4 Introduction

in semantics, syntax, morphology and phonology are built up from our cogni-
tion of specific utterances on specific occasions of use. This inductive process of
abstraction and schematization does not lose the conventionalized subtleties and
differences found among even highly specific grammatical constructions and word
meanings.
As we noted above, this hypothesis is a response to approaches to syntax and
semantics in which highly general and abstract schemas and categories, sometimes
claimed to be innately given, are assumed to govern the organization of linguistic
knowledge, and apparently idiosyncratic or anomalous patterns are relegated to
the periphery. Instead, cognitive linguists argue that the detailed analysis of subtle
variations in syntactic behavior and semantic interpretation give rise to a different
model of grammatical representation that accommodates idiosyncratic as well as
highly general patterns of linguistic behavior (see, e.g., the arguments in chapter 9).
In semantics, this model is manifested in Fillmore’s semantics of understanding
(chapter 2), and Cruse’s dynamic construal approach to categorization (chapter 4
and Part II; see also Croft 2000:99–114). In syntax, this hypothesis has given rise
directly to construction grammar as a new theory of syntax, and the usage-based
model, developed in greatest detail for morphology and phonology. These models
of syntax and morphology are described in Part III of this book.
PA RT I

A conceptual approach to
linguistic analysis
 2
Frames, domains, spaces: the
organization of conceptual structure

2.1 Arguments for frame semantics

What is it that words denote, or symbolize as cognitive linguists usually

put it? A simple assumption that has guided much research in semantics is that
words denote concepts, units of meaning. Concepts symbolized by words such
as stallion and mare can be compared and contrasted with one another. Compar-
isons of words is the approach taken by structural semantics, which analyzes
types of semantic relations among words, including hyponymy and antonymy.
Some approaches to (lexical) semantics have proposed that word concepts such
as STALLION and MARE1 are not atomic. Many concepts can be broken down
into semantic features, so that STALLION is [EQUINE, MALE], and MARE is
[EQUINE, FEMALE]. Finally, in the logical tradition that underlies much work
in semantics, concepts are ultimately defined by their truth conditions: the con-
ditions under which one can say that a concept does, or does not, appropriately
apply to a situation in the world.
In this widespread approach to semantics, it is recognized that concepts do not
simply float around randomly in the mind. First, there are the relations between
words and their corresponding concepts described by structural semantics. But
there has been a strong feeling that concepts are organized in another way as well.
Certain concepts ‘belong together’ because they are associated in experience. To
use a classic example (Schank and Abelson 1977), a RESTAURANT is not merely
a service institution; it has associated with it a number of concepts such as CUS-
TOMER, WAITER, ORDERING, EATING, BILL. These concepts are not related
to RESTAURANT by hyponymy, meronymy, antonymy or other structural seman-
tic relations; they are related to RESTAURANT by ordinary human experience.
The concept of RESTAURANT is closely tied to the other concepts, and cannot
be isolated from the other concepts.

1 We follow the practice of Fillmore (1982a) and Langacker (1987) in using lower-case italics to
represent the word form, and capitals to represent the concept underlying the word meaning.

7
8 A conceptual approach to linguistic analysis

The need for another means for organizing concepts has been felt by researchers
in cognitive psychology and artificial intelligence as well as in various branches
of linguistics, and has led to a variety of similar proposals, each typically with its
own name. Among these names are: frame, schema, script, global pattern, pseudo-
text, cognitive model, experiential gestalt, base, scene (Fillmore 1985:223, n. 4).
The most influential version of this proposal in cognitive linguistics has been the
model of frame semantics developed by Fillmore. We present Fillmore’s theory
and arguments in this section, and turn to extensions of Fillmore’s ideas by other
cognitive linguists in later sections.2
Fillmore views frames not as an additional means for organizing concepts, but as
a fundamental rethinking of the goals of linguistic semantics. Fillmore describes his
frame semantic model as a model of the semantics of understanding, in contrast
to a truth-conditional semantics: the full, rich understanding that a speaker intends
to convey in a text and that a hearer constructs for that text. Fillmore argues that in
the analysis of linguistic meaning, understanding is the primary data; truth-value
judgments and judgments of semantic relations such as synonymy and implication
are derivative and theory-driven (Fillmore 1985:235). Fillmore’s frame semantics
brings linguistic semantics back to that primary data and does not exclude any of
it from consideration.
Fillmore uses a tool metaphor to describe the understanding process (Fillmore
1982a:112): a speaker produces words and constructions in a text as tools for a
particular activity, namely to evoke a particular understanding; the hearer’s task
is to figure out the activity those tools were intended for, namely to invoke that
understanding. That is, words and constructions evoke an understanding, or more
specifically a frame; a hearer invokes a frame upon hearing an utterance in order
to understand it.
Fillmore uses a wide range of examples to demonstrate that there are signifi-
cant phenomena in linguistic semantics that cannot easily be captured in a model
of structural semantics, semantic features and/or truth-conditional semantics. We
survey his arguments here.
The analysis of semantic features is often justified on the basis of lexical sets
that appear to be analyzable in terms of a simple set of features. For example,
the lexical set in (1) can be analyzed in terms of the features [MALE/FEMALE],
[ADULT/YOUNG], and [UNMARRIED]:

(1) [MALE] [FEMALE]

MAN WOMAN [ADULT]
BOY GIRL [YOUNG]
BACHELOR SPINSTER [UNMARRIED]

2 The basic sources for Fillmore’s ideas are Fillmore 1975, 1977 (an expanded version of the first
paper), 1982a, 1985, 1986. Unfortunately all of these are difficult to access.
 Frames, domains, spaces 9

Yet our understanding of these concepts is more complex than this paradigm of
feature constrasts implies. The relation between man/boy and woman/girl is not the
same: for many people, the term girl is used for female humans at a significantly
higher age than the term boy is used for male humans (Fillmore 1982a:126).
Moreover, the attitudes towards the sexes that this linguistic behavior is assumed
to evoke has led to changes in the relationship and hypercorrection such that
the term woman is attested as being applied even to an eight-year-old girl (ibid.,
127). In a frame semantic analysis, man, boy, woman and girl evoke frames that
include not just the biological sexual distinction but also differences in attitudes
and behavior towards the sexes that would explain the traditional asymmetry in
the use of boy/girl and the more recent change in the use of woman, including
its hypercorrective use. Likewise, the difference between our understanding of
bachelor and our understanding of spinster involves much more than a simple
feature [MALE/FEMALE] (ibid., 131).
Many lexical contrasts contain semantic asymmetries that cannot be captured
by features (except in an ad hoc fashion), but lend themselves easily to a frame
semantic account. For example, the opposing terms used for the vertical extent of
an erect human being are tall and short, for vertical distance from a bottom baseline
(e.g. a branch of a tree) they are high and low, but for the vertical dimension of
a building they are tall and low (Fillmore 1977a:71). It would be difficult if not
impossible to come up with a unitary feature definition of tall that captured its
different contexts of use from high, and did the same for short vs. low. Instead,
one can simply describe the frames for humans, buildings and other objects, and
specify which words are used for vertical extent or distance in that frame.
Similarly, no simple unitary definitions would capture the contrast between the
adjectives live and alive given in (2)–(4) (Fillmore 1977a:76–77):

(2) a. Those are live lobsters.

b. Those lobsters are alive.
(3) a. Her manner is very alive.
b. She has a very alive manner.
(4) a. His performance was live.
b. He gave a live performance.

Moreover, one cannot define the features in terms of applicability to a semantic

class, such that the sense illustrated in (2a–b) applies to living things; this would
give an incorrect understanding to the theatre advertizing live naked girls than
the one intended (presumably, as opposed to naked girls on a film screen, not
dead naked girls; ibid.). In a frame semantic analysis, live and alive are simply
associated in different ways to three different frames: life in (2), personality in (3),
and mode of performance in (4). In other cases, there are outright lexical splits,
such as brother/brothers and brother/brethren, which represent a split in frames
10 A conceptual approach to linguistic analysis

including different plural forms; a unitary definition of brother would miss the
frame contrast (ibid., 76).
Fillmore notes that his frame semantic model shares significant properties with
lexical (semantic) field theory (Fillmore 1985:225–26; 1992:76–77). Lexical field
theory groups together words that are associated in experience, not unlike frame
semantics. However, lexical field theory differs from frame semantics in that words
are defined relative to other words in the same lexical field, whereas in frame
semantics, words are defined directly with respect to the frame. For example, in
lexical field theory, one would observe that large in the field of sizes of packages
of soapflakes is in contrast with jumbo, economy giant and family size and hence
describes the smallest size in the field, unlike uses of large in other lexical fields
(Fillmore 1985:227).
In frame semantics, the same observation can easily be captured: large labels
the smallest size in the SOAPFLAKES frame. But lexical field theory predicts that
the meaning of a word in a field can only be defined in contrast to neighboring
words in the field. Lexical field theory has difficulties if there are no neighboring
words, or a speaker does not know the neighboring words: it predicts that the term
has a different meaning. Fillmore notes that while German has a word for the
sides of a right angle triangle other than the Hypotenuse, namely Kathete, most
English speakers do not have such a word (ibid., 228–29). Yet the understanding
of English hypotenuse and German Hypotenuse is the same, provided the speaker
understands what a right angle triangle is. This is not a problem in frame semantics,
where the word concept is linked directly to the frame, in this case the RIGHT
ANGLE TRIANGLE frame.
Another argument in favor of a frame-based approach to lexical semantics are
words whose corresponding concepts inherently refer to other concepts extrinsic
to the concept denoted by the word. Some word concepts refer to a prior history of
the entity denoted. A scar is not just a feature of the surface of someone’s skin, but
the healing state of a wound; a widow is a woman who was once married but whose
husband has died (Fillmore 1977a:73). Other word concepts, especially for prop-
erties and actions, cannot be understood without understanding something about
the participant in the action or possessor of the properties: one cannot understand
gallop without knowing about the body of a horse, or hungry without understand-
ing the physiology of living things (ibid., 73–74). This is true of object concepts
as well: lap cannot be understood except in reference to a person’s posture and the
function of one’s lap in supporting another object (ibid.).
Another clear class of examples that requires reference to extrinsic entities are
deictic expressions that evoke the speech act situation (Fillmore 1982a:117). For
example, the past tense situates an event in a point or interval or time relative to
the speech act situation. The speech act situation, including its time of occurrence,
 Frames, domains, spaces 11

functions as the frame against which past time reference is profiled. Likewise, all
other deictic words and inflections, such as person deixis (I, you, he/she/it, we, they
and person-based agreement inflections) and spatial deixis (this, that, here, there),
evoke the speech act situation. Other types of grammatical words and inflections
also have meanings evoking the speech act situation. For example, the definite
articles the and a define the identity of the noun referent relative to the mutual
knowledge of speaker and hearer (the basically indicates mutually known, a not
mutually known, in most contexts). The meanings of the and a evoke the speech
act situation because they make reference to the mental states of speaker and hearer
(see also §3.4).
Above all, many word concepts cannot be understood apart from the intentions
of the participants or the social and cultural institutions and behavior in which
the action, state or thing is situated. For example, the concept VEGETARIAN
only makes sense in the frame of a culture in which meat-eating is common; the
concepts STRIKE or BORROW can only be understood in the frame of a culture
in which such actions occur (Fillmore 1982a:120). Even something as simple as
an apple core evokes a frame describing a particular way of eating apples: ‘an
apple-core is that part of the apple that somebody who eats apples the way most
of us do has left uneaten’ (Fillmore 1977a:73).
Another respect in which a word meaning makes reference to extrinsic entities is
that a word allows the speaker and hearer to focus their attention on only part of an
entire frame; no one word gives the full structure of the frame. The classic example
is the commercial transaction frame (Fillmore 1977a:58–59; 1977b); but a much
clearer case is the RISK frame (Fillmore and Atkins 1992). Fillmore and Atkins
identify the following elements of the RISK frame: Chance (uncertainty about the
future), Harm, Victim (of the Harm), Valued Object (potentially endangered by the
risk), Situation (which gives rise to the risk), Deed (that brings about the Situation),
Actor (of the Deed), (Intended) Gain (by the Actor in taking a risk), Purpose (of
the Actor in the Deed), Beneficiary and Motivation (for the Actor). The verb risk
occurs in many syntactic constructions, some of which are exemplified in (5a–e),
but none of them include all or even most of the elements of the RISK frame
(Fillmore & Atkins 1992: 83, 87, 89, 94, 96; all but the first are corpus examples):

(5) a. You’ve (Actor/Victim) risked your health (Valued Object) for a few cheap thrills
(Gain).
b. Others (Actor/Victim) had risked all (Valued Object) in the war (Situation).
c. She (Actor/Victim) had risked so much (Valued Object) for the sake of vanity
(Motivation).
d. Men (Actor/Victim) were not inclined to risk scalping (Harm) for the sake of
settlers they had never seen (Beneficiary).
e. I (Actor/Victim) didn’t dare risk a pause (Deed) to let that sink in (Purpose).
12 A conceptual approach to linguistic analysis

In a frame semantic analysis, any of the uses of risk evokes the entire RISK frame,
even if only part of that frame is overtly focused on by the construction in which
risk is used.
The semantics of understanding also allows Fillmore to account for linguistic
facts that do not lend themselves to a truth-functional analysis. For example, the
collocations in (6) could be reversed as in (7) without producing semantic anomaly
(Fillmore 1977a:75–76):

(6) a. A dog was barking.

b. A hound was baying.
(7) a. A dog was baying.
b. A hound was barking.

In other words, the difference between (6) and (7) cannot be accounted for by
semantic constraints. But the examples in (6a–b) sound much more natural because
the noun and the verb in each sentence both evoke the same frame.
Likewise, a truth-conditional semantics cannot capture many aspects of our
understanding of (8) (Fillmore 1985:230–31):

(8) My dad wasted most of the morning on the bus.

Fillmore notes that choosing father or dad (without the possessive) would express
a different relationship between the speaker and the speaker’s father; the morning
is understood to be defined against the frame of the working day (i.e, around 8am
to noon) rather than the calendar day (midnight to noon); waste frames the use of
time very differently from spend; and on the bus frames the speaker’s location in
terms of the bus being in service, rather than simply a physical container (which
would be evoked by in the bus).
A truth-conditional model also cannot account for the anomaly of frames that
are appropriate at one time of utterance but not at another because the world has
changed in the meantime. Fillmore uses the contrived example in (9), noting that
it could be said in 1984 but not in, say, 1919 (Fillmore 1985:238–39):

(9) During World War I, Ronald Reagan’s birth mother dropped his analog watch
into the sound hole of the acoustic guitar.

Such a sentence could be uttered in 1984, because World War II had occurred,
allowing the 1914–18 war to be renamed World War I; medical technology had
allowed the dissociation of the birth mother from the genetic mother (who donates
the egg); and electric guitars and digital watches had been invented. None of these
framings of the objects, persons or events was available in 1919, and so (9) would
be an impossible utterance at that time, even if true retrospectively.
Finally, frame semantics offers a natural account of a number of problematic
phenomena that seem to be caught between semantics and pragmatics, including
 Frames, domains, spaces 13

the nature of text coherence. A large class of presuppositional phenomena appear

to be tied to specific words, such as regret in (10) (Fillmore 1985:249):

(10) John regretted signing the letter.

Example (10) denotes (or entails) a particular mental state of John, namely his
feeling of regret, but (in one analysis) is said to presuppose that John did sign
the letter. The problem for truth-conditional semantics is that if John did not sign
the letter, (9) has no truth conditions. In order to preserve truth conditions one
may shunt the presupposition problem off to pragmatics. This seems odd since
the presupposition is associated with a particular word and its meaning, which
is semantic. But presuppositions display a further type of peculiar behavior, in
negative sentences such as (11):

(11) John didn’t regret signing the letter.

In one interpretation, the entailment is negated – John does not have any regrets –
but the presupposition is not – John signed the letter. However, there is another
interpretation of (11), namely that John did not regret signing the letter because
he did not sign it (Fillmore 1985:251).
Fillmore argues that the behavior of presuppositions can be easily accounted
for in a frame semantic analysis. The concept REGRET includes in its frame the
accomplishment of an action towards which the regretter has his/her regrets. If
the action is absent from the frame, understanding of the positive sentence fails.
Negation, on the other hand, can negate either the concept denoted or the frame
itself. Negating the state of affairs in the frame preserves the rest of the frame,
including the action that could have led to the regrets. This is the first interpretation
of (11) described above. Negating the entire frame, on the other hand, also negates
the action that could have led to the regrets. This is the second interpretation of
(11): the speaker denies the framing of the situation as including the action of John
having signed the letter.
We may compare (11) to other examples of frame negation, such as the one in
(12) (Fillmore 1985:245, from Wilson 1975:138):

(12) You didn’t spare me a day at the seaside: you deprived me of one.

In (12), the speaker denies the positively evaluated framing of the action as sparing
her, and replaces it with the negatively evaluated framing of depriving her.
Finally, the semantics of understanding plays a major role in text understand-
ing. For example, the well-known example of initial definite reference, e.g. the
carburetor in (13) (described as ‘bridging’ by Clark and Haviland [1977] and
‘evoked’ by Prince [1981a]), are in fact due to the frame evoked by the first
sentence (Fillmore 1977a:75):
14 A conceptual approach to linguistic analysis

(13) I had trouble with the car yesterday. The carburetor was dirty.

The car in the first sentence evokes a frame that allows the hearer to identify which
carburetor of the millions in the world the speaker was referring to. But frame
semantics contributes more than the resolution of definite reference to the analysis
of the coherence of texts. Fillmore contrasts (13) with (14) (ibid.):

(14) I had trouble with the car yesterday. The ashtray was dirty.

The second sentence in (14) is incoherent with the first, even though the definite
reference can be resolved (most cars have ashtrays). The reason for this is that there
is nothing evoked in the frame of having trouble with the car that has anything to
do with the ashtray – unlike (13), because dirty carburetors do cause problems for
cars.
Fillmore’s arguments present a wide range of data that justify the introduction
of frames to the analysis of linguistic semantics, and the replacement of a truth-
conditional semantics with a semantics of understanding. In the following sections,
we lay out more systematically the frame semantic model and follow its further
development in cognitive linguistics.

2.2 Concepts: profile-frame organization

In the preceding section, we described the frame as a coherent region

of human knowledge, or as a coherent region of conceptual space. The question
immediately arises: How does one identify a coherent region of conceptual space,
differentiating it from other regions? An a priori approach to this question, us-
ing one’s own intuitions to identify frames, would be highly subjective. A more
empirical approach to this question is to identify frames based on the words and
constructions of a human language such as English. This approach is taken by
Langacker (1987), which we will use as our starting point.
Langacker illustrates his approach to the problem with the meaning of the word
radius. The word form radius symbolizes (denotes) the concept RADIUS. We
begin here by assuming that concepts correspond to meanings of linguistic units
(words, complex expressions or constructions). One may also assume that concepts
exist that do not correspond to linguistic meanings. However, one would have the
same problems trying to identify concepts independent of linguistic meanings as
trying to identify frames independent of linguistic meanings, namely the lack of an
empirical basis for doing so. For this reason, we will restrict ourselves to concepts
corresponding to actual linguistic meanings in this chapter.
The first sense for radius in the American Heritage Dictionary is ‘a line segment
that joins the center of a circle with any point on its circumference.’ A RADIUS is