The Oxford Handbook of Cognitive Linguistics Part 48 pps

content, residing instead in certain basic cognitive abilities. A thing is any product
of grouping and reification. The grouping of constitutive entities is ubiquitous in
perception and cognition, effected (often quite automatically) on the basis of
similarity or contiguity. In the following display, for instance, we automatically
perceive four groups of three xs, not just twelve xs: [xxx xxx xxx xxx]. By reification
is meant the manipulation of a group as a unitary entity for higher-level cognitive
purposes. In the preceding display, the groups are reified when each is treated as a
single entity for the purpose of counting, so that four higher-order things are
observed. Many nouns profile things obviously formed from constitutive entities
in this fashion: group, set, stack, team, alphabet, orchestra, collection , and so on.
Physical objects, the category prototype, are precisely the case where grouping and
reification are too low-level and automatic to be consciously accessible.
31
A group results from conceptualizing entities together, as part of a single mental
experience. A mental operation that brings entities together, an assessment that
interconnects them in some fashion, constitutes the conception of a relationship.
32
While some relationships obtain at a single instant, others evolve (and typically
change) through time. A temporally evolving relationship is most naturally ac-
cessed by sequential scanning, in which the component states (the relationships
obtaining at successive points in time) are serially accessed, as in viewing a motion
picture. A relationship sequentially scanned through time is called a process, used
in Cognitive Grammar for the schematic definition of verbs. Verbs have a high
degree of temporality, for sequential scanning reinforces the profiled relationship’s
temporal evolution and thereby makes it salient. Relationships with a lesser degree
of temporality are said to be atemporal.
33
A relationship can be atemporal either
because it consists of just a single state (and can thus obtain at a single instant) or
because its temporal extension is viewed in summary fashion, with all its compo-
nent states being simultaneously active and accessible, as in a multiple-exposure

photograph. Adjectives, adverbs, and adpositions can be characterized as profiling
different sorts of atemporal relationships. Participles and infinitives are derived
from a verb by imposing a summary view on the process it designates, producing
either an atemporal relation or (with reification) an abstract thing.
An expression belongs to a particular grammatical class by virtue of instan-
tiating the schema describing it. Admire is thus a verb, and admirer a noun,
because—as determined by their highest-level profiles—they respectively instan-
tiate the verb and noun schemas: [process/ ] ? [admire/admire]; [thing/
] ? [admirer/admirer].
34
Like class schemas, grammatical markers are highly
schematic at the semantic pole, but phonologically they have specific content.
Semantically, for instance, the auxiliary verb do is equivalent to the verb class
schema [process/do]. When do combines with a full verb, as in They do admire
her, the schematic process profiled by do is equated with the specific process
designated by admire, hence the same process is symbolized twice. Being a sche-
matic verb, do can also function as a clausal pro form: They do. The derivational
suffix -er likewise evokes a schematic process, but only as its base; its profile is a
thing (prototypically identified as the more active participant in this process),
440 ronald w. langacker
which makes it a schematic noun. It derives a specific noun from a verb stem (e.g.,
admire) by imposing its profile on the specific process designated by the latter.
Grammar consists of combinatory patterns for assembling symbolically com-
plex expressions out of simpler ones. The traditional distinction between mor-
phology and syntax is just a matter of whether or not the expression formed is
larger than a word (e.g., admirer vs. do admire). There is otherwise no sharp dis-
tinction between them, and the same basic principles apply to both. A particular
complex expression consists of an assembly of symbolic structures, each phono-
logically specific. The constructional schemas describing their formation consist of
symbolic assemblies where some or all of the structures are both semantically and

phonologically schematic. Constructional schemas categorize (and are immanent
in) instantiating expressions, just as class schemas are.
Whether specific or schematic, symbolic structures are connected—and
thereby form assemblies—by correspondences and relationships of categorization. A
specific example, sketched in figure 17.9, is the nominal expression the table near
the door.
35
Correspondences are given as dotted lines. They indicate how symbolic
structures conceptually overlap by invoking entities construed as being the same.
The arrows for elaboration and extension (solid and dashed, respectively) indicate
that certain symbolic structures (or substructures thereof) are fully or partially
immanent in others and thus contribute to their emergence. In particular, what is
traditionally thought of as semantic and grammatical ‘‘composition’’ is viewed in
Cognitive Grammar as a matter of categorization. Two levels of composition are
shown in figure 17.9. At the ‘‘lower’’ level, two component structures, near and
the door, categorize the composite structure, near the door. At the ‘‘higher’’ level, the
component structures the table and near the door categorize the overall composite
structure, the table near the door. Observe that near is schematic with respect to
near the door, and the table with respect to the table near the door. On the other
hand, near the door constitutes an extension vis-a
`
-vis the door, and the table near the
door vis-a
`
-vis near the door, owing to discrepancies in the nature of their profiles.
At a given level of organization, ‘‘horizontal’’ correspondence lines specify
which facets of the component structures conceptually overlap and thus project to
the same substructure at the composite structure level. Here the landmark of near
corresponds to the profile of the door, which ‘‘unify’’ to form the composite con-
ception. At the higher level, the trajector of near the door corresponds to the profile

of the table. It is typical for one component structure to contain a schematic
element which corresponds to the profile of the other component and which is
elaborated by this component. This schematic substructure is called an elaboration
site (e-site), marked by hatching. The horizontal arrows thus indicate that the door
elaborates the schematic landmark of near, and the table the schematic trajector of
near the door. It is also typical for one component structure to impose its own
profile at the composite structure level. Thus, near contributes its profile to near
the door (which profiles the relationship of proximity, not the door), and the table
to the table near the door (which profiles the table). Called the profile determinant,
the prevailing component is marked with a heavy-line box.
cognitive grammar 441
Symbolic assemblies exhibit constituency when a composite structure (e.g., near
the door in figure 17.9) also functions as component structure at another level of
organization. In Cognitive Grammar, however, grammatical constituency is seen as
being variable, nonessential, and nonfundamental.
36
An expression can have the same
composite structure and the same grammatical relationships, with alternate orders of
composition (or even a totally ‘‘flat’’ structure). The information essential to gram-
mar does not reside in constituency but in the semantic characterizations of symbolic
structures and how these relate to one another. A structure’s grammatical class is
inherently specified by the nature of its profile. Various other aspects of grammatical
organization inhere in relationships of correspondence and categorization.
The head at a given level of organization is the profile determinant at that
level, that is, the component structure whose profile corresponds to the compos-
ite structure profile. In figure 17.9, near is thus the head in near the door,andthe table
in the table near the door. The distinction between a complement and a modifier hinges
on direction of elaboration vis-a
`
-vis the head. A complement elaborates asalient

e-site within the head. Hence the door is a complement of near. On the other h and, a
modifier contains a salient e-site elaborated by the head. Near the door is thus a
modifier with respect to the table. The grammatical notions subject and object are
characterized in terms of the semantic constructs trajector and landmark. A subject
or an object is a nominal expression whose profile corresponds, respectively, to the
trajector or to the landmark of a profiled relationship. The door is thus the object of
the preposition near in figure 17.9. And though it is not traditionally referred to as a
subject, by this general definition the table bears a subject relation to near the door.
37
Figure 17.9. Constructions
442 ronald w. langacker
A symbolic assembly is categorized by any number of constructional schemas
corresponding to various facets of it. Representing conventional patterns of the
language, the active schemas serve as templates guiding the formation of specific
assemblies, in which they are fully or partially immanent. Among those immanent
in the assembly of figure 17.9, for instance, are schemas for the prepositional phrase
construction and for the construction in which a prepositional phrase modifies a
head noun. Moreover, since the latter schema makes reference to a prepositional
phrase, it incorporates the former as one component. To the extent that complex
structural configurations recur, they give rise to constructional schemas of any size,
incorporating any number of smaller ones as substructures. Additionally, any
particular type of configuration—such as a prepositional phrase or a noun plus
modifier combination—is characterized not just by a single schema, but a family of
schemas representing constructional variants as well as generalizations at different
levels of abstraction. Like other aspects of linguistic organization, grammatical
constructions form complex categories, usually centered on a prototype. They can
be modeled as networks, where each node is an entire constructional schema.
Networks of constructional schemas provide the information that determines
the distribution of linguistic forms, that is, which elements conventionally appear
in particular constructions or constructional variants. Representing varying de-

grees of abstraction from usage events, the schemas in a network range from highly
specific structures, incorporating particular lexical items, to the maximally sche-
matic characterization of fully general patterns. These schemas also vary in their
degree of entrenchment and ease of activation for the categorization of new ex-
pressions. If high-level schemas are readily accessible, constructions are fully
productive. More commonly, it is lower-level schemas that are activated, so that
only the limited range of expressions they subsume (out of all those that would be
sanctioned by the highest level schema) are judged acceptable. Very often the
critical information resides at the lexical level. The specification that a particular
lexical item in a given construction takes the form of a low-level constructional
schema which incorporates that lexical item in the appropriate structural posi-
tion.
38
Even the most idiosyncratic distributional information can thus be cap-
tured in properly configured networks of symbolic assemblies.
5. Phonology

In principle, Cognitive Grammar embraces phonology to the same extent as any
other facet of linguistic structure. To date, however, there have been few attempts
to articulate the framework’s phonological pole or apply it descriptively. The the-
oretical discussion in Langacker (1987a, 1988b, 2000) is at best programmatic. More
substantive treatments of particular problems include Farrell’s (1990) usage-based
cognitive grammar 443
account of Spanish stress, Rubba’s (1993) description of discontinuous morphol-
ogy in modern Aramaic, F. Kumashiro’s (2000) examination of phonotactics (in
comparison to the Optimality Theory approach), and Tuggy’s (2003) analysis of
reduplication in Nahuatl. To these, of course, must be added certain substantial
and basically compatible research initiatives in the broader context of Cognitive
Linguistics (Nathan, this volume, chapter 23).
A fundamental Cognitive Grammar notion is that the semantic and phono-

logical poles of language show extensive parallelism, modulo the inherent differ-
ences between conceptualization and expression. Each pole is resolvable into a
number of separate yet interacting channels (figure 17.2). At each pole, moreover, a
distinction needs to be made between two sorts of elements and configurations:
those intrinsic to that pole (hence ‘‘unipolar’’) and those determined by symbolic
relationships (hence ‘‘bipolar’’). Certain parallels can also be observed in the kinds
of notions required for characterizing semantic and phonological structures. Be-
yond this, the strictures and basic theoretical concepts devised for language in gen-
eral are equally applicable to phonology (e.g., the content requirement, the usage-
based conception, complex categories centered on prototypes, rules as schemas).
Coexisting at either pole are two different organizational schemes, each with its
own raison d’e
ˆ
tre. Factors internal to each pole determine its unipolar organization.
At the phonological pole, some structures delimited in unipolar terms (represent-
ing various dimensions and levels of organization) are segments, syllables, words,
moras, and feet. At the semantic pole, unipolar organization can be identified with
conceptual structure viewed in its own terms, that is, independently of linguistic
expression. By contrast, bipolar organization is that imposed by symbolic structures
and assemblies. Lexical items, for instance, reside in pairings of phonological and
conceptual structures that do not necessarily coincide with structures naturally
delimited on unipolar grounds. It is precisely by virtue of participating in symbolic
relationships that their phonological and semantic representations are recognized
as linguistically significant elements.
39
More generally, each pole of a symbolic as-
sembly consists of phonological or conceptual structures delimited in bipolar terms.
The semantic and phonological structures directly referred to in grammar are those
reflecting bipolar organization. Phonology, on the other hand, is primarily concerned
with unipolar phonological organization

In the same way that semantic and grammatical analysis depends on the char-
acterization of conceptual structure, phonological analysis depends on the char-
acterization of sound structure. Although Cognitive Grammar does not yet make
any specific proposals in this regard, its basic philosophy dictates an approach that
straightforwardly reflects the phenomenon’s intrinsic nature. In this respect, some
version of ‘‘feature geometry’’ is closer to the mark than the representations of
classical generative phonology based on unordered sets of binary features. Certain
analogies (hopefully not too far-fetched) can be suggested between the descrip-
tions of sounds and meanings. For a sound, the action of each articulator can be
thought of as an ‘‘articulatory domain,’’ analogous to the cognitive domains
evoked by a lexical item as the basis for its meaning.
40
Like the domains of a matrix,
444 ronald w. langacker
articulatory domains are not disjointed but related in complex ways (e.g., voicing
in the larynx creates the conditions for the shaping of vowel quality by other
articulators). If articulatory domains provide phonological ‘‘content,’’ is there
something analogous to construal? At least one option seems not implausible: that
accentual prominence in a word is comparable to conceptual prominence, in par-
ticular profiling.
41
More secure, perhaps, is an analogy based on autonomy versus
dependence (Langacker 1987a: section 8 .3). A vowel is phonologically autonomous,
in the sense that it does not require the support of other sounds for its full
manifestation, whereas a consonant is phonologically dependent on a vowel. In the
same way, a thing is often conceptually autonomous, in that we can conceptualize
it independently of any relationship involving it, but since a relationship can-
not be conceived without invoking its participants, it is conceptually dependent on
them.
As with lexicon and grammar, phonology is approached in Cognitive Grammar

from a usage-based perspective (Langacker 1988b, 2000; Bybee 1994). It is subject to
the content requirement, which limits the structures one can posit to those occurring
in actual expressions and those derivable from these by means of schematization and
categorization. Like lexical and grammatical units, therefore, phonological units are
abstracted from usage events by the reinforcement of recurring commonalities. A
multitude of units are thus abstracted, of different sizes and representing various
levels and dimensions of schematization. Through relationships of categorization,
they are organized in complex categories centered on prototypes.
For example, a phoneme can be viewed as a complex category and modeled as
a network (Nathan 1986). Lower-level nodes in the network represent allophones,
the variants occurring in particular phonological contexts; the contextually least
restricted variant is the prototype (‘‘basic allophone’’).
42
Higher-level nodes rep-
resent further abstractions capturing whatever is common to different sets of
allophones. Various other notions of phonological theory correspond to still greater
degrees of schematization. A natural class of segments is defined by an abstracted
segment that is specific only in regard to certain properties, schematic in regard to
others. Corresponding to phonological features are abstracted segments that are
specific in regard to just a single property. A tier can be characterized as a pho-
nological sequence that is schematic except for properties of a certain sort.
The abstraction of phonological units from usage events is not limited to
structures of any particular size or nature. Among the units abstracted are schemas
representing syllables, words, prosodic patterns, and intonation contours. As with
segments, these are organized in complex categories comprising variants as well
as the generalizations emerging at higher levels of schematicity. Inherent in these
networks of abstracted units is a specification of phonological distribution and
well-formedness. For example, highly schematic representations of syllables specify
conventionally sanctioned syllable types (e.g., [CV], [CVC], [CCVC]), while more
specific schemas (e.g., [strV ], [pLV ], [ VNT]) enumerate permissible

consonant clusters. The phonotactic patterns of a language are thus embodied in
schemas for phonologically complex structures. As part of a dynamic processing
cognitive grammar 445
system, such units function as templates (routinized packets of processing activity)
with varying degrees of accessibility for the categorization of new expressions.
In Cognitive Grammar, linguistic ‘‘rules’’ are simply schemas. Phonotactic
rules are readily seen as schematic templates for the structures concerned. What
about phonological rules classically conceived as operations deriving superficial
forms from underlying representations? To the extent that these are justified, they
constitute patterns of phonological extension.
43
These patterns are simply sche-
matized representations of categorizing relationships. They conform to the con-
tent requirement because they emerge from actually occurring phonological
structures through the general processes of categorization and schematization.
Suppose, for instance, that a number of words ending in [d] start to manifest a
variant pronunciation in which the final stop is devoiced. For each such word,
usage events with [t] will be categorized as extensions from the established unit with
[d]: ([Xd#]
"
(Xt#)), ([Yd#]
"
(Yt#)), ([Zd#]
"
(Zt#)). Eventually, as shown in
figure 17.5, these recurring categorizations can themselves become entrenched
as conventional units, characterized at whatever level of schematization the data
permits: [[ d#]
"
[ t#]]. The structures related by these patterns of exten-

sion may be purely phonological, but they can also incorporate information about
symbolic relationships (e.g., the specification that [d] is suffixal), making them
equivalent to morphophonemic rules. Chains of extensions give the effect of rule
ordering.
In a symbolic assembly (e.g., figure 17.9), each component and composite
structure has both a semantic and a phonological pole. Internally, each symbolic
structure’s phonological pole manifests the unipolar phonological organization
of the language in question: its elements, their combination, and the patterns
they instantiate. External factors—the delimitation of phonological poles (by the
very fact of their symbolizing function) and their relationships to one another—
constitute bipolar phonological organization. In bipolar terms, the relation be-
tween component and composite phonological structures, or phonological com-
position, is the counterpart of semantic composition. Moreover, the phonological
composition at each level serves to symbolize the semantic composition at that
level. This is part of what it means to say that grammar is inherently symbolic.
Neither the phonological structures defined on a bipolar basis, nor the manner
of their combination, need be natural from the standpoint of unipolar organiza-
tion. It is for bipolar reasons that picnics is segmented into picnic and -s, whereas
internally—on unipolar grounds—its basic components are pic and nics. Note
further that in bipolar terms -s combines with picnic as a whole, while in unipolar
terms it is incorporated in the second syllable as part of its coda. More drastically,
when -s combines with the entire compound picnic table, its unipolar placement is
in the coda of the final syllable of the last word: picnic tables. There is no ‘‘mis-
match’’ here, for there is no reason to expect unipolar and bipolar organization to
coincide in the first place. In bipolar phonological composition, the composite
structure need only be some function of the component structures, with no re-
quirement that their combination be isomorphic to unipolar composition. Hence,
446 ronald w. langacker
there are numerous ways to form a composite phonological structure other than
simply juxtaposing two components. As a case in point, it is quite unproblematic

for one component structure to be placed inside the other.
44
6. Coverage

In assessing the empirical coverage of any linguistic theory, expectations have to be
realistic. The collective and cumulative efforts of all linguists, of all theoretical
persuasions, have come nowhere close to providing even minimally adequate
documentation of the world’s languages. For no single language have such efforts
provided anything even remotely approximating an exhaustive description. Nor
has any expression or phenomenon of any language been blessed with a full and
definitive description that all linguists would acknowledge as such. Even the best-
described and best-understood phenomena are subject to finer-grained description
and characterization at deeper levels of understanding.
The limitations are of course even more severe when considering a single theory
of fairly recent vintage and initially pursued by just a small group of scholars.
45
Of
necessity, therefore, research in Cognitive Grammar has followed a global strategy
with two main objectives. The first has been to establish the framework’s viability
(if not superiority) with respect to particular topics generally considered theoreti-
cally significant. To this end, a point has been made to tackle certain classic problems
dealt with in Generative Grammar, especially those supposedly demonstrating the
autonomy of syntax. Among such problems are passives (Langacker 1982), con-
straints on pronominal anaphora (van Hoek 1995, 1997), complementation (Achard
1998), so-called ‘‘raising’’ constructions (Langacker 1995c), and positive/negative
polarity items (Israel 1996a, 1998). A special effort has been made to specify the
meanings of grammatical formatives commonly taken as being semantically empty,
such as of (Langacker 1992b), markers for case (Smith 1987) and gender (Langacker
1988a), and all the elements of the English auxiliary (Langacker 1991).
The second objective is to show that these descriptive successes are not due to

a selective choice of topics, but that the theory can in principle be applied with equal
success to any phenomenon in any language. Here one can point to the framework’s
inherent generality and flexibility (and hopefully its avoidance of blatant language
bias). Still, there is no substitute for actually applying it to a large and representa-
tive sample of languages and linguistic phenomena. Cognitive Grammar is indeed
being applied to more and more languages of diverse types and genetic affiliations.
Branches of Indo-European where substantial work has been done include Romance
(Maldonado 1988, 1992;Vandeloise1991;DoizBienzobas1995; Achard 1996;Farrell
1998), Germanic (Smith 1987, 1993, 2001;Cornelis1997; Enger and Nesset 1999;
Mortelmans 1999), Slavic (Janda 1986, 1993;Cienki1995;Da˛browska 1997;Nesset
cognitive grammar 447
1998), and modern Greek (Manney 1995, 2000). Among the growing array of non-
Indo-European languages examined from a Cognitive Grammar perspective are
Basque (Doiz Bienzobas 1998), Finnish (Huumo 1998), Estonian (Vainik 1995),
modern Aramaic (Rubba 1993), Mandarin (Poteet 1987;Hsiao1991; Shen 1996);
Japanese (T. Kumashiro 1994, 2000; Matsumoto 1996; F. Kumashiro 2000;Nomura
2000), Korean (Lee 1999), Samoan (Cook 1988, 1993a, 1993b), and a variet y of native
languages of the Americas (Tuggy 1981, 1986, 1988, 1992, 2003;Casad1982;Beck1996;
Vela
´
zquez-Castillo 1996; Ogawa and Palmer 1999).
The range of linguistic phenomena investigated from a Cognitive Grammar
standpoint is likewise broad and steadily growing. On this front, a two-pronged
strategy has been pursued. On the one hand, an attempt has been made to examine a
wide spectrum of basic problems in preliminary terms. Here the objective is to show
that the framework can in principle accommodate them, and also to provide an
initial indication of what a Cognitive Grammar description might look like. An ex-
ample is Langacker (1991), which considers in turn many basic aspects of nominal
structure, clause structure, and complex sentences, primarily with respect to English,
but with numerous references and comparisons to other languages. The papers in

Casad and Palmer (2003) cover varied phenomena in non-Indo-European lan-
guages. The second prong of the strategy, complementary and clearly necessary, is to
investigate particular phenomena in great depth and detail. These in-depth probes
have thus far included such varied topics as Cora locatives (Casad 1982), French
complementation (Achard 1998), Samoan clause structure (Cook 1988), English
nominalization (Heyvaert 2003) and noun-noun compounds (Ryder 1994), dative
case in Polish (Da˛browska 1997), verb conjugation in Russian (Nesset 1998), middle
voice in Spanish (Maldonado 1992) and modern Greek (Manney 2000), as well as
double subject constructions (T. Kumashiro 2000; Kumashiro and Langacker 2003)
and internally headed relative clauses (Nomura 2000) in Japanese.
Unavoidably, coverage of the many facets of linguistic structure has been quite
uneven. While the widespread notion that Cognitive Grammar deals primarily
with locative expressions is completely erroneous, this has received its share of
attention (e.g., Lindner 1981, 1982; Casad 1982; Hawkins 1984; Taylor 1988; Van-
deloise 1991; Cuyckens 1995; Langacker 2002d). Quite a lot has been done on
nominal and clausal grounding, whereby a profiled thing or process is related to the
speech situation (Langacker 1991, 1997c, 1999d, 2001c, 2002b, 2002c, 2003c, 2003d,
2004a, 2004d; Mortelmans 1999; Brisard 2002). Possessive constructions, which
have a grounding function, have been dealt with extensively (Tuggy 1980; Lan-
gacker 1993c, 1995b, 2001d; Taylor 1994, 1996; Cienki 1995; Vela
´
zquez-Castillo 1996).
Overall, the greatest concentration of effort may well have been in the general area
of clause structure (Tuggy 1988; Langacker 1991, 1993a, 2001d; Cook 1993b; Smith
1993; T. Kumashiro 2000), including such topics as transitivity (Tuggy 1981; Rice
1987a, 1987 b; Cook 1988), voice (Langacker 1982, 2004c, forthcoming; Maldonado
1988, 1992; Manney 1995, 2000; Cornelis 1997), and the semantics of case markers
(Smith 1987; Cook 1993a; Janda 1993; T. Kumashiro 1994).
448 ronald w. langacker
The list of topics that have not been sufficiently investigated in Cognitive

Grammar can be as long as one cares to make it. Little has been done, for example,
on adverbs, comparatives, or serial verb constructions. Beyond an initial discus-
sion (Langacker 1991: section 11.2), coordination has not been dealt with. Though
morphology has not been ignored (Langacker 1987a, 1988a, 2000; Rubba 1993;
Nesset 1998; Tuggy 2003), the full-scale description of a highly elaborate mor-
phological system would be quite instructive. Also, the rather extensive efforts to
characterize the meanings of grammatical elements contrast with the limited at-
tention thus far devoted to lexical semantic description (Rice 1988; Vandeloise
1991; Farrell 1995; Shen 1996; Lee 1999; Langacker 2002a).
46
Obviously, expanding
the coverage of Cognitive Grammar into these and other areas is essential for its
continued development and empirical justification.
7. Directions

The word ‘cognitive’ should not obscure the fact that Cognitive Grammar is a
linguistic theory. Its analyses, descriptive constructs, and theoretical claims are all
inspired and supported by specifically linguistic considerations. The assumptions it
makes about mind and cognitive processing are general and fairly minimal. While
it is meant to be broadly compatible with secure findings of the cognitive sciences,
47
Cognitive Grammar does not simply adopt any extant psychological theory. Indeed,
it poses severe challenges for any processing model.
More extensive interaction with the cognitive sciences can be anticipated as
one direction in Cognitive Grammar’s future development. Hinting at the poten-
tial for such interaction is a certain amount of experimental and observational
work already carried out. Studies by Harris (1998) on entrenchment and by Tomlin
(1995, 1997) on the focal prominence of subjects, illustrate experiments bearing
on particular Cognitive Grammar notions. Basic ideas of Cognitive Grammar
are incorporated in Barsalou’s (1999) research on ‘‘perceptual symbol systems.’’ Its

usage-based nature dovetails with Tomasello’s (1992, 2003) observations on lan-
guage acquisition. Kellogg’s (1996) investigation of aphasia provides a measure of
support for its conceptual characterization of grammatical classes. In the future,
evidence from studies in psycholinguistics, neurolinguistics, language processing,
language acquisition, and aphasia ought to become increasingly important as an
empirical basis for assessing and refining Cognitive Grammar.
Though less straightforwardly empirical, applications of Cognitive Grammar are
significant as a potential source of validation. Its application to language pedagogy,
especially foreign language teaching, is starting to receive serious attention (Taylor
1993;Pu
¨
tz, Niemeier, and Dirven 200 1a, 2001b; Rudzka-Ostyn 2003). It offers a
cognitive grammar 449