Sanders, Ted, Joost Schilperoord, and Wilbert Spooren, ed. (2001) Text
Representation: Linguistic and Psycholinguistic Aspects. John Benjamins
Publishing Company, 363pp, hardback ISBN 1-58811-077-X, USD 86.00 /
90-272-2360-2, EUR 95.00, Human Cognitive Processing 8
Announced in http://linguistlist.org/issues/13/13-217.html
Laura Alonso Alemany and Irene Castellón
In what follows, we give an overview of the structure and contents of
the book, with a brief summary of each chapter. Then, each section is
presented in turn, with a more extensive review of each of the
articles. We finish with a critical overview of the volume as a whole.
SYNOPSIS
The theme of this book is the linguistic and psycholinguistic aspects
of text representation, from a cross-disciplinary perspective. It is
based on papers presented at the International workshop with the same
title held at Utrecht University in July 1997.
The book is divided in four sections. Section 1 deals with referential
coherence, focussing on accessibility. A second section, the most
extended one, discusses relational coherence by way of diverse
theoretical perspectives and empirical methods. Section 3 introduces
questions of knowledge representation related to text representation,
and section 4 presents aspects of text segmentation. Each section is
preceded by an overview from the editors, where they give a brief
summary of the following chapters, contextualise them and establish
relations between them.
The introduction (Chapter 1) presents the field of research and the
contents of the book.
Section 1 focuses on referential coherence. Chapter 2 gives an overview
of accessibility theory, which tries to explain the choice of one among
all possible referential forms in terms of accessibility of the
referent. Chapter 3 provide an account of how text cues guide the
reader's attention in reading and how they consequently contribute to
shaping the cognitive representation of the text. Finally, in Chapter
4, a modular hypothesis of lexical access in text production is
supported via experiments on familiarity of metaphors.
Section 2 deals with coherence relations. In Chapter 5, an intention-
based definition of pragmatic relations is proposed to adequately
address the distinction between the so-called 'semantic' and
'pragmatic' coherence relations. Chapter 6 presents empirical evidence
that supports a typology of concessive relations in terms of underlying
causality patterns and thematic continuity in the surrounding
discourse. Chapter 7 presents some problems of RST relation
'Elaboration' and proposes an alternative model of discourse coherence
where local dependencies are explained by RST relations except
Elaboration, and global coherence is based on focus-driven non-local
dependencies. Chapter 8 provides an account of the conjunction 'and',
claiming that it does not link segments but makes them jointly relevant
to the surrounding discourse. In Chapter 9 the separation of
propositional and illocutionary levels is questioned, and the pragma-
dialectical approach of Argumentation Theory is proposed as an
alternative analysis for linguistic cues in explanation and
argumentation.
Section 3 addresses knowledge representation. In Chapter 10 an
investigation about coherence relations and inferences is presented.
Chapter 11 proposes a quantitative model that represents how people
think about information technical or scientific. This model exploits
the relation between text and knowledge.
Section 4 focuses on text segmentation. In Chapter 12, a modular view
of language production processes is supported by an analysis of pauses
in dictations. The last chapter presents some discrepancies between
syntactical structure and discursive segmentation, and proposes the
distinction between two levels in discourse interpretation: cognitive
coordination and informational content.
INTRODUCTION
The first chapter of the book serves as an introduction. Ted Sanders
and Wilbert Spooren characterise the field of study of text
representation, they signal major research themes in the area and
present basic concepts that are going to be used throughout the rest of
the book. They put a special emphasis on the concept of coherence.
Coherence is considered as the mental correlate of textual
connectedness, which is in turn considered a text-constituting
characteristic. The distinction is made between referential and
relational coherence: while the first accounts for repeated reference
to the same object in a discourse, the second explains how coherence
relations like 'cause' connect text segments.
SECTION 1. Referential coherence: accessibility and text processing.
In the first chapter of this section, Mira Ariel presents an overview
of her accessibility theory. The central claim of accessibility theory
is that language users choose among all the possible referential forms
in the language depending on the accessibility of the referent, so that
more elaborate referential markers correspond to less accessible
referents. In other words, these markers signal the degree of
accessibility with which the mental representation to be retrieved is
held, so they can be ordered in an accessibility marking scale, going
from low accessibility markers (full names, definite descriptions) to
high accessibility markers (pronouns, verbal person inflections,
zeroes). It is the combination of accessibility factors (head
complexity, distance, grammatical role of the relativized position,
restrictiveness) that determines the referential form, and not any
single factor. However, accessibility considerations do not account for
the whole of the selection process: also contextual assumptions, such
as relevance-based considerations, play a role in determining
referential form.
Further research is presented that corroborates general accessibility
predictions and enriches the original theory. The author compares
accessibility theory to other theories of reference, such as Chafe
(1976), Givón (1983), Levinson (1987, 1991), Gundel et al. (1993) and
Centering (Grosz et al. 1986, 1995). She notes a common core, namely
that all theories offer some scale of referring expressions, that they
all agree in that pragmatic factors can override the principles they
propose and that they all converge on predictions about zeroes,
pronouns and lexical NPs. She discusses each of the mentioned theories
and comes to the conclusion that none can account for the full range of
distributional patterns of referring expressions as well as
accessibility theory. To finish, Mira Ariel presents sketches some
directions for further research, contextualising and justifying them.
In chapter 3, Michelle L. Gaddy, Paul van den Broek and Yung-Chi Sung
explain how heterogeneous text cues (linguistic cues, typographical
cues and text structural ones) guide the reader's attention in reading,
and how they consequently affect her mental representation of
discourse. They provide an account of the on-line reading process in
the framework of the Landscape Model, where the reading process and the
resulting mental representation is explained in terms of varying
activation degrees of the concepts throughout discourse. The studied
text cues are found to have definite activation functions in this
model. Within linguistic cues, function and relevance indicators (for
example, 'in summary') increase activation of the concepts they co-
occur with, favouring their subsequent retrieval in further discourse,
while anaphors and cataphors re-activate (or pre-active, respectively)
a concept that was in background. Typographical cues (italics,
boldface) have an effect comparable to that of linguistic relevance
indicators. Titles and headings are considered text-structure cues that
direct reader's attention to particular content, thus biasing their
processing of the text. In all the cases, activation of concepts can be
translated to higher attention from the reader, which has been proved
in various experiments that the authors refer to, mostly memory tests
or reading time experiments.
In Chapter 4, Rachel Giora and Noga Balaban address lexical access in
text production. Assuming comparability between text comprehension and
text production, they carry out an experiment on how literal and
metaphorical lexical meaning is accessed. In this experiment, subjects
rated metaphors in newspaper text according to their familiarity. It
was checked whether each metaphor was followed by a mention of its
literal meaning, which was taken as a signal that the coded meaning of
the word was activated. Results show that the coded meaning of a word
is usually activated, regardless of the familiarity of the metaphor,
that is to say, even if the context strongly evokes a 'figurative'
meaning. This rejects an interactionist hypothesis that assumes that
contexts directs lexical access so that only the appropriate meaning of
words is made available for comprehension. Consequently, a modular view
of lexical access, the 'graded saliency hypothesis' is supported. This
hypothesis assumes that all the meanings of a word that are coded in a
language are always activated. However, as the authors themselves
state, this experiment cannot be taken as conclusive evidence for a
modular view of lexical access, since the measure used was not on-line.
SECTION 2. Relational Coherence.
In Chapter 5, Alistair Knott discusses the distinction between two
kinds of coherence relations, the so-called 'pragmatic' and 'semantic'
ones. He presents problems with some previous proposals. He argues that
Sanders, Spooren and Noordman's (1992) pragma tic - semantic
distinction cannot satisfactorily account for the data. Sweetser's
proposal of dividing Sander's pragmatic relations into epistemic and
speech-act is found to solve some of the exposed problems, yet others
arise. Finally, Knott proposes an intention-based definition of
pragmatic relations that seems to satisfactorily account for the data,
while keeping Sander's simpler binary distinction of pragmatic and
semantic relations. However, some explanatory inadequacies are spotted
by the author himself that seem to favour Sweetser's tripartite
distinction. Future work, the author points out, should try to
generalise over the intentions of the protagonists of the discourse and
the participants in the speech act.
In Chapter 6, Leo Noordman presents a corpus study on concessive
(although) relations. He makes a distinction between asymmetric and
symmetric although-sentences, namely, those expressing a denial of
expectation and those expressing concessive opposition. In a denial of
expectation, the clause containing the expectation is syntactically and
discursively subordinate to the main proposition, which denies it. In
contrast, concessive oppositions establish a relation between two
clauses which are equally central to the discourse, even though one of
them is syntactically subordinate to the other. In addition, he takes
concession to be a complex thought, the combination of causation and
negation, and distinguishes two different concessives based on the
different causalities underlying although-sentences: 'default order',
if the cause precedes the consequence, and 'reversed order' if the
consequence is first. This second division is only applicable within
sentences expressing a denial of expectation.
The distinction between these different types of although-sentences is
supported by empirical evidence. In the first place, reading time
studies show that default order causal relations are processed faster
than reversed order ones, implying that the cause-consequence order is
more natural to human reasoning processes. This preference is also
supported by a corpus study showing that default order causal relations
are more frequent than reversed order ones. It was also found that
sentences expressing a denial of expectation have a tendency to the
subordinate clause - main clause structure, indicating another
reasoning preference, namely, to mention first the cause for the
expectation and subsequently the negation of that expectation.
What's more, the different relations were found to have different
behaviours in relation to the thematic development of discourse. To
describe how the different relations are embedded in their context,
four factors were taken into account: the main-subordinate order,
whether the text following and preceding an although-sentence was a
continuation of the main clause, whether the second clause in a complex
sentence is thematically related to the subsequent context and whether
the first clause is thematically related to the preceding context,
regardless of syntactical status. These factors constitute a model for
the thematic continuity of each of the three kinds of although-
sentences (default order and reversed order denials of expectation, and
concessions) with its context, with a fit of the data with the model
that ranges from 92% to 99%. Ultimately, this serves for establishing a
clear link between the mental representation of discourse and its
surface characteristics.
In Chapter 7, Alistair Knott, Jon Oberlander, Michael O'Donnell and
Chris Mellish argue that the Elaboration relation proposed by RST
(Rhetorical Structure Theory, Mann and Thompson 1988) is usually used
as a waste-paper basket, as the default relation in text analysis when
no other relation fits. They note that this relation is quantitatively
different to the rest of relations proposed by RST. Structurally,
elaboration can hold between non-adjacent spans, which is not the case
for other relations, whereas it is specially resistant to even the
simplest embeddings. Unlike the other relations in RST, elaboration is
not really a relation between propositions, but between components of a
proposition. Moreover, there are no linguistic signals that can be
consistently related to this relation, as there are for all the rest
(Knott and Mellish 1996).
The authors propose a new model of coherence, so that a coherent text
is considered as a 'sequence of focus spaces which succeed each other
in a legal manner'. Each focus space is constituted by a so-called
'entity-chain', a sequence of RS trees whose the top nucleus of each
tree is a fact about a common entity. A sequence of entity-chains is
coherent if the focused entity in each chain is mentioned in a
preceding chain which is not too far in the line of discourse.
In Chapter 8, Henk Pander Maat claims that the conjunction 'and' does
not link two segments directly, but makes them jointly relevant to the
surrounding context. His starting point is the relevance theoretic work
of Carston (1993) and Blakemore (1987), who were the first to propose
joint relevance as the adequate account for 'and'. He strengthens this
notion by considering 'and' as a topic continuity marker. Taking topic
to be the explicit or implicit question that is being answered by a
segment of discourse (Van Kuppevelt 1995), 'and' combines two segments
of discourse as a single topic. A corpus study of interclausal
conjunctions shows that the majority do present joint relevance. Two
kinds of joint relevance environments are distinguished: supporting or
elaborating an assumption or answering a single question, this being
the most frequent.
Some theoretical implications of his account of the conjunction are
discussed. First, the meaning of 'and' is claimed to be procedural,
because it constrains the possible implicatures between the joined
elements. Second, joint relevance relations are placed in an expansion
of the coherence relation classification of Sansders, Spooren and
Noordmant (1992), as a subtype of non-causal relations, namely additive
and comparative. Finally, this account is related to recent work on the
role of connectives in the construction of discourse representations,
suggesting that the differences between juxtaposed and coordinated
sentences should be investigated.
In Chapter 9, Francisca Snoeck Henkemans provides a different
perspective on coherence relations and connectives from Argumentation
Theory. She argues that the separation of propositional (content) and
illocutionary (means-end) levels in theories of coherence relations is
often inadequate. As an example, the author states that explanations
and argumentations can be confused, because they can be both based on a
causal relationship, despite their different illocutionary aim: while
explanations intend to facilitate comprehension, argumentations try to
increase acceptability of a certain standpoint. Linguistic cues, such
as connectives, are of little help, since most of them may signal both
argumentation or explanation.
The author argues that the pragma-dialectical approach of argumentation
provides a good basis for interpreting the linguistic cues in a well-
founded and systematic way. To support her claim, she gives an overview
of the main conditions that are taken into account in a pragma-
dialectical account of argumentation and explanation. These conditions
suggest that systematically linking the propositional and illocutionary
levels provides crucial information for the analysis of argumentative
discourse.
SECTION 3. From text representation to knowledge representation
Chapter 10 deals about the construction of inferences during text
comprehension. The aim of authors, A.C. Graeser, P. Wiemer-Hastings and
K. Wiemer-Hastings, is to show that knowledge plays a central role in
inference mechanisms. First, the constructionist theory of inference
generation is presented (Graeser et al 1994). Constructionist theory of
inference offers predictions about what Knowledge based inferences are
generated when readers construct a situational model. Secondly, a
three-pronged method for investigating inferences are explained, the
three prongs are: theoretical predictions, verbal protocols and on-line
behavioral measures. To finish, we find a catalogue of relations that
are used to relate text constituents and conceptual entities in world
knowledge structures. The author's assumption is that World Knowledge
is sufficiently constrained and it can be integrated in theories of
language processing based on lexicon, syntax and semantics.
This last section deals with coherence relations, distinguishing local
and global coherence; they present the Zwaan method 'Event indexing'
that assumes that the reader accesses five conceptual dimensions
(protagonist, temporality, spatiality, causality and intentionality).
Also the authors present a catalogue (appendix 1) of relations used in
a World Knowledge representation that can be worthwhile. Each relation
is defined by a name, a definition, composition rule and an example.
All these relations connect nodes of five categories (Concept, State,
Event, Goal, Style).
In Chapter 11 a model for thinking about bodies of knowledge is
presented. B. K. Britton, P. Schaefer, M. Bryan, S. Silverman and R.
Sorrells explain an investigation about the conceptual representation
of text; first, they introduce a way of representing knowledge as a
network of concepts. Based on expert knowledge on the subject of the
text, two structures are created: Expert structure (by qualified human
experts) and a Predicted Thought Structure (by an algebraic model).
This model is based in 'spreading activation and relaxation' that
permits to activate related concepts in a recursively way. Concepts and
relations between concepts compose these two structures.
In this way, two experiments are carried out in order to compare the
human thinking process with the expert structure and with the Predicted
Thought Structure. There are two groups of subjects: think condition
group and no think condition group (the difference is that the first
ones were asked to think about the text for a few minutes a day during
a week and the second ones were not. The authors point out that the
hypothesis of this work is that knowledge structures change as a result
of being thought about. Results of the experiment support the
Prediction Thought Model. The discussion is focused in three issues:
the meaning of the products of thought, their implications for memory
storage and retrieval and the implications of the results for the
validity of the representation.
SECTION 4. Segmentation
In Chapter 12, Joost Schilperoord investigates the soundness of
analysing pauses in dictations as empirical evidence for discourse
planning processes. Ample evidence is provided from previous studies
showing that length and location of pauses in discourse are good
correlates of the underlying production processes, and that they
succeed in distinguishing two levels in planning processing: conceptual
and linguistic. Long pauses at paragraph or sentence boundaries signal
conceptual planning, while short pauses at word or phrase level signal
linguistic planning. Clause level seems to be an intermediate between
the two.
Schilperoord analyses how variation in texts influences variation in
these correlates of production processes. The underlying hypothesis was
that, if the conceptual and linguistic processes interact, variation in
texts should cause the pausal correlates to co-vary. Three aspects of
variation in text were explored: differences between texts (length,
complexity), variation in pause lengths for different location types in
texts and 'real production time'. Results show that it was impossible
to predict any of the two pause patterns on the basis of the other, for
none of the three variables. Therefore, no empirical evidence was
gathered supporting the interaction hypothesis, which favours a modular
view of language production processes, where conceptual and planning
processing are independent of each other. Despite that, it was found
that processing at clause level is affected by paragraph and sentence
level processes, and that it affected linguistic level, thus partially
supporting the interaction hypothesis for this level. The author points
out that further research should be carried to support the empirical
results obtained so far, for example, exploring writing processes
requiring more planning activity than the ones in this study, lawyer's
dictations of informative letters.
In Chapter 13, Arie Verhagen addresses the delimitation of discursive
units, the so-called segments. He focuses on constructions where there
is a conflict between syntactical and discursive subordination, namely
complex sentences with embedded subject or complement clauses. RST
account of these constructions states that the subordinate clause was
to be considered as "part of its host clause" (Mann and Thompson 1988).
Verhagen claims that two different message dimensions of discourse
interpretation are necessary to adequately analyse these structures:
content and (intersubjective) coordination, similar to the distinction
between semantic and ideational relations in RST. Therefore, discourse
segments are not to be distinguished linearly, in one dimension, but in
two, so that main clauses provide the subject of consciousness for the
content of the subordinate clause. This is what the author calls the
'embedding construction'. Assuming this twofold account of discourse,
it is the main clause that is actually conceptually dependant on a
subordinate one. This conception provides a more adequate explanation
of some problematic examples, and succeeds in distinguishing subject
and complement clauses from restrictive relatives, which can never
constitute a separate discourse segment. In addition, evidence from
thematic continuity patterns supports this claim, as the distribution
of discourse anaphors is shown to be sensitive to the coordination and
content dimensions.
CRITICAL OVERVIEW
This book constitutes a good reference to a field which is quite
chaotic, because it is young and multidisciplinary. It is not a mere
collection of independent works, but it provides a joint overview of
the field. The editors emphasize the relations between each chapter, so
that it is easier to get a global picture of the area.
In an area in constant evolution, where no there are no established
methodologies, the works presented in this book provide a valuable
reference: the analysis of the problems is clear, and the proposed
methods and solutions are well argued. In addition, in most of the
cases, the solutions proposed are novel and sound. We would like to
remark a generalised effort to turn the implications of theoretical
models into empirically provable issues that can be objectively
supported with psycholinguistic experiments or corpus studies.
Moreover, the results of these empirical tests are not given as self-
sufficient evidence, but authors resort to statistical modelling to
provide unbiased interpretations of the data and avoid the fallacy of
fully supporting theories on raw data.
In each of the four subtopics of the book, different theoretical
perspectives are represented. The authors and editors have been able to
interrelate this diversified points of view so that diversity does not
lead to useless criticism, but results in mutual enrichment.
REFERENCES
Blakemore, D. (1987). Semantic constraints on relevance. London: Basil
Blackwell.
Carston, R. (1993). Conjunction, explanation and relevance. Lingua, 90,
27-48.
Chafe, W. L. (1976). Givenness, contrastiveness, definiteness,
subjects, topics and point of view. In C. N. Li (ed.), Subject and
topic (pp. 25-55). New York: Academic Press.
Givón, T. (1983). Topic continuity in discourse: An introduction. In T.
Givón (ed.), Topic Continuity in Discourse: A Quantitative Cross-
Language Strudy (pp. 1-42). Amsterdam: John Benjamins.
Grosz, B. J., Joshi, A., and Weinstein, S. (1995). Centering: A
framework for modeling the local coherence of discourse. IRCS Report
95-01, The Institute for research in cognitive science, University of
Pennsylvania.
Gundel, J. K., and Mulkern, A. E. (1993). Quantity implicatures in
reference understanding. Pragmatics and cognition, 6, 21-45.
Knott, A., and Mellish, C. (1996). A feature-based account of the
relations signalled by sentence and clause connectives. Language and
Speech, 39, 143-183.
Kuppevelt, J. van (1995). Discourse structure, topicality and
questioning. Journal of Linguistics, 31, 109-147.
Levinson, S. C. (1991). Pragmatic reduction of the binding conditions
revisited. Journal of linguistics, 27, 107-161.
Mann W.C. and Thompson, S. A. (1988). Rhetorical structure theory: A
theory of text organization. Text, 8, 243-281.
Sanders, T. J. M., Spooren, W. P. M. and Noordman, L. G. M. (1992).
Towards a taxonomy of coherence relations. Discourse Processes, 15, 1-
35.
ABOUT THE REVIEWERS
Irene Castellón is professor at the University of Barcelona, in the
Linguistics Department. Her main research area is Natural Language
Processing, in particular computational grammars and computational
lexicography.
Laura Alonso Alemany is a doctoral student at the CLiC, Centre for
Language and Computation at the University of Barcelona. Her main
research area is Discourse Processing for Automated Text Summarisation
in Spanish. She is currently working on a shallow rhetorical parser for
Spanish unrestricted text. Her areas of interest are discourse and
rhetoric, and natural language processing.
|
