Gilles Fauconnier2 and Mark Turner3
1. Introduction
Conceptual blending is a basic mental operation that leads to new
meaning, global insight, and conceptual compressions useful for
memory and manipulation of otherwise diffuse ranges of meaning. It
plays a fundamental role in the construction of meaning in everyday
life, in the arts and sciences, and especially in the social and
behavioral sciences. The essence of the operation is to construct a
The present paper presents a synopsis of work on language, form, and meaning in
conceptual integration theory. It consists of excerpts and summaries from
G. FAUCONNIER, "Conceptual Blending" in N. SMELSER and P. BALTES (eds.), The
Encyclopedia of the Social and Behavioral Sciences, 2000; G. FAUCONNIER and M.
TURNER , “Blending as a Central Process of Grammar”, in A. GOLDBERG (ed.),
Conceptual Structure and Discourse, Stanford, CSLI Publications, 1996;
G. FAUCONNIER and M. TURNER , The Way We Think: Conceptual Blending and the
Mind's Hidden Complexities, New York, Basic Books, 2002; M. T URNER and G.
FAUCONNIER, “Conceptual Integration and Formal Expression”, Journal of
Metaphor and Symbolic Activity, 1995. We thank Pierre Fastrez for his help in
choosing and putting together the appropriate materials.
Professor at the Department of Cognitive Science, University of California, San
Associate Director of the Center for Advanced Study in Behavioral Sciences,
Stanford University.
Recherches en communication, n° 19 (2003).
partial match between two input mental spaces, to project selectively
from those inputs into a novel 'blended' mental space, which then
dynamically develops emergent structure. Mental spaces are small
conceptual packets constructed as we think and talk, for purposes of
local understanding and action –they are very partial assemblies
containing elements, structured by frames and cognitive models. It
has been suggested that the capacity for complex conceptual blending
("double-scope" integration) is the crucial capacity needed for thought
and language.
In this article, we will look at language (and more specifically
grammar) as a culturally entrenched means of creating and
transmitting blending schemes. We will do so by considering the
relationships between linguistic forms and patterns of meaning
construction through conceptual integration and compression.
1.1. A simple example of conceptual blending: the boat race
A famous example of blending is "the boat race" or "regatta". A
modern catamaran is sailing from San Francisco to Boston in 1993,
trying to go faster than a clipper that sailed the same course in 1853.
A sailing magazine reports:
As we went to press, Rich Wilson and Bill Biewenga were
barely maintaining a 4.5 day lead over the ghost of the clipper
Northern Light, whose record run from San Francisco to
Boston they're trying to beat. In 1853, the clipper made the
passage in 76 days, 8 hours1 .
Informally, there are two distinct events in this story, the run by
the clipper in 1853 and the run by the catamaran in 1993 on
(approximately) the same course. In the magazine quote, the two runs
are merged into a single event, a race between the catamaran and the
clipper's "ghost". The two distinct events correspond to two input
mental spaces, which reflect salient aspects of each event: the voyage,
the departure and arrival points, the period and time of travel, the
boat, its positions at various times. The two events share a more
schematic frame of sailing from San Francisco to Boston; this is a
"generic" space, which connects them. Blending consists in partially
”Great America II”, Latitude 38, vol. 190, April 1993, p. 100.
matching the two inputs and projecting selectively from these two
input spaces into a fourth mental space, the blended space:
Generic space
cross-space mapping
Input space 2
Input space 1
selective projection
Blended space
Figure 1: the four-space model
In the blended space, we have two boats on the same course, that
left the starting point, San Francisco, on the same day. Pattern
completion allows us to construe this situation as a race (by importing
the familiar background frame of racing and the emotions that go with
it). This construal is emergent in the blend. The motion of the boats is
structurally constrained by the mappings. Language signals the blend
explicitly in this case by using the expression "ghost-ship." By
"running the blend" imaginatively and dynamically –by unfolding the
race through time– we have the relative positions of the boats and
their dynamics.
Crucially, the blended space remains connected to the inputs by
the mappings, so that real inferences can be computed in the inputs
from the imaginary situation in the blended space. For example, we
can deduce that the catamaran is going faster overall in 1993 than the
clipper did in 1853, and more precisely, we have some idea ("four and
a half days") of their relative performances. We can also interpret the
emotions of the catamaran crew in terms of the familiar emotions
linked to the frame of racing.
The "boat race" example is a simple case of blending. Two
inputs share structure. They get linked by a cross-space mapping and
projected selectively to a blended space. The projection allows
emergent structure to develop on the basis of composition (blending
can compose elements from the input spaces to provide relations that
do not exist in the separate inputs), pattern completion (based on
background models that are brought into the blend unconsciously),
and elaboration (treating the blend as a simulation and “running” it
1.2. The network model
Conceptual blending is described and studied scientifically in
terms of integration networks. In its most basic form, a conceptual
integration network consists of four connected mental spaces: two
partially matched input spaces, a generic space constituted by
structure common to the inputs, and the blended space. The blended
space is constructed through selective projection from the inputs,
pattern completion, and dynamic elaboration. The blend has emergent
dynamics. It can be "run", while its connections to the other spaces
remain in place. Neurobiologically, it has been suggested that
elements in mental spaces correspond to activated neural assemblies
and that linking between elements corresponds to neurobiological
binding (e.g. co-activation). On this view, mental spaces are built up,
interconnected, and blended in working memory by activating
structures available from long-term memory. Mental spaces can be
modified dynamically as thought and discourse unfold.
Four main types of integration networks have been distinguished:
Simplex, Mirror, Single-Scope, Double-Scope. In Simplexes, one
input consists of a frame and the other consists of specific elements. A
frame is a conventional and schematic organization of knowledge
such as "buying gasoline." In Mirrors, a common organizing frame is
shared by all spaces in the network. In Single-Scopes, the organizing
frames of the inputs are different, and the blend inherits only one of
those frames. In Double-Scopes, essential frame and identity
properties are brought in from both inputs. Double-Scope Blending
can resolve clashes between inputs that differ fundamentally in
content and topology. This is a powerful source of human creativity.
The main types of networks just mentioned are actually prototypes
along a continuum that anchors our intuitive everyday notions about
meaning to a unified understanding of the unconscious processes at
work. Varieties of meaning traditionally considered unequal or even
incommensurable –categorizations, analogies, counterfactuals,
metaphors, rituals, logical framing, grammatical constructions– can
all be situated on this continuum. Conceptual blending has been
shown to operate in the same way at the highest levels of scientific,
artistic, and literary thought, and at the supposedly lower levels of
elementary understanding and sentence meaning, as we will show in
rest of this article.
In crucial respects, the construction of meaning is like the
evolution of species. It has coherent principles that operate all the
time in an extremely rich mental and cultural world. Partial crossspace mappings, selective projection to the blend, development of
emergent structure in the blend are the constitutive principles of
conceptual integration. Constitutive principles already place strong
constraints on the relevant processes, but additional governing
principles limit their scope much further. They characterize strategies
for optimizing emergent structure. Often, satisfying one goes part way
toward satisfying another, but governing principles also frequently
compete with each other. Governing principles identified to date
• Topology: Other things being equal, set up the blend and the
inputs so that useful topology in the inputs and their outer-space
relations is reflected by inner-space relations in the blend.
• Unpacking: Other things being equal, the blend all by itself should
prompt for the reconstruction of the entire network.
• Web: Other things being equal, manipulating the blend as a unit
must maintain the web of appropriate connections to the input
spaces easily and without additional surveillance or computation.
• Integration: achieve an integrated blend.
More complex integration networks ("multiple blends") allow
multiple input spaces, and successive blending in which blends at one
level can be inputs at another.
1.3. Range of the cognitive operation
We started to study conceptual blending systematically in 1993,
when we discovered the structural uniformity and wide application of
the notion. Since then, important work has been done on the theory of
conceptual blending1 , and its empirical manifestations in
mathematics 2 , social science3 , literature4 , linguistics5 , and music6 .
There have been proposals for the mathematical and computational
modeling of the operation7 , and experimental research within
neuroscience on the corresponding neural and cognitive processes8 .
Additional information and an extensive bibliography can be found on
the blending website9 .
S. COULSON, Semantic Leaps: Frame-shifting and Conceptual Blending in Meaning
Construction, New York and Cambridge, Cambridge University Press, 2001.
A. R OBERT, "Blending in the interpretation of mathematical proofs", in J.P. KOENIG (ed.), Discourse and Cognition, Stanford, Center for the Study of
Language and Information (CSLI) [distributed by Cambridge University Press].
E. S WEETSER, "Blended Spaces and Performativity", Cognitive Linguistics, 11
(3/4), 2000, pp. 305-333. M. T URNER , "Backstage Cognition in Reason and
Choice", in A. L UPIA, M. MC CUBBINS and S. POPKIN (eds.), Elements of Reason:
The Science of the Mind and the Limits of Political Rationality, Cambridge,
Cambridge University Press. M. TURNER , "Conceptual Blending and
Counterfactual Argument in the Social and Behavioral Sciences", in Ph. T ETLOCK
and A. BELKIN (eds.), Counterfactual Thought Experiments in World Politics,
Princeton, N.J., Princeton University Press, 1996.
M. FREEMAN , “Grounded spaces: Deictic-self anaphors in the poetry of Emily
Dickinson”, Language and Literature, 6/1, 1997, pp. 7-28. T. O AKLEY,
"Conceptual blending, narrative discourse, and rhetoric", Cognitive Linguistics, 9,
1998, pp. 321-360. M. TURNER , The Literary Mind, Oxford University Press, 1996.
S.K. LIDELL, "Grounded blends, gestures, and conceptual shifts", Cognitive
Linguistics, 9, 1998. N. MANDELBLIT, Grammatical Blending: Creative and
Schematic Aspects in Sentence Processing and Translation, Ph.D. dissertation, UC
San Diego, 1997.
L. Z BIKOWSKI, "The blossoms of 'Trockne Blumen': Music and text in the early
nineteenth century", Music Analysis, 18/3, October 1999, pp. 307-345. Manuscript.
Lawrence Zbikowski offers a graduate seminar on conceptual mapping and
blending in the department of music at the University of Chicago.
J. GOGUEN, "An Introduction to Algebraic Semiotics, with Application to User
Interface Design.", in C. N EHANIV (ed.), Computation for Metaphor, Analogy, and
Agents, A volume in the series “Lecture Notes in Artificial Intelligence”, Berlin,
Springer-Verlag, 1999, pp. 242-291. T. V EALE, "Pragmatic Forces in Metaphor
Use: The Mechanics of Blend Recruitment in Visual Metaphors", in C. NEHANIV,
op. cit., pp. 37-51.
S. COULSON, op. cit.; R. G RUSH and N. MANDELBLIT, "Blending in language,
conceptual structure, and the cerebral cortex", in P. Å. BRANDT, F. GREGERSEN, F.
STJERNFELT and M. SKOV (eds.), Acta Linguistica Hafniensia, vol. 29, The Roman
Jakobson Centennial Symposium: International Journal of Linguistics,
Copenhagen, C.A. Reitzel, 1997, pp. 221-237.
Website for Conceptual Blending:
1.4. Compression
One of the central benefits of conceptual blending is its ability to
provide compressions to human scale of diffuse arrays of events.
We do not establish mental spaces, connections between them,
and blended spaces for no reason. We do this because it gives us
global insight, human-scale understanding, and new meaning. It
makes us both efficient and creative. One of the most important
aspects of our efficiency, insight, and creativity is the compression
achieved through blending. Certain conceptual relations, such as time,
space, cause-effect, identity, and change, show up again and again in
compression under blending. We call these all-important conceptual
relations “vital relations”. Compression maximizes and intensifies
vital relations. Double-scope blending capacity provides human
beings with the ability to do remarkable compressions, and every
language provides a systematic array for types of compression.
Various highly useful compression patterns become conventional and
are associated with specific grammatical forms. In this article, we will
consider the general cognitive phenomenon of associating a
compression pattern with a linguistic form.
2. Double-scope compression in a two-word nutshell
We can bring two things together mentally in various ways.
Blending them is one subset, and blends that satisfy the governing
principles are a much smaller subset. The subset below that consists
of those core compression patterns that are entrenched in a culture.
The next subset down consists of those entrenched compression
patterns that have associated grammatical forms.
It is easy to think that a simple form corresponds to a simple
meaning. But blending can perform massive compressions and
express them in simple forms. So, by virtue of the power of
compression and decompression, a simple form can prompt for the
construction of an extremely complicated meaning.
Some of the apparently simplest forms in the language consist of
putting two words together: noun-noun compounds like "boat house,"
adjective-noun combinations like "angry man," noun-adjective
combinations like "child-safe or "sugar-free" .
Through decompression, we can get quite different complicated
integration networks out of words like "dolphin-safe," "shark-safe,"
and "child-safe." In all of these cases, there is successive blending.
We first blend the mental space for the current situation –involving
dolphins, sharks, or children– with an abstract frame of danger. This
yields a specific counterfactual mental space in which dolphin, shark,
or child is assigned to a role in the danger frame. This mental space of
specific harm is disanalogous with the mental space for the current
situation. These two disanalogous spaces are inputs to a new blend, in
which the disanalogy is compressed into the property safe.
"Dolphin-safe," as it currently appears on cans of tuna, means
that measures were taken to avoid harming dolphins during the
harvesting of the tuna. "Shark-safe," as applied to, say, swimming,
refers to conditions under which swimmers are not vulnerable to
attack by sharks. "Child-safe," as applied to rooms, means the rooms
are free of typical dangers for children. In every case, from simple
forms the understander must construct elaborate integration networks.
How one does this mental work may differ from case to case. In
"dolphin-safe tuna," the role of the dolphin as potential victim is taken
to be useful. In "dolphin-safe diving," said of mine-seeking human
divers who are protected by dolphins who are not themselves at risk,
the blend uses dolphins in the role of agents of the safety. In "dolphinsafe diving," said of diving that imitates the way dolphins swim and is
therefore safe, the blend uses the manner of swimming associated
with dolphins. If we assume that dolphins eat goldfish, then "dolphinsafe goldfish" would cast dolphins in the role of predators. Genetic
engineers who are concerned not to produce anything resembling a
dolphin might refer to a technique that is known never to lead to a
dolphin embryo as "dolphin-safe". The dolphin here does not fill the
role of victim, victimizer, causal agent, or role model. In a world in
which the most humiliating thing for a shark is to resemble a dolphin,
behaviors that are unquestionably sharkly might be called "dolphinsafe". And of course, a compositional theory of meaning immune to
our dolphin examples would be hailed as "dolphin-safe".
If the adjective "safe" comes before instead of after the noun
"dolphin," then we find another multiplicity of potential meanings.
"Safe dolphin" can mean a dolphin that is protected, a dolphin that
will not inflict the kind of harm other dolphins might cause, the role
of swimmer-at-the-front-of-the-school whose responsibility it is to
keep the rest of the school safe from running into obstructions, or the
decoy dolphin robot which behaves so as to communicate to other
dolphins a situation of complete safety and so lulls them into being
A more specific blending pattern for the compounding of a noun
with "safe" arises when the noun refers to an endangered species.
Now, we can talk about "turtle-safe nets" or "salamander-safe
landscaping" or "hawk-safe agriculture." In order to make appropriate
sense of these phrases, one must know the compression pattern
chosen by the culture for thinking about endangered species, and also
know that it is associated with the Noun-safe form where the Noun
picks out an endangered species. Without this knowledge, the
meaning cannot be predicted compositionally from the noun and the
adjective. What the culture ends up with in this case is a very
powerful compression –a maximally simple two-word form that
points to an Integrated blend and that satisfies the Unpacking
governing principle (i.e. that prompts for the reconstruction of the
entire integration network) by evoking mapping schemes and
counterfactual spaces that go with "safe." The particular culture that
has entrenched this compression has double-scope creativity, a
language (English), ecological concerns, food packing corporations,
and grocery stores. Double-scope creativity is universal for our
species. The language English offers a syntactic noun-adjective form
and entrenches certain blend types that go with it, like those prompted
for by Noun-safe. The ecological concerns are relatively recent,
certainly as applied to dolphins. They make salient and important the
scenario of dolphins being harmed by fishing methods, and make it
desirable that the scenario be counterfactual. Consequently, the
scenario is a good candidate for Noun-safe expression, and a good
candidate for marketing departments seeking to induce shoppers to
buy their products. This convergence of syntax and commerce creates
the compression that appears on cans of tuna fish: "dolphin-safe".
Examples like "dolphin-safe" are useful because they highlight in
a transparent and uncontroversial way, the nature of the blending
process. Furthermore, they abound. Think of "cruelty-free" on bottles
of shampoo, or the variety of noncompositional integration running
across "waterproof," "tamper-proof," "foolproof," and "child-proof,"
or "talent pool," "gene pool," "swimming pool," "football pool," and
"betting pool".
Familiar compositions like "dirt-brown," "pencil-thin," "red
pencil" and "green house" work in the identical fashion, but because
they are very deeply entrenched, it is possible to misinterpret them as
somehow different in their operations from the examples above.
Following Charles Travis, we observe that noncompositional
conceptual integration is just as necessary in these "core" cases1 . "Red
pencil" can mean a pencil whose wood has been painted red on the
outside, a pencil that leaves a red mark (the lead is red, or the
chemical in the pencil reacts with the paper to produce red, or . . .), a
pencil used to record the activities of the team dressed in red, a pencil
smeared with lipstick, not to mention pencils used only for recording
deficits. For a set of houses that differ only in location and in the color
of the kitchen linoleum, "green house" can mean the house with green
linoleum, where "green linoleum" means the one with spots that are
green, where "green spots" means spots created with a green pencil,
where "green pencil" means . . . .
The scenarios needed for these integrated meanings are no
simpler than those needed for "dolphin-safe" and "fool-proof." The
cognitive capacities needed to construct these integrated meanings are
the same as those needed to interpret the supposedly exotic examples,
and these cognitive capacities apply as well to the supposedly central
examples like "green house" used to mean a house whose exterior
walls, exclusive of windows, shutters, trim, porches, flashing,
foundations, and fascia boards, are mostly green on the weather
surface. That some interpretations stand out more than others
–especially when the phrases are taken in isolation– stems from the
existence of strong defaults. This difference has to do with the
conceptual and linguistic defaults most likely to be activated in any
given situation, not with the mechanisms of integration.
If we look across the central cases like "red pencil" and
"government bond," we find that some of the relevant defaults are
provided by cultural frames with rich structure, others by generic
roles that run across many frames, others by the local situation at the
moment of utterance. This last case includes elicitation by linguists
and philosophers: the subject is asked to judge an expression in a
supposedly context-free way, but in fact must construct a minimal
context in which to interpret it. These minimal contexts typically use
the strongest defaults.
How do we go from the linguistic units to the conceptual
elements or from the conceptual elements to the linguistic units? In
C. T RAVIS, The True and the False: The Domain of the Pragmatic, Amsterdam,
John Benjamins, 1981.
the case of nominal compounds, the formal unit names two elements
in two different spaces, and directs the understander to find the rest.
We will call these conceptual elements the named elements. Consider
"land yacht" as a reference to a large, luxurious automobile. Clearly,
"land" and "yacht" come from different domains: yachts are
associated with water as opposed to land. "Land yacht" gives us land
from one space and yacht from another, and asks us to perform a
mapping between these spaces. In this mapping, yacht corresponds to
luxury car, land corresponds to water, driver corresponds to skipper,
and the road for the car corresponds to the course for the boat.
Figure 2: Land yacht
Figure 2 shows how the conceptual blend depends on building an
analogical mapping, and how, in the corresponding integrated
syntactic form "land yacht", "land" and "yacht" name elements that
are not counterparts in the mapping. "Land yacht" now names the new
element in the blend, even though it names nothing in either of the
inputs, and although land is not a counterpart of yacht. Formal
expression, in this case a two-word combination, prompts for the
construction of the blend and provides a way of naming part of the
emergent structure.
Consider "Language is fossil poetry". "Fossil poetry" works just
like "land yacht": fossil comes from the domain of paleontology and
poetry from the domain of expression. In the mapping, poetry
corresponds to the living organism, while language corresponds to the
fossil of that organism. The conceptual elements named in the
integrated syntactic form "fossil poetry" are not counterparts in the
conceptual mapping.
Now let us look at the kinds of compressions that can be
provided by such two-word noun-noun nutshells. Consider "jail bait,"
a phrase used to refer to an under-age girl whom an adult man finds
sexually attractive. "Jail" comes from the domain of human
criminality, while "bait" comes from the domain of fishing or
trapping. In the mapping between them, attraction to the girl
corresponds to attraction to the bait, initiating sex corresponds to
swallowing the bait, and ending up in jail (for sex with a minor)
corresponds to being caught. The conceptual elements named in the
integrated syntactic form "jail bait" are not counterparts in the
conceptual mapping. Here, obviously, we are prompted to borrow the
compressions and intensities of the fishing frame for the purpose of
compressing the "sex with a minor" frame and intensifying many of
its vital relations. For example, the causal chain in the "sex with a
minor" space, which runs from perception to incarceration, can be
long and diffuse, whereas the fishing frame has direct human-scale
causation: a single bodily action results in immediately being caught.
There is extraordinary emergent structure. In the blend, the man is not
to blame. In the space of fishing, the fish does not know that the bait
is bait. In the space with the man and the minor, the man certainly
does know about laws and jail and he recognizes that sex with the girl
is legally forbidden. But in the blend, he is blameless for the action,
indeed even the prime victim, even though he understands the law, the
prohibition, the possible punishment, and the reasons for it. The "jailbait" blend may acquire further emergent structure through the
principle of Intensification of vital relations. In the fishing space, the
intentionality is in the fisherman's attempt to trick the fish and catch
it. In the other input ("sex with a minor") the intentionality is in the
man's attraction. There is no counterpart for the fisherman in the space
with the minor, but it is nevertheless possible to project something
like the fisherman's intentionality into the blend. One corresponding
interpretation holds the girl herself responsible for what befalls the
man. Another might bring in the Devil. Another might bring in the
injustice of society and its laws.
"Jail bait" is an example of a two-word nutshell that prompts for
compression through borrowing of the inner-space relationships in the
fishing frame. In that compression, Time is scaled down and a diffuse
interpersonal interaction with many actions is compressed to a single
action –swallowing the bait. This compression can create relations in
the blend, such as the attribution of intentionality to the young
woman. There is also Highlights Compression –the sequence in the
human story of perception, greeting, seduction, doing the deed,
having it become known, being arrested and tried and sentenced and
jailed is all compressed in the blend into seeing and doing, where,
because taking the bait is automatically taking the hook, there is no
separation between committing the act and being punished. This is an
intense Cause-Effect compression. In the blend, the Effect is literally
in the Cause because the hook is literally inside the bait. "Jail bait" is
said as advice: the compression is meant to focus the man on the
Effect by making it part of the Cause, and thus give him powerful
Global Insight.
By contrast, there are many cases where it is outer-space vital
relations between the inputs that are compressed in the blend, as in
"caffeine headache," "money problem," and "nicotine fit," where the
disanalogy between the inputs is compressed into a property in the
blend1 . For example, money problems are a certain kind of problem,
the ones caused by absence of money. Again, we see the simplest
possible linguistic form prompting for remarkably complicated
integration networks. Communicating through simple grammatical
forms is possible because cognitively modern human beings can bring
to bear on those forms all of double-scope integration and its
governing principles and overarching goals. The language itself does
not have to carry such operations as compression or pattern
completion because human brains supply those operations at no
linguistic cost.
One common aspect of these compounds is that someone
attempting to "unpack" the linguistic form does not begin from the
Cf. G. FAUCONNIER and M. T URNER , “Chapter eleven: the Construction of the
Unreal”, in The Way We Think, op. cit., pp. 217-247.
assumption that the named elements are necessarily conceptual
counterparts. When presented with such a linguistic form, we cannot
predict, a priori, the relationship between the named elements. Notice
that the generic roles of these elements are different in "land yacht,"
"fossil poetry," and "jail bait". "Land" is a locative, "fossil" is a
product of a process, and "jail" is a result. "Yacht" is a means,
"poetry" an activity and its product, and "bait" an instrument.
Now consider "boat house." Again the same operations are
involved. As in "land yacht," we have a connection between the two
different spaces of land and water-houses are associated with land,
boats with water. In the mapping between them, the residents of the
house correspond to the boats, the house itself corresponds to a
protective shelter for storing the boats, and leaving the house
corresponds to being launched. "Boat" and "house" name elements
that are not counterparts in this mapping.
Of course, there is no restriction that prevents the named
elements from being counterparts. Consider "house boat," which
again evokes two different spaces of land and water. In the space of
land, the resident lives in the house; in the space of water, the sailor is
aboard the boat. In the formal integration "boat house," "boat" and
"house" are not conceptual counterparts; but in "house boat," the boat
and the house are conceptual counterparts, and they map onto a single
element in the blend. Similarly, "jail house" evokes a domain of
domestic residence and a domain of criminal punishment. In the
mapping between them, the jail and the house are conceptual
counterparts, and they map onto a single element in the blend. As
Christine Brooke-Rose showed in great detail, NP of NP can name
metaphoric counterparts, such as "fire of love"1 . Charles Fillmore
gives the example "One needn't throw out the baby of personal
morality with the bathwater of traditional religion"2 . These
counterparts need not be metaphoric: "the nation of England," "the
island of Kopipi", "the feature of decompositionality", "the condition
of despair".
In all of these cases, including those in which the syntactic form
names elements that are blended conceptually –"house boat" and "jail
house"– the blend is both less and more than the composition of the
input spaces. In "land yacht", we ignore that yachts have cooking and
C. BROOKE-ROSE, A Grammar of Metaphor, London, Secker & Warburg, 1958.
Charles Fillmore, personal communication.
sleeping facilities and require no manufactured course. On the other
hand, the blend contains more than the inputs: for example, the inputs
may supply the knowledge that we are dealing with a vehicle, but not
that it is a car as opposed to something else, or that many specific
features that we link with luxury cars belong to the land yacht: electric
windows, leather upholstery, opera windows, and suspension built for
comfort rather than handling.
In "fossil poetry", on the one hand, we ignore that fossils are
typically associated with extinct species, and generally that poetry is
not physical or biological. On the other hand, the conception of
language as a derivative of poetry is the central inference of the blend
but absent from the input space. In "jail bait," we ignore that someone
intends to lure the fish while perhaps no one intends to lure the man.
We ignore that the man is neither a fish nor (in the case where he
merely admires) a criminal. In the blend, we make use of the
particular social frame according to which the world is full of pitfalls
and traps for the man, teasing him with what it forbids. In the space
with the fish, fish are not capable of such a perspective, while in the
space of criminal action, the world does not necessarily tempt people
to commit crimes.
The situation is no different when the named elements happen to
be conceptual counterparts. In "house boat," we ignore that houses
have yards and are stationary or that boats are designed principally for
travel. We also know many things from background knowledge about
house boats that are not derivable from the inputs. We know that a
house boat cannot be simply a regular boat put on land that happens to
have people living in it, or a regular boat at mooring that someone has
been living in; but there is nothing in blending or the use of language
to prompt for integration networks that would forbid these meanings.
We see in these examples the falsity of the general view that
conceptual structure is "encoded" by the speaker into a linguistic
structure, and the linguistic structure is "decoded" by the hearer back
into a conceptual structure. An expression provides only sparse and
efficient prompts for constructing a conceptual structure.
The problem then is to find the relations between formally
integrated linguistic structure on the one hand and conceptually
integrated structures built by the speaker or retrieved by the hearer on
the other. In general, we will find that the conceptual integration is
detailed and intricate, while the formal integration gives only the
briefest indication of a point from which the hearer must begin
constructing this conceptual integration.
3. Formal blending
Novel conceptual blends do not generally need novel forms of
expression. A language already has all of the grammatical forms it
needs to express almost any conceptual blend. For example, all the
two-word expressions we have considered in this article use existing
syntax: compound forms like Noun-Noun, Adjective-Noun, and
Noun-Adjective, as well as particular existing nouns or adjectives
filling those forms. The concept land yacht may be a new blend, but
the phrase uses existing grammar and vocabulary to prompt for the
land yacht integration network.
Forms are mental elements, and they can be blended just like any
mental elements. Sometimes, this blending will align with the
blending of conceptual structure to which the forms attach. We can
see this pressure to achieve formal organization to express conceptual
blending across a range of constructions, from morphemes to
sentences. Consider single word integrations like "Chunnel," referring
to the tunnel that runs under the English Channel. Clearly, there is a
conceptual construction that integrates structure from both the abstract
frame of a tunnel and the specific frame of the body of water between
England and France. This integrated unit can serve as the site for
integrating a great range of knowledge, from the relevant geology to
problems of engineering, from the history of relations between
England and France to issues of quarantine, disease, and ecology.
This integration of content has as a corresponding grammatical form,
"the tunnel under the English Channel". This is already tightly
integrated. English makes available an even more compact compound
noun construction, "the Channel tunnel." By fortuitous accident, a
further integration of form is possible, given the phonemes in
"Channel" and "tunnel". This integration is a formal blend, triggered
by a partial phonological and orthographic mapping between the two
words. Pressure to integrate produces (in the case of English)
"Chunnel"; the corresponding accidents are lacking in French, leaving
as the most integrated form "tunnel sous La Manche." This shows
another important aspect of integration: it is opportunistic. That this
opportunism depends in any specific case upon apparently peripheral
accidents can lead to the mistaken view that the operation is
peripheral. Actually, the most central events and structures can arise
exactly by opportunistic exploitation of accidents. Evolution teaches
us that this is not paradoxical.
Suzanne Kemmer reports the following example of integration:
"The Whiners' most common complaint is that they've been
relegated to what Mr. Coupland calls 'McJobs.'"
"Mc" evokes a space of fast food and employment in that
industry. "Jobs" evokes the more general frame of seeking
employment. They have roles in common –workers, employers,
wages, benefits, possibilities for advancement, and so on– providing a
generic foundation for the blend. Two input spaces are set up, one for
aspects of McWorld and another for seeking employment. A
straightforward mapping links common roles. But that mapping does
not in itself provide the central inference of "McJobs." Specific
aspects of the McDonald jobs –no prestige, no chance of
advancement, no challenge, no future, boredom, a certain kind of
social stigma– are blended with the more general notion of low-level
service jobs. Absent this blend, we would be free to associate lowlevel service jobs with other stereotypes-altruistic and even saintly
devotion to others, climbing of the social ladder, small-town serenity
and routine, or freedom from avarice and grueling ambition. So this
simple blend brings in analogical mapping, the construction of a
generic space, and, in the blend, such new categories as the McJob,
the type of person who has the McJob, and the pay scale for a McJob.
One purpose is to bring inferences from the blend to the conception of
crucial realities –such as the plight of young people in the modern
economy– and to influence legislation and government policy. The
blend's power and efficiency seem to derive from its homogeneous
internal structure and its corresponding formal compression into a
single word. The striking thing about this unit is that it creates new
conceptual structure while all the conceptual engineering that went
into building it can be retrieved by a member of the relevant linguistic
and cultural community from the single word, "McJobs". We see that
the blend satisfies governing principles at the conceptual and formal
level: it offers maximal formal and conceptual Compression and
Integration and satisfies both Web (i.e. it maintains the web of
appropriate connections to the input spaces easily and without
additional surveillance or computation) and Unpacking (i.e. it allows
for the reconstruction of the entire integration network).
Formal blending can occur independently of whether there is any
background conceptual blending. An ad on the back of a bus for the
Del Mar racetrack, whose post time is 2:00 pm, reads: "Hunch hour.
2pm." To get the pun, we must access both "hunch" and "lunch hour"
simultaneously. This means doing pattern completion from "hunch
hour" to "lunch hour." It means partial mapping of "hunch" to "lunch
hour": "hunch" and "lunch" are noun phrases that differ only in their
initial phoneme. It means projecting the "unch" of "hunch" and the
"unch" of "lunch" both onto the "unch" of "hunch hour"; it means
projecting the nominal compound structure (N1 N2) of "lunch hour"
onto the corresponding nominal compound structure of "hunch hour".
It means projecting the common noun status of the counterparts
("hunch" and "lunch") onto the noun status of "hunch" in the blend.
Suppose one speaker in a group wonders aloud whether the new
mall down the road is still open at 9pm, and another speaker responds,
"Well, they should be, since everyone knows about Amahl and the
Night Visitors." Here, "Amahl" in the title of the opera is blended with
the noun phrase "a mall," but there is no conceptual blend. Or
consider a caption in Latitude 38 above pictures of the 1994 Vallejo
boat race, which had two legs, with the upwind leg particularly sunny:
"Vallejo 94 - Two legs, sunny side up"1 . This caption requires formal
blending, partial projection, mapping between forms, pattern
completion, and so on, but no conceptual blending: the race is not
blended with a particular breakfast dish.
Sometimes, the formal blend parallels the conceptual blend very
closely. The Atlanta Constitution of 17 February 1994 carried a frontpage caption reading, "Out on a Limbaugh", followed by a summary
of the story to be found inside: "Critics put the squeeze on Florida's
citrus industry for its $1 million deal with broadcaster Rush
Limbaugh". To get the punning effect of "Out on a Limbaugh"
requires accessing "out on a limb" and "Limbaugh" simultaneously.
Behind this formal blend is a conceptual blend with two input spaces,
one with an agent who climbs out on a limb of a tree, another with the
deal between the Florida citrus industry and Rush Limbaugh. Just as
limb and Limbaugh are blended conceptually, so "limb" and
"Limbaugh" are blended formally. In this particular unusual case,
Latitude 38, June 1994, vol. 204, pp. 116-117.
conceptual counterparts that are conceptually blended have formal
expressions that are formally blended. Even more, the formally
blended element in fact refers to the conceptually blended element.
The formal blend, as is standard for a blend, contains formal structure
that is not calculable from the formal inputs. Let us look at these
formal inputs. "Out on a limb" has an indefinite article with a
common noun. "Limbaugh" is a proper surname. While a proper
surname in English can become a common noun indicating one of a
group of people with that surname ("She's a Kennedy", "She's the
poorest Kennedy") or one of a group of people analogically
equivalent to a particular person ("He's an Einstein"), this is not what
is going on here. Nor is "a Limbaugh" a case of an indefinite article
used with a proper noun referring to an unknown person who happens
to have that label, as in "There is a Fidelia Cumquat on the telephone
for you". In "Out on a Limbaugh," "Limbaugh" has not become a
common noun, referring to namesakes or analogs of Limbaugh, nor is
it picking out an unknown person. On the contrary, it is picking out an
extremely well-known particular person. Nonetheless "Limbaugh"
follows an indefinite article; this is a property of its counterpart "limb"
in the other input to the blend. As a result the formal blend has new
syntactico-semantic structure, that is, indefinite article + proper name
associated with a known person.
In the Limbaugh example, we see formal blending and
conceptual blending in close parallel, but this mirroring is very rare.
We have already seen cases of formal blending that have no
corresponding blend at the conceptual level. Conversely, most
conceptual blending happens with no corresponding formal blend. For
example, "Look before you leap" is a standard prompt for
metaphorical blends that has no formal blending. There are even cases
where conceptual blending and formal blending are both at work, but
the formal blend runs contrary to the conceptual blend. For example,
contestants on the BBC game show "My Word" were challenged to
come up with an intelligible expression identical to the title of any
popular show song except for a single letter in the last word, for
example, "When you wore a pink carnation / and I wore a big red
nose". One contestant's response was "Why can't a woman / be more
like a mat?" (The original, from My Fair Lady, is "Why can't a
woman / be more like a man?") At the formal level, "man" and "mat"
are blended, but at the conceptual level, man is to be blended not with
mat but with its opposing element: person who walks on the mat.
4. Complex grammatical structures
We have already seen in some detail how many grammatical
constructions work:
—single words like "safe,"
—nominal compounds like "boat house," "house boat," and "jail bait,"
—adjective-noun compounds like "guilty pleasures," "likely
candidate," and "red ball,"
—Morphological combinations in a single word like "Chunnel."
In all these cases we saw that the construction in the language
had a stable syntactic pattern that prompts for a specific blending
scheme. The blending scheme carries with it a particular kind of
A language is a powerful culturally developed means of creating
and transmitting blending schemes. The capacity for language
depends intricately on the capacity for blending and compression1 .
The patterns we find in a language are the surface manifestation of
blending schemes that have emerged within a culture and that have
wide applicability. Over the last fifty thousand years, cultures have
developed many systems for saving people from the work of
inventing all the useful blending schemes from scratch. The most
obvious and perhaps the most powerful way cultures provide children
with useful blending schemes is through language.
In linguistics, the forms in language have been studied under the
term "syntax". It turns out that, however one looks at them, these
language forms are exceptionally complex, much more so than we
ever realize. So, notoriously, the study of syntax is also complex. Yet
that study is essentially incomplete if we do not simultaneously study
the blending schemes for which these language forms prompt. Even
when the language forms are simple –as in "jail bait"– the
corresponding integration networks can be very complicated.
We will now look in some detail at more elaborate constructions
that are central on anybody's account and analyze the central role of
blending. This will give an idea of the general inseparability of
language, blending, and compression. Language has elaborate formal
Cf. G. FAUCONNIER and M. TURNER , “Chapter nine: the Origin of Language”, in
The Way We Think, op. cit., pp. 171-193.
patterns because it prompts for powerful blending schemes. But the
formal patterns and blending schemes are so deeply entrenched as to
be almost invisible to consciousness.
4.1. Caused motion
One of the most common and familiar human scenes is moving
an object –we throw it, kick it, hurl it, or nudge it, and it moves in a
direction and lands somewhere. This is the "caused motion" scene. It
contains an agent who does something, and that act causes an object
to move. There are verbs whose entire job is to indicate some specific
version of the caused motion scene, like "throw," "hurl," and "toss."
Many, perhaps all languages have such verbs and a pattern into which
they fit. In English, that pattern is Subject-Verb-Object-Place, as in
"Jack threw the ball over the fence". But in English, unlike most
languages, this pattern can also be used with verbs that in themselves
do not express caused motion, verbs like "walk," "sneeze," and
"point". So we get, "I walked him into the room," "He sneezed the
napkin off the table," and "I pointed him toward the door". This
pattern can even accommodate verbs that involve no physical motion
at all –verbs like "tease," "talk," and "read". So we get, "They teased
him out of his senses," "I will talk you through the procedure," and "I
read him to sleep". In these examples, the subject performs some
action and this causes the object to "move" either literally or
metaphorically, in a "direction" either literal or metaphoric. The
caused-motion construction has been studied in great detail by Adele
Goldberg 1 .
In all of these cases, the form is prompting for an integration
network that has the caused motion scene as one input and some other
diffuse scene (involving walking, or throwing, or sneezing, or talking,
or reading, and so on) as the other input. In the case of "He sneezed
the napkin off the table," the diffuse input consists of a sequence of
events in which there is a person, a napkin, and a table, the person
sneezes, the sneeze moves the ambient air, the air stream impacts the
napkin, the napkin because it is light, moves under pressure from the
air stream, and typically reaches the edge of the table where gravity
causes it to move in a quasi-parabolic path (except for air resistance)
A. G OLDBERG, Constructions: A Construction Grammar Approach to Argument
Structure, Chicago, University of Chicago Press, 1994.
to land on the floor, which it cannot penetrate, because, after all, it is
only a napkin. In the diffuse input, we have an action, sneezing, with
an agent and a motion by an object, the napkin, in a direction. The
action is causally related to the motion. In the compressed causedmotion input, we have an agent, an action-motion, an object, and a
direction. Conceptually, there is a natural mapping from the causedmotion scene to the diffuse input: the agent maps to the agent, the
object to the object, the direction to the direction, and the actionmotion to any of a number of distributed candidates –the action, the
causal relation, or the motion.
In the compressed input, there is a syntactic form associated with
the conceptual compression. In the diffuse input, particular words like
"sneeze" and "napkin" and "off" and "table" are associated with
individual events and elements. In the full integration network, the
conceptual compression and the syntactic form come from the
compressed input, while some individual words come from the diffuse
input through selective projection. In the case of "He sneezed the
napkin off the table," we integrate the conceptual structure of the
caused-motion scene with many elements and events from the diffuse
input: the agent, the action, the causation, the object, the motion, and
the direction. The single action-motion in the caused-motion input
maps to at least three different elements –the action, the causal link,
and the motion in the diffuse input. We also integrate the causedmotion syntax (Subject-Verb-Object-Place) with a few of the words
available for the diffuse input, in this case a word for the agent ("he"),
a word for the action ("sneeze"), and some words for the object and
the direction.
But notice that we do not bring in a word for the causal link from
the diffuse input, or a word for the motion of the object. Because the
mapping from the action-motion to the diffuse input is not one-to-one,
we have many possibilities for projecting words from the diffuse
input. We might project a word for the causal link rather than the
action or the motion, as in "Sarge let the tanks into the compound".
This example typically evokes a military situation in which Sarge's
permission is needed for the tanks to enter the compound. The
sentence does not specify the particular causal action performed by
Sarge (waving his hand, signing a paper, giving a verbal OK by
telephone) or the motion of the tanks (being carried in on trucks,
airlifted by helicopters, moving on their own power). We might
project a word for the motion of the object, as in "He rolled the barrels
into the warehouse," where it is the barrel that rolls, not the agent. The
sentence does not specify the particular casual action (pushing the
barrels, kicking them down a ramp, pressing a button to release the
queue of barrels…), nor does it specify a causal link. In fact, the
single integrated action-motion in the caused-motion input includes
manner of both action and motion, as we see in specifically causedmotion verbs like "throw," "push," and "hurl," which indicate
something about both the manner of the action and the manner of the
motion of the object. So the single action-motion connects to the
manner of the action and the manner of the motion in the diffuse input
as well, and we can project words for those manners into the blend, as
in "He floated the boat to me" and "He wiggled the nail out of the
hole". The scenes evoked by these examples include action, motion,
and manner, but the words that are projected, "floating" and
"wiggling" do not themselves require motion along a path or external
action. They focus on the manner of being.
Such constructions offer ready-made and powerful blending
schemes. A tightly compressed frame and a corresponding syntactic
form from one input can be recruited into a blended space linked to a
diffuse input. Constructing a network based on that scheme for a
particular case depends crucially on being able to construct a generic
space that applies to the two inputs. In the case of the caused motion
examples we have seen, this generic space has agent-action, objectmotion, and direction. This description also fits the very striking
example, "They prayed the two boys home". Here the blend is
performing an extreme compression: the scene with the prayer and the
boys contains many causal steps over an expanse of time, and
intermediate agents, with relatively weak or vague causality, but in
the blend there is a single action that is directly causal for the boys'
coming home.
As we have seen many times in this book, metaphoric mappings
provide one of the standard ways of locating a cross-space mapping
between inputs and constructing an integration network. If the diffuse
input has causation of change of states, then there is already an
existing template for blending states and locations and changes of
states with changes of locations, and that template can be recruited
wholesale to provide much of the cross-space mapping and much of
the projection to the blend. For example, "I pulled him out of his
depression" prompts for a network where one input is the caused
motion scene with its syntax, and the other input has a complicated
interpersonal causation involving a change of psychological state. We
automatically construct much of this particular caused-motion
integration network by recruiting the states-locations blending
network. "Pull" is a prototypical caused-motion verb, but the analysis
works identically for "I talked him out of his depression" or "He drank
himself into oblivion".
We have seen cases where the cross-space mapping is a bundle
of Role-value mappings in a Simplex network (as in "He threw the
ball over the fence," where throw is a value for the role causal actionmotion) and cases where the cross-space mapping is analogical or
metaphorical ("He drank himself into oblivion"). The caused-motion
network also operates in the interesting ways when there are
counterfactual connections. Adele Goldberg offers the example "Pat
blocked Chris out of the room"; others include "We barred him from
the building," "John forbade him from participating," and with the
metaphorical connection, "We kept him out of trouble." Such
examples show that the compressed input is more general than just
caused-motion. The more general frame is that of an agent exerting a
force and of an object undergoing a force in the direction of a goal. In
the simple case, the agent exerts force on the object in the direction of
the goal, as in "we moved the wolf to the door." In this case, "move,"
indicates the application of a force by the agent; "to" indicates that the
force is in the direction of the goal; and "door" indicates the goal. But
alternatively, the force exerted by the agent can oppose the movement
of the object, as in "we kept the wolf from the door". Here, the verb
"keep" indicates that a certain force is applied in opposition to an
existing force; "from" indicates that this applied force is in the
direction away from the goal; and "door" indicates the goal.
We see then that the general compressed input in a "causedmotion" network is force-dynamic and applies equally well to the
caused-motion and blocked-motion examples. The blocked-motion
examples have an implicit counterfactual space: "keep" implies that
the agent's force opposes the object's force, so if the agent's force
disappears, the object moves to the goal.
4.2. Emergent syntax
In the caused-motion construction, the syntactic component
comes entirely from the compressed input space of integrated caused
motion, while words come from the space of the events associated
with the causal sequence. But there are other constructions in which
the syntactic form used for the blend does not come entirely from one
space. Part of it comes from one space, part of it from the other, and
part of it develops specifically for the blend. In this case, the blend
has emergent syntax relative to the inputs. Consider causatives in
French, which are formed using the verb faire ("do").
[Pierre fait
[Pierre makes eat
(meaning "Pierre feeds Paul")
[Paul is the agent of "manger"]
le paquet.
[Pierre makes send
the package]
(meaning "Pierre has the package sent")
["le paquet" is the object of "envoyer"]
Pierre fait
manger la soupe
à Paul.
makes eat
the soup
(meanig "Pierre feeds Paul the soup)
[Paul is the agent of "manger", "la soupe" is the object]
à NP
to Paul]
By these double-verb forms, French provides its speakers a way
to evoke an integration network that delivers a compressed, humanscale scene in which at least two agents (Pierre and Paul), a causal
action, a causal link, and a caused action (eat) are integrated into one
event. Now, the French language already has single-verb forms that
are good for parts of this integration. For example, "Jean fait le pain"
is fine for evoking a scene in which Jean performs a causal action
involving another element (le pain). And "Paul mange la soupe" is
fine for evoking a scene in which Paul performs some action on
another element (la soupe). French has several basic single-verb
clausal constructions. How shall we express the scene in which Pierre
does something that causes Paul to eat the soup? None of the basic
single-verb clausal constructions quite serves. French offers three
complex blends for doing the job1 . Each has as one input one of three
compressed basic single-verb clausal constructions, and as the other
input the diffuse chain of causal events with intermediate agents that
we want to compress. The blend takes much of its clausal syntax from
G. FAUCONNIER and M. TURNER , "Blending as a Central Process of Grammar”, op.
the compressed first input, but, and this is crucial, it has additional,
emergent syntax. In that syntax there are now two verbs. Additionally,
there are novel positions for clitic pronouns (like le, lui, and se, as in
"Paul se fait tuer par Jean" versus "Paul fait se tuer Jean") and various
complements (as in "Paul fait envoyer le paquet à Marie à Jean,"
meaning "Paul has the packet sent by Jean to Marie"). In these
double-verb causatives we see double-scope integrations at the
conceptual level: the conceptual frame for the basic construction does
not match in a one-to-one fashion the complex and diffuse causal
chain in the other input. We also see double-scope integrations at the
formal level, delivering new, emergent syntactic forms for expressing
the blend. In the caused-motion construction, the syntactic form came
entirely from the compressed input. Here, in the double-verb
causatives, the syntactic form comes only partly from the compressed
input. Some words and their grammatical categories come from the
other, diffuse input. And the full syntactic form is emergent in the
The double-verb causative constructions may be a response to an
integration problem faced by humans everywhere: one person does
something that is causal for someone else's action. A blending
template achieves appropriate compressions in these instances, and
language forms can prompt for these compressions. Many languages
have arrived independently at the solution of double-verb causative
constructions, using blends like the ones in French but having
different emergent syntax. The development of these language forms
is an example of the way in which human cultures, using the
cognitively modern capacity for double-scope integration, evolve
blending templates, including language forms, that are transmitted to
subsequent generations. Elsewhere 1 , our principal argument was that
language could not be developed as an efficient set of forms without
the capacity for double-scope integration. Here we make that point
more explicit by highlighting the crucial role of blending in a few
particular constructions. Languages, with their grammars, are great
cultural achievements produced and perpetually transformed through
the exploitation of the capacity for double-scope integration.
Cf. G. FAUCONNIER and M. TURNER , “Chapter nine: the Origin of Language”, in
The Way We Think, op. cit., pp. 171-193.
5. Cultural evolution of language
Language changes over centuries. Latin gave rise to French. It is
in fact a remarkable universal feature of all languages that they
change over cultural time. For the most part, deep changes in the
structure of a language take so long that we do not see them
happening in a single lifetime, although we sometimes see lesser
changes taking place in the space of a few years, as when new words
come into the language, slang and idioms are created, or existing
words acquire new extensions, such as "virus" for the nefarious
program that ruins your computer. Linguists agree that this change is
not a matter of either improvement or deterioration of a language.
Languages do not change because they are deficient or unstable.
Perfectly fine systems in a language routinely evolve into different
perfectly fine systems. Why should this be? In most grammatical
theories, powerful principles have to be invoked.
We suggest that the central role of conceptual blending,
compression, and double-scope creativity in grammar and
grammatical constructions induces language change very naturally
and, in fact, unavoidably. One reason for this is the natural emergence
of new syntax under pressure from borrowed compressions, as we
saw in the case of the French double-verb causatives. Another reason
for progressive change in languages is that conceptual blending
networks are underspecified. Because such networks prompt for
mapping schemes without specifying the mappings and projections
exhaustively, grammatical constructions leave users, singly and
collectively, some leeway in the actual implementation of the
mapping schemes. This kind of language change stems from variation
of the underspecified aspects of the selective projections and
mappings within the network. Suzanne Kemmer and Michael Israel
have shown conclusively, in their extensive studies of the "way"
construction, that within a relatively stable blending scheme over
several centuries, usage during a certain period will emphasize certain
projection patterns and not others, and that this usage will change over
time 1 .
See S. K EMMER and M. ISRAEL, "Variation and the Usage-Based Model", in K.
BEALS et al. (eds), Papers from the Parasession on Variation and Linguistic
Theory, Chicago, Chicago Linguistic Society, 1994; M. ISRAEL , "The Way
Modern examples of the "way" construction are "He found his
way to the market," "He made his way home," "He elbowed his way
through the crowd," "He jogged his way along the road," "He talked
his way into the job," and "He whistled his way through the
graveyard". This usage developed from a Middle English go-yourpath construction, which accepted almost any noun that meant
something like "way," as in "He lape one horse and passit his way"
(1375) and "Tho wente he his strete, tho flewe I doun" (1481). Later,
a new syntax developed that allowed for a complement indicating
direction, as in "He went his way home". The construction developed
by allowing ever more kinds of verbs indicating something about the
manner of the movement. As Israel remarks, these verbs "tend to
cluster around certain well-defined semantic prototypes." Between
1826 and 1875, a large number of verbs coding difficult motion or
tortuous paths become acceptable. They include plod, totter, shamble,
grope, flounder, fumble, wend, wind, thread, corkscrew, and
serpentine. Dickens writes in 1837: "Mr. Bantam corkscrewed his
way through the crowd". It is not until the end of the nineteenth
century that we begin to find verbs like crunch, crush, sing, toot and
pipe in the construction. Israel explains that they encode "not motion
per se, but rather the noise that inevitably accompanies certain forms
of motion". A different development of the construction, the means
thread, comes up fairly late at the end of the sixteenth century. In this
thread, verbs for creating a path become acceptable, as in "Arminius
paved his way" (1647). A familiar example would be "Every step that
he takes he must battle his way" (1794). Israel writes, "By 1875,
examples include uses with push, struggle, jostle, elbow, shoulder,
knee, beat and shoot. In the nineteenth century, as the manner thread
experiences a rapid expansion, the means thread begins to allow verbs
encoding increasingly indirect ways of reaching a goal," as in "Not
one man in five hundred could have spelled his way through a psalm"
or "He smirked his way to a pedagogal desk". The role of blending is
It is useful, in this light, to consider the construction . . . as an
example of a syntactic blend—that is, as a specialized
grammatical pattern serving to combine disparate conceptual
Constructions Grow", in A. GOLDBERG (ed.), Conceptual Structure, Discourse and
Language, Stanford, Center for the Study of Language and Information, 1996, pp.
contents in a single, compact linguistic form. Essentially, the
modern construction provides a way to blend the conceptual
content of an activity verb with the basic idea of motion along
a path. The trend toward verbs coding activities which are
increasingly marginal to the achievement of motion thus
reflects the construction's gradually increasing power to blend
different types of events into a single conceptual package1 .
Israel has succeeded in showing that some underspecified
projection patterns for the integration network associated with the
“way” construction were extended over time. He also shows that
another underspecified projection pattern for the construction
diminished over time: "The third thread . . . involved usages with
verbs like keep, hold, take, snatch, and find for coding the acquisition
or maintenance of possession of a path. These usages were very
common in early stages of the construction. But unlike the other two
threads, this usage shrank rather than expanded over time, so that now
only find . . . and a few other verbs remain to represent it"2 .
6. Conclusion
All along, we have stressed creativity and novelty as
consequences of conceptual integration. But creativity and novelty
depend on a background of firmly anchored and mastered mental
structures. Human culture and human thought are fundamentally
conservative. They work from the mental constructs and material
objects that are already available. Conceptual integration too has
strong conservative aspects: it often uses input spaces, blending
templates, and generic spaces that are anchored in existing conceptual
structure; it also has governing principles driving blends in the
direction of familiar, human scale structures; and it readily anchors
itself on existing material objects. Emergent structure –both
conceptual and formal– can arise through conceptual integration
within basically conservative integration networks.
This general pattern of development in culture and thought has
the evolution of grammar as a special case. We have seen sample
evidence of the central role of blending and compression both in
superficially simple constructions like Noun-Noun and in those that
Ibid., p. 226.
Ibid., p. 221.
are acknowledged as highly complex, like French double-verb
causatives. Advanced conceptual integration operates simultaneously
for both conceptual and formal structure. It requires double-scope
capacity, and by its very nature promotes both continuity and change
at the conceptual and formal levels. Indeed, for conceptual blending to
happen at all, continuity is essential. We have shown how novel
constructions or variants of constructions draw –thanks to conceptual
integration– on deeply entrenched constructions, conceptualizations,
and blending templates. But because blending involves selective
projection, composition, completion, and elaboration under a set of
governing principles, it can produce new, well-anchored conceptual
and formal structures. In his article on “way” constructions, Israel
points to just these features of simultaneous conservativism and
novelty. He writes, "Utterances should sound like things the speaker
has heard before". There are also, he notes, forces for innovation. The
world is rich, both physically and culturally, and it evolves, and this
places pressures on us to create new conceptions and expressions. We
do this through double-scope integration, but the products are not
wholly novel. Constitutive and governing principles ensure that the
network is in many ways deeply familiar, not the least in using
familiar frames, a canonical set of vital relations, an easily accessible
initial cross-space mapping, and human-scale organization and
compression in the blended space. For grammar, this delivers slightly
new expressions that, however novel, are intelligible precisely
because they are for the most part strongly anchored to existing
constructions. When we hear an expression, we try to construct an
integration network, but to do so we have to do some selective
projection, composition, completion, and elaboration that is not
specified by what we hear, so there is yet more room for creativity
and novelty. We do as much blending as we need to do to make sense
of the utterance, and this work is simultaneously conservative and
As we have argued in some detail in this article, blending turns
out to be a central feature of grammar. Far from being an
independently specified set of forms, grammar is an aspect of
conceptual structure and its evolution.