Review of Mental Models and the Interpretation of Anaphora
|Date: Wed, 12 Nov 2003 12:57:27 -0500
From: Umarani Pappuswamy
Subject: Mental Models and the Interpretation of Anaphora
Garnham, Alan (2001) Mental Models and the Interpretation of Anaphora,
Psychology Press Ltd.
Umarani Pappuswamy, Research Associate, Learning Research and
Development Center, University of Pittsburgh, USA.
BOOK'S PURPOSE AND CONTENT
The origin of the term 'mental model' can be traced back to Kenneth
Craik (1943). Craik pointed out that the mind constructs "small-scale
models" of reality that it uses to translate external events into
internal models and also reason about these representations. There has
been a considerable amount of research in this area following Craik.
Some of the significant works are by cognitive scientists such as
Miller, Galanter and Pribram 1960, Newell and Simon 1976, Johnson-Laird
1980, Johnson-Laird and Garnham 1986, Garnham, Oakhill, and Johnson-
Laird 1982, Marr 1982 Garnham 1987, Garnham and Oakhill 1989. Though
all these studies (from visual perception to discourse processing) have
their own diversities, they all invoke the mental-model hypothesis in
various forms. In processing a discourse, it is very important to have
a better comprehension of a text in order to make inferences necessary
for a coherent interpretation. Humans construct 'mental models' of the
situations in the discourse as they perceive them. The mental model,
thus constructed, should be able to resolve an anaphoric expression, if
any, present in the discourse. The research outlined in this book
presents an overview of the author's research on anaphor interpretation
set within the context of mental models on anaphora from the
disciplines of psycholinguistics, philosophy, linguistics, and
computational linguistics. He argues that '..the notion of a mental
model is essential in the detailed description of the processes of
anaphor resolution' (p. ix).
The entire book is presented in eleven chapters. A brief
description/summary about the chapters are given below:
Chapter 1: This chapter gives an introduction to the theme of the book.
The author talks about Marr (1982)'s theory of visual processing and
compares his own goals with that of Marr's with regard to mental
processing. Marr focussed on visual processing whereas Garnham
concentrates on text comprehension. The main focus is about the
interpretation of anaphoric expressions in text, such as definite and
indefinite pronouns, and elliptical verb phrases. The author claims
that in the interpretation of anaphoric expressions, the construction
of a representation of part of a world is essential. Marr attempted to
focus on the "what" of visual processing though he does give some hints
about his answer to the "why". Garnham, on the other hand, tries to
seek answers for both these questions in his book.
Chapter 2: Garnham studies the nature of mental representations of a
text, and in particular, the question of what components they are built
from. The chapter discusses at length as what constitutes a 'mental
model' of a text and what kinds of things are represented in the mental
model. He turns to linguistic semantics for answers starting from
Montague (1973)'s model of formal semantics. For Montague, texts are
about individuals, and entity and truth values (e & t) are the two
fundamental semantic types. After a critical observation of these
types, Garnham points out that Montague has not given importance to
'events and properties' as basic semantic types. Though Montague did
not deny the existence of properties, he regarded it be derivative
(from e & t), and he did not have a place for events in his system.
Montague's system is based only on individual sentences. This approach
seems to be inadequate and this reviewer also agrees with Garnham on
this aspect. The author also provides evidences from various other
scholars, Davidson (1967) and Parsons (1990) to argue that Montague's
model is inadequate when discourse is considered. He quotes Prior
(1968), Kamp (1978), Kamp and Rohrer (1983) and emphasizes that both
properties and events are represented directly in mental models. He
also provides facts that events are related to one another (spatial,
temporal, logical, causal, intention and moral) to support his claim.
Chapter 3: This chapter introduces mental models theory as a
psychological theory, focussing on its application to language
processing. As is well known, many aspects of language use depends on
an individual's cognitive ability. Garnham discusses the scope, the
symbolic nature, and assumptions of mental models theory. In addition
to the assumptions of mental models theory, he describes some
assumptions that are particular to the theory as applied to language
processing (incremental processing). A theory of comprehension is
worked out - the detailed mechanisms that extract information from the
current piece of text, and add it to the mental model should be
specified. This interpretation keeps on updating the model and provides
the context for interpreting the next piece of text. This forms the
core of the book in interpreting anaphoric expressions. The author
points out that 'integration' and 'construction' are two important
aspects in text comprehension and that mental models are considered to
be central to both these aspects. He also gives an outline of what a
theory of language processing should explain: 'explicit content of a
text, orgranisation of knowledge to make inferences'. He also hints at
the complexities involved in constructing mental models with syntactic
and semantic knowledge of the text in a study.
Chapter 4: In this chapter, the author presents an overview of
linguistic approaches to anaphora. He looks into the way linguists
handle nominal and verbal anaphora, and states that the goals and
methods/approaches to anaphora of psycholinguists differ from linguists
in many ways. The psychologistic approach should explain why sometimes
people use anaphoric expressions in ways that linguistic descriptions
forbid. He also discusses the morphological and syntactic issues
related to anaphora and cataphora using English language as an example.
The syntactic constraints and discourse constraints on anaphora are
also examined at depth.
Chapter 5: The fifth chapter introduces the methodologies used in the
psycholinguistic analysis of anaphors. It discusses the techniques such
as self-paced reading, eye-movement monitoring, and priming. According
to Garnham, self-paced techniques are straightforward to set up and the
analyses can be carried out in a much easier way than the other
methods. Eye-monitoring technique has an advantage over self-paced
reading in that the text can be presented in a normal form to the
user. Priming and paper-pencil tests are also mentioned here. He feels
that all these techniques have their own problems and suggests that it
is very secure to use a combination of several techniques in studying
anaphoric expressions of a text than using a single technique. The
author also makes an attempt to study the processing of anaphoric
expressions on a corpus ,collected by himself and Jane Oakhill, over a
period of years from published and publicly "broadcast" material. The
results of the study showed that 'many uses of anaphoric expressions
fall foul of linguistic prescriptions, but do not cause overt
difficulties in comprehension' (p. 63). it was concluded that there
is no straightforward relation between linguistic prescription about
how anaphoric expressions should be used and the way they are
Chapter 6: This chapter focuses on various empirical studies on
anaphora from a psycholinguistic perspective. He turns his attention to
three important aspects to the interpretation of anaphoric expressions
as outlined by him in 1987: 1. identifying the anaphor 2. recognizing
the context under which the anaphor occurs and 3. deriving the meaning
of the anaphor from its semantic content and from the associated
context. Once the anaphoric expression has been identified, it is easy
to fix the appropriate meaning to it based on the context using
linguistic cues. Psycholinguists have observed many factors that
influenze the interpretation of anaphors. These include morphosyntactic
factors (gender, number, case and animacy, and the binding relations of
syntactic theories) Stylistic factors, discourse-based factors (local
and global foci), Clause and sentence boundaries and knowledge-based
(pragmatic) factors. It should be noted that the author talks only
about the interpretation of 'pronominal' anaphors, and that too only
'definite' pronouns and nouns. Abstract anaphors such as 'this, that
and it' may pose challenging issues for constructing appropriate mental
models for resolving the same.
Chapter 7 discusses the two classes of anaphoric expressions (deep and
surface) linguists use. The linguistic theory behind this postulated by
Sag and Hankamer (1984) is presented in detail. They renamed 'surface
anaphors' as 'ellipses' and 'deep anaphors' as 'model-interpretive'
anaphors. The criterion under which these anaphors can be distinguished
from each other is outlined . Garnham presents summaries of some of the
psycholinguistic studies on deep and surface anaphors and also observes
that though the linguistic theory provides a good account of
grammaticality judgement, there is no adequate experimental evidence
to support the processing theory (as laid out by Sag and Hankamer ).
Chapter 8 discusses anaphoric islands. Referring back to Ross (1967) and
Postal (1969), the author provides explanation for the notion of an
anaphoric island. This kind of situation arises only when a definite
pronoun, has a meaning that does not correspond in a straightforward
way to the meaning of another expression in the surrounding text. The
author summarizes works of Postal and other linguists, each emphasizing
morphological or syntactic or semantic or a combination of these
processes to account for the acceptability of the references. Postal
(1969)'s arguments were from a generative semantic point of view and
his claim was that references into anaphoric islands were always
unacceptable, even when the intended interpretation of the anaphor and
the antecedent are morphologically related (cf. this author, p. 112).
However, Corum 1973, Lakoff & Ross 1972 counter-argued Postal's claim.
Furthermore, Lakoff and Ross also argue that the acceptance of an
anaphoric reference relied on the syntactic relation between the
antecedent and the anaphor in addition to their morphological relation.
Browne (1974) showed that reference into an anaphoric island required a
semantic relation between the anaphoric island and the natural
interpretation of the pronoun besides their morphological relation
(contrary to Postal's and Corum's claims, as a part-whole relation
between the meaning of the anaphor and the antecedent trigger). Mohanan
(1986) explained the anaphoric island data within the theories of
lexical morphology and phonology. The author also summarizes many other
works related to this concept (see the chapter for details).
Chapter 9: Garnham presents a neat overview of the notion of implicit
causality and empirical studies on the same. 'Implicity causality' was
first introduced by Garvey and Caramazza (1974) who suggested that
this phenomenon was primary associated with verbs and using sentence
completions they divided the verbs into three categories (p. 124).
Although Garvey et. al (1975) considered, and rejected, the use of
Fillmore's case roles to explain this phenomenon for two reasons (the
effect of social status and the effects of passivisation cannot be
explained by case roles) other authors like Brown and Fish (1983) used
'semantic roles' to describe them. They identified two patterns:
action verbs impute causes to their agents, rather than their patients,
and that mental state verbs impute causes to their stimuli, rather than
to the experiencer (p. 126). Both approaches have their own
limitations. The rest of the chapter summarizes the empirical studies
(including his own studies) related to consistency between implicit and
exlicit causes, the notion of early and late effects of implicit
causality and a section on implicit consequentiality.
Chapter 10: This chapter is about stereotypes in texts. 'There are a
set of nouns that, while not specifying the sex of the person that they
apply to as part of their meaning, more commonly apply to people of one
sex than the other, nouns such as "nurse" and "engineer". Such nouns
can be referred to as stereotyped. (p.140)". Garnham presents the
results of many studies (including his own) on stereotyping whose claim
was to show that, when a text containing a stereotyped noun was
introduced, a following clause containing the pronoun that matches the
stereotype noun is read more quickly than one in which the pronoun does
not match the stereotype.
The final chapter, namely, Chapter 11 reviews some of the themes
associated with the mental models theory of anaphoric interpretations
of text presented in the earlier chapters. It concludes by saying that
mental models are a crucial component of the theory of text
comprehension, and they indeed play a significant role in the
interpretation of anaphoric expressions.
Lastly, the author presents a neat compilation of the references
running from pages 149 - 162, followed by an Author Index and a Subject
There is no doubt that this book is an excellent source of
psycholinguistic research within the context of anaphoric
interpretation of texts. All the chapters are well-written with a good
coverage of a broad range of the topics and empirical studies (if any)
related to the subject matter. Each chapter can be read and understood
independently. The author also provides a short summary at the end of
each chapter. There are no footnotes or endnotes (which of course,
doesn't make the reader to see the end of the page or flip pages to
continue the thread of comprehension). It would have been wonderful if
co-indexing of pronouns with the possible antecedents was made (as
found in most of the works related to anaphoric resolution). This work
focuses only on pronominal anaphors - one or two chapters on the
resolution of abstract entity anaphors (this, that and it) would have
thrown some light on how the theory of mental models can be applied to
the interpretation of texts involving abstract objects. However, this
reviewer feels that this book has a lot to offer and will be of
interest to linguists, psycholinguists, philosophers and cognitive
scientists working in this field of research. This book is also a good
source of information for a lay person - to be more precise, the review
and originality of the works presented here are accessible to those not
already familiar with the subject content, and would definitely prove
to be a stepping stone for further research in this area. It could
also be used as a text book in psycholinguistic courses.
(I list only the references cited under the section 'Book's purpose and
content'. All other references cited in the body of this review can be
found in the book).
Craik, K. 1943. The nature of explanation. Cambridge, UK: Cambridge
Garnham, A., Oakhill, J. V., and Johnson-Laird, P. N. 1982. Referential
continuity and the coherence of discourse. Cognition 1: 29-46.
Garnham, A.1987. Mental Models as Representations of Discourse and Text.
Chichester: Ellis Horwood.
Garnham, A., and Oakhill, J. V.1989. The everyday use of anaphoric
expressions: Implications for the 'Mental Models' theory of text
comprehension. In N.E. Sharkey ed., Modelling Cognition: All Annual
Review of Cognitive Sciences. Vol. 2. Norwood, NJ:Ablex.
Johnson-Laird, P. N. 1986. Conditionals and mental models: In
E. C. Traugott, A. ter Meulen, J. S. Reilly and C. A. Ferguson, eds. On
Conditionals. Cambridge, MA: Harvard University Press.
Johnson-Laird, P. N., and Garnham, B. G. 1980. Descriptions and
discourse models. Linguistics and Philosophy 3:371-393.
Marr, D. 1982. Vision: A Computational Investigation in the Human
Representation of Visual Information. San Francisco: Freeman.
Miller, G.A. , Galanter, E., and Pribram,K. 1960. Plans and the
Structure of Behaviour. New York: Holt, Reihart, and Winston.
Newell, A., and Simon, H. A. 1976. Computer science as empirical inquiry:
Symbols and search. Communications of the ACM 19:113-126.
| ABOUT THE REVIEWER:
ABOUT THE REVIEWER
Dr. Umarani Pappuswamy is a research associate at LRDC, University of
Pittsburgh, USA. She has a Ph.D. in Linguistics with specialization in
computational linguistics. Her main areas of research interests are:
computational semantics - knowledge representation and knowledge reasoning,
discourse analysis, typology - syntax, morphology and semantics, and corpus