The object is then said to satisfy the constraints set by the theory.
Inthepresentcollection,anumberofauthorscontributedtodefineconstraintsthus
understood to specify properties that are relevant in the context of research on dis-
course. The multiplicity of identified constraints mirrors the multiple facets of this re-
search area itself. To give a rough understanding of major issues in discourse research,
we will lay out three paradigms in this introduction and relate them to each other and
to the texts in this volume.
The three paradigms we selected share a focus on rhetorical relations: a discourse
is conceived as such only if every part of it is connected to the rest via certain relations

 Constraints in discourse — an introduction
that specify its role. This property of discourse is classically related to coherence and
cohesion and can be used as a constraint to distinguish well-formed discourses from
arbitrary sets of objects.
The paradigms were developed during the last 20 years and within their frame-
works, a number of such constraints have been proposed for the description and
explanation of the multiplicity of dependencies between units of discourse. Segmented
Discourse Representation Theory (sdrt), for example, posits a selection principle over
interpretations of discourse: among possible interpretations of a discourse the one is
selected that renders the discourse as coherent as possible. This is operationalised via
the number of rhethorical relations that connect parts of the discourse and an order-
ing over preferences for those relations: the more the better, given their type for some
discourse. This principle is called Maximise Discourse Coherence (mdc) and of course
is a constraint over the selection of interpretations as well as discourses: of those
interpretations that can be generated for a given discourse only those are acceptable
that have the highest possible degree of coherence. And among objects generally only
those count as discourse for which some interpretation establishes coherence. Con-
sider what would happen if (1b) and (1c) were exchanged in example (1), taken from
(Asher and Lascarides 2003); the resulting discourse would clearly be less acceptable,
and one might well argue that this would be due to the loss of coherence.
(1) a. One plaintiff was passed over for promotion three times.
b. Another didn’t get a raise for five years.
c.
A third plaintiff was given a lower wage compared to males who were
doing the same work.
d. But the jury didn’t believe this.
One prominent constraint that is recognised by almost all theories of discourse is the
so-called Right Frontier Constraint (rfc), see especially the chapters in Part I of this
book. This constraint amounts to a restriction over attachment points in a discourse.
(We will give a short characterization here and discuss the rfc a little more extensively
in Section 3.) Consider Example (1) again. Under any reasonable interpretation, (1d)
can only be related to either the immediately preceding utterance (1c) or to the totality
of the preceding utterances (1a–1c). In the first case, what the jury didn’t believe was
just the fact that one plaintiff was given a lower wage compared to males who were
doing the same work. In the second case, the jury wouldn’t believe any of the reported
facts. What should not be possible—and that is the claim connected with the rfc—is
an attachment of (1d) to (1a) or (1b) alone. These two utterances should be blocked as
attachment points.
The name Right Frontier Constraint derives from an assumption over representa-
tions stating that more recent utterances, or, more general, constituents in a discourse
are graphically represented to the right of less recent ones. Discussion of formal repre-
sentations of discourse structure and measures of anaphoric distances can be found in
the chapters of Part II of this book. The most recent constituents in discourse (1) prior

Constraints in discourse — an introduction 
to the utterance of (1d) are either (1c) or the compound constituent (1a–1c), which
makes these two being situated on the right hand side of the representation given
this assumption. As accordingly all and only those constituents that are accessible for
pronominal anaphoric attachment are on the right hand side of the representation,
this constraint is called rfc.
As a reaction to the variety of constraints, there will be discussions on a broad
spectrum of restrictions on well-formedness, be these universal, language indepen-
dent restrictions, like the two mentioned seem to be, or language specific constraints.
It is one interesting property of constraints that they can be more or less specific, and
their effects can add to each other. Thus, one can end up with a very strong filter over
admissible structures by combining constraints that pertain to different properties of
objects. Exemplarily, there are discussions on language-specific constraints that don’t
seem to be readily transferable to other languages from, e.g., German. For more on
language specific constraints, see the chapters in Part IV of this book.
Other chapters, Part III, deal with psycholinguistic or neurolinguistic reflexes of
constraints and their empirical testing. During the processing of discourses by human
participants, the linguistic constraints can be expected to produce effects and generate
preferences for strategies or solutions. These predictions of course should be empiri-
cally testable.
2. The cognitive status of rhetorical relations
The theory of rhetorical relations is a cornerstone of discourse analysis. In general, it
is undisputed that the meaning of text is more than the conjunction of the meanings
of its sentences, but there are different opinions about the cognitive status of rhetori-
cal relations. One position assumes that rhetorical relations are part of the linguistic
inventory of language users and therefore of their linguistic competence. When faced
with a sequence of two text segments, the hearer or reader searches a closed list of
Figure 1. A graphical representation of what it means for a node to be on the right frontier:
node α represents the last utterance in a discourse. α and every node dominating α (like β) is
thus on the right frontier and available for attachment for a subsequent utterance γ.
α
β
γ

 Constraints in discourse — an introduction
rhetorical relations and chooses that relation which fits best, where the criterion for
fitting best varies from theory to theory. From this we may distinguish positions that
assume that the extra information that the reader infers from the concatenation of two
text segments is derived e.g., from assumptions about the speaker’s intentions, com-
monsense world knowledge, and conversational maxims alone. Rhetorical relations
are then not part of our basic linguistic inventory. We may call the first position a
non–reductionist position and the second position a reductionist position. Within re-
ductionist positions we may roughly distinguish between approaches that take their
starting point in plan-based reasoning, and approaches that take their starting point in
Gricean pragmatics. The most important frameworks of discourse analysis discussed
in this volume are non–reductionist in character, e.g., the Linguistic Discourse Model
(Polanyi 1986), Rhetorical Structure Theory (Mann and Thompson 1987), and Seg-
mented Discourse Representation Theory (Asher and Lascarides 2003). As an illustra-
tion, we discuss the following example:
(2) Ann calls a taxi service.

Ann: (1) I need a taxi now. (2) Pick me up at the Dortmund railway station and
(3) drop me at Haus Bommerholz.
The first sentence is a directive speech act asking the taxi service to supply a transpor-
tation to Ann. Propositions (2) and (3) provide more information about the lift. They
elaborate the content of the first sentence. A non–reductionist would assume that there
exists a rhetorical relation Elaboration that is inferred by the addressee. The inference
of text coherence begins with an interpretation of the sentences (1), (2) and (3). The
addressee then searches a mental library of rhetorical relations. We may assume that
it contains the entries Elaboration, Explanation, and Result. Each rhetorical relation
defines constraints that must be fulfilled by text segments which are connected by the
relation. For example, a text segment β can only elaborate a text segment α if β denotes
a sub-eventuality of α, whereas Explanation and Result assume that the eventualities
are non-overlapping and that one is the result of the other. Hence, the addressee can
infer Elaboration, and therefore text coherence, from the fact that the propositions in
(2) and (3) refer to sub-eventualities of the event mentioned in (1). (For more on this
cf. Section 6.)
A reductionist tries to show discourse coherence without reference to a predefined
set of rhetorical relations. Instead, the explanation may for example rest on assumptions
about the speaker’s domain plans. Taking a lift with a taxi is an activity which can be bro-
ken down into being picked up by the taxi at a certain place, the taxi ride, and being dropped
at the destination. Schematically, we can describe this decomposition as follows:
→
1
(S ) TakingTaxi(P) PickUp(P, Time1, Place1), TaxiRide, Drop(P, Time2, Place2)
An analysis of Example (2) may proceed as follows: Sentence (1) states the speak-
er’s domain intention. This activates schema (S1), which is shared knowledge in the

Constraints in discourse — an introduction 
relevant language community. In order to make the directive in (1) felicitous, some of
the parameters in (S1) have to be specified. This is done in sentences (2) and (3); they
state the place of departure Place1 and the destination Place2. Coherence is achieved
by direct reference to a schema like (S1). Discourse becomes incoherent if the hearer
cannot find a domain schema which connects the text segments, as seen in the follow-
ing example:
(3) Ann calls a taxi service.
Ann: (1) I need a taxi now. (2) I grew up in Bielefeld, Ostwestfahlen–Lippe.
A reductionist position which is based on plan recognition is widespread among
approaches in artificial intelligence, e.g., (Grosz and Sidner 1986; Litman and Allen
1990).
The assumption that rhetorical relations are part of our linguistic inventory has
consequences for our understanding of both pragmatics and, especially, conversation-
al implicatures (Grice, 1975). For an example we look at:1
(4) Ann: Smith doesn’t seem to have a girl friend.
Bob: He’s been paying lots of visits to New York lately.
Implicature: Smith possibly has a girl friend in New York (p).
In order to understand Bob’s utterance as a contribution to the ongoing conversation,
Ann has to find a rhetorical relation that connects his utterance to her contribution.
We may assume that there exists a rhetorical relation of Counterevidence. The infer-
ence of Counterevidence can proceed from the semantic content of the utterances and
their prosodic and other linguistic properties. It is not necessary that the inference
takes into account the interlocutors’ intentions.
If Counterevidence holds between Ann’s and Bob’s utterances, then Bob’s utterance
must provide evidence for the negation of Ann’s claim, i.e., it must provide evidence
for the claim that Smith has a girl friend. This is the case if one assumes that Smith
possibly has a girl friend in New York. Hence, the construction of a rhetorical relation
between the two utterances leads to an accommodation of the implicature (p).
We may contrast this reasoning with the standard theory of conversational impli-
catures (Grice 1975), (Levinson 1983, Ch. 3), which assumes that the implicatures are
derived by reasoning about each other’s intentions. According to Grice, interlocutors
adhere to a number of conversational principles which spell out how discourse par-
ticipants should behave in order to make their language use rational and efficient. In
particular, Grice assumes that each contribution to the ongoing conversation serves a
joint goal of speaker and hearer. A possible derivation of the implicature may proceed
1. For a more thorough discussion of this example and the relation between Grice’ theory
of conversational implicatures and the assumption of rhetorical relations see (Asher and
Lascarides 2003, Sec. 2.6).

 Constraints in discourse — an introduction
as follows: (1) Ann’s utterance raises the question whether Smith has a girl friend; (2)
Bob’s contribution must be relevant to this question; (3) Bob’s contribution can only be
relevant if Smith possibly has a girl friend in New York; (4) as Bob has done nothing
in order to stop Ann from inferring that (p), it follows that she safely can infer that (p).
In contrast to the first explanation, this explanation infers implicatures directly from
joint intentions and a general principle of relevance.2
3. Topics in the analysis of discourse constraints
In the previous section, we were introduced to different positions concerning the status
of rhetorical relations. Rhetorical relations provide the backbone of some of the most
important formal frameworks in discourse analysis. In this section, we want to address
some topics in discourse analysis which are related to the investigation of discourse
constraints. We start with constraints related to rhetorical relations and the discourse
structures constructed by them. In this context, we introduce, for example, the Right
Frontier Constraint as first codified by Livia Polanyi (1986) in her ldm (for more detail
see Section 4).
Text coherence is the result of interconnectedness of text segments. The analysis
using rhetorical relations naturally leads to a representation as a graph. The terminal
nodes of the graph can be identified with elementary illocutionary acts. The graph in
Figure 2 shows an analysis of the following example, in which Ann tells how she came
to Haus Bommerholz:
(5)
Ann: (1) I arrived at 10 am. (2) I took a taxi then. (3) It picked me up at the
Dortmund railway station and (4) dropped me at Haus Bommerholz. (5)
I thought it might be quite complicated to get to this place but (6) it wasn’t.
A natural question that arises concerns the general structure of these graphs. First
we may ask, what kind of branches are associated with the different rhetorical rela-
tions. Are they always of the same kind or can we distinguish between different types
of relations? Closely related to this question is that for the types of graphs that can be
generated. For example, the graph in Figure 2 has a tree like structure and only binary
branches. A third question concerns the comparability of different representations.
The tree in Figure 2 is an rst graph (Mann and Thompson 1987). These trees are dif-
ferent from trees which we usually find in syntax. In syntactic trees, the relations that
connect two constituents are normally attached to the branching nodes. In rst graphs
2. Asher and Lascarides (2003) point out that any existing theory of conversational implica-
tures in the tradition of Grice, has to assume that interlocutors carry out costly computations
about each other’s intentions. Hence, a theory of conversational implicatures which is based on
the theory of rhetorical relations is attractive from a cognitive point of view as it makes weaker
assumptions about the inference capabilities of the interlocutors.

Constraints in discourse — an introduction 
they are labels to the edges connecting the nodes. We will see syntax like graphs in the
section about the Linguistic Discourse Model. The answers to the above questions im-
pose more or less strict constraints on discourse. These topics are especially discussed
in the contributions by Danlos (Chapter 4) and Egg Redeker (Chapter 6).
In Figure 2, we can find two types of relations: relations like Elaboration which are
attached to an arch and relations like Narration which are attached to branches starting
from a shared node. Text segments connected by Narration are intuitively on the same
level, whereas a text segment that is attached to another text segment by Elaboration
or Evidence is subordinated to this segment. The distinction between coordinating and
subordinating discourse relations became very influential with (Grosz and Sidner 1986).3
One way of conceptualising the distinction between subordinating and coordinating
rhetorical relations is based on the discourse intentions of the speaker. In Example (2),
the sentences ‘Pick me up at Dortmund railway station’ and ‘Drop me at Haus Bommerholz’
provide information without which the addressee cannot successfully perform what
was asked from him in the first sentence ‘I need a taxi now’. In a coordinated sequence
like ‘(1) I arrived at 10 pm. (2) I took a taxi then.’ neither (1) is uttered in order to sup-
port (2), nor is (2) uttered in order to support (1). Each sentence can stand alone, and
none needs the other in order to justify its occurrence. In contrast, the utterance of (2)
CONTRAST
(5)
EVIDENCE
NARRATION
(1) ELABORATION
(2) NARRATION
(3) (4)
(6)
Figure 2. An analysis of Example (5). The graph shows the rhetorical relations that hold
between text segments.
3. rst distinguishes between multi-nuclear and nucleus-satellite relations. This distinctions is
closely related to Grosz and Sidner’s (1986) distinction between coordinating and subordinating
relations.

 Constraints in discourse — an introduction
‘Pick me up at Dortmund railway station’ in Example (2) cannot be justified without
the information that Ann needs a taxi.
The distinction between coordinating and subordinating discourse relations is in-
corporated in most formal frameworks and in all frameworks which we will present
in the next sections. There are differences how subordination and coordination are
defined. In particular, there are different ways of thinking about the nature of these
relations. For example, they may be defined in terms of discourse plans and intentions,
or in a purely syntactic way.
Subordination and coordination are the properties of rhetorical relations that de-
fine the right frontier. Roughly, the right frontier denotes the zone in a graph where
new text segments can attach. It is on the right side of the discourse graph if we assume
that the graph is a tree and that the order from left to right corresponds to the natural
order of discourse segments in text or dialogue. We consider the following example,
where Ann tells another story:
(6)
Ann: (1) I took a taxi to Haus Bommerholz. (2) It picked me up at the railway
station. (3) The ride took more than half an hour. (4) The taxi driver didn’t
know his way. (5) This was very annoying.
To which proposition does (5) refer? Sentences (2) and (3) are coordinated to each
other and subordinated to (1). Sentence (4) is subordinated to (3). The right frontier
consists of the segments (1), (3), (4), and (2+3). It is defined as follows: the top node
of a tree is always on the right frontier; if a sequence of coordinated nodes is subordi-
nated to a node on the right frontier, then the sequence itself and its rightmost coordi-
nated node are also on the right frontier.4
The right frontier constraint states that new discourse segments can only attach
to segments that are positioned on the right frontier. This means that in our example
(5) can only attach to (1), (3), (4), or the compound (2+3). This does not follow from
expectations about annoying things:
(7)
Ann: (1) I took a taxi to Haus Bommerholz. (2) I had to wait very long for it.
(3) Then, the ride took more than half an hour. (4) The driver didn’t know
his way. (5) This was very annoying.
Again, (5) can only attach to the segments on the right frontier, i.e., to (1), (3), (4),
and the compound (2+3) but not to (2).
The claim that new discourse segments can only attach to the right frontier needs
some qualification. What can attach are anaphoric expressions, i.e., discourse elements
which need a previous discourse element in order to receive a truth value. Examples
of anaphoric expressions are pronouns like ‘he,’ ‘she,’ or ‘it’, but also abstract object
4. The precise definition of the right frontier and its associated constraint is, of course, frame-
work dependent; see especially sections 4 and 6.

Constraints in discourse — an introduction 
anaphora (Asher 1993) like ‘this’ in sentence (5) which refers to an preceding event.
Furthermore, we can think of a complete sentence like (5) as an anaphoric expres-
sion that needs a previous discourse segment to which it can be linked by a rhetorical
relation.
Not all anaphoric expressions are bound by the right frontier constraints. For ex-
ample definite descriptions can pick up objects which were introduced in segments
left to the right frontier. Here is a slight variation of an example from (Asher and
Lascarides 2003):
(8)
(1) One plaintiff was passed over for promotion three times. (2) Another didn’t
get a raise for five years. (3) A third plaintiff was given a lower wage compared
to males who were doing the same work. (4) But the jury didn’t believe this.
(4ʹ) But the jury didn’t believe the first case.
‘This’ in sentence (4) can only refer to either the compound of (1), (2), (3) or (3) alone.
In contrast, ‘the first case’ in (4ʹ) refers to (1), which is not on the right frontier.
An obvious problem for the right frontier constraint are cataphors, i.e., pronouns
that refer to objects that are introduced later in discourse. The graph in Figure 2
shows another potential problem: The last coordinated sentences (5) and (6) are
superordinated to the previous discourse (1)–(4) in such a way that (1)–(4) are at-
tached to the last sentence (6). This is not possible if we assume that sentences (5) and
(6) are attached sequentially to the previous graph for (1)–(4). It is possible to analyse
the discourse in Example (5) in other ways which avoid this problem. The right frontier
constraint is discussed especially in the papers by Asher (Chapter 2) and Prévot Vieu
(Chapter 3). Consten Knees (Chapter 9) discuss abstract object anaphora. Chiarcos
Krasavina (Chapter 5) discuss different methods to measure the distance between
anaphors and their antecedents in discourse graphs.
Another important constraint connected to rhetorical relations and the structures
defined by them is the Maximize Discourse Coherence (mdc) constraint introduced by
(Asher and Lascarides, 2003). rst graphs, for example, connect discourse segments
Figure 3. An analysis of Example (6).
ELABORATION
(1)
NARRATION
EXPLANATION
(2)
(3) (4)

 Constraints in discourse — an introduction
by a single rhetorical relation. The mdc constraint represents the contrary position.
It states that as many rhetorical relations as possible are realised between discourse
segments. This can be understood best from the interpretation perspective. The
addressee tries to connect the different segments by as many discourse relations as
possible. Coherence is defined by connectedness through rhetorical relations. Maxi-
mising the number of relations that hold between segments is then the same as max-
imising discourse coherence. An intuitive example is the following one taken from
(Asher and Lascarides 2003, p. 18):
(9)
(1) John moved from Brixton to St. John’s Wood. (2) The rent was less
expensive.
There are two possibilities to resolve the bridging anaphora in (2). ‘The rent’ can relate
to the rent in Brixton or St. John’s Wood. In both cases, (2) provides background infor-
mation, hence (2) can attach to (1) by a relation named Background. But if we assume
that ‘the rent’ refers to St. John’s Wood, then we get in addition also an explanation for
whyJohnmoved.ThisisthepreferredreadingofExample(9).Wegetthisinterpretation
if we maximise the number of discourse relations as the preferred reading allows to
connect (2) with Background and Explanation to (1), whereas the dispreferred reading
allows a connection only with Background.
So far, we presented phenomena and constraints directly related to the discourse
structure defined by rhetorical relations. But not all discourse constraints are con-
nected to these relations. We here mention two important principles: centering (Grosz
et al., 1995) and DRT subordination (Kamp and Reyle 1993).
Grosz and Sidner (1986) distinguished between three components of discourse
structure: the linguistic structure, the intentional structure, and the attentional state.
The linguistic structure is defined by discourse segments and the relations holding
between them. The intentional structure is defined by the speaker’s intentions that
underlie the discourse segments and the relation between these intentions. The atten-
tional state is defined by the immediate focus of attention at each point of the discourse.
Grosz and Sidner distinguish between local and global discourse coherence. Global
discourse coherence roughly corresponds to the coherence defined by the discourse
relations holding between discourse segments, i.e., it is associated with the linguistic
structure. Local coherence refers to coherence among the utterances of one discourse
segment.5 Centering Theory (CT) explains, for example, why the discourse in Example
(10a) is more coherent than the discourse in (10b) (Grosz et al., 1995, p. 206).
(10) a.
(1) John went to his favourite music store to buy a piano. (2) He had
frequented the store for many years. (3) He was excited that he could
finally buy a piano. (4) He arrived just as the store was closing for the day.
5. Here, discourse segment has to be understood roughly as meaning a sequence of coordinated
utterances.

Constraints in discourse — an introduction 
b.
(1ʹ) John went to his favourite music store to buy a piano. (2ʹ) It was a
store John had frequented for many years. (3ʹ) He was excited that he
could finally buy a piano. (4ʹ) It was closing just as John arrived.
CT assigns to each utterance a set of forward looking centres and a unique backward
looking centre. Forward and backward looking centres are semantic domain entities
like persons, things, and events. The backward looking centre is the immediate focus
of attention. The forward and backward looking centres of two consecutive utterances
are related to each other as follows: The backward looking centre of the second ut-
terance must be an entity from the forward looking centre of the first utterance. The
elements of the forward looking centre are ranked according to salience. The subject
is most likely to be ranked highest. CT formulates several discourse constraints that
are derived from forward and backward looking centres. One rule states that the back-
ward looking centre of a sentence must be realised as a pronoun if any element of the
forward looking centre of the previous utterance is also realised by a pronoun. This
predicts that (11a) is better than (11b):
(11) a. John met Mary. He loves her.
b. John met Mary. John loves her.
Another rule states, for example, that a continuation of backward looking centres is
preferred over a change. This explains the observation in Example (10) and explains
why the use of the pronoun ‘he’ in sentence (5) of Example (12) is misleading (Grosz
et al., 1995, p. 207).
(12)
(1) Terry really goofs sometimes. (2) Yesterday was a beautiful day and he was
excited about trying out his new sailboat. (3) He wanted Tony to join him on
a sailing expedition. (4) He called him at 6 AM. (5) He was sick and furious at
being woken up so early.
DRT subordination likewise imposes restrictions on anaphoric accessibility of
discourse objects. In contrast to the constraints presented so far, DRT subordination is
derived from the logical form of utterances. It explains why, for example, the following
uses of pronouns are infelicitous:
(13) a. In the cage there was no lion. *It was snoring and sleeping.
b. If a farmer owns a donkey, he beats it. ?He is my neighbour.
Discourse is interpreted incrementally by constructing Discourse Representation
Structures (drss). Several construction algorithms have been proposed. One sugges-
tion is to construct a unique drs for each new sentence and merge it with a drs rep-
resenting discourse old information. A drs consists of a pair 〈U,Con〉 of a discourse
universe U and discourse constraints Con. The universe U contains discourse referents,
which correspond to the familiar variables in first-order logic. U represents the set of
entities introduced by the discourse. The discourse constraints in Con are, in the most

 Constraints in discourse — an introduction
simple case, a set of first-order formulas that represent the truth conditionally relevant
content of the discourse. A drs representing the sentence ‘It was snoring and sleep-
ing’ is 〈{y}, {snoring(y), sleeping(y)}〉: y is the discourse referent introduced by ‘it.’ In
order to interpret the sentence in a given context, y has to be linked to a discourse old
referent. Let’s consider the case where the first sentence would have been ‘In the cage
there is a lion.’ This sentence can be represented by a drs 〈{x}, {lion}(x), in-cage(x)}〉.
We can see that y can only be linked to x. We acquire a drs representing the meaning
of the whole discourse by merging the two drss into one. This can either be achieved
by building the unions of the universes and conditions and adding the constraint
x = y, or by replacing y by x in the first drs and then building the unions. This leads
either to 〈{x, y}, {lion(x), in-cage(x), x = y, snoring(y), sleeping(y)}〉, or to 〈{x}, {lion(x),
in-cage(x), snoring(x), sleeping(x)}〉.
In our example (13a), the context is given by the sentence ‘In the cage there was
no lion.’ In DRT, this sentence is represented by a drs of the form 〈0,{¬ 〈x, {lion(x),
in-cage(x)}〉}〉, or in graphical notation:
¬
x
lion(x),
in-cage(x)
We here encounter a negated drs 〈x, {lion(x), in-cage(x)}〉 in the condition set of
a larger drs. The negated drs corresponds to the first-order formula ¬∃x, (lion(x)∧
in-cage(x)). In addition to the truth-conditions, the drss represent information about the
accessibility of discourse referents for subsequent anaphors. Anaphors in a new drs D can
only be linked to discourse referents contained in the universe of the drs with which D is
merged. In our example, we see that y introduced by ‘it’ cannot be linked to x because x is
not an element of the universe of the drs representing the first sentence of (13a). The uni-
verse of the subordinated negated drs is not accessible. In contrast to Grosz and Sidner
(1986), subordination and coordination are not defined in terms of discourse goals but in
terms of the logical form of sentences. Psycholinguistic evidence about the accessibility of
discourse referents will be discussed by Burkhardt in chapter 7 and by Kaup Lüdtke in
Chapter 8. Burkhardt studies definite determiner phrases and Kaup Lüdtke discuss the
accessibility of discourse referents which were introduced in negated contexts.
4. The ldm
One of the first theories of discourse structure and interpretation that explicitly
acknowledged the rfc was formulated in (Polanyi 1986). This paper was based on

Constraints in discourse — an introduction 
earlier work (Polanyi and Scha 1983a, b) and remains influential. We will here describe
a recent version of the ldm, taken mainly from Polanyi (2001).
Following the idea that it is possible to build up discourse recursively, the ldm
identifies basic or elementary discourse units; we will use the abbreviation “e-dcu” here
for those units, taking up the notation from (Polanyi 1986). The criterion for a stretch
of discourse to be counted as an e-dcu is that it is (part of) an utterance that describes
a single event or event type in a Davidsonian sense as stated in, for example, (Davidson
1980). The characterisation is thus twofold: on the one hand, it is syntactical to the ex-
tent that it gives a criterion that can be tested purely by the form of the constituent. On
the other hand there is a semantic/pragmatic criterion that might give rise to the iden-
tification of dcus which are only parts of utterances. We will discuss an example below.
But note that the ldm does not make use of, for example, the notion of a speech act.
I like to read SF I like to sleep late
I like to ski
C
Figure 4. [C[e-DCU I like to read SF], [e-DCU I like to ski], [e-DCU I like to sleep late]] as a tree
representation.
The claim that discourse can be constructed recursively means, on the other hand,
that there have to be (syntactic) rules of combination for the constituents at various
levels. The ldm knows three of those rules:
Coordination According to ldm, coordination is an n-ary conjunction of the co-
ordinated dcus, where the semantics of the conjunction corresponds to the inter-
section of the semantics of the constituents. The coordinated dcus are subordinated
to a freshly introduced or already existing C-node. Thus, to quote an example
from Polanyi (1986), [C[e-DCU I like to read SF], [e-DCU I like to ski], [e-DCU I like to
sleep late]], consisting of a conjunction of three e-dcus, is interpreted as ∩([[I like
to read SF]], [[ I like to ski]], [[I like to sleep late]]), where “[C. . .]” expresses the
top-level node, ∩ has the same arity as the conjunction and “[[f]]” maps f into its
meaning. (For a tree representation of the discourse structure, see Fig. 4.) Thus, all
the information of subordinated nodes (here, e-dcus) is collected and inherited
by the superordinate C-node.
Subordination Accordingtotheldm,subordinationisalwaysbinary,andtheseman-
tics of the superordinate S-node is the semantics of the left sister. For example, the
structure [S[e-DCU I like to do fun things on vacation], [e-DCU I like to read SF]] re-
ceives an interpretation that is just [[I like to do fun things on vacation]] (“[Ŝ. . .]”
expresses the top-level node again). The semantics of the right sister doesn’t

 Constraints in discourse — an introduction
contribute to the meaning of the S-node. It is appropriate to distinguish between
subordination of nodes in the hierarchy and subordination of dcus with respect
to one another. As explained in the previous paragraph, the daughters of a C-node
are coordinated, while in the present case, the daughters of an S-node are not.
(The left sister is superordinated.)
n-ary relations There is one case of n-ary relation that is different from coordina-
tion.6 Polanyi (2001) doesn’t give a general characterisation of this type of relation,
but states that logically, rhetorically, or interactionally related pairs of dcus are
n-arily connected. [B[e-DCU If John goes to the store], [e-DCU he’ll buy tomatoes]] is an
example for such a binary structure; here, the relation between the dcus is the
logical if/then. The semantics of such nodes is complex. According to Polanyi
et al. (2004), what is available at those nodes is information about each constituent
and the relationship connecting them.
The analysis of ‘if John goes to the store he’ll buy tomatoes’ already is the announced
example for sub-sentential e-dcus. This analysis is triggered by the presence of the log-
ical connective if ___ then ___. There is no general characterisation for the triggering
class of constructions in the texts on the ldm. But as propositional logical connectives
take expressions as arguments that can classically be interpreted as sets of possible
worlds in which the respective argument is evaluated as true, this seems to mirror the
desire to allow for a finely-grained semantic analysis of discourse.
Here might be the natural point to have a closer look at the semantics as utilised
by the ldm. As Polanyi (2001) puts it, the central concern of the ldm is the setting and
resetting of contexts. This emphasis on contexts is quite in the spirit of Kaplan (1978)
and distinguishes the otherwise rather similar semantic concerns from those found in
(Heim 1983; Kamp and Reyle 1993; Groenendijk and Stokhof 1990) and elsewhere. Of
course, as Polanyi stresses, the semantics utilised by the ldm is dynamic.
Figure 5 shows a representation of an already complicated case, namely the se-
mantics of a dcu expressing reported speech. What is summarily described as “index”
or “indexes” for dcus in the upper list actually comes in different flavours; Polanyi
(2001) gives a list that spells it out as the partial ordering interaction speech
event genre unit modality polarity point of view of contexts.
Polanyi (2001) demonstrates the explanatory power of the ldm by giving an in-
teresting analysis of a Yiddish anecdote. We will here just briefly analyse the short taxi
ordering event from Example (2), p. 5.
As discussed above, Ann’s utterance of I need a taxi now gets further elaborated
by the subsequent utterance Pick me up at the Dortmund railway station and drop me at
Haus Bommerholz. According to the ldm principles set out, the first utterance is analy-
sed as [e-DCU I need a taxi now], expressing a single event. The second utterance has to be
6. Polanyi (1986) in an earlier version of the ldm acknowledged only binary rather than n-ary
relations.

Constraints in discourse — an introduction 
rendered as a binary construction including [e-DCU Pick me up at the Dortmund railway
station] and [e-DCU drop me at Haus Bommerholz]. We arrive at the intermediate analy-
sis for the second utterance [n-ary[e-DCU Pick me up at the Dortmund railway station],
[e-DCU drop me at Haus Bommerholz]]; and finally, using abbreviations for the utterances,
for the whole discourse we get [S[e-DCU u1], [n-ary[e-DCU u2], [e-DCU u3]]].
Note that there might be a choice in the analysis; one might decide to first attach
[e-DCU u2]] to [e-DCU u1] as [S[e-DCU u1], [e-DCU u2]] and only later introduce the node
[n-ary] to enable the attachment of [e-DCU u3]. This, however, depends on the actual
implementation of the discourse parser and is only relevant for the processing, while it
makes no difference to the result.
With regard to the interpretation of this short discourse, what we arrive with at the
top level node is just the interpretation of [e-DCU I need a taxi now], according to the
rule for interpreting subordinating nodes. The interpretation of the subordinated dcu
⟦[n-ary[e-DCU Pick me up at the Dortmund railway station], [e-DCU drop me at Haus
Bommerholz]]⟧ plays no role for the semantics of the discourse. This is a little surpris-
ing and might need some further investigation; however, given the way the semantics
for the ldm is set up it makes sense: all the relevant indexes seem to be set for the taxi
ordering event. The events described in the subordinate dcus carry different indexes
that should not have direct impact on the current ones.
At this point it may be beneficial to introduce a distinction between parts of se-
mantic information with respect to their availability. There should probably be parts
that percolate through the tree (e.g., information relating to individuals) and others
that don’t (like the indexes). We will see how other theories handle the passing of in-
formation in the following sections of this chapter.
5. Rhetorical Structure Theory
The origins of Rhetorical Structure Theory (rst, Mann and Thompson (1987)) lie in
text generation. It soon developed into a general linguistic framework for analysing
indexes of reporting DCU
e1 at t1
···
event of reporting
indexes of DCU reported
event(s) reported
Figure 5. The representation of dynamic semantic interpretation of a case of reported speech.

 Constraints in discourse — an introduction
text in terms of the rhetorical relations that hold between text segments. The minimal
units of texts are the speech acts produced on sentence level. Units linked by rhetorical
relations themselves form a text segment that can again be linked to other text seg-
ments. rst distinguishes between two types of relations: relations that connect text
segments which are of equal importance to the text, so-called multi-nuclear relations,
and relations that connect text segments of differing importance, so-called nucleus-
satellite relations. A nucleus is a text segment which can stand alone as a coherent
text. In contrast, a satellite would not form a coherent text without its super-ordinated
nucleus. The following Example (14) shows three nuclei in (1), (2) and (5), and two
satellites in (3) and (4) which elaborate sentence (2):
(14)
Ann: (1) I arrived in Dortmund by train. (2) I took a taxi to Haus Bommerholz.
(3) It picked me up at the railway station. (4) The ride took more than half an
hour. (5) When I arrived at Haus Bommerholz, I checked in immediately.
If we delete (3) or (4), then the remaining text is still coherent; whereas a deletion
of (2) produces an incoherent text because the two satellites cannot be subordinated to
any other text segment. The sentences (1), (2) and (5) form a temporal sequence. Each
of them could be deleted independently of the other without disrupting text coher-
ence. A graphical rst representation of Example (14) is shown in Figure 6.
As mentioned before, rst distinguishes between two types of relations: multi-
nuclear and nucleus-satellite relations. The definition of a multi-nuclear relation is
divided into two parts: constraints on the combined nuclei and a definition of the
intended effects of the relation. Definitions of nucleus-satellite relations are divided
into three parts: constraints on the nucleus and satellite individually, constraints on the
nucleus-satellite combination, and again a definition of the intended effects. We here
show some examples for each relation type.7
Figure 6. An analysis of Example (14).
ELABORATION
SEQUENCE
LIST
(1)
(2)
(3) (4)
(5)
7. The definitions are taken from (Mann and Thompson 1992). They can also be found on the
rst homepage http://www.sfu.ca/rst/01intro/definitions.html. The page provides an extensive
list of relations together with examples and many elaborated text analyses.

Constraints in discourse — an introduction 
Multi-nuclear relations
The following tables show the definitions of the relations Contrast and Sequence. The
relation Sequence corresponds to the sdrt relation Narration, see Section 6.
Contrast
Constraints on nuclei Intended effects
No more than two nuclei; the situations in
these two nuclei are (a) comprehended as the
same in many respects (b) comprehended as
differing in a few respects and (c) compared
with respect to one or more of these
differences.
Receiver recognizes the comparability and
the difference(s) yielded by the comparison.
Sequence
Constraints on nuclei Intended effects
There is a succession relationship between
the situations in the nuclei.
Receiver recognizes the succession
relationships among the nuclei.
Nucleus-satellite relations
In the following tables, N stands for nucleus, S for satellite, and R for Receiver.
Evidence
Constraints on N or S Constraints on N+S Intended effects
on N: R might not believe N
to a degree satisfactory to W
on S: R believes S or will find
it credible
A’s comprehending S
increases A’s belief of N
A’s belief of N is
increased
Preparation
none S precedes N in the text; S tends to
make R more ready, interested or
oriented for reading N
R is more ready, inter-
ested or oriented for
reading N
Elaboration
none S presents additional detail about
the situation or some element of
subject matter which is presented
in N or inferentially accessible in
N in one or more of the ways listed
below. In the list, if N presents the
R recognizes S as
providing additional
detail for N. R
identifies the element
of subject matter

 Constraints in discourse — an introduction
first member of any pair, then
S includes the second:
• set : member
• abstraction : instance
• whole : part
• process : step
• object : attribute
• generalization : specific
for which detail is
provided.
Graphical convention
There is a graphical convention how to draw rst trees. We have seen examples in
Figures 2, 3, and 6. Corresponding to the two types of relations, these graphs are built
up by two types of graphical components. Figure 7 shows the graphical convention for
depicting multi-nuclear relations. Figure 8 shows the corresponding convention for
nucleus-satellite relations.
rst is mainly a theory of discourse coherence based on rhetorical relations. Im-
portant for the understanding of rst is the role of intentions. A text is only coherent
if the receiver can recognise behind each text segment an effect intended by the text’s
author. This contrasts, for example, with the treatment of rhetorical relations in sdrt.
As an effect, the inclusion of author’s intentions in the definition of rhetorical rela-
tions leads to a more finely grained distinction within these relations than in other
approaches. The explicit inclusion of intentions makes it an attractive framework for
discourse generation.
In rst, it is assumed that only one relation can hold between two segments, but
multiple analysis of one text may be possible. Maybe, the strongest drawback com-
pared to other theories is rst’s lack of a theory of anaphoric restrictions. Likewise,
Figure 8. Graphical component for nucleus-satellite relations.
Nucleus Satellite
RELATION
RELATION
Nucleus 1 Nucleus 2
...
Nucleus n
Figure 7. Graphical component for multi-nuclear relations.

Constraints in discourse — an introduction 
there is no direct connection to theories of discourse interpretation, like Discourse
Representation Theory (Kamp and Reyle 1993). As a consequence, rst is weak as a
framework for discourse interpretation.
6. sdrt
One of the most recently developed theories of discourse meaning is sdrt, promi-
nently defended by Asher and Lascarides (2003). It can be thought of as building on
two cornerstones: the first is rst, as layed out in section 5. The other one is drt, which
already was touched upon in section 3. drt had been around in the form of grey pa-
pers in the 80s and canonized in (Kamp and Reyle 1993).
For an example of how drt deals with the construction of meaning representa-
tions of discourse, let us look at Ann’s calling a taxi service again. “I need a taxi now”
can be taken to introduce four discourse markers: one for each Ann and a taxi, one for
the event (or state) of needing the taxi, and one for the moment denoted by “now”. In
drt, this would be written as in Figure 9(a):
i x e1 n
taxi(x)
need(e1, i, x, n)
j y r e2 m
Dortmund-railway-station(r)
pick-up(e2, j, y, r, m)
(a) (b)
Figure 9. Representation of I need a taxi now and Pick me up at Dortmund railway station in
standard drt.
The continuation in the above example (2) “Pick me up at Dortmund railway
station” might be given a semantic representation in the form of the drs 9(b) by a simi-
lar reasoning. Using a standard merging operation like in (van Eijck and Kamp 1997),
the two drss turn into the representation in Figure 10.
Notice that this representation actually allows to identify all the discourse enti-
ties even after the merge, and the truth conditions of the discourse are rendered quite
nicely: the whole discourse can be said to be true with respect to some model if the
referents can be mapped to that model such that the properties and relations expressed
by the predicates are satisfied, just as required by standard model theory. However,
some information is not preserved by the obtained drs: “Pick me up at Dortmund railway
station” is a request or command, whereas “I need a taxi now” is a statement. The utterance
mood (or illocutionary force) is not preserved. Further, the two utterances stand in a
certain relation (cf. Section 1.5): the second utterance elaborates the first. Note that
their order is not arbitrary. This information likewise is not preserved because the
discourse segmentation is abstracted over in drt.

 Constraints in discourse — an introduction
sdrt preserves discourse segmentation. The drss that represent discourse seg-
ments are labelled by prepending tags (here, greek letters with indices) as shown in
Figure 11. The labels can be used to refer to the drss representing the meanings of dis-
course segments. Whereas it might be best to think of earlier discourse theories—and
very obviously so in the case of (Grosz and Sidner 1986)—as adhering to a metaphor
of stack execution, sdrt might best be reconstructed along the lines of relational da-
tabases. (Discourse referents are distinguished entries in that they don’t point to other
entities etc.) It is thus possible to express relations that hold between them and proper-
ties they have. Thus, the representations for the meanings of the two utterances from
Ann’s taxi call turn into a structured representation from which it is possible to recover
much more information than from the plain drs.
π1:
i x e1 n
taxi(x)
need(e1, i, x, n)
π2:
j y r e2 m
(a) (b)
Figure 11. Representation of I need a taxi now and Pick me up at Dortmund railway station in
sdrt.
This representation of course is called an sdrs. As can be seen in Figure 12, the sdrs
not only contains the two sub-sdrss representing the meaning of each utterance, but
also a predicate denoting the kind of relation the two meanings stand in: π2 is inter-
preted as elaborating π1. sdrt has adopted the insight that it is necessary to express
structural underspecification from earlier approaches, e.g., udrt (Underspecified
drt), cf. Reyle (1993), or clls (Constraint Language for Lambda Structures (Egg et al.,
2001)). The way sdrt does so with respect to discourse relations is by introducing
variables for relations.
i x e1 n j y r e2 m
taxi(x)
need(e1, i, x, n)
i = j
Figure 10. Representation of I need a taxi now. Pick me up at Dortmund railway station.

Constraints in discourse — an introduction 
π1 π2 i
π1:
x e1 n
taxi(x)
need(e1, i, x, n)
π2:
y r e2 m
pick-up(e2, i, y, r, m)
Elaboration(π1, π2)
Figure 12. Representation of I need a taxi now. Pick me up at Dortmund railway station
in sdrt.
In Figure 12, the Elaboration relation is fully specified, following the analysis above.
If, however, it were unclear exactly which relation(s) held between the two discourse
units, it would be feasible to just write R1 in place of Elaboration, thus introducing a
variable for a relation. Note that sdrt, in marked contrast to rst, allows for multiple
relations to hold between discourse units. This is relevant not only for trivial cases,
like inverse relations: Explanation and Cause are such trivial cases. The sdrt analysis
of Explanation is that of a causal explanation: the occurrence of one event explains
(or can be employed to explain) why another event occurs. Non-trivial cases might be
some Elaboration in parallel to Background as in Example (9). Combinations of pos-
sible relations holding between discourse consituents (or rather, the sdrss expressing
their meaning) are constrained, however, by the outcomes that are predicted by the
relations: if, e.g., R1 predicts temporal overlap of the related events, and R2 predicts a
temporal sequence of them, then R1 and R2 can not simultanously hold between two
constituents. According to the database metaphor, the criterion is that the table con-
taining the outcomes may not become inconsistent. If it were the case that by adding
a relation Rn to a table containing the outcomes predicted by the relations R1 . . . Rm
became inconsistent, a decision would have to be made which (set of) relation(s)
would have to be dropped. The principle guiding such a decision would be that of
Maximize Discourse Coherence, cf. p. 9, i.e., drop the minimal number of relations that
is necessary to render the outcomes maximally consistent (taking scalar relations etc.
into account).
The introduction of labels for discourse constituents allows to reason about the
structure of discourse without having to bother with the meaning of the constituents.
Thus, the complexity of modal dynamic predicate logic that is needed for the inter-
pretation of the discourse constituents is not introduced into the reasoning about the

 Constraints in discourse — an introduction
structure of the discourse. However, when needed all of the information can be recov-
ered and used. To accomplish this, the language (and corresponding logic) to describe
the rhetorical structure of a discourse is combined with the language used to describe
(and logic used to reason over) underspecified logical forms. There is a very detailed
description of this so-called glue language (and corresponding logic) in (Asher and
Lascarides, 2003, 184ff) and we will not go into the details here.
Compared to rst, sdrt is definitely stronger as a theory of discourse interpreta-
tion. Because of its rich formal inventory, sdrt allows a detailed description of dis-
course meaning, whereas rst in the described version must be said to be definitely
more restricted. The ldm, on the other hand, seems to have a coverage that is similar
to that of sdrt. While sdrt can be said to be semantically driven and syntactically
informed, the ldm ought to be characterized as being syntactically driven. It would
definitely be worth to have a look at which approach is cognitively more appropriate,
but this can not be done here.
7. About the papers
The book divides into four parts. The first part contains two chapters by Asher and
Prévot Vieu which discuss the right frontier constraint. The second part contains pa-
pers which compare different frameworks according to the discourse structures which
can be generated in these frameworks. It includes chapters by Danlos, Chiarcos Kras-
avina, and Egg Redeker. The third part approaches the topic of discourse constraints
from the cognitive perspective. The chapters are based on experimental studies con-
ducted by Burkhardt and Kaup Lüdke. The last and largest part contains work which
applies discourse theory to language specific phenomena. It inludes five chapters by
Consten Knees, Nishiama Koenig, Avarintseva-Klisch, Holler, and Speyer.
In view of the importance of the Right Frontier to the various frameworks as a re-
striction on anaphoric accessibility, it is well justified to open this book on constraints
in discourse with a discussion of the RFC. The first contribution, Troubles on the Right
Frontier by Nicholas Asher, discusses some challenges to the Right Frontier Constraint
and proposes a refinement of rfc that meets them. It is argued that different anaphors
behave differently with respect to rfc depending on their presupposed information
content. In general anaphoric expressions with little to no presuppositional content
appear to obey rfc without exception. However, anaphors with a ‘heavier’ presup-
positional content (definite descriptions and complex demonstratives) have a better
chance of remaining felicitous even though the relations between them and their ante-
cedents violate the Right Frontier Constraint.
The right frontier constraint is also at the heart of the paper contributed by Lau-
rent Prévot Laure Vieu, The Moving Right Frontier. The authors argue that the co-
ordinating or subordinating nature of discourse relations plays a major role in certain
cases of revision of the discourse structure. They focus in particular on a relation typi-
cal in narratives, Result, as well as on a family of dialogue relations: content-relations

Constraints in discourse — an introduction 
introduced by interrogatives. They conclude that the observed complex behaviour
shows that the rfc need to be “handled with care”.
Closely related to those discussions of the rfc is Laurence Danlos’ Comparing rst
and sdrt Discourse Structures through Dependency Graphs, since the rfc is a restric-
tion over docking points in discourse structures. Danlos discusses the rst distinction
between Nucleus and Satellite arguments and the sdrt distinction between coordinat-
ing and subordinating discourse relations. She proposes a third mode of representa-
tion for discourse structures, called Dependency Graph. She argues that rst is far
too restrictive with respect to generative capacity, sdrt a little bit too restrictive, and
dependency DAG formalism a little bit too powerful.
Christian Chiarcos Olga Krasavina, in their Rhetorical distance revisited — A
parameterised approach, develop some notion of rhetorical distance. This notion
allows for a comparative representation of different theories; the effect is a scalar
concept of accessibility. The authors reconstruct three theories of discourse-structural
accessibility that differ in their assumptions on discourse structuring and its limiting
force on the search for the antecedent. Results of an empirical study on the compara-
tive predictivity of the reconstructed theories for the use of pronouns in German and
English newspaper corpora are discussed.
Underspecied discourse representation is at the center of Markus Egg’s Gisela Re-
deker’s paper. An approach to discourse structure that builds on syntactic structure to
derive that part of discourse structure that can be captured without taking recourse to
deep semantic or conceptual knowledge is proposed. The authors claim that this con-
tribution is typically only partial. They develop a notion of underspecified constraints
that describe the structures a given discourse might have. This results in an interface
from syntax to discourse and a clean interface to modules of discourse resolution.
Still dealing with discourse structure, but from a processing perspective, Petra
Burkhardt’s Dependency Precedes Independence: Online Evidence from Discourse Pro-
cessing, investigates the integration of definite determiner phrases (dps) as a function of
their contextual salience. dps depend on previously established discourse referents
or introduce a new, independent discourse referent (and see Asher’s paper in this
volume). A formal model that explains how discourse referents are represented in
the language system and what kind of mechanisms are implemented during dp in-
terpretation is presented. Experimental data from an event-related potential study are
discussed that demonstrate how definite dps are integrated in real-time processing.
Two distinct mechanisms Specify R and Establish Independent File Card and a model
that includes various processes and constraints at the level of discourse representation
are assumed to explain the data.
Barbara Kaup Jana Lüdtke in their Accessing Discourse Referents Introduced in
Negated Phrases: Evidence for Accomodation? present an investigation into negation.
According to standard theories of dynamic semantics, a discourse referent introduced
by a noun phrase (np) in the scope of a negation should be inaccessible to subsequent
anaphoric reference. Kaup’s paper presents empirical findings on anaphora resolution
in the context of negations.

 Constraints in discourse — an introduction
Lexical or pronominal nps that refer to propositionally structured referents (such
as events, processes, states and facts) while introducing them as unified entities into a
discourse representation, called complex anaphors, are at the focus of Manfred Con-
sten’s Mareille Knees’ contribution Complex Anaphors in Discourse. They describe
anaphoric complexation processes and constraints on them in terms of ontological
categories. Additionally, they provide a resolution model for complex anaphors and
discuss different kinds of disambiguation processes based on ontological and lexical
features as well as conceptual knowledge.
Atsuko Nishiyama Jean-Pierre Koenig report two corpus studies of the present
perfect in English and Japanese in The discourse functions of the present perfect. They argue
that the inferences required to interpret the present perfect follow from general default
rules or commonsense entailment rules, and that the use of the perfect is relevant for dis-
course coherence in two ways. First, the presence of the state which the perfect introduces
helps establish discourse relations, or allows the establishment of additional discourse
relations between discourse segments. Second, the pragmatic inferences required to in-
terpret the perfect can indirectly trigger the rules needed to establish discourse relations.
German right dislocation, according to Maria Averintseva-Klisch in her German
Right Dislocation and Afterthought in Discourse, subsumes two distinct constructions,
dislocation proper and afterthought. These differ in a number of prosodic, syntactic and
semantic characteristics and also have different discourse-functional properties. Right
dislocation marks a discourse referent as especially salient on the current stage of the
discourse. This requires the fulfilment of certain anaphoric constraints on the follow-
ing discourse. Afterthought is a local reference clarification strategy and has no impact
on the global discourse structure.
Anke Holler’s contribution A discourse-relational approach to continuation draws
upon the distinction between two classes of non-restrictive relative clauses in German:
continuative and appositive ones. The paper investigates whether the notion of
communicative-weight assignment first introduced by Brandt can be couched in
discourse-structural terms by exploiting the distinction between coordinating and
subordinating discourse relations in the sense of Asher and Vieu (2005).
The filling of the vorfeld in German, following Augustin Speyer in his German
Vorfeld-filling as Constraint Interaction, depends on information structural rather than
strictly syntactic constraints. He argues for a ranking among scene-setting elements
(which are said to be most likely to appear in the vorfeld), followed by contrastiveele-
ments and finally by topics. The difference in likelihood to be in the vorfeld are argued to
be best modelled by an Optimality Theoretic account that is sketched out in the paper.
Bibliography
Asher, N. (1993). Reference to Abstract Objects in Discourse. Kluwer Academic Publishers.
Asher, N. and Lascarides, A. (2003). Logics of Conversation. Cambridge University Press.

Constraints in discourse — an introduction 
Bäuerle, R., Schwarze, C., and Stechow, A. v., editors (1983). Meaning, Use, and Interpretation of
Language. Foundations of Communication—Library Edition. de Gruyter.
Blache, P. (2000). Constraints, linguistic theories and natural language processing. In Proceedings
of NLP-2000.
Davidson, D. (1980). Essays on Action and Events. Clarendon Press.
Egg, M., Koller, A., and Niehren, J. (2001). The constraint language for lambda structures.
Journal of logic, language and information, 10:457–85.
Grice, H.P. (1975). Logic and Conversation. In Cole, P. and Morgan, J. L., editors, Syntax and
Semantics, volume 3, pages 41–58. Academic Press.
Groenendijk, J. and Stokhof, M. (1990). Dynamic Montague Grammar. Technical report, ILLC,
University of Amsterdam. Obtainable via FTP from http://www.wins.uva.nl/research/illc/.
Grosz, B.J., Joshi, A.K., and Weinstein, S. (1995). Centering: A Framework for Modelling
the Local Coherence of Discourse. Technical Report TR-18-95, Center for Research in
Computing Technology, Harvard University.
Grosz, B.J. and Sidner, C. (1986). Attention, intentions, and the structure of discourse. Compu-
tational Linguistics, 12(3):175–204.
Heim, I. (1983). File change semantics and the familiarity theory of definiteness. in: (Bauerle
et al., 1983).
Kamp, H. and Reyle, U. (1993). From Discourse to Logic—Introduction to Modeltheoretic Seman-
tics of Natural Language, Formal Logic and Discourse Representation Theory, volume 42 of
Studies in Linguistics and Philosophy. Kluwer.
Kaplan, D. (1978). Dthat. In Cole, P., editor, Pragmatics, volume 9 of Syntax and Semantics,
pages 221–43. Academic Press.
Kempen, G., editor (1987). Natural Language Generation. Number 135 in NATO Advanced
Science Institutes—Applied Sciences. Martinus Nijhoff Publishers.
Levinson, S.C. (1983). Pragmatics. Cambridge University Press.
Litman, D.J. and Allen, J.F. (1990). Discourse processing and commonsense plans. In Cohen,
P.R., Morgan, J., and Pollack, M.E., editors, Intentions in communication, pages 417–44.
MIT Press.
Mann, W.C. and Thompson, S.A. (1987). Rhetorical Structure Theory: Description and
Construction of Text Structures. in: (Kempen, 1987). pp. 85–95.
Mann, W.C. and Thompson, S.A., editors (1992). Discourse Description — Diverse linguistic
analyses of a fund-raising text. John Benjamins Publishing Company.
Mann, W.C., Matthiesen, C. M. I. M., and Thompson, S.A. (1992). Rhetorical Structure Theory
and Text Analysis. In Mann, W. C. and Thompson, S.A., editors, Discourse Description —
Diverse linguistic analzses of a fund-raising text, pages 39–78. John Benjamins Publishing
Company.
Polanyi, L. (1986). The linguistic discourse model: Towards a formal theory of discourse struc-
ture. Techn. Report TR-6409, BBN Laboratories Inccap.
Polanyi, L. (2001). The Linguistic Structure of Discourse. In Schiffrin, D., Tannen, D., and
Hamilton, H.E., editors, Handbook of Discourse Analysis. Blackwell.
Polanyi, L., Culy, C., van den Berg, M.H., Thione, G.L., and Ahn, D. (2004). Sentential structure
and discourse parsing. In Webber, B. and Byron, D., editors, Proceedings of the ACL2004
Workshop on Discourse Annotation. ACL/SIGDial.
Polanyi, L. and Scha, R. (1983a). On the Recursive Structure of Discourse. In Ehlich, K. and
van Riemsdijk, H., editors, Connectedness in Sentence, Discourse and Text, pages 141–78.
Tilburg univ.

 Constraints in discourse — an introduction
Polanyi, L. and Scha, R. (1983b). The Syntax of Discourse. Text, 3(3):261–70.
Reyle, U. (1993). Dealing with Ambiguities by Underspecification: Construction, Representa-
tion and Deduction. Journal of Semantics, 10:123–79.
van Eijck, J. and Kamp, H. (1997). Representing discourse in context. In Benthem, J. F. v. and Ter
Meulen, A. G., editors, Handbook of Logic and Language. MIT Press.

Troubles on the right frontier
Nicholas Asher
CNRS, Laboratoire IRIT
1. Overview
The Right Frontier Constraint (RFC), originally proposed by Polanyi in the eighties
(Polanyi 1985) is one of the central empirically motivated constraints on discourse
update and anaphora resolution in Segmented Discourse Representation Theory or
SDRT, a formal theory of discourse interpretation that integrates dynamic seman-
tics and a conception of rhetorical function into the analysis of discourse content.1
In SDRT these two tasks are codependent; anaphora resolution offers constraints on
discourse attachment and vice-versa. This is because in SDRT attachment with or
without discourse connectors is a matter of resolving an underspecified antecedent for
a term in a discourse relation, which is exactly what is involved with pronoun resolu-
tion. While there is considerable support for the the Right Frontier Constraint or RFC,
there are also some challenges. in this paper I will look at these challenges in detail,
and propose a refinement of RFC that meets them. In particular, I’ll argue that differ-
ent anaphors behave differently with respect to RFC depending on their presupposed
information content. In general expressions anaphoric expressions with little to no
presuppositional content appear to obey RFC without exception. This would predict
that then discourse attachments understood as anaphors with no presupposed con-
tent must obey RFC, which also is in accord with the facts. However, anaphors with a
‘heavier’ presuppositional content (definite descriptions and complex demonstratives)
have a better chance of remaining felicitous even though the relations between them
and their antecedents violate the RFC. This points to a special role that such expres-
sions play in discourse, a role noticed by Ariel, Gundel and others: such expressions
permit the reader to focus on a discourse entity that was not at this point in processing
the discourse salient. Integrating such observations into a theory of discourse struc-
ture will enable us to refine RFC appropriately.
To set the stage, I’ll begin with an overview of SDRT and its formalization of
RFC.
1. I would like to thank Laure Vieu, Laurent Prévot and Laurence Danlos for helpful
comments on this paper and for helpful discussions on the topic in general.

 Nicholas Asher
2. An introduction to the Right Frontier Constraint and its formalisation
Anaphors in natural language are subject to several constraints governing their possible
antecedents.Whilesyntacticandsemanticconstraintsofthesortdiscussedingenerative
syntax and dynamic semantics respectively are widely accepted, discourse constraints
on anaphora are less well known at least in the philosophical community. The semantic
constraintofaccessibilityofantecedentsindynamicsemanticseasilymakessenseincon-
ceptual terms: if the constraint of semantic accessibility is violated when one identifies a
variable introduced by an anaphor with some antecedently introduced variable or dis-
course referent v, there is no value assigned to v in the local context of the anaphor and
so the identification is uninterpretable. On the other hand, the syntactic constraints
from Binding Theory and RFC are structural constraints. Schlenker (2005) has argued
for a pragmatic reinterpretation of the Binding Theory and one can explain RFC in
similar terms; RFC is a presentational constraint that, together with other principles of
SDRT, is a refinement of the Gricean constraints of relevance and orderliness. But to
see precisely how this is the case, I need to give a little background about SDRT.
A discourse structure in SDRT or sdrs is a triple 〈A, F, Last〉, where:
• A is a set of labels.
• Last is a label in A (intuitively, this is the label of the content of the last clause that
was added to the logical form); and
• F is a function which assigns each member of A a member of a formula of the
SDRS language, which includes formulas of some version of dynamic semantics
(DRT, DPL, Update Semantics, Martin Löf Type Theory, among others.)
This notion of discourse structure is very abstract and very general. One important
distinction for SDRT (and for many other theories of discourse structure) that needs
to be added to understand the notion of a right frontier is the distinction between two
types of discourse relation. There are subordinating discourse relations and coordinat-
ing discourse relations. Asher and Vieu (2005) provide some theory internal tests as
to whether a given discourse relation is subordinating or coordinating. These tests
confirm that the discourse relation of Narration is a prime example of a coordinating
relation, while the relation of Elaboration is a prime example of a subordinating rela-
tion. The difference between coordinating and subordinating relations for defining the
right frontier constraints is best understood by moving from the abstract definition of
an SDRS to a graphical representation of an SDRS. Here’s the algorithm for construct-
ing a graph from an SDRS understood as above.
• Each constituent (or label) is a node
• Each subordinating relation creates a downward edge
• Each coordinating relation creates a horizontal edge.
This graphical representation immediately imposes some constraints on what sort
of SDRSs are possible.

Troubles on the right frontier 
• No two nodes can be connected by both a subordinating and coordinating rela-
tion.
• Several edges (of the same type) are possible between 2 constituents.
• Many SDRSs can be represented as trees (Baldridge and Lascarides 2005) but
some cannot (Danlos 2003).
• Anaphora resolution and SDRS update are dependent on the graph structure.
These graphs also make explicit a dimension of discourse coherence. Discourse coher-
ence is dependent on the connectedness of the graph; the degree of connectedness
of the graph is one measure of coherence. However, SDRT allows for underspecified
graph connections as well as underspecified anaphoric connections. These lead to a
scalar notion of coherence, Maximize Discourse Coherence, or MDC. Roughly a dis-
course structure is maximally coherent if it has the fewest underspecifications, the
maximal number of connections, the strongest connections between constituents.2
Here is a simple example of a discourse structure, familiar from Asher and
Lascarides (2003):
(1) π1. John bought an apartment
π2. but he rented it.
Here is (1)’s discourse structure:
(1ʹ) • A = {π0, π1, π2}
• F(π1) = ∃x∃e(e ⊰ now ∧ apartment(x) ∧ buy(e, j, x))
• F(π2) = ∃eʹ(eʹ ⊰ now ∧ rent(eʹ, j, x))
• F(π0) = Narration(π1, π2) ∧ Contrast(π1, π2)
• Last = π2
Here’s another familiar, but slightly more complex example.
(2) π1. John had a great evening last night.
π2. He had a great meal.
π3. He ate salmon.
π4. He devoured lots of cheese.
π5. He then won a dancing competition.
π6. # It (# the salmon) was a beautiful pink.
Here’s the SDRS for (2):
(2ʹ) 〈A, F, Last〉, where:
• A = {π0, π1, π2, π3, π4, π5, π6, π7}
• F(π1) = Kπ1
, F(π2) =Kπ2
, F(π3) = Kπ3, F(π4) = Kπ4, F(π5) = Kπ5
,
F(π0) = Elaboration(π1, π6)
2. MDC also involves a notion of minimization of discourse constituents beyond those intro-
duced by the clauses of a text. But that will not be an issue here.

 Nicholas Asher
F(π6) = Narration(π2, π5) ∧ Elaboration(π2, π7)
F(π7) = Narration(π3, π4)
• Last = π5
Here’s the corresponding graph of (2ʹ):
Elaboration
Elaboration
Narration
He ate salmon He devoured cheese
Narration
He had a
great meal
He won a
dancing competition
John had a lovely evening
With these examples, we can now describe the “right frontier” as it’s defined in
SDRT–it is more general and more precise than the notion of a right frontier in other
discourse theories. This governs where new information can attach in SDRT. We define
the set of available nodes for attachment as falling under the following possiblilites.
1. The label α = Last;
2. Any label γ ≥*
D α where ≥*
D is defined recursively:
(a)
R(γ, α) is a conjunct in F(l) for some label l, where R is a subordinating
discourse relation (like Elaboration, Explanation or ⇓);
(b)
R(γ, δ) is a conjunct in F(l) for some label l, where R is a subordinating
discourse relation and F(δ) contains as a conjunct Rʹ(δʹ, α) or Rʹ(α, δʹ), for
some Rʹ and δʹ; or
(c)
R(γ, δ) is a conjunct in F(l) for some label l, where R is a subordinating dis-
course relation and δ≥*
Dα.
For all relations other than structural relations, we can now also use the notion
of the available nodes to constrain the resolution of anaphoric conditions in SDRT.
Imagine the following situation:
• β:Kβ;
• Kβ contains anaphoric condition φ.
The available antecedents then are:
1. in Kβ and drs-accessible to φ
2. in Kα, drs-accessible to any condition in Kα, and there is a condition R(α,γ) in the
sdrs such that γ = β or γ≥*
D β (where R isn’t structural).

Troubles on the right frontier 
The upshot of these definitions is that an antecedent for an anaphoric expression must
be drs-accessible on the right frontier as defined in SDRT.
The predictions of the Right Frontier Constraint largely confirm intuitions. For
instance, the availability constraint on anaphors predicts that (2π1 – π6) is infelicitous;
the relation between pronoun or the definite description and its antecedent violates
the right frontier condition. Also the attachment doesn’t make sense. So this discourse
is doubly damned according to the principle MDC.
Why does RFC exist? It derives from the idea that the author should present in-
formation in an orderly way. If one wishes to comment or modify the information in
some discourse constituent, one should do it before one has closed off that part of the
story. This is an intuitive idea that SDRT makes quite precise. One cannot just simply
go back to elaborating on or commenting on the salmon once one has moved on to
talk about the rest of John’s evening in (2). The fact that it’s difficult to attach π6 to the
rest of the discourse structure also shows that something like the Gricean constraint of
Relevance is being violated.3
Matters, however, are more complicated than this straightforward picture would
suggest. If we replace π6 in (2) with (π7) the discourse is much better. Why?
(2)
(π7) The entire next day John kept remembering what a beautiful color his
salmon had been.
SDRT and its formalisation of RFT don’t at all explain why (2π1 – π5, π7) sounds quite
adequate. This is the challenge for RFC that I want to examine here.
How prevalent are these violations in real texts? I’ve made a preliminary study
using Wall Street Journal news stories, editorials, and letters annotated with SDRT dis-
course structure (each by 2 annotators), I’ve found 10 out of 173 cases of anaphoric
definites that violate RFC. 2/3 of the cases can be resolved by a choice of attachment
point in the structure (annotators had trouble with this). Less than 2% of the cases
look like definite violations of the Right Frontier. So RFC seems to be a real constraint
even for definites.
Nevertheless, there is a difference between anaphoric expressions that may be ref-
erentially equivalent. The use of definites as opposed to pronouns often improve the
ability of speakers to recover anaphoric connections. The following example adapted
from Laure Vieu and Laurent Prévot (2005) shows this:
(3) a. This morning, in the subway, I almost got robbed.
b. At some point, I noticed that a man was pulling at my purse.
c. I just froze, I couldn’t say a word.
d. Suddenly, a woman screamed.
3. For a discussion of the relation between Relevance and SDRT see Asher and Lascarides
(2003).

 Nicholas Asher
e. The pickpocket (The man, ?He) let go of my purse and ran away.
f.
I wanted to thank the woman (?her) but she had already disappeared into
the crowd.
Though for some the use of the simple pronouns is passable, the use of the definites
markedly improves the discourse for most speakers of English. There is certainly
enough information available to the interpreter given the different gender of the two
antecedents to find the intended antecedents for the two pronouns. (3) isn’t a discourse
where interpreters lack relevant information and hence are simply unable to pick out
the intended antecedents of the anaphoric expressions. The awkwardness noticed with
pronouns in (3 is not a case of simple pronoun ambiguity as in (4):
(4) a. John called Jim a Republican and then he insulted him.
b. Pat invited Sandy over. She cooked dinner for her.
We observe the same differential behavior between pronouns and definite descrip-
tions when we consider a slight variant of (2π7):
(2)
(πʹ7) The entire next day John kept remembering what a beautiful color it had
been.
(2πʹ
7), when appended to (2π1 – π5), is no better than (2π1 – π6). Somehow definites
enable us to pick up non salient antecedents in a way that pronouns don’t.
The differences between the use of pronoun and of the definite description in the
variations of (2) or in (3) are striking. Since nothing else in the variations changes,
it is logical to try to explain the difference by taking a look at the difference between
pronouns and definite descriptions. The perspective of generation here is instructive.
Why would someone choose to use a definite description or a complex demonstrative
over a simple anaphoric pronoun? I think we can find the beginnings of an answer in
the work of Ariel (1988) and Gundel et al. (1990).
Gundel and Ariel hypothesize a hierarchy of referential expressions, according
to which certain expressions require a more salient antecedent than others. We can
express their observations as follows using to represent ‘requiring a more salient
antecedent than’:
• The hierarchy of referential expressions:
0 anaphors pronouns definite descriptions proper names.
From the perspective of the use of a referring expression, we can put the point in a
slightly different and perhaps more illuminating way. Using an expression from the
right hand side of the hierarchy makes salient a discourse entity that was not salient-
before.
One problem with these observations is that the authors provide no precise model
of discourse salience. Furthermore, existing models of salience like Centering The-
ory (Joshi, Weinstein and Grosz 1986; Beaver 2004) or the numerical algorithms of

Troubles on the right frontier 
Mitkov (1994) provide no explanation of the differential behavior between pronouns
and definites.4 So it’s difficult to use these theories to account for the observations.
RFC, in effect, offers a model of salience, and a way of putting the Referential Hi-
erarchy to the test: any discourse referent introduced within a constituent on the right
frontier is salient; salient antecedents must occupy a position on the right frontier. The
point then about definites and complex demonstratives is that they in effect change
discourse structure by putting their constituents on the right frontier. On the other
hand, 0 pronouns should not be able to alter discourse structure and so the notion
of discourse structure and discourse update in Asher and Lascarides (2003) or Asher
(1993) should suffice. Thus, we will coarsen the hierarchy of referential expressions, by
putting a division between pronouns and the expressions to their left and other expres-
sions to the right on the hierarchy:
• Salience Hypothesis: expressions on the referential hierarchy that require at least
as salient antecedents as pronouns must have SDRT available antecedents.
3. An Application of the Right Frontier to Ellipsis
Let’s see how this proposal works out. Asher (1993) provides considerable evidence for
the Right Frontier constraint with regard to propositional anaphora and to VP ellipsis.
To give one example,
(5)
One plaintiff was passed over for promotion three times. Another didn’t get a
raise for five years. A third plaintiff was given a lower wage compared to males
who were doing the same work. But the jury didn’t believe this (any of this),
Here it is difficult to impossible to get any other antecedents to the simple demonstra-
tive, except the proposition expressed by the penultimate clause or the proposition ex-
pressed by the first three clauses. The Right Frontier Constraint and SDRT’s semantics
for anaphors referring to abstract entities explains these facts. But there I did not distin-
guish between the behavior of various sorts of anaphors that refer to abstract entities. To
look at how the Salience Hypothesis fares, I want to look at another form of 0 anaphora
or ellipsis, sluicing. Sluicing is a kind of ellipsis that has always been thought to be gov-
erned by syntactic constraints. But a recent paper of Romero and Hardt (2004) suggests
4. Centering Theory exploits a number of features like whether the antecedent was mentioned
in the previous clause, what grammatical role it plays in the previous sentence and so on, to
determine these transitions. Centering theory doesn’t make any predictions about what is and
isn’t possible in terms of anaphoric connections, though it ranks the anaphoric links in terms
of transitions and so at least implicitly imposes a preference ordering on the set of antecedents.
Mitkov’s model is a much simpler account which provides just a partial ordering of candidate
antecedents based on a variety of superficial and easily recoverable features.

 Nicholas Asher
that discourse constraints may also be at work in this phenomenon. The following are
typical examples. The material that should follow the wh elements has been deleted or
is missing in (6); this material must be recovered from the context.
(6) a. John ate, but I never figured out what 0 [John ate].
b. John ate. Sam ate. But I never figured out what 0 [John ate and Sam ate].
c. John ate. But I don’t know what.
d. Mary kissed somebody. You’ll never guess who.
Sluicing can occur across separate sentences; so traditional syntactic theories, whose
domain of inquiry is the syntactic structure of an individual sentence, can’t impose any
relevant constraints on the phenomena we will be studying.
The Right Frontier Constraint makes interesting predictions concerning sluicing.
Consider (7)
(7) a. John left and then Mary kissed someone. You’ll never guess who.
b. Mary kissed someone and then John arrived. #You’ll never guess who.
c.
Mary kissed someone and then John arrived. You’ll never guess from where.
d.
John arrived and then Mary kissed someone. #You’ll never guess from where.
By using the expression and then and using SDRT’s rules for inferring discourse rela-
tions (Asher and Lascarides 2003), we’ve forced a Narration relation on the discourses
in (7). Given SDRT’s rules, this forces the right frontier to contain just the second
clause of the first sentence as well as the constructed topic, required by the axioms for
Narration (Asher and Lascarides 2003). The upshot of this is that only material in the
clause or in the topic is available for reconstructing the ellipsis; and since topics must
generalize over the clauses they span, we can conclude that only the second clause will
furnish material for reconstruction. The examples in (7) bear out this prediction: in
(7a, c) the second clause furnishes an appropriate antecedent; in (7b, d) it does not.
Further evidence that the Right Frontier Constraint is operative (and not some
simpler constraint like adjacency of discourse units) come from the following data:
(8) a.
Mary kissed someone because John left for some other party. You’ll never
guess who.
b. ??Because Mary kissed someone, John left early. You’ll never guess who.
c. Mary kissed someone, He’s a student here. You’ll never guess who.
d. Mary kissed someone. You know him. But you’ll never guess who.
SDRT predicts (8a, c, d) to be OK, whereas a constraint of adjacency would not. The
reason is that in these examples subordinating discourse relations obtain between the
first two clauses, and according to RFC either the first or the second clauses furnish
available antecedents for the ellipsis. SDRT, however, predicts (8b) to be bad, since the
only appropriate antecedent is not available according to RFC.
Let’s now turn to single sentence examples like:
(9) a. *Mary arrived after John ate but it’s unclear what.
b. Mary arrived after John ate but it’s unclear what John ate.

Other documents randomly have
different content

O' visan jag sjöng har jag själv satt ihop,
:,: Skål o' smutt :,:
På den ska du ha dej ett leverop!
Rosa Lilja o' grön Persilja
o' krusade Myntha o' Hjärtans Fröjd!
Laulaessaan hän heitti kauniin harmaatukkaisen päänsä
taaksepäin ja silmät säteilivät. Hän oli oikeassa elementissään ja
tempasi aina kuulijat mukaansa.
Ja selvää oli kaikille jotka häntä kuuntelivat, että hän täten
seuraelämän puitteissa tyydytti synnynnäisiä taiteellisia
taipumuksiansa. Hänen kodissaan palvottiin iloisia kaunotaiteita
vanhanaikaisella hartaudella. Ja se on nykyään harvinaista.
Mutta sitten yksin jäätyään katsoi hän joskus eteensä synkin ja
surumielisin katsein. Ja tottunut silmä, joka sattui hänet näkemään,
huomasi helposti, että kärsimys oli hänelle hyvin läheinen tuttu.
Mutta siitä ei nuoriso tiennyt mitään. Heistä täti kenraalitar oli
maailman rattoisin ja iloisin täti.
Sillä jos hän tiellä keksi Kirstin, loistivat hänen äsken synkät
kasvonsa veitikkamaista hyväntuulisuutta ja kuului heti:
— Hyve peive, pikku rouva — mite rouvale kuulu?
Ja jos Yrjö tuli vastaan alempana kauppapuodin kohdalla, pysähtyi
täti hänen kanssaan juttelemaan siitä mikä häntä huvitti, ja he
sopivat jostakin hauskasta tilaisuudesta, jonka täti pian panisi
toimeen heille kaikille.

Ja Yrjö tuli kotiin hymy huulillaan ja sai Kirstin jännityksellä
odottamaan luvattua hetkeä.
Sellainen oli Lehtolan täti, Koivulan-lasten hyvä ystävä.
20.
Kodin piiriin sai Yrjö laskea myöskin mummon.
Hän asui yläkerroksen läntisessä huoneessa, joka oli aurinkoisin
kaikista, koska iltapäiväaurinko on lämpimin. Se paistoi sinne,
kunnes se vaipui metsän taakse, ja vielä maille mentyänsä se loi
punaisen hohteen mummon seinälle.
Siellä oli vanhanaikaiset huonekalut mummon lapsuuden kodista
ja paljon kuvia seinillä, hänen vanhemmistaan nuorimpiin
lastenlapsiin asti. Mummo istui siellä ikkunan ääressä muistojensa
keskellä kutoen tai ommellen ja katsellen vuoroin ulos ja vuoroin
sisään.
Sieltä hän seurasi maailman menoa vilkkaalla katseellaan. Hän
tiesi mitä talossa häärättiin ja hoiti kukkia ja liinakaappia. Jos
puutarhassa kevättyöt olivat kesken, oli hän mukana kylvämässä ja
polkuja suoraksi astumassa lavojen välissä, tai jos syksyisin satoja
korjattiin, ripusti hänkin korin käsivarrelleen ja lähti papumaahan tai
hernemaahan tähystellen taivaanrantaa, tulisiko sieltä ehkä halla
yöksi, joka panisi kiirettä. Mummon päivätyö oli milloin mitäkin lajia,
ja hän lähetti sen ohessa valmistuneita käsitöitään ympäri maata

lapsilleen ja lastenlapsilleen itsekunkin syntymäpäiville tervetulleeksi
lahjaksi.
Suuren suuri kirjeenvaihto oli mummolla myöskin, sillä kaikki
pienet nallikatkin ja hienohelmat hänen suuressa perheessänsä
olivat mummon kanssa suoranaisessa kirjeenvaihdossa ja heidän
variksenjalka-kirjeisiinsä ja ukko-akka piirustuksiinsa piti tunnollisesti
vastata, puhumattakaan lasten ja muitten omaisten kirjeistä.
Entä päivän lehdet! Ne oli tarkkaan luettava, sillä ei saanut
tapahtua mitään tärkeämpää omassa maassa eikä paljon
muissakaan mummon siitä tietämättä. Oli hänellä surunsa tai
ilonsakin siitä, mitä lehdet milloinkin kertoivat, mielipiteensä hänellä
oli siitä myöskin, vaikkei hän niitä juuri muille tyrkyttänyt, hän kun
enimmäkseen tyytyi toteamaan, miten kukin asia maailmassa
kehittyi. Ne eivät hänen puolestaan muuta kaivanneet kuin että hän
ne tiesi.
Se oli aika mukavaa muille, sillä jos he kiireitten sattuessa
ennättivät vain ohimennen lehtiin vilkaista, niin ei tarvinnut muuta
kuin pistäytyä ammentamaan mummon tietovarastosta pääpiirteet
rientävän ajan tapahtumista. Tutkintojen ja kokeitten aikana Yrjö
käytti säännöllisesti tätä tapaa hyväkseen ja sillä välin muutenkin.
Mummon kautta hän siis tuli perehtyneeksi moniin oman aikansa
tapahtumiin.
Varmaan hän sai toisen käsityksen vanhuudesta kuin olisi saanut
ilman tätä jokapäiväistä seurustelua. Sillä huomasihan hän, että
mummo eli rikasta ja täyteläistä elämää, että hänen luonaan asui
maailma pienoiskoossa rauhaisan ilta-auringon valossa. Mutta eikö
lähenevä yö luonut varjoaan vanhuksen tielle? — Se ei tuntunut
paljoakaan merkitsevän. Mummo näytti olevan huoleton sen

suhteen, se näytti olevan hänelle sen yön kaltainen, joka illoin nousi
metsän takaa peittäen seudun vaippaansa. Kaikki yöt olivat
Korkeimman kädessä, myöskin viimeinen. Siitä ei tarvinnut itse
huolehtia. Hänelle kuului ilta, tämä hiljainen lauantai-ilta, jolloin viikon
työt oli loppuun suoritettu, ja oli aikaa mietiskellä ja löytää itsensä
kaiken kiireen ja askartelun jälkeen. Ja jolloin oli suloista koota
omaiset ympärilleen nauttiakseen heidän kanssaan lämmöstä
kotilieden ympärillä ja illan leppoisasta rauhasta.
Niin eli mummo vanhuuden päiviään eikä voi ajatella parempaa
tapaa. Ja moni, joka hänen luonaan kävi, tuli ajatelleeksi, että se
koti, jossa sellainen vanhuus säteilee iltahohdettaan ympärillensä,
on suuren rikkauden siunaama.
Se oli hänelle rakas, tuo hänen huoneensa, tuo hänen hiljainen
sopukkansa, johon hän kuitenkin sai niin paljon mahtumaan
yksinäisinäkin hetkinä. Koko hänen nykyinen pieni maailmansa
koivuineen ja puutarhoineen, muistoineen ja jokapäiväisine pikku
puuhineen.
Sinne tuli usein Kirsti aapiskirja kädessään tai lankakerä pyörien
puikkojen varassa, ja mummon silmien alla pujottautuivat silmukat
toisiinsa verkalleen ja kielen suussa kääntyessä. Se oli molemmille
yhtä paljon kärsivällisyyttä kysyvää, mutta oppia otettiin ja annettiin
kuten sitä on ammoisista ajoista otettu ja annettu, ja ystävyyttä
rakennettiin kuten sitä on ennenkin sukupolvien kesken rakennettu.
Ja isä naputti joskus ovelle ja kysyi mielistellen:
— Saako tulla, Sofi?

Vaikkei mummon nimi ensinkään ollut Sofi — isä oli hänet vain
siksi ristinyt, ja isä ja mummo pitivät mielellään keskenään
kaikenlaista pientä ilvettä.
Yrjö oli melkein ottanut tavaksi lukea läksynsä mummon
mukavassa keinutuolissa, minne aurinko niin iloisesti paistoi ja mistä
näki kauemmaksi kuin hänen omasta ikkunastaan. Ja mummo pysyi
silloin niin ääneti, ettei kuulunut muuta kuin hänen puikkojensa
keveätä pakinaa.
Mummo ja Yrjö ymmärsivät toisiansa mainiosti. He olivat
hengenheimolaisia tuon aikojen läpi periytyneen sielujen heimouden
tietä, joka kulkee salattuna ja yhdistää toisiinsa usein ventovieraat,
vieläpä eri kansojen ja eri aikakausien yksilöt. Mummon ja Yrjön
kesken vallitsi selvä sukulais-yhteys.
— Mitä nyt ajattelet, mummo? — kysyi Yrjö nähdessään mummon
käsien vaipuvan helmaan ja hänen katseensa pysähtyvän yhteen
pisteeseen.
— Mitä milloinkin — vastasi mummo — kuka niitä ajatuksia voi niin
tarkoin selvitellä!
— Niin, mutta sano nyt sentään!
— Minä tässä vain ajattelin Koivurinteen väkeä — kuulin heidän
äänensä äsken hallissa.
— Pidätkö heistä mummo?
— Paljon heistä pidän — kukapa ei heistä pitäisi — he ovat hyviä
ja herttaisia ihmisiä.

— Onko sinusta sitten niin tärkeätä olla hyvä ja herttainen?
— Tietysti. En pane mihinkään ominaisuuteen niin paljon arvoa
kuin hyvyyteen, oikein tuohon aitoon, välittömään
sydämenhyvyyteen. On viisauskin suuri lahja, mutta se jättää
ihmisen kylmäksi, jos se yksin jää vallalle. Hyvyys se on, joka tuo
mielenrauhaa ja onnea ihmisten keskuuteen. Se on sitä paitsi
itsessäänkin viisautta, sillä hyvä ihminen näkee kaikki oikein päin —
hän ymmärtää kaiken oikeammin kuin pahansuopa, joka näkee
kaikki kierosti, olkoonpa kuinka lahjakas tahansa.
— Sitähän sinä aina, mummo, siitä hyvyydestä — — —.
— Niin, katsos, jos hyvyys pääsisi vallalle, niin siinähän samalla
olisi se Jumalan valtakunta, jota odottelemme. Silloinhan olisimme
perillä. Ja kun en nyt enää voi mitään muu takaan toimittaa, niin
istun tässä toivomassa sitä tulevaksi ja rukoilemassa sen puolesta
— ja tiedäthän minun uskovan, että sekin voi auttaa hyvää
pääsemään edes hiukankin eteen päin maailmassa.
Kyllä. Sen Yrjö kyllä tiesi. Harvoin mummo siitä näin mainitsi,
mutta tiesihän tuon muutenkin. Se tuntui kaiken läpi.
Siinä mummo sitten illan viimeiset hetket itseksensä vietti kirjojen
kirjan ääressä tai menneitä muistellen. Paljon hänellä olikin
muistettavaa, koko pitkä elämän taival. Koko hänen entinen sukunsa
lepäsi jo nurmen alla, hän oli yksin enää jäljellä lapsuutensa
perheestä. Mutta he olivat hyvässä tallessa kaikki, ja hän siirtyi päivä
päivältä heitä lähemmäksi. Olihan heidän seurassaan jo hänen
lapsiaankin, ja ne ne vasta olivatkin liittäneet hänet läheisesti tuohon
toiseen maailmaan, joka oli hänelle niin tutunomainen, ettei tuntunut
olevan kuin askel tästä sinne.

Ja aurinko meni mailleen kellertävien peltojen helmaan ja loi
punaisen hohteensa mummon harmaalle seinälle.
21.
Muistatko, Yrjö rippikouluaikaasi? Olit silloin kuudentoista vuoden
vanha ja pitempi kuin isä. Ja sinulla näkyi selvästi haivenia
ylähuulessa.
Se syksy ei ollutkaan mikään tuollainen toivoton kausi, jolloin
sataa ropisee viikkokausia ja kärryjen pyörät tarttuvat liejuun. Päin
vastoin kauniista päivistä ei tullut loppuakaan. Koivujen lehvistö hohti
kullan keltaisena ikkunaan sinistä taivasta vastaan. Maa tuli routaan
jo ennenkuin lehdet karisivat puista ja lunta satoi jo marraskuussa.
Syystyöt puutarhassamme olivat kesken sinun mennessäsi
ensimmäiselle tunnillesi, ja kun palasit viimeiseltä, valmistimme jo
joulua.
Kävit mielelläsi rippikouluasi. Sinulla olikin opettajanasi sellainen
pappi, joka nuoria ymmärsi ja oli aikansa virtauksia seurannut. Hän
oli lämmin ja vapaamielinen eikä vaatinut oppilailtaan sokeata uskoa
siihen, minkä suhteen heillä oli epäilyksiä. Oikeastaan hän oli vain
vanhempi ystävä, joka uskoi nuoremmille mitä hän itse oli etsinyt ja
löytänyt.
Näitten tuntien jälkeen astuit kotiin päin valaistuja katuja pitkin.
Kävelit hitaasti kirja kainalossa, ja äsken kuultu sana täytti niin
kokonaan mielesi, ettet tullut kiinnittäneeksi pienintäkään huomiota

ihmistuoksinaan Ylioppilastalon edustalla. Kirjakauppojen uutuudet
eivät houkutelleet sinua pysähtymään, ja musiikkikauppojen kuvat
sivuutit niihin katsahtamatta. Kuljit suoraa päätä Rautatientorin poikki
ja aseman odotussalin läpi vilkaisematta ympärillesi.
Ja toistit mielessäsi sen, mitä olitte äsken lukeneet.
Sehän oli jo sanottu se merkillinen totuus, joka saattoi maailman
pelastaa. — Miksei sen vaikutusta missään huomattu?
Juna nykäisi jo kulkemaan ja ratavallin lyhdyt alkoivat vilahdella
vaununikkunoiden ohitse.
Koulussa olitte enimmäkseen saaneet kiviä leivän asemesta,
ajattelit vielä siellä istuessasi. Miksi ihmisille tyrkytettiin niin paljon
sellaista, mikä oli pelkkää muotoa?
Koneellisesti astuit pois junasta. Kelloa et huolinut vilkaista, tuskin
tarkastaa, oliko juna jo kulkenut sillan yli ja oliko asematalo suuri ja
valkoiseksi maalattu vai oliko se pieni ja keltainen. Se pieni oli se
oikea. Mutta vaisto sanoi milloin sen kohdalla pysähdyttiin.
Ratavalli oli kovassa jäässä. Liu'uit siitä alas ja seurasit sitten
huvila-alueen lautakäytävää.
Näit kaiken sen, minkä olit viime aikoina kuullut, kuin ilmestyksen
edessäsi. Kai se oli sitä samaa, jota mummokin oli tarkoittanut, kun
hän oli puhunut hyvyydestä ja Jumalan valtakunnasta. Niin kai pitäisi
olla. — Ei kadehtia. Ei etsiä omaansa. Ei katkeroitua. Ei muistella
kärsimäänsä pahaa. Ei saisi kerskailla eikä pöyhkeillä — vaan pitäisi
kaikki peittää. Jos voisi olla sellainen — tai edes vähän sinne päin,
silloinpa olisi ihmisellä voimaa!

Silloin hän myös voisi ymmärtää kaikki. — Ymmärtää! Siinä se oli:
ymmärtää! Joka ymmärtää, hän ei tuomitse. Joka näkee, hän ei ole
suvaitsematon. Nähdä —! Mutta ihmiset eivät näe — he ovat sokeita
— ja raivoavat sokeina toisiansa vastaan. On ollut olemassa yksi,
joka on nähnyt enemmän kuin kaikki muut, ja hänen katseensa
mahtoi olla ihmeellinen. Se näki kaiken läpi. Hän näki Jumalan ja
hän näki ihmisen. Ja siinä oli silta heidän välillään — voimavirta —
säde, joka sitoi taivaan ja maan toisiinsa.
Sehän oli niin yksinkertaista ja niin vanhaa, mutta se oli kuitenkin
aina uutta. Oli olemassa jotakin, joka voisi maailman vapauttaa ja
uudistaa, mutta elämä näkyi kulkevan juuri päinvastaista latua.
Taikka näyttikö vain siltä? Täytyikö kaiken kehityksen tapahtua
taistelun kautta? Ja siinä oli ehkä tuon kamppailun alla jokin
näkymätön virtaus, joka kerran veisi voittoon. Kyllähän sen
epämääräisesti tunsikin siellä olevaksi — ja joskus se pulpahti esille.
Sillä olihan sentään rakkautta elämässä.
Oli ollut olemassa yksi — eikä enempää tarvittu kuin että oli yksi
ollut. Hänessä oli ollut se voima, johon kaikkien tuli uskoa. Hän tunsi
itse, että hän oli maailman pelastus.
Ja hän oli sitä varmaan. Maailma ei häntä tosin tuntenut, mutta
hän oli sentään sen pelastus.
Muistatko, Yrjö, miten usein niihin aikoihin puhuimme
keskenämme näistä asioista? Kerroit kotiin tultuasi mistä oli ollut
kysymys, ja ihmettelimme yhdessä, ettei sellainen oppi, josta niinkin
paljon saarnattiin, kyennyt sen enempää oloja muuttamaan. Sitä oli
opetettu vuosisatojen ja vuosituhansien kuluessa ja suuri osa
ihmiskuntaa sitä muka tunnusti, mutta missä oli sen vaikutus?

Istuimme pienessä huoneessasi, jonka edustalla kuuset huojuivat, ja
juttelimme elämän syvistä salaisuuksista.
Mutta jos nyt jokainen koettaisi parastansa omalla tavallansa —
jos jokainen olisi hyväntahtoinen kaikille, soisi muille hyvää, palvelisi
muita?
Niin. Silloin se mullistaisi koko nykyisen elämän.
Mutta kukaan ei voi vastata siitä, mitä toiset tekevät tai jättävät
tekemättä; jokainen vastaa vain itsestänsä.
Ja muistelimme siinä kirjoituspöytäsi ääressä yhdessä tuota
merkillistä kohtaa, joka muodoltaan vaihteli sitä lausuessamme:
Ellei sinulla ole rakkautta, niin olet kuin kilisevä vaski — niin tyhjä.
Ja vaikka olisit kuinka oppinut ja kuinka viisas —ja vaikkapa
sinulla olisi usko, joka voisi vuoria siirtää — niin ei sekään vielä olisi
mitään.
Ja vaikka antaisit pois kaiken minkä omistat ja tekisit kaikesta
lahjoituksia ja säätiöitä, niin ettei jäisi mitään itsellesi — ja vaikka
ponnistaisit viimeisetkin voimasi ja antaisit ruumiisi palaa — mutta
sinua ei lämmittäisi hyvyyden lämpö, niin tuo muu ei hyödyttäisi
mitään.
Sillä tärkeintä ei ole se, minkä tekee, vaan se, mitä on.
Ja meistä tuntui siinä yhdessä tuumiessamme kuin olisi jossain
ikkuna auennut ja kuin siitä ikkunasta olisimme nähneet elämän
aivan uudessa valossa. Siitä näki kauas ja laajalti. Siitä ikkunasta

näkisimme yhä laajemmalle ja laajemmalle ajan kuluessa, kuka ties
lopulta miten kauas —! Kuka ties!
22.
Tuli sitte juhlapäivä. Sunnuntai.
Hanget hohtivat ja huurteet putoilivat puista. Koirat eivät
haukkuneet, lapset eivät toisiansa huudelleet. Oli aamusta alkaen
sunnuntai.
Se oli Yrjön ripillepääsöpäivä. Mutta se oli samalla isän juhlapäivä,
sillä hän oli sitä odotellut. Hän oli suunnitellut sitä mielessänsä,
aamuhämärissä laatikoita aukoillut ja jotakin niistä etsinyt, kirjoittanut
jotakin vanhan kirjan sisäkannelle ja asettanut sen huoneensa
pöydälle. Ja aamiaisen jälkeen ennen kirkkoon lähtöä kutsui hän
Yrjön ja äidin ja mummon ja pikku siskon huoneeseensa ja pani
kätensä vanhan kirjan kannelle.
Mutta ennenkuin hän sen enemmän siihen kajosi, otti hän pienen
rasian pöydältä ja sieltä kantasormuksen, jossa kiilsi suuri
karneolikivi keskellä.
Sen sormuksen, sanoi hän, oli hän saanut äidiltään ripille
mennessään. Mutta koska äiti oli sen hänen sormeensa painanut,
niin saisi Yrjön äiti nyt painaa sen poikansa sormeen.

Ja minä painoin sen silloin sinun sormeesi, Yrjö, ja sinä aina sen
jälkeen kannoit sitä.
— Se olkoon merkki meidän välillämme — sanoi isä — että aina
kuulumme yhteen.
Mutta isällä oli vielä jotakin muuta annettavaa. Vanha, kulunut,
nahkakantinen raamattu. Sen päälle hän laski kätensä ja sanoi
vakavasti:
— Tämän sinä saat. Sen haluaisi jokainen omakseen meidän
suuressa suvussamme — myöskin toisessa sukuhaarassamme.
Sillä se on ainoa muistoesine mikä meillä on jäljellä yhteisestä
lähtökohdastamme, pienestä pappilasta, josta olen sinulle kertonut
— ja josta kaikki polveudumme. Sain tämän kirjan isältäni lääkäriltä
— ja hän oli sen saanut isältään everstiluutnantilta — ja hän papilta
— ja nyt sinä saat sen lahjana isältäsi, koska kirja nyt on minun ja
minä tahdon että sinä, ainoana poikanani, sen perit.
Kun selailet tämän kirjan lehtiä, niin tiedät, että sitä ovat selailleet
ennen isämme ja esi-isämme elämänsä erilaisina ajankohtina,
sellaisina kuin ne jokaiselle eteen tulivat. Se oli papin työpöydällä
viikot läpeensä, mutta sunnuntaina hän otti sen mukaansa
saarnatuoliin. Vanhempana hän levitti sen eteensä ikkunalaudalle,
jotta näkisi paremmin, kiinnitti silmälasinsa ja luki sitä pitkät hetket
itseksensä. Se onkin sangen kulunut. Perästäpäin sitä ovat
kuluttaneet muut, hänen poikansa koskenperkkausretkiltään palattua
ja pojanpoikansa sairaan luota kotiin tultua. Jokainen heistä on
vuorostaan tarvinnut sitä — ellei ennemmin niin vanhetessaan.
Mutta luulen, että he ovat siitä myös jotakin löytäneet.

Nyt sinä sen saat — pidä se hyvin, ja ole nimellesi kunniaksi!
Se oli isän ensimmäinen ja viimeinen puhe Yrjölle.
Mutta Yrjö näytti siltä kuin hän olisi saanut omakseen kalleinta
mitä oli olemassa. Jotakin joka oli arvokkaampaa kuin herraskartano,
kuin suvun kantatalo, jossa muistot säilyivät itsestänsä, ja jonka
hyllylle tämäkin vanha kirja olisi ollut asetettava kaiken perustaksi.
Sen kirjan lehdillä eli se vanha henki, josta vanhukset olivat
voimansa saaneet ja josta nuoret lähtökohtansa löytäisivät
pyrkiessään vuoroonsa jaloon ihmisyyteen.
Ja nyt oli isä asettanut Yrjön tuon kaiken perijäksi. Hän oli
työntänyt ponnahduslaudan hänen jalkojensa alle — ja sinä, Yrjö,
sinä sen kaiken hyvin oivalsit.
23.
Saanhan puhua siitä, joka oli elämäsi kevät?
En kajoa sellaiseen, joka oli yksin sinun ja hänen keskistä ja jota
ette toivoneet muitten tietävän. Sanon vain sen, mitä ette
keneltäkään peitelleet. Ja kun sitä ajattelen, on kuin näkisin nupulla
olevan oksan, jollaisia kukkivien kirsikkapuitten maassa muutamin
pehmein sivellinvedoin niin mielellään maalaillaan valkoiselle
silkkikankaalle.
Ja siinä paljaassa paisuvassa oksassa on koko keväimen voima ja
runsaus.

Ethän ensinkään salannut, että pidit hänestä enemmän kuin
muista ja että hän vastasi tunteeseesi. Sehän oli helppo kenen
tahansa huomata. Koko koulu sen tiesi, mutta ei kukaan ollut sitä
tietävinään. Oli kuin itsestään selvä asia, että olitte keskenänne
ystäviä.
Marja istui edessäsi koulussa, ja näit hänen valkoisen niskansa
puseron kauluksen yläpuolella. Hänen tapansa istua oli päiväsi
ilmapuntarina. — Jos kaikki oli hyvin ja sujui tasaisesti, näytti hän
sinulle raikkaan profiilinsa ja oli iloinen ja huoleton, mutta jos olit
jostakin pahoillasi tai jos sattui jokin ikävyys tai harmi, silloin pisti hän
heti kätensä sinun käteesi pöydän alla ja kääntyi miltei ympäri sinuun
päin. Eikä hän siitä välittänyt, että sai huomautuksia
tarkkaamattomuudesta. Hän oli ystäväsi ja tahtoi sen sinulle
osoittaa.
Vastapalvelukseksi autoit häntä fysiikan ja geometrian ongelmissa
ja laadit hänelle valmiiksi pitkät sanaliuskat hänen käännöksensä.
Mutta tämä kaikki oli teistä vain itsestään selvää, eikä se
mielestänne oikeastaan mitään ollut. Ja kuitenkin se oli kaikki. Se oli
teidän tapanne puhua toistenne kanssa siitä, mistä ette koskaan
saaneet sanaa sanotuksi. Mitä tuskin selvästi ajattelittekaan, mutta
minkä tunsitte olevan olemassa suurimpana rikkautenanne ja
onnenanne.
Suothan mielelläsi, Yrjö, että häntä kiitän hänen ystävyydestään
sinua kohtaan? Olen jo häntä monta kertaa kiittänyt.
Tiesimmehän silloin hyvin kotonasi, että viimeisenä kouluvuotenasi
miltei joka päivä saatoit hänet kotiin tai menit häntä vastaan ennen
koulutunnin alkamista. Juna-ajatkin olivat niin mainiosti järjestetyt,

että tullen mennen jäi juuri sen verran aikaa kuin siihen tarvittiin. Ja
ellei jäänyt, niin myöhästyit junasta. Kuinka monta kertaa odotinkaan
turhaan ja tiesin syyn.
Olihan se sinusta harmillista, kun huomasit junan juuri porhaltavan
tiehensä ja siinä olivat edessäsi vain tyhjät kiskot. Mutta sinä pistit
kädet taskuihisi ja palasit odotussaliin, jonne jäit odottamaan
seuraavan junan lähtöä, suomatta ympäristöllesi silmäystäkään. Olit
kokonaan itseesi vajonnut. Viimeisten päivien tapahtumat kulkivat
ohitsesi sellaisina hetkinä, kaikki nuo pikkupiirteet, jotka jäivät
muistoina mieleesi. Ja varsinkin se viimeinen hetki juuri erotessanne
tänään. Hän oli puristanut kättäsi niin tukevasti jäähyväisiksi ja
katsonut sinua suoraan silmiin niin suurella ja pehmeällä katseella.
Kuinka se oli suloisesti hivellyt! Tunsit sen kiireestä kantapäähän
asti. Et päässyt sen lumoista, etkä tahtonutkaan päästä, et nyt etkä
koskaan.
Samassa juolahti jokin kaukainen muisto mieleesi. Isän sanat
aikoja sitten eräänä iltana, jolloin hän oli istuutunut vuoteesi laidalle
hyvää yötä sanoakseen — ja puhunut siitä, että piti odottaa siksi
kunnes se oikea kerran tulisi hän, jolla oli tumma tukka kuin äidillä —
jolla oli pehmoinen käsi — ja pehmoinen katse. Ah, isä! — Se oikea
on tullut!
Häntä olit sinä, Yrjö, odotellut, ja nyt istui hän joka päivä edessäsi
koulussa.
Hänellä oli hiukset solmussa niskassa ja pari pehmeätä suortuvaa
riippumassa avonaisella kauluksella. Niin lähellä hän istui, että miltei
kuulit hänen hengityksensä. Muistit, että olitte ranskan tunnilla
jotakin kuiskailleet — ja näit vieläkin hänen silmänsä — hänen
leukansa pyöreyden — hänen valkoiset hampaansa huulten

auetessa Kaiken elit uudestaan siinä istuessasi ja ihmisten
edestakaisin liikkuessa.
Vihdoin otit esille muistiinpanokirjasi ja lisäsit siihen tapasi mukaan
muutamia sanoja viime päivien merkillisistä kokemuksista:
»Olin eilen heillä. Marjan äiti istui ompelu pöytänsä ääressä ja
puhuin hänen kanssaan. Silloin tuli Marja ulkoa ja istuutui viereeni.
Me katselimme toisiamme koko ajan — uskon varmaan, että hän
pitää minusta.
»Hän tuli väliajalla vastaani Martti veljensä kanssa. Martti on aina
niin ystävällinen minulle — luulen että hän aavistaa jotakin. Marja
katsoi taaksensa. Mahtaako hän tietää, kuinka paljon hänestä
pidän? Ja mahtaako hän pitää yhtä paljon minusta?
»Istuimme tänään käsikädessä maantiedetunnilla. Hän tuli vain
viereeni, kun Jussi oli poissa, eikä opettaja mitään sanonut. Voi
Marja sentään, kun olet minulle hyvä ystävä, ja miten aina olemme
samaa mieltä kaikesta! Olet aivan kuin sisar ja kuitenkin rakkaampi
kuin sisar voisi olla. Paljon rakkaampi. Kuten isä ennusti.»
Sitten paljon myöhemmin löysin tuon pienen mustakantisen
muistikirjasi pöytälaatikostasi ja luin sen läpi. Olithan sen sinne
jättänyt — tervehdyksesi perästäpäin meille ja Marjalle. Hän sen
sitten sai. Eikä siinä ollut mitään muuta kuin tuon keväisen
rakkautenne tarina pikkupiirteittäin, ei mitään, jota ette voisi näyttää
koko maailmalle. Mutta me emme sitä näytä — se on teidän
pyhättönne.
Tänään, tässä suuressa sunnuntairauhassa, tulevat muistot
mieleeni hetkistä, jotka silmäini edessä ennen välkkyivät ja häipyivät

pois menneisyyteen.
* * * * *
Oli aikainen kevättalvipäivä ja silloinkin sunnuntai.
Muistelmani näkyvät kulkevan sunnuntaista sunnuntaihin. Omaksi
ihmeekseni. Ja kuitenkin se on aivan luonnollista, koska sunnuntait
meillä olivat merkkipäiviä jo sen kautta, että silloin isä ja sinä olitte
kotona koko päivän. Minusta tuntuu todella nyt perästäpäin, kuin
kaikki tärkeät asiat olisivat tapahtuneet sunnuntaisin ja kuin
elämämme kokemukset muodostaisivat helminauhan, jonka
suurimmat helmet ovat sunnuntaita.
Marja oli tullut luoksemme aikaisella aamujunalla, ja te olitte olleet
hiihtämässä yhdessä. Olin sillaikaa kattanut kahvipöydän
ruokahuoneessa, kääntänyt vaaleanpunaisen pelargonian valoa
kohti, nostanut uunin laidalta ikkunalle muutamia tulppaaneja ja
hyasintteja, joiden kukat jo olivat puhjenneet paperitötteröjensä alla,
ja katsellut auringonpaistetta, joka heijastui ruudullisena suurten
huoneitten matolla. Isän askelia kuului toisesta kerroksesta ja Kirsti
leikki pallollaan ylemmässä hallissa. Hän kuului juttelevan isän
kanssa.
Silloin te palasitte hiihtoretkeltänne terveyttä ja nuoruutta uhkuen.
Olitte puhuneet toistenne kanssa jostakin, mitä lie ollut, juuri sisään
astuessanne, ja jatkoitte siitä vielä eteisessä leikillänne väitellen ja
raikkaasti nauraen tömistäessänne lumen jaloistanne.
— Eipä, kuin Pariisista Lontooseen!
— Älähän, vaan Lontoosta Pariisiin! —

— Niin, kumpikohan matka lienee pitempi?
Ja siitä väitellessänne pyrähditte sisään, ja sinä, Yrjö, tahdoit
saada minutkin vakuutetuksi siitä, mikä suunnaton merkitys
lentokoneilla tulisi olemaan tulevaisuudessa. Tulisi aika, jolloin
halveksien puhuttaisiin niistä matosista, jotka tyytyivät ryömimään
maassa.
— Minä puolestani ainakin aion oppia lentämään — ja sitten
nähdään,
Marja!
Olit asettunut kädet selän takana nojaamaan kaarevaa uunia
vastaan väliseinän nurkkauksessa ja huojuit siinä edestakaisin
puhuessasi. Minä kuuntelin sinua matalalta sohvalta sitä vastapäätä.
Silloin tuli Marja, pisti kätensä kainaloosi ja huojui hänkin
edestakaisin, loistaen iloa ja katsoen sinua silmiin sinun siinä
kertoessasi tulevaisuuden ihmeistä.
Näin ettet rävähyttänyt silmiäsi etkä hievahtanut paikaltasi. Teillä
oli molemmilla aivan itsetiedoton ilme kasvoillanne, ikäänkuin tuo
kosketus olisi ollut luonnollinen ilmaus ystävyydestänne. Sinä jatkoit:
— Me olemme Marjan kanssa päättäneet, että joskus menemme
jotakin opiskelemaan jonnekin ulkomaille, Berliiniin —?
— Eipäs, vaan Lontooseen!
— Vaikkapa Pariisiin — ja silloin on pääasia, että lennämme —!
Eikö totta, Marja — hei! Korkealla pilvien yläpuolella katsomme teitä
siellä alhaalla ylenkatseellisesti näin —!
— Eipäs, vaan näin —!

— Sinä —! — No niin! — Ja leijailemme ilmassa ja minä hoidan
konetta ja ymmärrän pienimmänkin vivahduksen sen surinassa. Ja
aurinko paistaa avaruudessa —. Se olisi jotain, Marja!
— Olisi se! Jos sinä vain pyydät, niin kyllä minä tulen. Varmasti!
Istuin ja katselin teitä molempia. Saattoiko olla olemassa kahden
ihmislapsen kesken suhde, joka oli noin itsetiedoton, noin raikas ja
noin kaunis! Se oli kuin ilmestys jostakin, jota en ollut ennen nähnyt
ja johon tuskin olin uskonut. Sillä ei ollut mitään nimeä eikä se
mitään tarvinnut. Se oli vain olemassa ja iloitsi olemassaolostaan.
Mutta tiedän, että se vähitellen varttui yhä enemmän tietoiseksi ja
että se pyöri mielessäsi yhä polttavampana kiintymyksenä, joka ajoi
sinut Marjan läheisyyteen. Se tuli päiväsi ensimmäiseksi ja
viimeiseksi kaipaukseksi.
Eräänä sunnuntaina viikkoja edellisen jälkeen tulitte meitä vastaan
palatessamme kotiin asemalta. Se on mielestäni kuin eilen. Oli
satanut paksulta lunta, tiet olivat aivan tukossa, niin ettette voineet
kävellä tasaisesti, vaan kuljitte käsi kädessä eteenpäin jäätynyttä
harjannetta pitkin hoiperrellen molemmille puolin. Te nauroitte ja
huikkasitte ja teidän silmissänne oli jokin aivan erikoinen
yhteisymmärrys. Oli ikäänkuin juuri olisitte uskoneet toisillenne
salaisuuden ja kuin jokaisella katseellanne olisitte siitä toisillenne
viitanneet. Miksi minä sen niin tarkkaan muistan? En sillä kertaa
pannut siihen niin paljon huomiota. Mutta perästäpäin minulle selvisi,
että te sinä päivänä todellakin olitte solmineet jonkinlaisen liiton
keskenänne. Ja että suhteenne senjälkeen oli entistään vielä paljon
läheisempi ja varmempi.

Se oli onnellisimpia päiviäsi, Yrjö. Marja on siitä perästäpäin
minulle kertonut.
Mutta keväthanget hohtivat auringossa ja lumi suli vähitellen pois.
* * * * *
Pari päivää tuon tuiskupäivän jälkeen tulla tömähdit kotiin koulusta
väliajalla, mitä ei milloinkaan muuten pitkän välimatkan vuoksi
tapahtunut. Mutta sinä päivänä tulit asemalta, juosta karautit portaita
toiseen kerrokseen ja jäit seisomaan huoneesi ovelle. Luulin, että olit
jotakin unohtanut, jota välttämättömästi tarvitsit. Tarjosin sinulle
aamiaista ja kahvia — olin miltei säikähtynyt kiireestäsi — mutta
seisoit vain ovellasi ja palasit sitten juosten takaisin rautatieasemalle.
Mitä varten olit tullut? Mitä olit oveltasi tarkastanut?
Kirjoituspöydälläsi oli pieni punainen tulppaani, jonka Marja
edellisenä päivänä oli sinulle lähettänyt.
Sitäkö olit tullut katsomaan?
Sitä, Yrjö! Muistat hyvin sen pienen kukkasen vaatimattomassa
tuohiruukussaan. En voi enää sellaista yksinäistä pientä viestiä
nähdä muistamatta sinua. Ja tiedän nyt, että Marja oli sen sinulle
antanut eilisen liittonne merkiksi. Ette ajatelleetkaan vielä muita
merkkejä. Ei. Tunteenne oli vielä niin herkkää, että sanoihin
puhkeaminen olisi sitä vain häirinnyt. Sen symbooliksi sopi pieni
veripunainen kukkanen. Ja siitä kukkasesta, siitä Marjan
ystävyydenosoituksesta olit niin onnellinen, että sinun täytyi käydä
tuo pitkä matka edestakaisin vain vilahdukselta nähdäksesi sen.

Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade
Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!
ebookultra.com

Get Constraints in Discourse 172nd Edition Anton Benz (Ed.) free all chapters

More Related Content

Similar to Get Constraints in Discourse 172nd Edition Anton Benz (Ed.) free all chapters (20)

Recently uploaded (20)

Get Constraints in Discourse 172nd Edition Anton Benz (Ed.) free all chapters