Townsend, David J., and Thomas G. Bever (2001) Sentence Comprehension: The
Integration of Habits and Rules. MIT Press, 445pages, $24.95, paperback
Stanka A. Fitneva, Cornell University
[For another review of this book, see
Why are words better perceived in sentences than in jumbled sequences?
Townsend and Bever's response is the following: The perceptual salience of
words in sentences derives from matching a temporary "pseudosyntactic"
meaning-form structure and a proper grammatical structure. Imagine looking
up your tentative answer to a crossword puzzle question in the answer key.
The epiphany "I got it!" I believe is akin to what Townsend and Bever
suggest is happening in sentence perception and what results in the
perceptual salience of words in sentences. Their solution to the word
salience puzzle - the Late Assignment of Syntax Theory- integrates
symbolic computations and behavioral associations in sentence
After the introductory Chapter 1, Chapter 2 presents an argument for the
fundamental role of the sentence in comprehension. The chapter charts the
history of the problem motivating the development of the Late Assignment
of Syntax Theory: the mysterious perceptual effects of sentences. In a
delightful excursion through the history of reasoning about the sentence,
the authors visit Wundt, Bloomfield, Osgood to reach the experimental
results of George Miller and colleagues from the 50s and 60s. The puzzle
Miller's research presented to researchers foregrounded the role of
abstract linguistic knowledge in speech perception. The authors review the
rise and waning of various theories, e.g., theories postulating a direct
link between perception and transformational rules and the development of
alternative grammatical theories.
Extensive but parenthetical, Chapter 3 provides an overview of grammatical
theory for psychologists and others not in the midst of syntactic
theorizing and sentence processing work. The chapter provides a useful
overview of minimalist syntax. Most of the evidence in favor of LAST
reviewed in the following chapters however predates minimalism and it is
not critical for understanding the proposal. Importantly, there is a
convergence between LAST's and minimalist assumptions about the input to
grammatical derivations. (It also voices the frustration of
psycholinguists with the overnight changes of syntactic theories, some
versions more, some less related and concerned with overt behavior.)
Extensive and less optional is Chapter 4: an overview of theories of
sentence comprehension from the last 20 years. The authors discuss
structural theories emphasizing the independence of syntax in
comprehension (Parcifal, Minimal attachment , Construal, etc.). Next is
the set of statistical theories (competition, constraint satisfaction,
dynamical systems, and hybrid models), the discussion of which begins with
foregrounding the similarities and differences between associationism and
connectionism. The chapter concludes with an assessment of the two types
of theories with regards to identifying the sentential unit of analysis.
The implications that the authors draw are that an adequate theory of
sentence comprehension should take into account frequency of argument
structure, that lexical frequency information by itself is not enough to
define a sentence, and that the grammar defines what is an acceptable
sequence of words.
Responding to these adequacy constraints, Chapter 5 outlines the
analysis-by-synthesis, or Late Assignment of Syntax Theory of sentence
comprehension (LAST). The model proposes that comprehension starts with
bottom-up processes, called 'pseudosyntax', establishing an initial
conceptual and functional structure of the input. Pseudosyntax operates on
superficial grammatical cues, e.g., morphology, and applies form and
meaning templates to the input (e.g., NVN and agent-action-patient). The
initial meaning-form hypothesis is the input to a grammatical derivation
that results in the surface syntactic form. This form is matched against
the sequential input. The existence of two surface representations of the
input, one syntactically derived and one perceptual, explains for the
authors the perceptual salience of words in sentences.
Some basic evidence for the model (Chapter 6) consists in early access to
function words and access to meaning before access to syntactic structure.
Chapter 7 is a detailed presentation of the research on the processing of
reduced relative clause ambiguity (e.g., "The horse raced past the barn
fell."). The empirical research on this construction serves to illustrate
the application of canonical templates, such as NVN, to the sequential
input, a central feature of the pseudosyntax idea. Chapter 8 elaborates on
other implications of the model, e.g., modularity (suggesting boundaries
between associative and rule-based computational domains) and the
importance of proposition boundaries. An important disclaimer found in
this chapter is that LAST comprehension processes apply incrementally, not
at sentence boundaries as most of the text might make the reader think.
Chapters 9 recruits further evidence for the model from language
acquisition and neurological disorders touching on the interface between
language learning and language processing. Chapter 10 opens up the
discussion to broader problems in the cognitive sciences, e.g.,
consciousness and reality and the grain problem in inductive learning.
This book synthesizes an enormous amount of research and shows how LAST
can explain the constellation of empirical results. The effort to reach an
audience outside the sentence-processing school is pervasive. To help
readers follow the discussion and arguments, the authors use text boxes to
describe experimental methodology. The experimental evidence is described
with detail and care allowing non-experts to get into the material.
Sections of the book can certainly be used as teaching materials, e.g.,
the discussion of contemporary models of sentence processing in Chapter 4
and the detailed summary of research on reduced relative clause ambiguity
in Chapter 7. On the other hand, for anybody involved in psycholinguistic
research, the detail might seem somewhat excessive.
The authors mention that several of their theoretical choices might be
unsavory for either the connectionist or the symbol manipulation schools.
Critical then is the question: If a researcher adopted this theoretical
framework, what would his/her research agenda look like? The pointers from
the text seem to suggest a focus on the points of integration of the
information coming from the two computational streams, rule- and
habit-based, and on the different speed of information processing and
integration at different points of the sentence. Being persuaded of the
theoretical originality and fruitfulness of such an agenda is the key for
this theory taking off.
About the reviewer:
Stanka Fitneva is a graduate student at Cornell University studying
sentence processing and evidentiality.