Lexical concepts, cognitive models and

meaning-construction
VYVYAN EVANS

Abstract

In this paper I address the role of words in meaning-construction. My start-

ing point is the observation that the ‘meanings’ associated with words are
protean in nature. That is, the semantic values associated with words
are flexible, open-ended and highly dependent on the utterance context in
which they are embedded. In attempting to provide an account of meaning-
construction that coheres with this observation I develop a cognitively-
realistic theory of lexical representation and a programmatic theory of
lexical concept integration. My fundamental claim is that there is a basic
distinction between lexical concepts, and meaning. While lexical concepts
constitute the semantic units conventionally associated with linguistic
forms, and form an integral part of a language user’s individual mental
grammar, meaning is a property of situated usage-events, rather than
words. That is, meaning is not a function of language per se, but arises
from language use. I present an account of lexical concepts and the concep-
tual knowledge structures, cognitive models, with respect to which they are
relativised. I also situate this theory within a usage-based account. I then
develop a theory of lexical concept integration which serves to provide
an account of how lexical concepts are combined in service of situated
meaning-construction. As the constructs lexical concept and cognitive
model are central to the theory of lexical representation and meaning-
construction I present, I refer to the approach developed here as the Theory
of Lexical Concepts and Cognitive Models, or LCCM Theory.

Keywords: lexical concept; cognitive model; meaning-construction;

semantic composition; lexical representation; word-meaning;
LCCM theory; encyclopaedic knowledge.

Cognitive Linguistics 17–4 (2006), 491–534 0936–5907/06/0017–0491

DOI 10.1515/COG.2006.016 6 Walter de Gruyter
492 V. Evans

1. The protean nature of word ‘meaning’

The traditional view of meaning-construction is based on the assumption
that words have sense-units, or ‘meanings’, which are typically conceived
as static ‘lexical entries’ (Allwood 2003; Pustejovsky 1995; Tyler and
Evans 2001). Lexical entries are thought of in many formal and computa-
tional approaches to linguistic semantics as being tagged with syntactic,
morphological and semantic features. These lexical entries combine,
together with the grammatical structure of the sentence, to produce
sentence-meaning, known technically as a ‘proposition’. The combinato-
rial property of language that facilitates the integration of word ‘mean-
ings’ with syntactic structures producing sentence-meaning is referred to
as the principle of compositionality.
However, words in use do not behave in the straightforward manner
assumed by this received view (Clark 1983; Coulson 2000; Croft 2000;
Evans and Green 2006; Sweetser 1999). That is, the ‘meaning’ associ-
ated with a word in any given utterance appears to be, in part, a func-
tion of the particular linguistic context in which it is embedded. Put an-
other way, word ‘meaning’ is protean, its semantic contribution sensitive
to and dependent on the context which it, in part, gives rise to (Croft
2000).
To illustrate consider the following ‘meanings’ of fast.

(1) a. That parked BMW is a fast car.

b. That car is travelling fast.
c. That doddery old man is a fast driver.
d. That’s the fast lane (of the motorway).

In each of these examples the semantic contribution of fast, what I

will later refer to as its informational characterisation, is somewhat dif-
ferent. In (1a) fast has to do with the potential for rapid locomotion.
In (1b) it has to do with rapid locomotion. In (1c) it relates to ‘caused’
motion beyond an established norm: a speed limit. And in (1d) fast
concerns a venue for rapid locomotion. Examples such as these show
that the view of open class words, as possessing fixed meanings, is un-
tenable on closer scrutiny. The precise semantic contribution of any
word is a function of the utterance context in which it is embedded,
and, moreover, the sorts of (conceptual) knowledge these lexical enti-
ties provide access to, as I shall argue in detail. In other words, words
don’t have ‘meanings’ in and of themselves. Rather meaning is a func-
tion of the utterance in which a word is embedded, and the complex
processes of lexical concept integration, an issue which is developed
below.
 Lexical concepts, cognitive models and meaning-construction 493

2. Towards a cognitive compositional semantics

Key to providing an account of the protean nature of words, in so far as
they contribute to meaning-construction, is to provide a descriptively ade-
quate account of i) the sorts of knowledge that words provide access to
and ii) an account of how words, and their knowledge structures, are in-
tegrated or ‘composed’. In the literature on lexical representation and se-
mantic compositionality, three sorts of approaches have been invoked by
scholars who have attempted to provide accounts which recognise that
word-meaning is protean in nature. These three sorts of approaches can
be characterised as follows:
i) Fine-grained polysemy (e.g., Lako¤ 1987): involves a vast number of
distinct senses which attempt to exhaust the possibilities that actu-
ally occur in language;
ii) Abstract/underlying lexical conceptual paradigms (i.e., monosemy;
e.g., Pustejovsky 1995): employs cognitive/linguistic ‘devices’ that
operate on underlying conceptual entities in order to ‘‘generate’’ sur-
face interpretations of words;
iii) Underlying lexical entries (semantics) together with specific
principles/rules of interpretation (pragmatics) (e.g., Herskovits
1986).
There are two di‰culties common to each of these approaches as
they attempt to account for (situated) variation in word-meaning. First,
each of the accounts assumes that word-meanings are relatively stable,
circumscribed knowledge structures which can be (relatively) straightfor-
wardly identified. That is, they assume that word-meanings are relatively
rigid, discrete sense-units, qua mental entities. More recently, a number of
scholars have suggested that in fact word-meaning is less a discrete body
of circumscribed knowledge. Rather, words serve as points of access to
larger-scale encyclopaedic knowledge structures, which are potentially
vast in scope as argued in detail below. On this view, words provide ac-
cess to what I will refer to as a semantic potential, with di¤erent sorts of
knowledge being potentially activated.
The second di‰culty associated with the three views mentioned above,
notwithstanding their attempt to handle variation in word-meaning, is
that they do not constitute usage-based accounts of word meaning. That
is, they make no attempt to relate their theoretical claims to the nature
of situated meaning, and thus how words are used in context in order
to express local communicative intentions. What a word ‘means’, which
is to say, which part of its encyclopaedic knowledge potential is activated
will always be a function of how it is being used in any given context.
This includes both the linguistic context (the surrounding words and
494 V. Evans

grammatical constructions), and the extra-linguistic context, including the

situated communicative intention of the language user. Thus, we require
an account of the nature of the semantic potential that words provide ac-
cess to, and an account of how this semantic potential is constrained by
virtue of the way in which words are combined and their contexts of use
(i.e., the cognitive operations that facilitate di¤erential activation of a
word’s semantic potential). The purpose of this section is to sketch such
an approach. The remainder of the paper will develop the details.

2.1. The semantic potential of words

Recent work in linguistics (e.g., Haiman 1980; Fillmore 1982, 1995; Lan-
gacker 1987), and cognitive psychology (e.g., Barsalou 1992a, b, 1999,
2003; Zwaan 2004) suggests that an encyclopaedic account of word-
meaning is required for a descriptively adequate account of lexical repre-
sentation and psycholinguistic processing. For instance, both Langacker,
and Fillmore make the point that word-meaning is always a consequence
of a larger knowledge structure with respect to which a given word is re-
lativised. Fillmore models such knowledge structures in terms of what he
refers to as semantic frames. These are conceptual entities that are, in
part, defined by virtue of the linguistically encoded concepts that make
them up, and which a given word necessarily evokes. Langacker (1987)
makes a similar point arguing that a word’s scope of predication (infor-
mally, its ‘meaning’) involves a profile and a base, where the profile des-
ignates a conceptual substructure within a larger unit, namely the base.
For instance, the lexical item hypotenuse designates a substructure, the
longest side, within a larger unit, namely a right-angled triangle. Thus,
the ‘meaning’ of hypotenuse involves both the entity designated and the
larger unit with respect to which the designated entity, the profile, is a
substructure.
In addition to background knowledge structures of these kinds, words
provide ‘points of access’ (in Langacker’s terms) to large-scale encyclo-
paedic knowledge networks. Langacker models such knowledge struc-
tures in terms of domains, with domains being organised in hierarchical
fashion forming a domain matrix (see Langacker 1987; see also Croft
1993; Evans and Green 2006: Ch. 7). A similar idea is developed in the
work of Cruse (e.g., 2002; Croft and Cruse 2004) who refers to the ency-
clopaedic knowledge that words provide access to as purport.
There are two recent representative models of encyclopaedic semantics
which attempt to account for how the semantic purport that a word
provides potential access to is constrained. Both models take account of
the utterance context (described in more detail below) which serves to
 Lexical concepts, cognitive models and meaning-construction 495

constrain that part of a word’s semantic potential which is activated. To

illustrate the ideas just touched upon, I briefly present these models here.
The first is a language processing model developed by Zwaan (2004).
Zwaan’s model, which he refers to as the Immersed Experiencer frame-
work, is concerned with modelling how language provides access to
perceptual simulations (i.e., mental sensory rehearsals based on stored
perceptual representations), which are integrated in service of language
understanding. For instance, the lexical item red designates di¤erent
things depending on the sorts of entities and experiences it relates to.
Zwaan makes the point in the examples in (2), which are indicative of
those he uses:
(2) a. The teacher scrawled in red ink all over the assignment.
b. The red squirrel is in danger of becoming extinct in the British
isles.
red designates two di¤erent sorts of sensory experience precisely because
the context constrains the sort of perceptual simulations we derive. That
is, while the hue associated with the use of red in (2a) is quite a vivid red,
the hue of the simulation prompted for by (2b) is likely to be closer to a
dun/browny colour. In present terms, red has a large semantic potential,
which relates to a range of di¤erent possible hues (one dimension along
which the colour spectrum varies). That aspect of the word’s potential
which is activated is a consequence, in part, of the way it is constrained
by the utterance context, and specifically the scene evoked by the utter-
ance context.
A further important lesson from the work of Zwaan, and indeed others
who take a perceptual simulation approach to language understanding
(e.g., Bergen and Chang 2005), is that the semantic potential associated
with words is perhaps largely non-linguistic in nature. That is, the seman-
tic potential of red is not ‘there’ in the word itself. That is, whatever red
designates, we are not dealing with purely linguistic knowledge. Rather,
the form red provides access to perceptual information and knowledge,
which can be reconstructed or simulated.
The second recent model I briefly discuss here is the proposal put for-
ward by Allwood (2003). Allwood provides an account of what he refers
to as meaning potential (see also Zlatev’s 1997, 2003 related notion of use
potential). Allwood explicitly argues that a word’s meaning potential is

all the information that the word has been used to convey either by a single
individual, or on the social level, by the language community. A consequence
of this approach is that no attempt is made to distinguish between lexical and
encyclopedic information in terms of the kind of information that is contained in
496 V. Evans

the meaning potential. Meaning potentials contain both kinds of information—

information deriving from use of language and information deriving from other
experience with the world. (Allwood 2003: 43)

Central to Allwood’s proposal is the position that a word’s meaning

potential is ‘activated’ providing a situated interpretation. Thus, meaning
is always contextually determined, and is selected for from among the
knowledge ‘potential’ that a word provides access to. To illustrate, con-
sider the following examples adapted from Allwood (2003: 45):
(3) a. A carburettor is a part of a car.
b. A car need not have a carburettor since petrol can be directly
injected.
Allwood makes the point that the use of carburettor in (3a) probably ac-
tivates less detailed information than the use in (3b). Moreover, precisely
what is activated is subject to individual language users, as di¤erent indi-
viduals will have di¤erent encyclopaedic knowledge structures, and thus
di¤erent meaning potentials which can be both accessed and activated.
One of the purposes of the present paper is to provide an account of the
sorts of cognitive and linguistic operations which must be in place in order
for activation of a word’s semantic potential to occur. However, the
account of word-meaning I present in this paper di¤ers in two important
respects from the view presented by Allwood. First, I prefer the term ‘se-
mantic potential’ to Allwood’s notion of ‘meaning potential’. This fol-
lows, as I shall argue in the next section, as words don’t in fact have
meanings. I argue that meaning is not a property of language per se, but
rather is a function of language use, and thus, a characteristic of a process
of meaning-construction, rather than relating to mental entities/units
stored in memory. Meaning-construction is not an unpacking of stored
information, as assumed in more traditional accounts. Rather, it is a con-
structive process, in which integration of lexical units involves di¤erential
access to the conceptual knowledge which lexical entities potentially af-
ford access.
Second, I also argue for a distinction between purely linguistic knowl-
edge, and the encyclopaedic knowledge (the semantic potential) that
words provide access to (Allwood appears to assume that there is no prin-
cipled distinction between lexical and encyclopaedic knowledge, a posi-
tion that is hard to maintain). This I model in terms of the distinction be-
tween lexical concepts, stored linguistic knowledge units, and cognitive
models, conceptual knowledge structures which constitute the semantic
potential that lexical concepts provide access to. I use the term cognitive
model profile to refer to a word’s semantic potential. That part of the
 Lexical concepts, cognitive models and meaning-construction 497

word’s cognitive model profile that is activated in a given usage context I

refer to as its informational characterisation—to be described in detail
below).

2.2. Meaning and use

Having introduced the idea that a word relates to a semantic potential, I
now discuss the relationship between language use and meaning. I do so
by adopting a version of the usage-based thesis employed in cognitive lin-
guistics (see Evans and Green 2006 for discussion).
Language use is integral to our knowledge of language, our ‘language
system’ or ‘mental grammar’. The organisation of our language system is
intimately related to, and derives directly from, how language is actually
used (Croft 2000; Langacker 2000; Tomasello 2003). Through processes
of abstraction and schematisation (Langacker 2000), based on pattern-
recognition and intention-reading abilities (Tomasello 1999, 2003), lan-
guage users derive linguistic units. These are relatively well-entrenched
mental routines consisting of conventional pairings of form and meaning
(Langacker 1987; see Evans and Green 2006: Ch. 4, for a review).
However, the range of linguistic units available to the language user
massively underdetermine the range of situations, events, states, relation-
ships and other interpersonal functions that the language user may poten-
tially seek to use language to express and fulfil. One reason for this is that
language users live in a socio-physical ‘matrix’ that is continually shifting
and evolving. No two situations, feelings or relationships, at any given
point in time, are exactly alike. We are continually using language to ex-
press unique meanings, about unique states of a¤airs and relationships, in
unique ways. While language has a range of ‘ready made’ schemas, or lin-
guistic units which can be combined to express a representative range of
scenarios we may wish to refer to and describe, these necessarily under-
determine the mutability of human experience. As Langacker puts it,
‘‘linguistic convention cannot provide a fixed, unitary expression for
every conceivable situation that a speaker might wish to describe.’’ (Lan-
gacker 1987: 278). Accordingly, the linguistic units employed by language
users can only ever partially sanction (in Langacker’s e.g., 2000 terms)
the situated way in which they are used. As Croft argues, language use
involves solving a co-ordination problem, in which language users must
employ non-conventional co-ordination strategies and devices. That is,
language users typically employ the conventional repertoire of linguistic
units, including patterns of assembling linguistic units (such as word
order conventions, which are themselves linguistic units), in non-
conventional ways.1
498 V. Evans

On this view, meaning, which is associated with the utterance (or

usage-event), is a consequence of combining the conventional schemas
or linguistic units in novels ways in order to solve the particular co-
ordination problem at hand, thereby facilitating communication. The rea-
son, then, for the apparently protean nature of the ‘meanings’ associated
with words is that linguistic units are only ever realised as part of linguis-
tic utterances, which are necessarily (i.e., by definition) situated, and thus
part of an act of communication. But in being so realised, they have neces-
sarily undergone context-induced ‘shifts’ in their semantic value, and so are
never exactly the same as the lexical representations that sanction them.
Those scholars, however, who adhere to some form of Fregean compo-
sitionality are assuming, incorrectly, that the conventional semantic rep-
resentations associated with linguistic units such as words are realised in
language (use). They are not. As with the distinction between allophones
and phonemes in phonological theory, linguistic representations, by anal-
ogy akin to phonemes, are never actually perceived, but are inferred
based on how ‘sense-shifted’ words appear to behave in (situated) usage-
events, as judged over many instances of use. In this, then, the job of the
lexical semanticist is to employ meaning in usage-data, by analogy akin
to allophones, in order to infer the existence of the underlying lexical con-
cepts (stored mental schemas), akin to phonemes, which partially sanc-
tion the semantic contributions which surface.
This said, we are now in a position to provide some basic distinctions.
First of all we need to provide a definition of an utterance. This is less
straightforward a task than one might assume. As I will define it, a usage-
event or utterance has a unit-like status, in that it represents the expres-
sion of a coherent idea, making (at least partial) use of the conventions of
the language (informally, the ‘norms’ of linguistic behaviour in a par-
ticular linguistic community). In other words, an utterance is a somewhat
discrete entity. However, I use the expressions ‘unit like’ and ‘somewhat
discrete’ because an utterance is not an absolutely discrete nor a precisely
identifiable unit. This follows as utterances involve grammatical forms
such as word order, semantic structures, speech sounds, patterns of into-
nation such as pitch contours, slight pauses, and accelerations and decel-
erations, and so forth. While these properties converge on discreteness
and unity, they do not co-occur in fixed patterns, and therefore do not
provide a set of criteria for collectively identifying an utterance. In this
respect, utterances di¤er from the related notion of sentence.
A sentence, as defined in particular by formal linguists, is an abstract
entity. In other words, it is an idealisation that has determinate proper-
ties, often stated in terms of grammatical structure. For example, one def-
inition of (an English) sentence might consist of the formula: S ) NP VP.
 Lexical concepts, cognitive models and meaning-construction 499

The notion of a sentence, while based on prototypical patterns found in

utterances, is not the same as an utterance. Utterances typically occur
spontaneously, and often do not conform to the grammaticality require-
ments of a well-formed sentence, as understood in formal linguistic
theory. For example, in terms of structure, an utterance may consist of a
single word (Hi!), a phrase (No way!), an incomplete sentence (Did you
put the . . . ?), or a sentence that contains ‘errors’ of pronunciation or
grammar because the speaker is tired, distracted, or excited, and so on.
While much of formal linguistics has been concerned with modelling the
properties of language that enable us to produce grammatically well-
formed sentences, utterances often exhibit graded grammaticality (see
Langacker 1987; see also Evans and Green 2006). As this discussion re-
veals, while a sentence can be precisely and narrowly defined, an utter-
ance cannot. While sentences represent the structure associated with a
prototypical utterance, utterances represent specific and unique instances
of language use. Once a sentence is given meaning, context and phonetic
realisation, it becomes a (spoken) utterance. Although the theoretical
construct of ‘sentence’ might suit the aesthetic tastes of linguists of certain
persuasions, as I am concerned with an account of lexical representation
and meaning-construction that reflects how language is used, it is ulti-
mately the utterance, rather then the idealised notion of the sentence,
which is most relevant for present purposes.
Having provided this (qualified) definition of an utterance, we are now
in a position to distinguish meaning from lexical representation. My claim
is that the essential distinction between lexical representation and mean-
ing is that while meaning is a property of the utterance, lexical represen-
tations are the mental abstractions which we infer must be stored as part
of the language user’s knowledge of language, in order to produce the
range of novel uses associated with situated instances of a particular
word (or construction). The meaning associated with an utterance I will
refer to as a conception. Thus, conceptions are a function of language
use. Lexical representations, or rather more technically, lexical concepts,
represent the semantic pole of linguistic units, and are the mentally-
instantiated abstractions which language users derive from conceptions
and the specific semantic contribution perceived to be associated with
particular forms.

2.3. An architecture for meaning-construction

The conclusions to emerge from this discussion suggest a number
of requirements for a theory of lexical semantics and the role of
words in meaning-construction. We require both an account of lexical
500 V. Evans

representation and a theory of concept integration, which together should

contribute to a descriptively adequate and psychologically realistic ac-
count of meaning-construction. We require a theory of lexical representa-
tion which provides a descriptively adequate account of the kind of lin-
guistic knowledge that language users appear to possess. We also require
an account which provides a means of understanding how lexical repre-
sentations interface with conceptual knowledge, which is to say, their se-
mantic potential. That is, we require a theory that takes an encyclopaedic
perspective on linguistic meaning. We also require an account of how lex-
ical representations, together with the informational characterisations de-
rived from the semantic potential available combine in order to provide
situated meanings. Finally, as meanings associated with words are a func-
tion of specific utterances, and thus a consequence of discrete usage-events,
our theories of lexical representation and lexical concept integration must
be thoroughly usage-based in nature. As the two aspects of the theory I
present are complex, I present a summary of the architecture here. Each
of the constructs introduced are argued for in detail in later sections.

2.3.1. LCCM theory As the theory I present in the remainder of the

paper centrally invokes the theoretical constructs of the lexical concept
and the cognitive model, I refer to the model as the Theory of Lexical
Concepts and Cognitive Models, or LCCM theory for short.
LCCM theory consists of i) an account of lexical representation
(lexical concepts and cognitive models) and ii) an account of meaning-
construction (composition) of lexical concepts in a way which activates,
or, in my terms, provides an access route through the cognitive models
accessed by a given lexical concept. This can serve to highlight particular
facets and relations of a given cognitive model at the expense of other
facets and relations. As noted, the fundamental assumption is that mean-
ing (more technically a conception), is a property of an utterance (a situ-
ated instance of language use), which is formed, in part, by cognitive op-
erations which apply to the lexical representations (lexical concepts and
the cognitive model profiles to which lexical concepts provide access sites)
deployed by language users. Thus, meaning arises by virtue of language
users forming interpretations based on the lexical concepts employed, the
way lexical concepts are combined, and the access routes through the cog-
nitive model profiles accessed by given lexical concepts. Moreover, these
interpretations are always guided by background knowledge and extralin-
guistic context.

2.3.2. Lexical representation Knowledge of language includes i)

lexical concepts, and ii) cognitive models. Lexical concepts constitute
 Lexical concepts, cognitive models and meaning-construction 501

linguistically encoded concepts—that is conceptual knowledge encoded in

a form that can be externalised via language. Thus, lexical concepts con-
stitute the semantic pole of symbolic assemblies of form and meaning (see
Langacker 1987), and are conventionally associated with linguistic forms
of all kinds including words (the focus in this paper), bound morphemes,
idiomatic phrases and grammatical constructions. Accordingly, lexical
concepts, by definition, concern purely linguistic knowledge, as discussed
in some detail later. A second important part of the lexical representation
is the notion of the cognitive model, which is a large-scale coherent body
of non-linguistic knowledge which lexical concepts provide access sites to.
The range of cognitive models which are accessed, either directly or indi-
rectly by a lexical concept, I refer to as a cognitive model profile. Individ-
ual cognitive models consist of facets (attributes) and relations (structural
invariants) which hold between facets.

2.3.3. Lexical concept integration The meaning-construction process

takes place by virtue of lexical concept integration (‘composition’).
This process involves two component processes: i) lexical concept selec-
tion and ii) fusion. Lexical concept selection involves selecting the most
appropriate lexical concepts associated with each form in an utterance,
guided by utterance and extra-linguistic context. Fusion, the second com-
positional process consists of two further constituent processes which are
held to occur in tandem: i) integration and ii) interpretation. Integration
involves the construction of larger lexical entities, driven by linguistic
knowledge (lexical concepts). These larger lexical units, which I term lex-
ical conceptual units, are then interpreted. That is, the larger unit receives
an informational characterisation, which is to say, those parts of the cog-
nitive model profiles (semantic potential) associated with each lexical con-
cept in the larger unit is interpreted in a way that is in keeping with the
larger unit. Put another way, integration provides (linguistic) instructions
which serve to determine how the various lexical concepts are collectively
interpreted, and thus, the access route that each individual lexical concept
a¤ords through its cognitive model profile. The result is that any given
word will provide a unique activation of part of its semantic potential on
every occasion of use. This follows as every utterance, and thus the result-
ing conception, is unique.
Accordingly, this view of compositionality is radically di¤erent from
the received Fregean view. While Fregean compositionality assumes that
each usage of a word recruits stable, context-independent information,
LCCM theory assumes that the semantic value associated with a word
will vary slightly every time it is used. An overview of the architecture is
presented in Figure 1.
502 V. Evans

Figure 1. An overview of the architecture of LCCM Theory

3. Lexical representation
I now turn to a more detailed account of the first aspect of LCCM theory,
namely how lexical representation is modelled. I defer a more detailed ac-
count of lexical concept integration until section 4.

3.1. Lexical concepts

First, it is important to make clear that lexical concepts are not units of
meaning. To reiterate a point central to the present argument, meaning
(a conception) is a function of language use, and thus a property of
an utterance. Lexical concepts are linguistically encoded concepts or
mental representations that additionally serve as access sites to concep-
tual knowledge (cognitive model profiles). I now present six fundamental
properties associated with the lexical concept qua mental entity.

3.1.1. Lexical concepts are form-specific First, lexical concepts are

form-specific. That is, they are conventionally associated with specific lin-
guistic forms. A corollary of this is that lexical concepts, the semantic pole
of a form-meaning pairing, are necessarily language specific. Thus, each
language consists of an inventory of language-specific lexical concepts.

3.1.2. Lexical concepts are associated with di¤erent sorts of forms Sec-
ond, as lexical concepts are conventionally associated with a given
 Lexical concepts, cognitive models and meaning-construction 503

linguistic form, it follows that lexical concepts are conventionally associ-

ated with a wide range of forms. The range of forms with which lexical
concepts are conventionally associated includes overt forms: those which
have resolved phonetic forms, such as cat, and implicit forms: those
which have unresolved phonetic form, such as the ditransitive construc-
tion (subject verb obj1 obj2), e.g., John baked Mary a cake; John gave
Mary the cake; John refused Mary the cake, etc. (see Goldberg 1995).
Moreover, overt forms that have distinct lexical concepts conventionally
associated with them include bound morphemes, ‘simplex’ words, ‘com-
plex’ or polymorphemic words, and idiomatic expressions and phrases.
In addition to grammatical constructions, implicit forms include gram-
matical ‘relations’ such as subject and object, and lexical classes such as
noun and verb.

3.1.3. Forms are not lexical concept-specific Third, although lexical

concepts are form-specific, a single form can be conventionally associated
with a potentially large number of distinct lexical concepts which are re-
lated to degrees, as attested by the phenomenon of polysemy.2 That is,
forms are not lexical concept-specific. A consequence of this is that the
lexical concepts which share the same form can be modelled in terms of
a semantic network (see Evans and Green 2006, for discussion). As any
given lexical concept potentially provides access to other lexical represen-
tations it is associated with, I refer to the lexical concepts B, C, D . . . in
the same semantic network as lexical concept A as the semantic network
profile of that particular lexical concept. Put another way, lexical con-
cepts, which are linguistic entities, exhibit polysemy, and yet each lexical
concept provides a distinct semantic potential.

3.1.4. Lexical concepts have a lexical profile Fourth, the definitional

property of any given lexical concept is that it has a unique lexical profile,
its unique ‘biometric’ identifier. A lexical profile is an extension of criteria
presented in Evans (2004a), and akin to the notion of an ID tag (Atkins
1987) and behavioural profile (Gries 2006). While a lexical concept associ-
ated with a particular form possesses a particular semantic value—I will
use the mnemonic of small capitals inside square brackets as a gloss—
determining whether a particular usage of a form relates to one lexical
concept rather than another is a matter of examining the selectional ten-
dencies (i.e., semantic, collocational and formal patternings) associated
with a given usage. While any given usage of a lexical concept will have
its own unique selectional requirements, general patterns (‘tendencies’)
can be established, and form part of the conventional knowledge associ-
ated with a particular lexical concept.3 General selectional patterns in
504 V. Evans

terms of semantic, collocational and grammatical tendencies are what I

refer to as a lexical profile.
Two sorts of information form a lexical concept’s lexical profile. The
first relates to semantic selectional tendencies. In Evans (2004a) this was
referred to as the Concept Elaboration Criterion. The second relates to
formal or grammatical selectional tendencies. In Evans (2004a) I referred
to this as the Grammatical Criterion.
To get a sense of how the lexical profile can be used to identify dis-
tinct lexical concepts which underlie specific usage-events, I present
the lexical profiles for a number of distinct lexical concepts encoded by
the polysemous forms time (a noun) and fly (a verb). This discussion
is based on Evans (2004a; see also Evans 2004b, 2005), and Evans
and Green (2006). Nevertheless, I emphasise that the methodology I
introduce here is primarily meant to serve as a guide for future research.
Ultimately, establishing the existence of a given lexical concept is an em-
pirical question, which can be investigated using psycholinguistic and
corpus-based tools and methodologies (see e.g., Cuyckens et al. 1997;
Gries 2006).
Consider the following examples illustrating examples sanctioned by
distinct lexical concepts for time:
(4) Time flies when you’re having fun.
(5) The time for a decision is getting closer.
(6) The old man’s time [¼ death] is fast approaching.
(7) Time flows on (forever).
These instances of the lexical form time all appear in the ‘subject’ phrase.
Moreover, the verb phrase which complements the subject phrase relates
to a motion event. Thus, motion is being ascribed to the entities time con-
tributes in prompting for, in each example. Yet, the semantic contribution
associated with time appears to be distinct in each example. In the first
example, the semantic contribution associated with time appears to relate
to an assessment of temporal magnitude. Thus, we might gloss the seman-
tic value associated with this instance of time as [duration]. In (5) the se-
mantic contribution of time might be glossed as [moment]. This follows
as the conception associated with the utterance as a whole relates to a
specific temporal moment when a particular decision is to be taken. Thus,
the contribution of time in this example appears not to relate to a dura-
tional elapse, but rather a discrete instant. In (6) the semantic contribu-
tion associated with time appears to relate to an event, which extra-
linguistic context informs us is death. Thus, this instance might be glossed
by [event]. Finally, in (7), the semantic contribution associated with time
appears to relate to an unending temporal elapse. In earlier work, I
 Lexical concepts, cognitive models and meaning-construction 505

described this as the ‘matrix’ sense associated with time, in which we un-
derstand Time to be ‘the’ event in which all other events occur. Thus, the
gloss we might apply to describe this instance of time is [matrix].
Based on the quite distinct semantic contributions associated with this
range of usages of time, I argued in Evans (2004a) that there are a range
of distinct lexical concepts associated with time identified by the glosses
introduced above. Moreover, each of these distinct usages has a distinct
lexical profile associated with it which supports this perspective. Let’s il-
lustrate for each.
I begin by examining the grammatical tendencies associated with each
use of time. To do this, let’s consider the kind of noun phrase in which
each semantically-distinct use appears. We first note that the examples in
(4) and (7) appear, on the face of it, to be similar. Neither is pre-modified
by a determiner. However, further examples reveal that the [duration]
lexical concept of time as in (4) can be determined by the definite arti-
cle when the assessment of temporal magnitude is specific rather than ge-
neric, while the [matrix] lexical concept cannot be.
(8) During the dinner date, the time seemed to fly. [duration]
(9) *The time flows on (forever). [matrix]
Indeed, this patterning appears consistent with the semantics of these
uses. While the [matrix] lexical concept already relates to a unique refer-
ent, the event which subsumes all others, and thus further specification
which the definite article would provide is superfluous, with assessments
of temporal magnitude both specific and more generic readings are avail-
able, encoded by determiner patterns exhibited in (4) and (8) respectively.
Thus, we can say that while both the [duration] and [matrix] lexical con-
cepts appears to pattern formally like mass nouns (see Evans 2004a for
evidence that they fail to allow determination by the indefinite article),
the [duration] lexical concept, but not the [matrix] lexical concept,
allows determination by the definite article.
The examples in (5) and (6) also exhibit unique patterns in terms of
grammatical collocational tendencies, both from each other and from the
examples in (4) and (7). The [moment] lexical concept appears to pattern
straightforwardly as a count noun, allowing determination by the definite
article, as in (5), or by the indefinite article, as in (10).
(10) A time will come when we’ll be forced to make a decision.
[moment]
The [event] lexical concept in (6) appears to require a pre-modifying gen-
itive noun phrase followed by the enclitic possessive ‘-s’, or else an attrib-
utive pronoun, serving a similar function.
506 V. Evans

(11) His time [¼ death] is fast approaching.

Thus, in subject position, these uses of time all appear to have quite dis-
tinct formal selectional tendencies. Let’s now turn to the semantic selec-
tional tendencies associated with these uses of the lexical concepts associ-
ated with time. The point here is that the nature of the motion event
encoded by the verb phrase is distinct for each of the semantic uses.
Moreover, the choice of motion event type is compatible with the seman-
tic value of the lexical concepts which underlie each use (see e.g., Evans
2004a, for detailed discussion). For instance, the [duration] lexical con-
cept, and this particular variant—which in previous work I refer to as
[temporal compression] as this instance relates to an assessment of tem-
poral magnitude which proceeds more ‘quickly’ than usual—is comple-
mented by verb phrases which encode motion events which are rapid in
nature, as evidenced by the example in (4).4 In contrast, the [moment] lex-
ical concept appears to allow a wider range of motion events, including
imperceptible motion as in (12), rapid motion, as in (13), and terminal
motion, as in (14).
(12) The time for a decision has gone/vanished/disappeared.
(13) The time for decision is racing towards us/fast approaching.
(14) The time for a decision is approaching/getting closer/has arrived.
The [event] lexical concept appears to restrict the range of motion events
which can collocate with it to terminal motion events, i.e., motion events
which terminate ‘at’ the experiential locus, typically a human experiencer.
Finally, the [matrix] lexical concept appears to require motion events
which are non-terminal in nature. That is, it requires motion events which
are on-going, a paradigm example being ‘flow’. Thus, each of the examples
represent specific instances of distinct lexical concepts which exhibit dis-
tinct semantic and formal selectional tendencies: distinct lexical profiles.
Now consider a further illustration of the lexical profile relating to dis-
tinct lexical concepts associated with a single form. This time we consider
just a few of the many lexical concepts associated with a particular verb:
(to) fly.
(15) The plane/bird is flying [self-propelled aerodynamic
(in the sky). motion]
(16) The pilot is flying the plane [operation of entity capable of
(in the sky). aerodynamic motion]
(17) The child is flying the kite [control of lightweight entity]
(in the breeze).
(18) The flag is flying (in the [suspension of lightweight object]
breeze).
 Lexical concepts, cognitive models and meaning-construction 507

For convenience I have glossed the apparent semantic contribution of

each of the instances of fly. The glosses appear in square brackets along-
side the relevant examples. This data, and the glosses suggest that each
instance is sanctioned by a distinct lexical concept associated with fly. If
so, we should expect to be able to adduce a distinct lexical profile associ-
ated with each. Unlike nouns, for which a salient grammatical feature is
how they are determined, a salient grammatical feature for verbs is tran-
sitivity. In terms of formal dependencies then, we see that the hallmark of
the lexical concepts which license the uses in (15) and (18) is the lack of a
direct object (an intransitive verb). This contrasts with the lexical con-
cepts which sanction the examples in (16) and (17) which both require
a direct object (a transitive verb). This distinction in transitivity fails to
distinguish (15) from (18) and (16) from (17). For this we must rely on se-
mantic tendencies. The hallmark of each of these lexical concepts is that
they require distinct semantic arguments.
For instance, the [self-propelled aerodynamic motion] lexical con-
cept which is held to sanction the use of fly in (17) only applies to
entities that are capable of self-propelled aerodynamic motion. Entities
that are not self-propelled, such as tennis balls, cannot be used in this
sense (*the tennis ball is flying in the sky).
The lexical concept which underlies the use of fly in (16): [operation of
entity capable of aerodynamic motion], is restricted to the operation
by an entity which can be construed as an agent, and moreover, to enti-
ties that can undergo self-propelled aerodynamic motion. Further, the en-
tity must be able to accommodate the agent and thereby serve as a means
of transport. This explains why planes and hot air balloons are compati-
ble with this sense, but entities unable to accommodate an agent are not.
This is illustrated by example (19).

(19) ??He flew the sparrow across the English Channel.

Nevertheless, entities which can be construed as being guided, or at least

susceptible to being trained by a volitional agent which cannot accommo-
date an agent, are partially sanctioned by this lexical concept, as the fol-
lowing example illustrates.

(20) He flew the homing pigeon across the English channel.

In the case of [control of lightweight entity] as evidenced by the use

of fly in (17), this lexical concept appears to be restricted to entities that
are capable of becoming airborne by turbulence, and can be controlled by
an agent on the ground. This lexical concept appears to be specialised for
objects like kites and model/remote controlled aeroplanes.
508 V. Evans

Nevetherless, particular instances of lexical items such as fly appear to

rely on what I refer to as multiple sanction. In the following example:
(21) The kite is flying (in the sky).
this use of fly(ing) appears to be partly sanctioned by both the [self-
propelled aerodynamic motion] and the [control of lightweight
entity] lexical concepts. It adopts the grammatical patterning of the for-
mer lexical concept, but we understand that it must be controlled by an
agent, rather then being self-propelled.
The final lexical concept, glossed as [suspension of lightweight
object], selects for entities that can be supported by virtue of air turbu-
lence, but remain ‘connected to’ the ground. This lexical concept applies
to flags as well as hair and scarves, which can ‘fly’ in the wind.

3.1.5. Lexical concepts can be combined Fifth, lexical concepts can be

combined in a range of ways, as determined by the nature of the lexical
concepts involved (their semantic value as discussed below). This is the
principle mechanism by which language functions in order to create flexi-
ble prompts for the process of interpretation (the informational character-
isation derived by virtue of integrating lexical concepts in a way which
provides di¤erential activation of the semantic potential available via
each individual lexical concept).

3.1.6. Lexical concepts possess a semantic value The final property

associated with lexical concepts relates to the semantic value provided,
in principle, by the lexical concept. While the lexical concepts mentioned
thus far have been provided with semantic glosses, given in small capitals
inside square brackets, these are simply shorthand labels for the complex
conceptual knowledge structures that lexical concepts provide. Before
considering the sort of linguistic and conceptual knowledge that lexical
concepts provide or access, it is important to reiterate that although lexi-
cal concepts are conceptual in nature, they are knowledge structures which
are specialised for symbolic representation (i.e., in language). Accord-
ingly, they are of a quite di¤erent kind from the sorts of conceptual rep-
resentations described by psychologists. In other words, while lexical con-
cepts are conceptual in nature, they are not the same sort of entity that
psychologists refer to as ‘concepts’.
The semantic value associated with a lexical concept has (at least)
five dimensions, treated in turn below. While the first relates to non-
linguistic knowledge, the remaining four are all unique to language, and
thus constitute the linguistic knowledge associated with a given lexical
concept.
 Lexical concepts, cognitive models and meaning-construction 509

3.1.6.1. Semantic potential First, lexical concepts possess a semantic

potential (in the sense discussed above). This semantic potential, via the
process of interpretation (discussed in detail later), gives rise to an infor-
mational characterisation. This occurs by virtue of lexical concepts pro-
viding access to conceptual (or ‘‘encyclopaedic’’) knowledge structures.
In LCCM theory these knowledge structures are characterised in terms
of cognitive models, which collectively form a cognitive model profile for
a given lexical concept. Cognitive models are described in detail later. Ac-
cordingly, lexical concepts provide access sites or, in Langacker’s (1987)
terms, ‘‘points of access’’ to conceptual knowledge. Indeed, it is this as-
pect of a lexical concept’s semantic value which is often most important
in accounting for the protean nature of word ‘meanings’ in language use.
However, the informational characterisation a¤orded by lexical con-
cepts is not equally rich for all lexical concepts. For instance, there is a
full distribution in terms of degree of specificity running from those lexi-
cal concepts which provide access to a highly detailed semantic potential
to those which provide access which is highly impoverished. For example,
even within a single lexical class, the lexical concepts [automobile] asso-
ciated with the form car and [thing] associated with the form thing pro-
vide very di¤erent levels of detail in terms of their semantic potential.
While [automobile] is richly detailed, which is to say the nature of the
conceptual knowledge to which it provides access, [thing] is poorly de-
tailed (or impoverished), in relative terms.
One obvious way in which levels of detail di¤er is in terms of the dis-
tinction between lexical concepts encoded by so-called open versus closed
class forms. However, as Gentner and Boroditsky (2001) observe, this bi-
furcation is in fact better thought of in terms of a continuum. In present
terms, this continuum in fact relates, in part, to di¤erential level of detail
in semantic potential.

3.1.6.2. Encapsulation The second dimension of the semantic value

associated with a lexical concept relates to the notion of encapsulation.
That is, lexical concepts serve to encapsulate often complex and informa-
tionally di¤use ideas. While it is the cognitive model(s) that the lexical
concept provides access to which provides a coherent informational char-
acterisation, a lexical concept nevertheless serves to encapsulate complex
ideas by providing access at the relevant point in the knowledge ‘matrix’,
i.e., the cognitive model profile. Evidence for this encapsulating function
comes from culture-specific lexical concepts which cannot be easily cap-
tured in another language. An example from Korean is the lexical con-
cept encoded by the form nunchi, which might be translated into English
as ‘eye-measure’. This lexical concept relates to the idea that one should
510 V. Evans

be able to judge how others are feeling, such as whether a guest in one’s
home is hungry or not, and thus be in a position to o¤er food so that the
guest is not embarrassed by having to request it. Lexical concepts serve to
encapsulate complex ideas which are di¤usely grounded in an intricate
cultural web of ideas and information. They achieve this by providing ac-
cess sites at particular ‘points’ in conceptual knowledge.

3.1.6.3. Relational vs. non-relational The third dimension relates to

whether a lexical concept is relational or non-relational (Langacker
1987). For instance, while lexical concepts associated with forms labelled
as ‘nouns’ are non-relational, lexical-concepts associated with forms
which are labelled ‘verb’, ‘adjective’ or ‘preposition’, for instance, are re-
lational. A consequence of this is that part of the semantic value associ-
ated with relational lexical concept includes information relating to the
sorts of lexical concepts which the relational lexical concept can relate,
i.e., ‘argument structure’ or ‘valence’. For instance, the [contact] lexical
concept encoded by on encodes relational information relating to a figure
and reference object (‘ground’), as exemplified by the following preposi-
tional phrase: the cat on the mat.

3.1.6.4. Temporal structure The fourth dimension concerns the tempo-

ral structure of a lexical concept. That is, some lexical concepts, i.e., those
that are relational, encode how the temporal structure of the relation is
being accessed, i.e., whether it evolves in time, as encoded by verbs, or
whether it is provided as a ‘gestalt’, as encoded by prepositions. Lan-
gacker (1987) refers to this distinction as sequential scanning versus sum-
mary scanning.

3.1.6.5. Referentiality The final dimension relates to referentiality.

Lexical concepts refer to or index entities of di¤erent sorts. Some lexical
concepts provide denotational reference, referring to entities which are
conceived as objectively existing, or at least as having some objective ba-
sis in the socio-physical world of experience, such as physical artefacts.
Other lexical concepts provide deictic reference. They serve to refer to or
index entities understood with respect to some deictic centre, such as the
speaker’s physical location or social status, etc. Other lexical concepts
have anaphoric or cataphoric reference, referring to entities in the linguis-
tic context itself. Indeed, while obvious examples of such lexical concepts
include those encoded by, for instance pronouns, others include ‘‘shell’’
nouns (Schmid 2000), such as thing and aim, which take their reference
from the propositions which they simultaneously serve to encapsulate
and mark as coherent propositions.
 Lexical concepts, cognitive models and meaning-construction 511

Table 1. Summary of properties of lexical concepts

Property Details

lexical concepts are form-specific i.e., lexical concepts are conventionally associated with
a specific form. Thus, lexical concepts are necessarily
language specific.
lexical concepts are associated i.e., lexical concepts constitute the semantic pole of
with di¤erent sorts of forms symbolic assemblies of form-meaning (linguistic units)
of all kinds
forms are not lexical concept- i.e., are associated with a ‘semantic network’ of related
specific lexical concepts, and thus exhibit polysemy
lexical concepts have a lexical i.e., a unique ‘biometric’ identifier associated with
profile formal and collocational tendencies
lexical concepts can be combined i.e., lexical concepts can be combined in various
predictable ways, due to valence relations (relational
vs. non-relational lexical concepts and larger lexical
conceptual units), in service of activating semantic
potential and thus meaning-construction
lexical concepts possess a i.e., in context, lexical concepts provide an
semantic value: informational characterisation.
i) semantic potential i.e., access to a cognitive model profile (encyclopaedic
knowledge).
ii) encapsulation i.e., provides access to the cognitive model profile at a
particular access site (in the cognitive model profile).
iii) relational vs. non-relational i.e., relates to specific information concerning valence,
and thus combinatorial abilities and properties.
iv) temporal structure i.e., encodes how the temporal structure of the relation
is being accessed, i.e., whether it evolves in time, as
encoded by verbs, or whether it is provided as a
‘gestalt’, as encoded by, for instance, prepositions.
v) referentiality i.e., lexical concepts refer to or index entities of di¤er-
ent sorts. Some lexical concepts refer to entities which
are conceived as objectively existing. Other lexical con-
cepts refer to entities understood with respect to some
deictic centre, such as the speaker’s physical location.
Others refer to entities in the linguistic context itself.

A summary of the various properties associated with lexical concepts is

provided in Table 1.

3.2. Cognitive models

Having provided an overview of (at least some of ) the key properties as-
sociated with lexical concepts, we now return to a key dimension associ-
ated with their semantic value. This concerns the semantic potential af-
forded by lexical concepts. This section is concerned with introducing
and describing the construct of the cognitive model in more detail.
512 V. Evans

My claim is that cognitive models, while related to the notion of frame

(Barsalou 1999), semantic frame (Fillmore e.g., 1982; 1985; Fillmore and
Atkins 1992) and domain (Langacker 1987), are distinct from all three.
The distinct notion of cognitive model is necessary for understanding the
way lexical concepts contribute to meaning-construction. The main claim
is that lexical concepts provide sites of access to cognitive models and are
relativised with respect to them.
In previous work (Evans 2004a, 2004b), I referred to the larger-scale
knowledge structures with respect to which lexical concepts are relativised
as cognitive models. The reason for preferring this term over the related
notions of domain/base or semantic frame is that a cognitive model, in
my sense, is a coherent, in large-part non-linguistic, knowledge structure,
similar to what Langacker seems to have in mind, and in some statements
to what Fillmore has in mind, particularly when he tends towards the view
of a semantic frame as incorporating an (experiential) scene. That is, it is
a richly specified conceptual entity, akin to what Barsalou (1999) refers to
by his use of the term ‘frame’. However, as with both Fillmore’s notion of
a semantic frame and Langacker’s notion of a domain or a base, a cogni-
tive model is accessed, at various points, by distinct lexical concepts,
which are thus relativised to it, and in part, collectively constitute it.
In other words, a cognitive model represents an interface between richly-
specified conceptual knowledge and nodes of access at particular points in
the cognitive model provided by specific lexical concepts. Thus, lexical
concepts provide particular established (i.e., conventional) perspectives
or construals with respect to the set of cognitive models (cognitive model
profile) accessed via a given lexical concept. Yet, a cognitive model is far
richer than the sum of the lexical concepts which provide access sites to it.
This follows as while lexical concepts are conceptual units specialised for
symbolic representation in language, cognitive models are not. Rather,
they are multi-modal conceptual entities, which can be used as a basis
for perceptual simulations (see Barsalou 1999; and others, e.g., Prinz
2002 and Zwaan 2004).
Cognitive models relate to coherent bodies of knowledge of any kind.
For instance, they include knowledge relating to specific entities, such as
the complex knowledge associated with a specific entity such as ‘car’, or a
more specific entity such as ‘my car’. They include information such as
whether the car needs filling up and when I last cleaned its interior. Cog-
nitive models can relate to ‘procedural’ bodies of knowledge, such as ‘cul-
tural scripts’ which form templates for how to interact in restaurants in
order to be seated and secure a meal (cf. Sharifian’s 2003 notion of
cultural conceptualisations which are culturally ‘distributed’). Cognitive
models also include bodies of knowledge relating to more abstract entities
 Lexical concepts, cognitive models and meaning-construction 513

such as containment, love and physics. They operate at varying levels

of detail, and while stable, are dynamic being in a perpetual state of
modification and renewal by virtue of on-going experience, mediated
both by linguistic and non-linguistic interaction with others and one’s
environment.
Lexical concepts may be relativised with respect to more than one, typ-
ically many, cognitive models. As cognitive models are necessarily con-
nected to and inherit structure from many others, a lexical concept can
potentially be connected to a vast network of conceptual knowledge.
Those cognitive models which are directly accessed by a lexical concept
are referred to as primary cognitive models. The set of such primary cog-
nitive models for a given lexical concepts is termed the primary cognitive
model profile. The sorts of structured knowledge which is a subset of a
given primary cognitive model I refer to as a secondary cognitive model.
Such cognitive models relate to what I term a lexical concept’s secondary
cognitive model profile. This distinction is important to distinguish be-
tween literal and figurative language use (see Evans and Zinken to appear
for detailed explication).
By way of illustrating the relationship between a lexical concept and its
cognitive model profile, consider the lexical concept [france]. This lexical
concept provides access to a large number of cognitive models (its cogni-
tive model profile) at a particular access site, which is to say, a particular
point in the cognitive model profile. Accordingly, it provides an encapsu-
lating function. A very partial cognitive model profile for this lexical con-
cept is provided in Figure 2.

Figure 2. Partial cognitive model profile for [france]

514 V. Evans

In Figure 2, the lexical concept [france] provides access to a potentially

large number of knowledge structures. As each cognitive model consists
of structured knowledge, providing access to other sorts of knowledge,
we can distinguish between cognitive models which are directly accessed
via the lexical concept, and those cognitive models which form sub-
structures of the directly accessed (i.e., primary) cognitive models. That
is, such secondary models are indirectly accessed via the lexical concept.
Accordingly, a cognitive model profile is a structured inventory of knowl-
edge (a semantic potential) which lexical concepts a¤ord access to.
For instance, the primary cognitive model profile consists of (at the
very least) the following cognitive models: geographical landmass,
nation state and holiday destination. Each of these cognitive models
provides access to a sophisticated and large body of knowledge. In Figure
2 a flavour of this is given by virtue of the various secondary cognitive
models which are accessed via the nation state cognitive model. These
include national sports, political system and cuisine. For instance,
we may know that in France, the French engage in national sports of par-
ticular types, e.g., football, rugby, athletics and so on, and take part in
competitions of various kinds including the FIFA football world cup,
the six nations rugby competition and the rugby world cup, the Olympics,
and so on. That is, we may have access to a large body of knowledge con-
cerning the sorts of sports French people engage in. We may also have
some knowledge of the funding structures and social and economic con-
ditions and constraints that apply to these sports in France, France’s in-
ternational standing in these particular sports, and further knowledge
about the sports themselves including the rules that govern their practice,
and so on. This knowledge is derived from a large number of sources.
With respect to the secondary cognitive model of political system,
Figure 2 illustrates a sample of further secondary cognitive models which
are accessed via this cognitive model. In other words, each secondary cog-
nitive model has further secondary cognitive models which it provides ac-
cess to. For instance, (French) electorate is a cognitive model accessed
via the cognitive model (French) political system. In turn the cognitive
model (French) political system is accessed via the cognitive model
nation state. Accordingly, nation state is a primary cognitive model
while electorate and political system are secondary cognitive models.
Finally, it is worth reiterating that the cognitive model profile accessed
via a lexical concept, while conceptual in nature, relates to knowledge
that is non-linguistic. That is, it derives from our interaction with the
world about us, including sense-perception. In contrast, lexical concepts
are linguistic entities as described in the previous section, which also pro-
vide access to non-linguistic knowledge. In situated use, lexical concepts
 Lexical concepts, cognitive models and meaning-construction 515

are always interpreted (a process described later), such that they provide
an informational characterisation. This is achieved via activation of part
of the semantic potential available.

4. Lexical concept integration

The discussion of lexical concepts and how they relate to cognitive
models in the preceding section now allows us to examine how meaning-
construction occurs. That is, we are now in a position to see how lexical
representations contribute to the formation of a conception. This process
of meaning-construction I will refer to as composition. However, we are
not dealing with Fregean compositionality. Rather, we are dealing with
composition that makes use of lexical concepts, and the cognitive models
with respect to which they are relativised.
Composition involves two processes, termed selection and fusion.
Composition also accounts for semanticality—semantic well-formedness.
I address each of these issues below.

4.1. Lexical concept selection

This is the process in which linguistic or extra-linguistic context selects for
a particular lexical concept. Selecting the ‘correct’ lexical concept is re-
quired by fusion, the operation in which lexical concepts are integrated
and the resulting integrations are interpreted. One of the complexities
associated with meaning-construction, however, is that many processes
occur at the same time, and thus, it is far from clear that the processes
involved are, in psycholinguistic terms, sequential (Gibbs 1994).
An example of extra-linguistic context giving rise to selection follows.
(22) That recent hike is killing me.
The form hike has at least two lexical concepts associated with it. One
relates to a long, typically cross-country walk, while another relates to
an increase in financial charges or payments of some kind. In the context
of a recent cross-country walk, the example in (22) might relate to sore
body parts. In the context of, for instance, a recent central bank base-
rate increase, the speaker might be referring to the di‰culty of keeping
up with mortgage repayments. Thus, the extra-linguistic context provides
a means of selecting the most appropriate lexical concept.
Now consider the following examples, which illustrate the role of lin-
guistic context in selection.
(23) The judge asked the defendant to approach the bar.
(24) The customer ordered her beer at the bar.
516 V. Evans

The form bar has a number of distinct lexical concepts associated with
it, including the ‘bar of a court’ at which the judge sits, and a ‘bar in
a public house’ at which alcohol is purchased and served. The appropri-
ate lexical concept is selected in these examples due to the linguistic
context.
Another example of linguistic context selecting the appropriate lexi-
cal concept is illustrated by ellipsis of the following kind, often termed
zeugma.
(25) On the day my driving licence expired, so did my old dad.
In this example, a di¤erent lexical concept for expired is selected in each
clause, despite the verb not appearing in the second clause. In the first
clause, the lexical concept selected for can be glossed as [expiration of
permission], in the second [death].
Despite being able to di¤erentially identify the separate contributions
of extra-linguistic and linguistic context with respect to lexical concept se-
lection, the typical arrangement appears to involve a process I refer to as
co-selection, involving both linguistic and extra-linguistic context. To il-
lustrate, consider the following utterance adapted from a recent news-
paper headline.
(26) France shot down the EU constitution
One of the points made by Croft (1993) in discussing a not dissimilar ex-
ample, was that words often appear to take on ‘meaning’ only when it is
clear what the ‘meaning of the whole’, what I refer to as the conception,
relates to. That is, it is only by knowing what the entire utterance relates
to that the ‘parts’ can be interpreted. For instance, France might relate to
the geographical landmass, the political ‘nation-state’ entity, the govern-
ment, the head of state, the people, a national sports team, a delegation
from France, or something else. Similarly, shot down has a number of
conventional interpretations associated with it, including the ‘literal’
meaning plus other conventional readings such as ‘forcefully reject’. Sim-
ilarly, EU constitution could relate to the membership of the EU, the
health of the EU, or the new treaty and charter of rights and social provi-
sion recently presented to European Union member states for ratification.
However, co-selection relies upon selecting the most mutually appro-
priate readings associated with each of these expressions. That is, to
understand the semantic contribution of one, we have to understand the
semantic contribution of all. Thus, co-selection has to do with the inter-
dependence of lexical concept selection. Selecting the most appropriate
lexical concept associated with a given form is a mutually-involving ‘sym-
biotic’ process.
 Lexical concepts, cognitive models and meaning-construction 517

Nevertheless, how is this co-selection process guided? I suggest that it

proceeds by virtue of a discourse ‘‘anchor’’, which I refer to as the utter-
ance topic. This is related to what Fauconnier (1997) terms the ‘base
space’. The topic might be inferred based on the preceding discourse, the
extra-linguistic context, or indeed, the utterance context itself. However,
the topic constitutes a ‘mini-theory’ concerning the general nature of the
conception, and is informed by the hearer’s assumptions regarding the
speaker’s communicative intention. Thus, co-selection crucially relies on
knowing the topic, which guides the co-selectional process. Indeed, this
is what Croft was referring to when he talks of the whole leading to our
understanding of the parts, what he describes in terms of the ‘unity of the
domain’. That is, unless we understand that the topic of the conception
illustrated by (26) relates to European Union politics, we will be unable
to make the most appropriate lexical concept co-selections, and thus be
unable to build the conception that the newspaper headline writer has in
mind.
Thus, co-selection can only proceed once we understand that the refer-
ences that need to be assumed relate to a complex body of current a¤airs
knowledge relating to the politics of the European Union. Indeed, to con-
struct a conception similar to the one the headline writer presumably has
in mind requires not only understanding the EU as a political and eco-
nomic entity consisting of 25 European member states, it also requires
knowledge relating to the raging debate that has held sway in many Eu-
ropean countries about the direction of the EU, and the recent ratification
process relating to a new treaty, labelled ‘EU constitution’, which in-
volved referenda being held in a number of European countries. The ut-
terance in (26) relates to the rejection of the EU constitution by a major-
ity of French voters in a recent referendum.

4.2. Fusion
Fusion concerns the process in which selected lexical concepts are com-
posed such that they give rise to a particular conception. Fusion involves
two component processes: integration and interpretation. I address each
in turn.

4.2.1. Integration Integration is the process in which selected lexical

concepts are incorporated into larger structures, what I refer to as com-
posite lexical-conceptual structures. One way in which this process occurs
is due to a process which Langacker refers to as elaboration. For instance,
the conceptual representation associated with a verb such as kick encodes
schematic roles for ‘kicker’ and ‘kickee’. Indeed, this relates to the view
518 V. Evans

that lexical concepts can be relational or non-relational as discussed ear-

lier. These roles, what Langacker refers to as elaboration sites, can be in-
tegrated with, or, in Langacker’s terms, elaborated, by lexical concepts
encoded by other lexical forms, e.g., He kicked me.
Of course, integration can become more complex when it involves lex-
ical concepts which have more than two elaboration sites. A case in point
is the so-called ditransitive or double-object construction. Goldberg (1995)
shows that this grammatical form has, in present terms, a conventional
lexical concept associated with it which can be glossed as [x causes y to
receive z]. The letters X, Y and Z correspond to distinct elaboration sites
(in Langacker’s terms).
(27) She kicked him the ball.
Of course, there is more to integration than elaboration. Composite
lexical structures can themselves be conjoined with other composite lexi-
cal structures. This process I refer to as constituency reanalysis. Essen-
tially, this is the process in which composite lexical structures once estab-
lished are treated as unitary entities or constituents at the next level of
processing. For instance, the lexical concept encoded by and is specialised
for integrating composite lexical structures. Thus, integration involves
elaboration (building of composite constituents), and constituency reanal-
ysis (building utterances from smaller ‘meaning’ constituents). However,
and crucially, it is important to emphasise that what licenses these pro-
cesses, providing coherence to the integration, is compatibility of the
lexical concepts involved (rather than, for instance, semantically ‘blind’
syntactic processes, as in many formal approaches). Indeed, semantic
incompatibility relates to utterances which are semantically anomalous,
an issue discussed later.

4.2.2. Interpretation Interpretation is a process that proceeds in con-

junction with integration. While integration serves to conjoin lexical con-
cepts giving rise to composite lexical conceptual structures, interpretation
serves to activate part of the semantic potential (cognitive model profile)
that each lexical concept provides access to. It does so in a way that is
consistent with the other lexical concepts of the composite lexical concep-
tual structure. In other words, it is not enough for meaning-construction
to select an appropriate lexical concept, and integrate lexical concepts
into a composite lexical-conceptual structure. In addition, the selected
lexical concept(s) must then be interpreted within the composite lexical-
conceptual structure, the new linguistic context, in which it occurs. This
process of interpretation, then, provides the crucial break between lexi-
cal representation and meaning-construction. It is as a consequence of
 Lexical concepts, cognitive models and meaning-construction 519

interpretation that a conception arises. In this section I will focus on in-

terpretation as it applies to lexical concepts encoded by word classes of
the following kind: nouns, adjectives, and verbs.
Throughout the paper I have referred to a construct termed access.
This concerns the ability that lexical concepts have to interface with non-
linguistic knowledge, that is, the cognitive model profile. Thus each lexi-
cal concept has a unique cognitive model profile to which it a¤ords ac-
cess, by virtue of serving as an access site to non-linguistic knowledge at
a unique, and thus di¤erent, point in the human knowledge matrix.
However, language use, and the consequent construction of a concep-
tion involve interpretation, and thus activation of just part of the cogni-
tive model profile accessed via a lexical concept. This occurs by virtue of
an access route through the cognitive model profile being established.
This activation process is a consequence of situated language use, which
I describe below. In addition to the establishment of an access route, there
is a further kind of activation which I refer to as highlighting. While the
establishment of an access route involves activation of a (typically lim-
ited) number of cognitive models in the cognitive model profile, highlight-
ing involves activation of a distinct facet or facets within a single cogni-
tive model. As highlighting is a consequence of situated access by a
lexical concept, and thus, is prompted for by language, it can be further
divided into two specific kinds, a function of the distinction between rela-
tional and non-relation lexical concepts described earlier. These two
forms of highlighting I refer to as perspectivisation and adjustment, which
I discuss in detail below. Figure 3 captures the distinct sorts of activation
of a lexical concept’s semantic potential that access a¤ords.

4.2.2.1. Cognitive model profiles In order to see how an access route is

established, I first briefly elaborate on the earlier discussion of a cognitive
model profile. In order to do so, consider the lexical concept [book] as en-
coded by the form (a) book, and a very partial cognitive model profile
that it a¤ords access to (Figure 4).
The partial cognitive model profile in Figure 4 provides a diagram-
matic representation of part of the semantic potential which the lexical
concept [book] provides an access site to. That is, knowledge accessed
by [book] includes, at the very least, that a book is a physical entity, and
is interacted with via a process of reading. These two distinct sorts of
knowledge, knowledge relating to an artefact, and the process of reading
are captured in Figure 4 by the two cognitive models book and reading
respectively. The two cognitive models are related by virtue of a reader
who interacts with the physical artefact by reading the printed text.
This relation holds between cognitive models (or facets), and is what
520 V. Evans

Figure 3. Activation processes within a cognitive model profile

Figure 4. The relationship between lexical concepts, cognitive models, and facets and
relations

Barsalou (1992b) refers to as a structural invariant. I capture this relation

in Figure 4 by a double-headed arrow, and the specific relation involved
is signalled by the mnemonic reader. In addition, cognitive models consist
of a large detailed, but structured, body of knowledge. The various attrib-
utes that make up this body of knowledge I refer to as facets. Figure 4
provides two facets for each of the cognitive models which [book]
provides access to. The cognitive model book relates to the physical arte-
fact, consisting of, at the very least, knowledge as to the physical structure
and organisation of a given book. This includes detailed knowledge
 Lexical concepts, cognitive models and meaning-construction 521

concerning the material aspects of the artefact, including its dimensions,

weight, binding (paper or cloth), and so forth. This facet of our knowl-
edge about books I refer to as the tome facet. In addition to the physical
organisation and construction of a book, books consist of text which is
interacted with through the process of reading. his I refer to as the text
facet.
The reading cognitive model relates to the process involved in interact-
ing with books, especially the nature of the interaction with the text itself.
One consequence of this interaction is that reading takes up a period of
time, which I refer to as the duration facet. That is, depending on the
amount of text involved, reading can take lesser or greater amounts of
time. Another consequence of interaction with books is the level of inter-
est that a given book holds for the reader. This I refer to as the level of
interest facet. That is, while the reader might judge the book to be inter-
esting another might be judged to be boring, and so on.

4.2.2.2. The establishment of an access route Access routes are estab-

lished by virtue of situated language use. In order to see how, consider
the following examples.
(28) That’s a long book.
(29) That’s a heavy book.
Each of these utterances involves a slightly di¤erent interpretation associ-
ated with the form book. That is, a consequence of each of these distinct
utterances is that book achieves a distinct informational characterisation.
This is achieved by virtue of each use of book being interpreted in a
way consistent with the utterance context such that a slightly di¤erent ac-
cess route is established through the cognitive model profile accessed via
[book].
For instance, the conception that results from (28) has to do with an
assessment of a relatively extended duration required in order to read the
book in question. In contrast, the conception associated with (29) has to
do with an assessment as to the weight associated with the book in ques-
tion. The process of interpretation then involves di¤erential activation of
aspects of the semantic potential accessed via [book], in a way that is con-
sistent with the lexical concepts encoded by the other forms which make
up the utterance context. Thus, the example in (28) involves an access
route which serves to activate the reading cognitive model profile. In
contrast, (29) involves an access route which involves the activation of
the book cognitive model.
Now let’s consider a more complex example, involving an access route
involving access in several cognitive models. To do so reconsider the
522 V. Evans

partial cognitive model profile accessed via the lexical concept [france] in
Figure 1. Now re-consider the utterance in (26), above, re-produced
below:

(26) France shot down the EU constitution

In this example interpretation requires that an access route is established

through the cognitive model profile accessed via [france] in a way that is
consistent with the lexical concepts associated with the other linguistic
forms and units in the utterance. The interpretation associated with
France in this example is that we are concerned with the French elector-
ate, and specifically that part of the French electorate which voted against
ratification of the EU constitution. In other words, [france] in this exam-
ple achieves an informational characterisation which is attained by acti-
vating the cognitive models which are shown in bold in Figure 5.
What this example shows is that an access route typically involves the
activation of a number of distinct cognitive models, in order to facilitate a
situated interpretation, i.e., the informational characterisation of a lexical
concept.
It is also worth pointing out that this sort of example feels like a
more figurative instance of language use than say an example involving
[france] as in the following:

(30) France is a country of outstanding natural beauty.

Figure 5. Access route established by the interpretation of [france] in the utterance France
shot down the EU constitution
 Lexical concepts, cognitive models and meaning-construction 523

The use of France in (30) constitutes an instance of what might be

thought of as metonymy (see Evans and Zinken to appear for detailed
discussion of how LCCM theory accounts for figurative language such
as metonymy).

4.2.2.3. Highlighting As previously noted, in addition to the general

activation process which involves the establishment of an access route
across cognitive models, there is a further activation process. This I refer
to as highlighting. This process involves activation within a single cog-
nitive model. There are two variants of highlighting which I term per-
spectivisation and adjustment. The distinction between the two relates to
a fundamental distinction between the kinds of lexical concepts encoded
in language: those that are relational (as encoded by, for instance, verbs,
adjectives and adverbs), and those that are non-relational (as encoded
by noun forms). The basic insight is that while non-relational lexi-
cal concepts determine what is activated—the highlighting process of
perspectivisation—relational lexical concepts can influence how the con-
ceptual structure is activated—the highlighting process of adjustment.

4.2.2.3.1. Perspectivisation This is an interpretation process involving

activation by virtue of lexical concepts which are non-relational in nature,
and thus as encoded by noun forms.5 That is, lexical concepts of this kind
activate structure within a single cognitive model, such as determining
which facets are highlighted, (rather than determining how the facets are
highlighted). To illustrate, consider the following examples.

(31) a. That’s a heavy book to carry around in your school bag all
day.
b. That antiquarian book is so old that it is illegible in places.
(32) a. That book is really long.
b. That book is really boring.

Reconsider the partial cognitive model profile accessed via the lexical
concept [book] presented as Figure 4. The examples in (31) give rise to a
conception which establishes an access route in the book cognitive model.
The di¤erence between the informational characterisation associated with
book in each of the examples in (31) relates to the two distinct facets tome
versus text. That is, the process of interpretation that gives rise to the two
distinct conceptions (or readings) associated with each utterance comes
from utterance context informing which facet is highlighted. In (31a) the
utterance context, particularly the lexical concept associated with heavy,
serves to highlight the tome facet, and especially those aspects of our
524 V. Evans

knowledge relating to the physical properties associated with heavy books

which mean that they are physically demanding to carry around over an
extended period of time. In contrast, the informational characterisation
associated with book in (31b) relates to the text facet. This is highlighted
due to our understanding that because of the ageing process, and wear,
old books may have text which is illegible. Thus, the informational char-
acterisation associated with [book] is slightly distinct in each of these ex-
amples by virtue of the process of perspectivisation, which serves to per-
spectivise a distinct facet in each example.
The examples in (32) illustrate perspectivisation of distinct facets in the
cognitive model of reading. The example in (32a) perspectivises the dura-
tion facet. This follows as the conception involves understanding [book]
as activating structure relating to the reading process, and in particular,
an assessment that the book in question takes a relatively great period of
time in order to be read. In contrast, the informational characterisation
associated with [book] in (32b) results from perspectivising the level of
interest facet. That is, the semantic value of [book] is slightly di¤erent
in this example. As with the examples in (31), the examples in (32) reveal
that the ‘meaning’ associated with book is a function of di¤erent ‘paths’
of activation, involving distinct access routes, and distinct sorts of activa-
tion within a single cognitive model.

4.2.2.3.2. Adjustment We now turn to the second of the two activation

processes known as highlighting. This relates to relational lexical con-
cepts, as encoded by adjectives, prepositions and verbs. This process of
interpretation I refer to as adjustment. I do so in order to distinguish the
process from that of perspectivisation. The reason for selecting a distinct
term is that the process involved appears to be slightly di¤erent, a con-
sequence of the relational nature of these lexical concepts. The main
di¤erence appears to be that while perspectivisation a¤ects the nature
of the knowledge structure(s) being highlighted (i.e., ‘what’, e.g., tome
vs. text facet), adjustment a¤ects the quality of, and thus ‘how’ the entity
in question is being highlighted. Consider some examples by way of
illustration.

(33) a. a small mouse

b. a small elephant
(34) a. a red pen
b. a red squirrel
(35) a. a good man
b. a good meal
 Lexical concepts, cognitive models and meaning-construction 525

With the examples of small, and red in (33) and (34) the sensory qualities
interpreted will depend on the lexical concept with respect to which it is
fused. That is, the informational characterisation associated with small
varies not in terms of the notion of relative size (what), but rather how
we interpret this. Thus, we interpret the absolute dimensions that apply
to small in (33a) to be quite distinct from those in (33b). Yet, there is no
confusion that ‘small’ can apply equally to an elephant or a mouse, each
of which (a small mouse versus a small elephant) are radically di¤erent in
terms of their absolute dimensions.
Similarly, the interpretation of good is adjusted depending on the com-
posite lexical-conceptual structure it is involved in. For instance, a good
man might possess attributes such as physical beauty, honour, providing
for his family, and so on, depending upon context. The sorts of qualities
associated with a good meal, however, are more likely to include the size
of the portions, how tasty the food is, that it consists of wholesome ingre-
dients, and so on. Thus, we adjust how the knowledge associated with
good is being activated rather than what is being activated, a consequence
of the relational (here attributive) nature of the lexical concept associated
with good.
Now consider the lexical concepts encoded by the verb bake: the well-
known change-of-state/creation ‘alternation’.
(36) Fred baked the potato.
(37) Fred baked the cake.
In these examples the informational characterisation associated with bake
is adjusted in the light of the informational characterisation accessed by
virtue of the other lexical concepts integrated in the composite lexical-
conceptual structure. Previous researchers have referred to this process
of adjustment by terms such as ‘accommodation’ (e.g., Talmy 1977) or
‘coercion’ (e.g. Pustejovsky 1995; Goldberg 1995). However, such schol-
ars have emphasised the role of other aspects of language in ‘coercing’
the ‘meaning’ of the verb. For instance, Goldberg argues that sentence-
level verbal argument constructions coerce verbal meanings. The reason
for selecting the term adjustment here is that the process I am describing
relates not primarily to ‘one-sided’ coercion by the grammatical construc-
tion. Rather, the adjustment is a consequence of a mutually interdepen-
dent process of interpretation in which bake provides access to an infor-
mational characterisation associated with rich conceptual knowledge,
part of which is highlighted in conjunction with and in response to the
informational characterisation provided by other lexical concepts in the
composite lexical conceptual structure. Thus, in addition to a process of
lexical concept integration, there are, additionally, activation processes:
526 V. Evans

perspectivisation and adjustment, which are processes of interpretation.

These necessitate accessing cognitive models which lexical concepts pro-
vide access sites to, but the nature of the information accessed must be
‘calibrated’ with respect to the contribution of the other lexical concepts
in the composite lexical conceptual structure. Moreover, adjustment is
also guided by other ‘contextualisation cues’ such as speaker communica-
tive intention, extra-linguistic context, accompanying gestures (see Ken-
don 2004 for an overview) and so forth. In the light of the discussion in
this section, Figure 6 presents an overview of the lexical concept integra-
tion process.

4.3. Semanticality
Finally, having provided a programmatic sketch of how the constructs
of lexical concepts and cognitive models might relate to a cognitively re-
alistic approach to compositional semantics, I now address the semantic
‘well-formedness’ of conceptions. Indeed, conceptions are, by definition,
semantically coherent. We will see that this is the case by considering sit-
uations in which conceptions fail.
The term semanticality, introduced into linguistics by Pustejovsky
(1995), relates to the semantic well-formedness of an utterance. Semanti-
cally well-formed utterances give rise to conceptions. Utterances that fail
in this regard are semantically anomalous. However, as with the related
notion of grammaticality, semanticality or semantic anomaly is a matter
of degree rather than an either/or distinction.
The principle reason for semantic failure appears to be a failure in
matching semantic selectional dependencies, discussed earlier. Of course,
semantic selection tendencies, or collocational patterns, are a consequence
of semantic compatibility. Even lexical concepts that are potentially
dissonant and can be said to clash, need not result in failure to form a
conception. This follows as the semantic potential which lexical concepts
provide access sites to is vast and extremely richly detailed, allowing the
possibility of novel access routes, perspectivisations and adjustments re-
sulting in a semantically well-formed conceptions, particularly with ap-
propriate extra-linguistic context to assist in the co-selection process. In-
deed, this is the strategy that prevails in so-called ‘figurative’ language use
(see Evans and Zinken to appear).
To consider this phenomenon consider some examples involving the
verb began.
(38) a. He began the book
b. ?He began the dictionary
c. ??He began the rock
 Lexical concepts, cognitive models and meaning-construction 527

While the first example evidences a semantically well-formed utterance,

the second example is semantically odd. This follows as dictionaries are
not something we ‘begin’, as their function relates to reference and look-
up. Thus, there is a mismatch between the informational characterisations
provided by the cognitive model profiles as accessed by the lexical con-
cepts in this utterance.
However, in certain situations extra-linguistic context can help, as
pointed out by Pustejovsky (1995). For instance, Malcolm X, the African
American civil rights activist who promoted violent struggle, is famously
known to have read a dictionary while in prison ‘like a book’. As the only
book available to him was a dictionary he began at the letter A and read
through to Z. In such a situation, the example in (38b) becomes semanti-
cally acceptable.
The example in (38c) is semantically anomalous as a rock is not an en-
tity that has internal structure that is subject to a sequential process that
can be construed as having a starting point. Thus, while a dictionary is a
book, that can, under certain novel contexts, be construed as an entity
that can be read sequentially, (38b) is less semantically anomalous that
(38c).6
Interestingly, notice that where another form has a somewhat related
lexical concept as when two lexical concepts collocate: start the car, the
lexical concept encoded by begin cannot be used: ?begin the car, unless
by ‘begin the car’, we mean something like ‘begin its construction’.
The point then, is that both the lexical profile associated with lexical
concepts and the semantic value of lexical concepts, as defined, are neces-
sary for understanding the phenomenon of semanticality and semanti-
cally anomalous composition.

5. Conclusion
In this paper I have made a number of proposals in order to de-
velop a cognitively-realistic account of lexical semantics and meaning-
construction: the Theory of Lexical Concepts and Cognitive Models,
and thus to develop an account which is consonant with the facts of lan-
guage use. I argued that meaning is not a property of words, but rather of
the utterance: that is, a function of situated use. Words, as such, don’t
have ‘meanings’. The representational aspects of language that contribute
to meaning involve two dimensions: lexical representations, including ac-
cess to non-linguistic, conceptual knowledge, and a cognitively-realistic
account of compositionality. I modelled lexical representation by devel-
oping the construct of the lexical concept, and the conceptual structures
that lexical concepts provide access to. Lexical concepts are relatively
528 V. Evans

Figure 6. Meaning-construction in LCCM theory

schematic units of knowledge which are relativised to (and thus provide

access to) conceptual knowledge at particular ‘sites’ in the knowledge sys-
tem. Conceptual knowledge is organised into cognitive models which form
an encyclopaedic knowledge network. Lexical concepts are integrated,
 Lexical concepts, cognitive models and meaning-construction 529

guided by a number of principles, giving rise to utterance meaning: a

conception.
The significance of LCCM theory developed here is that we are
provided with a distinction between meaning and representation. More-
over, the present approach provides a way of connecting the concerns
of lexical and compositional semantics with cognitive approaches to
grammar (e.g., Goldberg 1995; Langacker 1987) and the conceptual pro-
cesses which underpin these, as studied, for instance by Fauconnier and
Turner (e.g., 2002). Current and future research aims to develop the
approach, apply it to a wider array of data, and develop a lexically-based
account of metaphor and metonymy (e.g., Evans and Zinken to appear).
In particular, one important area that I have not addressed in any detail
in this paper is the nature of the lexical profile associated with lexical con-
cepts (although see Evans 2004a). In particular, future research aims to
examine the formal (i.e., syntactic) reflexes of the semantic value of lexi-
cal concepts such as the constructs of relationality, temporal structure
and referentiality (although see Goldberg 1995; and particularly Lan-
gacker 1987).
Communication employing language can succeed or fail because of the
complex possibilities involved in meaning-construction. While I have at-
tempted to sketch some of the linguistic processes involved, I have neces-
sarily missed out the sorts of interpretative principles that relate to infer-
ential processes as discussed in the ‘pragmatic’ tradition ranging from
work by scholars such as Searle (e.g., 1969), to Grice (e.g., 1975), and
particularly Sperber and Wilson (e.g., 1995). I have also not addressed
the ‘social’ psychological perspective as represented in work on social
roles, contexts, and settings ranging from that of Erving Go¤man (e.g.,
1981) to the role of speaker/hearer interaction, and common knowledge
and context as developed in the work of Herbert Clark (e.g., 1996), or
Chris Sinha (e.g., In print). I have also not addressed in any detail the
contribution of closed-class structures such as grammatical constructions,
as addressed in the Construction Grammar tradition (e.g. Goldberg 1995,
2006; Michaelis 2003). Nor have I addressed the role of contextualisation
cues, including colloquial language use, and intonation, etc., as repre-
sented in the interactional sociolinguistics tradition associated with the
work of John Gumperz (e.g., 1982), and the discourse-based work of
scholars such as Deborah Schi¤rin (e.g., 1987), and Wallace Chafe (e.g.,
1994). A fuller account of meaning-construction must at least include all
these things.
Finally, the research presented here is programmatic. Accordingly, it
is fitting that I conclude with a caveat. A significant portion of the
account of meaning-construction presented here rests on the construct
530 V. Evans

of the cognitive model profile. As cognitive models are claimed to be psy-

chological entities, we require a fully fleshed out psychologically-based
account. Moreover, I have presented no experimental evidence for the
di¤erent levels of constructs I have posited. Clearly, psycholinguistic evi-
dence will be required in order to support the theory I have presented here.

Received 1 August 2005 University of Brighton, UK

Revision received 16 May 2006

Notes
* I gratefully acknowledge detailed comments on an earlier version of this paper by Alan
Cienki, Malgorzata Fabiszak, Shane Lindsay, and two anonymous reviewers for Cogni-
tive Linguistics. I am also indebted to the 2005–6 cohort of students on the MA in Cog-
nitive Linguistics at the University of Sussex, who worked through many of the ideas
presented here with me. Some of the proposals in this paper have also benefitted from
discussions with Jörg Zinken. Many of the specific theoretical suggestions have been
presented at a number of venues. These include the Department of Psychology seminar
series, University of Portsmouth, May 2005; the International Cognitive Linguistics
Conference held in Seoul, South Korea, July 2005; the New Insights in Semantics and
Lexicography (NISL) conference, held in Lublin, Poland, September 2005; the New Di-
rections in Cognitive Linguistics (NDCL) conference held at the University of Sussex,
October 2005; and the MetNet workshops 1 & 2, funded by the ESRC, which were
held at the Universities of Leeds and York in May 2006. I am grateful to members of
the audiences at these venues for their feedback which has informed the version of
LCCM theory presented here. Finally, I gratefully acknowledge the support and advice
of Adele Goldberg. The author is a‰liated to the University of Brighton. Author’s
E-mail address: 3vyv.evans@brighton.ac.uk4.
1. As Croft (2000) notes, this is precisely why language change is possible, and proceeds as
rapidly (in relative terms) as it does.
2. See Evans (2005) and Tyler and Evans (2001, 2003) for detailed discussion of polysemy.
3. Identifying such selectional tendencies is ultimately an empirical question. Important
techniques in this regard have been developed recently in corpus linguistics. See for in-
stance Gries (2006) and Stefanowitsch and Gries (2003).
4. The temporal compression variant of duration associated with time can also be struc-
tured in terms of motion events which relate to a lack of perceptual awareness, such as
the following: Where has the time gone? The time seemed to have vanished, etc.
5. Note that by non-relational I have in mind valence properties. Nouns are conceptually
‘independent’ in that they refer to conceptually autonomous entities. In Gentner and
Borodsitsky’s (2001) terms they are cognitively rather than linguistically oriented.
6. It has been pointed out to me by Shane Lindsay that the example in (38c) would not be
anomalous in the context of a sculpture.

References
Allwood, Jens
2003 Meaning potentials and context: Some consequences for the analysis of vari-
ation in meaning. In H. Cuyckens, R. Dirven and J. Taylor (eds.). Cognitive
 Lexical concepts, cognitive models and meaning-construction 531

Approaches to Lexical Semantics. Berlin/New York: Mouton de Gruyter,

29–66.
Atkins, Beryl T.
1987 Semantic ID tags: Corpus evidence for dictionary senses. Proceedings of the
Third Annual Conference of the UW Centre for the New Oxford English Dic-
tionary, 17–36.
Barsalou, Lawrence
1992a Cognitive Psychology: An Overview for Cognitive Scientists. Hillsdale, NJ:
Lawrence Erlbaum.
1992b Frames, concepts and conceptual fields. In Adrienne Lehrer and Eva Kittay
(eds.). Frames, Fields and Contrasts. Hillsdale, NJ: Lawrence Erlbaum,
21–74.
1999 Perceptual symbol systems. Behavioral and Brain Sciences 22, 577–609.
2003 Situated simulation in the human conceptual system. Language and Cogni-
tive Processes 5(6), 513–562.
Bergen, B. K. and N. C. Chang
2005 Embodied construction grammar in simulation-based language understand-
ing. In J.-O-Östman and M. Fried (eds.), Construction Grammars. Amster-
dam: Benjamins, 147–190.
Chafe, Wallace
1994 Discourse, Consciousness and Time: The Flow and Displacement of Conscious
Experience in Speaking and Writing. Chicago: Chicago University Press.
Clark, Herbert
1983 Making sense of nonce sense. In Levelt, W. J. M. and R. Jarvella (eds.), The
Process of Language Understanding. New York: Wiley, 297–331.
1996 Using Language. Cambridge: Cambridge University Press.
Coulson, Seana
2000 Semantic Leaps: Frame-shifting and Conceptual Blending in Meaning Con-
struction. Cambridge: Cambridge University Press.
Croft, William
1993 The role of domains in the interpretation of metaphors and metonymies.
Cognitive Linguistics 4, 335–370.
2000 Explaining Language Change: An Evolutionary Approach. London:
Longman.
Croft, William and D. Alan Cruse
2004 Cognitive Linguistics. Cambridge: Cambridge University Press.
Cruse, D. Alan
2002 Aspects of the micro-structure of word meanings. In Ravin, Y. and C. Lea-
cock (eds.), Polysemy: Theoretical and Computational Approaches. Oxford:
Oxford University Press, 30–51.
Cuyckens, Hubert, Dominiek Sandra and Sally Rice
1997 Towards an empirical lexical semantics. In Birgit Smieja and Meike Tasch
(eds.), Human Contact Through Language and Linguistics. Frankfurt: Peter
Lang, 35–54.
Evans, Vyvyan
2004a The Structure of Time: Language, Meaning and Temporal Cognition. Am-
sterdam: John Benjamins.
2004b How we conceptualise time. Essays in Arts and Sciences 33(2), 13–44.
2005 The meaning of time: Polysemy, the lexicon and conceptual structure. Jour-
nal of Linguistics 41(1), 1–39.
532 V. Evans

Evans, Vyvyan and Melanie Green

2006 Cognitive Linguistics: An Introduction. Edinburgh: Edinburgh University
Press.
Evans, Vyvyan and Jörg Zinken
to appear Figurative language in a modern theory of meaning construction: A lexical
concepts and cognitive models approach. In C. Makris and R. Christley
(eds.), Art, Body and Embodiment. Cambridge: Cambridge Scholars Press.
Available on-line at http://www.vyvevans.net.
Fauconnier, Gilles
1997 Mappings in Thought and Language. Cambridge: Cambridge University
Press.
Fauconnier, Gilles and Mark Turner
2002 The Way We Think: Conceptual Blending and the Mind’s Hidden Complex-
ities. New York: Basic Books.
Fillmore, Charles
1982 Frame semantics. The Linguistic Society of Korea (ed.), Linguistics in the
Morning Calm. Hanshin, 111–137.
1985 Frames and the semantics of understanding. Quaderni di Semantica 6,
222–254.
Fillmore, Charles, and Beryl T. Atkins
1992 Toward a frame-based lexicon: The semantics of risk and its neighbors’. In
Lehrer, A. and E. F. kittay (eds.), Frames, Fields and Contrasts. Hillsdale,
NJ: Lawrence Erlbaum, 75–102.
Gentner, Dedre and Lera Boroditsky
2001 Individuation, relational relativity and early word learning’. In Bowerman
M. and S. Levinson (eds.), Language Acquisition and Conceptual Develop-
ment. Cambridge: Cambridge University Press.
Gibbs, Raymond W.
1994 The Poetics of Mind. Cambridge: Cambridge University Press.
Go¤man, Erving
1981 Forms of Talk. Philadelphia: University of Pennsylvania Press.
Goldberg, Adele
1995 Constructions: A Construction Grammar Approach to Argument Structure.
Chicago: Chicago University Press.
2006 Constructions at Work: The Nature of Generalization in Language. Oxford:
Oxford University Press.
Grice, H. Paul
1975 Logic and conversation. In P. Cole and J. L. Morgan (eds.), Syntax and Se-
mantics 3: Speech Acts. New York, NY.: Academic Press, 41–58.
Gries, Stefan Th.
2006 Corpus-based methods and cognitive semantics: The many meanings of to
run. In Gries, Stefan Th. and Anatol Stefanowitsch (eds.), Corpora in Cogni-
tive Linguistics. Corpus-Based Approaches to Syntax and Lexis. Berlin/New
York: Mouton de Gruyter.
Gumperz, John
1982 Discourse Strategies. Cambridge: Cambridge University Press.
Haiman, John
1980 Dictionaries and encyclopedias. Lingua 50, 329–357.
Herskovits, Anette
1986 Language and Spatial Cognition. Cambridge: CUP.
 Lexical concepts, cognitive models and meaning-construction 533

Kendon, Adam
2004 Gesture: Visible Action as Utterance. Cambridge: Cambridge University
Press.
Lako¤, George
1987 Women, Fire and Dangerous Things: What Categories Reveal About the
Mind. Chicago: University of Chicago Press.
Langacker, Ronald
1987 Foundations of Cognitive Grammar, vol. I. Stanford: Stanford University
Press.
2000 A dynamic usage-based model. In M. Barlow and S. Kemmer (eds.),
Usage-based Models of Language. Stanford, CA.: CSLI Publications, 1–
64.
Michaelis, Laura
2003 Word meaning, sentence meaning, and syntactic meaning. In Cuyckens, H.,
R. Dirven and J. Taylor (eds.), Cognitive Approaches to Lexical Semantics.
Berlin/New York: Mouton de Gruyter, 163–210.
Prinz, Jesse
2002 Furnishing the Mind: Concepts and their Perceptual Basis. Cambridge, MA:
MIT Press.
Pustejovsky, James
1995 The Generative Lexicon. Cambridge, MA: MIT Press.
Searle, John
1969 Speech Acts: An Essay in the Philosophy of Language. Cambridge: Cam-
bridge University Press.
Schi¤rin, Deborah
1987 Discourse Markers. Cambridge: Cambridge University Press.
Schmid, Hans-Jörg
2000 English Abstract Nouns as Conceptual Shells: From Corpus to Cognition.
Berlin/New York: Mouton de Gruyter.
Sharifian, Farzad
2003 On cultural conceptualisations. Journal of Cognition and Culture 3(3),
187–207.
Sinha, Chris
In print Cognitive linguistics, psychology and cognitive science. In Geeraerts, Dirk
and Hubert Cuyckens (eds.), Handbook of Cognitive Linguistics. Oxford:
Oxford University Press.
Sperber, Dan and Deirdre Wilson
1995 Relevance: Communication and Cognition, 2nd ed. Oxford: Blackwell.
Stefanowitsch, Anatol and Stefan Th. Gries
2003 Collostructions: Investigating the interaction between words and construc-
tions. International Journal of Corpus Linguistics 8 (2), 209–243.
Sweetser, Eve
1999 Compositionality and blending: Semantic composition in a cognitively
realistic framework. In Janssen T. and G. Redeker (eds.), Cognitive Linguis-
tics: Foundations, Scope and Methodology. Berlin/New York: Mouton de
Gruyter, 129–162.
Talmy, Leonard
1977 Rubber sheet cognition in language. Chicago Linguistics Society 13.
Tomasello, Michael
1999 The Cultural Origins of Human Cognition. Harvard: Harvard University Press.
534 V. Evans

2003 Constructing a Language: A Usage-based Theory of Language Acquisition.

Harvard: Harvard University Press.
Tyler, Andrea and Vyvyan Evans
2001 Reconsidering prepositional polysemy networks: The case of ‘over’. Lan-
guage 77(4), 724–765.
2003 The Semantics of English Prepositions: Spatial Scenes, Embodied Meaning
and Cognition. Cambridge: Cambridge University Press.
Złatev, Jordan
1997 Situated Embodiment: Studies in the Emergence of Spatial Meaning. Stock-
holm: Gotab.
2003 Polysemy or generality? Mu. In Cuyckens H., R. Dirven and J. Taylor
(eds.), Cognitive Approaches to Lexical Semantics. Berlin/New York: Mou-
ton de Gruyter, 447–494.
Zwaan, Rolf
2004 The immersed experiencer: Toward an embodied theory of language com-
prehension. In Ross, B. H. (ed.), The Psychology of Learning and Motiva-
tion, vol. 44. New York: Academic Press, 35–62.