dickhudson.comdickhudson.com/wp-content/uploads/2015/11/cogling.d… · web view[cognitive...
TRANSCRIPT
[Cognitive Linguistics (2012, 23, 219-30)]
Richard Hudson. An Introduction to Word Grammar. Cambridge: C ambridge
University Press. 2010, 348 pp. Paperback, ISBN 9780521721646.
Reviewed by Chunshan Xu 1 and Haitao Liu 2
1. Department of Foreign Language, Anhui University of Architecture, China.
E-mail: ⟨[email protected]⟩2. School of International Studies, Zhejiang University, China.
E-mail: ⟨[email protected]⟩An Introduction to Word Grammar, which is Richard Hudson’s most recent
work, proposes a distinctive cognitive linguistic theory integrating modern
cognitive sciences and dependency grammar. This book is arranged into three
parts, respectively devoted to its theoretical foundation in cognitive sciences,
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AM222 Book reviews Cognitive Linguistics 23–1 (2012)
the cognitive interpretations of language in general and of English in p articular.
This book is logically coherent and thoughtfully structured. Each part of this
book serves as the theoretical basis upon which the next part is unfolded while
directions and summaries provided throughout the whole book help readers to
link the corresponding sections in different parts into a unified “network”.
The author takes a cognitive stand in linguistic study, which implies a close
relation between theories of cognitive sciences and theories of language — a
good reason why the first part is entitled How the mind works. Before dwelling
on Word Grammar, the author first introduces the cognitive theories on which
Word Grammar is built, a thoughtful arrangement so far neglected in many
works of cognitive linguistics. It is generally acknowledged in the field of
c ognitive linguistics that language is part of cognition, subjecting itself to
c ognitive mechanism. Hence, cognitive linguistic inquiry should start with the
question — what cognition is. The author’s answer to this question is that cog-
nition includes general knowledge. Thinking can best be understood in terms
of representational structures in the mind and computational procedures that
operate on those structures, which is the central hypothesis of cognitive s cience
(Thagard, 2005). In other words, cognitive sciences concern the acquisition,
presentation and the working mechanism of this knowledge as well as its rela-
tion with non-knowledge experiences. Addressing these questions is a neces-
sary preparation for the establishment of a cognitive linguistic theory because,
Hudson believes, grammar, which is undoubtedly a part of cognition, must be
acquired, presented and used in the same way as general knowledge.
The author holds that general knowledge is a conceptual network, a well-
established concept of numerous cognitive theories that answers two ques-
tions: what knowledge is and how it is presented, or organized. A conceptual
network is composed of nodes — the concepts, and edges — the relations among
nodes. The conceptual network is similar to symbolic-network models such as
hierarchical networks and spreading activation networks. However, it is obvious
at the same time that the author’s model also has a flavor of connectionism, as
can be clearly seen in Hudson’s definition of concept. According to Hudson, a
concept is nothing but its connections with other nodes and therefore, “the
network is firmly ‘anchored’ to external units such as perceptual categories”
(Hudson, 2007: 18), which is a typically distributed point of view. Hudson’s
model borrows a lot from both symbolic network model and distributed
n etwork model. The organization is hierarchical, the working mechanism is
spreading-activation and the concepts are stored and processed in a distributed
and parallel manner. In short, the model of the conceptual network has solid
theoretical foundations in cognitive disciplines.
In a conceptual network, the nodes of concepts are defined by features, or the
relations with other nodes, and these relations ultimately link to non-concept
nodes such as percepts, emotions and motor-skills. Hudson distinguishes be-
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AMBook reviews Cognitive Linguistics 23–1 (2012) 223
tween two types of relations: primitive relations and relational concepts. The
latter are concept nodes linking two other concept nodes via two primitive rela-
tions: argument and value. Since conceptual relations far outnumber the prim-
itive relations, these two primitive relations play vital roles. In a conceptual
network, the ways nodes are connected (organized) constitute the very knowl-
edge. To put it in another way, knowledge and organization of knowledge
c annot be distinctly differentiated. Another vital primitive relation is the Isa
relation, which leads to the hierarchical structure of the conceptual network.
Without this relation, there is no possibility for generalization — the fundamen-
tal operation of cognition, let alone the methodic structure of our knowledge,
which contributes significantly to the high efficiency in information process -
ing. In other words, this primitive relation is vital for the establishment of the
entire knowledge system where taxonomy is omnipresent. Owing to the Isa
relation, inheritance, another important cognitive operation, is possible, which
means that the feature links of a super node can be inherited, via Isa relations,
by its sub nodes. That is, all these features are stored as links to only one super
node, instead of links to each of the sub nodes, which is undoubtedly a great
advantage in terms of efficiency. In brief, since a node is itself nothing but the
converging point of links, knowledge is the structure of the entire conceptual
network.
Then how is the knowledge acquired? — another hot issue in cognitive sci-
ences. In terms of conceptual networks, this question can be reformulated as
following: how a specific network structure is formed. Hudson believes that
it results from the creation of new nodes, the creation of new links, and the
induction, or generalizations, that initially start with nothing but the percepts,
motor skills and feelings. When human beings encounter a new entity, they
build, for this entity, a new node in mind with links for important perceivable
features. Because of the novelty of this exemplar node, human attention will
keep it at a high activation level so that it will not quickly disappear. As a
r esult, next time when the same entity appears, this node can quickly be acti-
vated, transforming into a type node. For example, a node for robin with such
features as its profile, wing, and beak can be established in this way. If more
entities that share some features occur, nodes sharing these features will be
built. A robin, a sparrow and a nightingale all have features such as wing and
beak; therefore, the c orresponding nodes all link to these two features. As the
result of spreading activation, every time one such node is activated, other
nodes, owing to their shared features, are also activated. Repetition brings
about frequency effects, and thus, there forms a “tightly-knit little network
with each of the exemplar nodes linked to each of the properties” (85). This
background activation is circulated ‘off-line’ and can be spotted by the mind to
build a new node connected, via Isa relations, to existing nodes of this little
network. This mechanism is, as we can see, rather similar to the well-known
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AM224 Book reviews Cognitive Linguistics 23–1 (2012)
process of implicit learning. This process can be recycled again and again to
build nodes at different levels, or rather, generalizations at different levels.
First, the concepts for robin, sparrow and nightingale are built through in-
duction, then, on this basis, these nodes are recycled to establish an even
more general concept node. N evertheless, all the concepts are rooted in
concrete and specific experiences, a fundamental view of cognitive linguis-
tics. In other words, a concept, however abstract it may be, comes from
experiences.
Then, Hudson addresses the question of how to use the knowledge. Actually,
how to use knowledge means how to exploit the conceptual network to com-
prehend new experiences, which involves, according to Hudson, node build-
ing, attention, spreading activation, binding and best matching. That is, when a
new entity occurs, a node is built, together with its perceivable features. At the
same time, the activation randomly spreads from the features and finally con -
verges on the node which is currently most activated. Then the new node is
bound to the target node and the features of the target node are inherited. In this
way, we can make use of existing knowledge to understand the present situa-
tion. This process unfolds through the network, in the form of activation. This
model conforms to the physiological structure of the brain, as we know of it so
far.
Here are two noteworthy points. First, different from Hudson’s model
a dvanced in Language Networks: a new word grammar, this model attaches
much importance to attention. In fact, every new node can have numerous
feature links which on the one hand may well overwhelm the processing
c apacity of the brain and, on the other hand, will dramatically reduce the
efficiency of the whole system since not all these features are useful in
information processing. Similarly, since the target node may also link to
numerous other nodes, it is impossible and inefficient that all the features
should be inherited, or, activated. However, in this model, attention, which
serves as a filter allowing only significant input stimuli and output features
when new experiences are processed, is exploited to amend these two
flaws. Apparently, this new model integrates both bottom-up processing
and top-down processing while the one proposed in Language Networks is
a bottom-up one.
Secondly, the notion of ‘choice set’ is introduced into this new model, which
is fairly significant in the operation of overriding and multi-inheritance. Over-
riding is an important mechanism, endowing the conceptual network model
with considerable robustness because binding between a new node and the
target node, owing to this mechanism, doesn’t necessarily entail a 100 percent
matching. Exceptions, virtually unavoidable in all categories, have brought
troubles to classic category theories, which mostly hold a clear-cut definition
as the necessary basis for any concept. However, with the overriding mecha-
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AMBook reviews Cognitive Linguistics 23–1 (2012) 225
nism, exceptions can be satisfactorily accommodated within the framework
of conceptual network, since an exemplar node in a conceptual network is
a llowed to possess features, of course not too many, contradicting those of the
existing node to which it is to bind, as long as the matching between them is
the maximum within the entire network. Hudson calls it the best global candi-
date. In this case, there should be a mechanism to avoid the possible inheri-
tance of conflicting features. The author, to address this issue, puts forward the
concept of choice set: a set of inter-inhibiting alternatives. According to neuron
network theory, the links between nodes can be excitatory or inhibitory con-
nections (Solso et al., 2004: 142). We can regard the links within a choice set
as inhibitory. Rather similar to the famous prototype theory, this model is able
to cope with exceptional cases, and what’s more, it is potentially capable of
providing a computational model to process peripheral category members.
Apart from overriding, choice set also plays a vital role in multiple inheritance,
which is a pervasive phenomenon of one node inheriting features from more
than one super node and hence has the same risk of inheriting conflicting fea -
tures as overriding does.
In the first part, the author presents a detailed introduction of the cognitive
foundation of Word Grammar — conceptual network. The important ideas of
this theory, such as Isa relation, taxonomy, inheritance, frequency e ffect, spread-
ing activation, attention, etc., are well established in cognitive sciences. As has
been mentioned, the model of conceptual network is based upon both the sym-
bolic network and the distributed network (or connectionist network), which
are two important models in cognitive sciences. Of the numerous theories of
cognitive linguistics, not many bear so close a relation with modern cognitive
sciences. In brief, one of the striking features of this theory is its i ntimate rela-
tion with contemporary cognitive sciences, especially cognitive psychology.
The author devotes the entire first part to a convincing interpretation of the
organization, acquisition and employment of knowledge, which serves as the
theoretical preparation for the next topic: language knowledge. Since Word
Grammar is based on cognitive psychology, it is possible to test this theory
with empirical approaches, which is also an intriguing feature of Word Gram-
mar. In addition, such an a rrangement not only exhibits a solid foundation of
cognitive sciences but also paves a smooth way for the readers to go gradually
from general theory to specific topics, another feature still wanting in most
works of cognitive l inguistics.
The second part, entitled How Language Works, accounts for more than half
of the whole book — a reasonable arrangement considering the fact that this
book is concerned with cognitive linguistics. Lakoff (1990) argues that cogni-
tive commitment is a defining feature of cognitive linguistics, which is a com -
mitment to accord one’s account of human language with what is generally
known about the mind and the brain, from other disciplines as well as our own.
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AM226 Book reviews Cognitive Linguistics 23–1 (2012)
In other words, cognitive linguistics holds that language, which is not autono-
mous, subjects itself to the general cognition mechanism of human b eings: a
fundamental disagreement with Generative grammar. Therefore, a cognitive
linguistic theory narrows the questions raised at the very beginning of this
book down to the following: the presentation (organization) of linguistic
knowledge, the acquisition of this knowledge, the use of this knowledge and
the relation between linguistic knowledge and non-linguistic knowledge. Word
Grammar adopts the approaches of dependency grammar, which concerns the
links between words. Apparently, dependency theory is in accordance with the
network idea. Linguistic knowledge, undoubtedly a part of general cognition,
must be a part of the conceptual network. According to Word Grammar, lin-
guistic knowledge consists in the links among the nodes in a language network
and the links between language network and the rest of the human mind. Surely
enough, the nodes in this language network must be concepts. Nevertheless,
these concepts are not the traditional ones that we usually identify as meanings
of words. Hudson points out that a concept in fact is the result of generaliza-
tion. In this sense, with its meaning left out, a word, which is a type in our
mind, must be a concept itself. For example, in this sentence I have a small
dog, he has two big dogs, there are two tokens of the word type have and two
tokens of dog. Both dog and have, as two words in our mind, are clearly inde-
pendent of any text and situation and therefore are the generalizations, or con-
cepts, abstracted from experience.
Now that words themselves are concepts, they may well be treated as nodes
in conceptual networks. Actually, that is why this theory is named Word Gram-
mar. With words treated as nodes and the rest of linguistic knowledge as links,
of which lexical meaning is one, the outline of Word Grammar is drawn. With
dog as an example, Hudson illustrates that the meaning of dog is one of
such features as spelling, word class, pronunciation, etc, which define a word
node. Of course, some of these features do not belong to linguistic knowledge,
a fact revealing the close relation between linguistic nodes and non-linguistic
nodes.
The nodes of the language network are also structured in terms of
t axonomy — the basic structuring principle of conceptual network. In other
words, the numerous word nodes form a hierarchy via Isa relation for the sake
of economy and efficiency. According to Word Grammar, inflections, which
are results of generalization and have their own feature links, also stand as
concepts. Lexemes and inflections are both the sub nodes of word, so words are
divided into lexemes and inflections, which can be further divided into sub -
types, with each level reflecting a certain degree of generalization. In this way,
the important operation of inheritance and overriding can run smoothly in the
language network. For example, books Isa book (a sub node of noun) and
p lural (a sub node of inflection) and thus can multi-inherit their features. Obvi-
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AMBook reviews Cognitive Linguistics 23–1 (2012) 227
ously, linguistic knowledge lies in the links, or rather, the features, of nodes in
this network.
As mentioned above, some of these features such as word-class and syn-
tactic valence are within the scope of linguistic knowledge, i.e. the links
within the language network. At the same time, many features such as the
links to meaning, spelling, pronunciation and speaker (an important factor of
context), etc., indicate the close relations between language network and the
rest of our mind. To recapitulate, Word Grammar implies that there is no
clear-cut division between linguistic knowledge and general knowledge, which
closely relate to each other in the conceptual network. This is a stand entirely
coherent with the fundamental tenet of cognitive linguistics — language is part
of cognition.
Word Grammar presents network analyses in terms of morphology, syntax,
semantics, etc. Of course, from a network point of view, they are not separate
modules but embedded and interrelated in the whole network. The morpho-
logical analysis is mainly unfolded in terms of inflection and derivation, both
of which can be well dealt with from a network perspective. For instance,
walks is linked to walk via Isa relation, while walker is linked to walk via
agent-noun relation, a lexical relation instead of a hierarchical relation. Never-
theless, when it comes to word forms, the two pairs both exhibit a variant
r elation. Some irregular morphological variations are treated as sublexemes in
Word Grammar. Hudson (2007) presented a more detailed analysis of mor-
phology, to which those who are interested in morphology can refer. With
d ependency grammar as the theoretic foundation, Word Grammar reduces syn-
tax to the dependencies among words. In Word Grammar, syntactic rules are
the valences, lying in words. Every word (including inflections) has its own
valence, or, dependency pattern, which is either attached directly to words
themselves as features or inherited indirectly from super nodes. There are
two types of valence: parent-valence and dependent-valence. The grammati-
cality of a sentence then depends on whether valences carried by words in
this sentence can be filled. As Hudson puts it, word tokens are grammatical
if their properties are compatible with all the properties that they should
i nherit from the grammar. Since the valence of noun demands a parent, a word
in a sentence, if it is a noun, should satisfy this valence. Closely related with
valence is another important feature at the syntactic level: the landmark rela-
tion, which decides the word order in a sentence. The cognitive foundation of
this feature is the figure /ground theory advanced by Gestalt psychologists
(Koffka, 1935) and the subsequent landmark theory (Langacker, 2000). In
Word Grammar, the valence feature and landmark feature merge into a com-
plex feature, that is, the landmark feature depends on the valence feature. The
dependency between two words normally fixes their serial order with the
p arent, which is more prominent than the dependent, serving as the landmark.
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AM228 Book reviews Cognitive Linguistics 23–1 (2012)
For example, an o bject-valence link between two words entails a before-
landmark link between them. Certainly, exceptions are possible as the result
of the overriding mechanism. Hudson advances the Best Landmark Principle
to exclude tangled dependencies and chaotic sentential order. In this sense,
Word Grammar seems to be potentially able to provide a formal model as TG
grammar does.
Hudson holds that ‘meaning’, which he divides into social meaning and
r eferential meaning, is a relation. With this as a starting point, Hudson re-
spectively probes, from a network perspective, sense and referent, anaphora,
sentence meaning and social meaning. The network presentation of the differ-
entiation between referent and sense lays the foundation on which anaphora
and sentence meaning are explored. Hudson distinguishes between identity-of-
reference and identity-of-sense anaphora and gives a detailed account of mean-
ing and syntax, that is, how the meanings of words combine to produce the
compositional meaning of a sentence or a phrase.
Word grammar not only presents a network picture of linguistic knowledge
organized in taxonomy, but also gives a network account of the acquisition and
use of linguistic knowledge, which involves again those terms such as activa-
tion, spreading, binding and attention. For instance, learning a new word in-
volves the building of an exemplar node with its feature links, the retention of
high activation due to novelty and attention, and the survival of the new node
until it becomes a type when the same word appears again. Learning syntax is
a little more complicated, as it involves building nodes for syntactic schemas
derived from experiences, an inductive process similar to the one elaborated in
the first part. Apparently, learning syntax is a process of generalization and can
occur with different degree of generalization. With regard to using linguistic
knowledge, inheritance again plays an important role. For example, if one
p erson wants to say cat, activation spreads from the meaning of cat while his
attention keeps the nodes concerning pronunciation at a high activation level.
Consequently, the nodes connected to cat via its pronunciation feature receives
a double dose of activation and become the global best candidate. Hudson
calls this process ‘attention channeled activation’. As can be seen, the work-
ing and learning of linguistic knowledge can be interpreted in a network
manner, totally in accordance with the operation of the conceptual network
in general.
In this part, Hudson illustrates that everything related with language can be
described in terms of nodes and links. Virtually every operation of language,
according to Word Grammar, is a network operation. In addition, treating words
as the central elements of grammar brings one admirable advantage. That is,
with the rich links that a node is allowed, Word Grammar presents the possibil-
ity of dealing with all important branches of synchronic linguistics (such as
phonology, morphology, syntax, semantics, pragmatics and social linguistics)
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AMBook reviews Cognitive Linguistics 23–1 (2012) 229
within one unified theoretical framework — a significant achievement in the
field of theoretical linguistics, especially cognitive linguistics. Different from
other cognitive linguistic theories (Langacker, 2000; Goldberg, 1995; Croft,
2001), which are profoundly influenced by Gestalt psychology, Word Gram -
mar builds itself mainly upon modern cognitive sciences, especially cognitive
psychology, and inevitably, as cognitive psychology does, focuses on the orga-
nization and operating mechanism of linguistic knowledge, rather than abstract
schemes. Still owing to this feature, Word Grammar is potentially able to pro-
cess real language materials and to serve as an important model in such fields
as NLP. Modern computational linguistics attaches increasing importance to
the application of complex features to the models of language processing sys-
tems because researchers have realized the extent to which language opera-
tions depend on multiple factors, linguistic or non-linguistic. Word Grammar
obviously embraces this trend — the feature links of word nodes are by nature
complex features. That is, the multiple links of word nodes can endow sen-
tence parsing with whatever level of granularity as is needed. A sentence parser
can be equipped with different amounts of feature links to reap parsing results
at different granularity levels. Furthermore, the mechanism of overriding and
global best matching can make for a robust sentence parser. For example, the
feature links may well be endowed with different weights according to their
roles in composing an acceptable sentence (of course how to obtain those
weights is still a problem), then, instead of a simple acceptance or rejection, the
parser can compute and output a degree of acceptability by calculating the
weight of those overridden or inherited features. In short, Word G rammar
can be applied to the analysis of real corpora and contribute significantly to
the establishment of computational models in Natural Language Processing,
which is another distinctive feature so far lacking in many cognitive linguistic
theories.
In this part, Hudson draws a network picture of linguistic knowledge, which
is nothing but nodes and links. Though languages may be different, the struc-
turing principles and working mechanism are the same. Therefore, the differ-
ences among languages consist in the differences of nodes and links that
c ompose the language network. At the very beginning of the third part — How
English Works, Hudson describes English linguistics as the study of English
based on the ideas and methods of the ‘general linguistics’ that has pushed
forward our understanding of how language works. This shows that the third
part will be an illustration of how the general theory of Word Grammar is
e mbodied in a specific language like English.
It is reasonable to believe that the operations of the language network such
as activation, binding, inheritance, etc, are the same for different languages,
since the working mechanism must be the same. Owing to this reason, the third
part, which is the shortest one in this book, addresses only two major issues:
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AM230 Book reviews Cognitive Linguistics 23–1 (2012)
English words and English syntax. The acquisition and use of English is not
touched because the difference among languages mainly lies in the nodes and
links, not the working mechanism.
As explained above, the Isa relation makes a unique contribution to the
t axonomic structure and the ensuing high efficiency of a conceptual network.
Word Grammar centers on words, so the taxonomy of words comes as the first
topic. Word Grammar categorizes English words into noun, verb, adjective,
preposition, adverb and conjunction. Determiners constitute a subtype of pro-
noun, which is itself a subtype of noun b ecause determiners such as any and
this share features with pronouns, which themselves in turn resembles nouns in
some aspects. The overriding mechanism can well cope with the existing con-
flicting features among determiners, pronouns and nouns. Besides lexemes,
inflections are also treated as concepts. For example, in English, finite is
a node, of which verb inflection is the super node and tensed verb, the sub
node.
As concept nodes, word classes are defined by their feature links, which can
be inherited by their sub nodes, a mechanism contributing to high efficiency.
Hudson offers some ‘membership tests’ to decide the word class of a word.
These tests rely on the features of a word class such as tense, number and
v alence. Surely enough, exceptions are inevitable, and as we all know, linguis-
tic concepts like word class are generally fuzzy, a characteristic of not only
language but also human cognition. However, within the framework of Word
Grammar, the unavoidable exceptions can be reasonably construed as the
r esult of overriding.
One unique feature of Word Grammar is the integration of linguistic and
non-linguistic factors, which is clearly illustrated in the section entitled social
properties. The author raises two questions: “how do English words convey
social information?” and “what kinds of social information do English words
convey?” (277). According to Word Grammar, the links to meaning, pronun-
ciation and valence can all convey social information, but the ways these mes-
sages are linked to nodes are yet to be explored. As to the second question, the
author mentions formality but acknowledges that there is still much work to do
in this field.
On the syntactic level, one feature of English is, as Hudson points out, that
it is a head-medial language, which leads to the differentiation between pre-
dependents and post-dependents: two concepts devoid in many languages. To-
gether with adjuncts and valents, they give rise to pre-adjuncts, post-adjuncts,
subjects and complements.
Valences fall into parent-valences and dependent valences. In English, not
only finite verbs can be free from a parent-valence, but also some expressions
like how about, what if, etc. In Word Grammar, they are treated as special
s ublexemes with distinctive valences. Dependent valence is another important
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AMBook reviews Cognitive Linguistics 23–1 (2012) 231
feature at the syntactic level. With the analysis of their dependent valences,
a close resemblance can be found between English prepositions and ‘subordi-
nating conjunctions’. Possessive ‘‘s’ is seen as a word in Word Grammar
b ecause it has its own dependent valences as other independent words have.
According to Word Grammar, there are seven types of dependent valences:
pre- adjunct, post-adjunct, subject, complement, (direct) object, indirect object
and predicative.
Word Grammar treats the number agreement also as a feature link. For a
noun, the default number feature is singular which can be overridden by a sub
node whose number feature is plural — a technicality adopting the contrast
b etween ‘markedness’ and ‘unmarkedness’.
In short, this part, instead of discussing human cognitions, focuses on lan-
guage knowledge itself because English, just like any other language, are pre-
sented in our brain in terms of nodes and links. The differences simply lie in
what nodes are built and how they are linked. However, the ways in which this
network organizes itself and operates is no different from any other language
network.
This book presents a unique theory of cognitive linguistics that is solidly
based on cognitive sciences, especially cognitive psychology, and is poten-
tially able to subject itself to empirical tests. This theory is presented as a for-
mal system with such rigour that it can be used to analyze real language usage.
Considering these two features, Word Grammar as presented in this book is a
testable cognitive grammar. What’s more, as a textbook, this book is very
thoughtfully structured. Readers can go through the whole book in a linear
order as they usually do. Nevertheless, considering the correspondences among
the sections in different parts, the author provides a ‘lateral approach’ to read-
ing, linking the corresponding sections in all three parts together. At the begin-
ning and the end of a section, the author explicitly directs the readers to
c orresponding sections in others parts, which helps readers understand “how
various general ideas from cognitive science apply to language and explain its
characteristics”. In this book, the author even provides different reading links
to readers with different proficiency in linguistics. Moreover, the author gives
summaries at the end of each chapter and most sections, and these summaries
will be reintroduced at the beginning of the corresponding chapters in the
f ollowing parts. This is a considerate help for students to grasp the important
points after learning each chapter and to review them before learning another
corresponding one.
In brief, this book not only presents a testable cognitive grammar with a
solid foundation in cognitive sciences, which opens a new window for those
interested in cognitive and general linguistics, but also, owing to its reasonable
structural organization, serves as a quality textbook for both students and
teachers of linguistics.
Brought to you by | UCL - University College London
Authenticated
Download Date | 11/23/15 10:04 AM232 Book reviews Cognitive Linguistics 23–1 (2012)
References
Croft, W., 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective.
Oxford: Oxford University Press.
Goldberg, A., 1995. Constructions: A Construction Grammar Approach to Argument Structure.
Chicago: University of Chicago Press.
Hudson, R., 2007. Language Networks: a new word grammar. Cambridge: Cambridge University
Press.
Koffka, K., 1935. Principles of Gestalt Psychology. London: Lund Humphries.
Langacker, R. W., 2000. Grammar and Conceptualization. Berlin: Monton de Gruyter.
Lakoff, G. 1990. The invariance hypothesis: Is abstract reason based on image schema? Cognitive
Linguistics 1(1): 39–74.
Solso, R. L., MacLin, O. H., MacLin, M. K., 2005. Cognitive Psychology, 7th edition. Beijing:
Peking University Press.
Thagard, P., 2005. Mind: Introduction to Cogntive Science. Cambridge. MA: MIT Press.