Any mental space in the integration network can be modified at any moment in its construction. In particular, the inputs can be modified by reverse mapping from the blend. In the wedding example, backward projection from the daydream blend can alter the man’s sense of his actual relation to his girlfriend. Consider a second example. The clipper ship Northern Light sailed in 1853 from San Francisco to Boston in 76 days, 8 hours. That time was still the fastest on record in 1993, when a modern catamaran, Great American II, set out on the same course. A few days before the catamaran reached Boston, observers were able to say: (1) At this point, Great American II is 4 .5 days ahead of Northern Light. This expression frames the two boats as sailing on the same course during the same time period in 1993. It blends the event of 1853 and the event of 1993 into a single event. In one mental space, we have the event of 1853; in a second, we have the event of 1993. These are the input spaces to the blend. There is a partial cross- space mapping that connects counterparts in the input mental spaces. It connects the two boats, the two paths, San Francisco to San Francisco, Boston to Boston, and moment of departure to moment of departure. There is a generic mental space, which maps onto each of the inputs and contains what the inputs have in common: an ocean voyage from San Francisco to Boston. There is selective projection to the blended space, which brings in the two boats, the course, their actual positions and times on the course, and so on, but not the 1853 date or the 1853 weather conditions, the fact that the clipper ship was engaged in transporting cargo, and so on. In the blend, the geographies and times of the two input spaces are fused, but the boats are not fused. They are brought in as separate elements. The blend develops emergent structure not in the inputs. First, composition of elements from the inputs makes relations available in the blend that did not exist in the separate inputs. In the blend but in neither of the inputs, there are two boats traveling from San Francisco to Boston, instead of one. Second, completion brings additional structure to the blend. The scenario of two boats moving toward the same goal on the same course and having departed from San Francisco on the same day fits into an obvious and familiar frame, that of a race, which is auto- matically added to the blend by pattern completion. By virtue of that frame, we can now run the scenario dynamically: in the blend, the two boats are racing. Such ‘‘running of the blend’’ is part of elaboration. In this case, elaboration of the blend is constrained by projection of locations and times from the inputs. Running the blend modifies it imaginatively, delivering the actual winning and losing in the blend and the emotions associated with those relations. This is new struc- ture: there is no relation of being ‘‘ahead’’ or ‘‘behind’’ another boat in either of the input mental spaces, even if we run them dynamically. But those two boats in the blend are projected back to the ‘‘same’’ boats in the two input mental spaces. Their status in the blend projects back to their counterparts in the input spaces. In this case, the expression ahead of, used to express the conceptual relation between the two boats, prompts for a blend in which the two boats bear close spatial and temporal relations. 380 mark turner The race frame in the blended space may be invoked more noticeably, as in (2): (2) At this point, Great American II is barely maintaining a 4.5-day lead over Northern Light. Maintaining a lead is an intentional part of a race. Although in reality the cat- amaran is sailing alone and the clipper’s run took place 140 years before, the situ- ation is described in terms of the blended space. No one is fooled: the clipper has not magically reappeared. The blend remains solidly linked to the inputs, and inferences from the blend can be projected back to the inputs: in particular, if we know that Great American II is 4.5 days ahead of Northern Light in the blend, then we know that the corresponding location of Northern Light in its input is less far along the course than the corresponding location of Great American II in its input, and we know that it takes 4.5 days of sailing (by one of the boats) to get from the first location to the second. Another noteworthy property of the race frame is its emotional content. Sailors in a race are driven by emotions linked to winning, leading, losing, gaining, and so forth. This emotional value can be projected to the Great American II input. The solitary run of Great American II, conceived as a race against the nineteenth- century clipper, can be lived with corresponding emotions, which can in turn change the course of events. The crew of Great American II can draw courage and com- mitment from seeing themselves as engaged in a historic competition, or if they are daunted by Northern Light’s performance, may be cowed into failure. Conceptual integration networks routinely involve certain Vital Relations which can obtain between mental spaces in the network (‘‘outer-space vital relations’’) or within mental spaces in the network (‘‘inner-space vital relations’’) or in some cases both: Change Identity Time Space Cause-Effect Part-Whole Representation Role Analogy Disanalogy Property Similarity Category Intentionality Uniqueness In addition to the constitutive principles of conceptual blending, blending op- erates under a set of governing principles having much to do with Vital Relations. conceptual integration 381 a. Topology Principle: other things being equal, set up the blend and the inputs so that inner-space relations in the blend reflect useful topology in the inputs and their outer-space relations. b. The Pattern Completion Principle: other things being equal, complete ele- ments in the blend by using existing integrated patterns as additional inputs. Other things being equal, use a completing frame that has relations that can be the compressed versions of the important outer-space vital relations between the inputs. c. The Integration Principle: achieve an integrated blend. d. The Maximization of Vital Relations Principle: other things being equal, maximize vital relations in the network. In particular, maximize the vital relations in the blended space and reflect them in outer-space vital relations. e. The Intensification of Vital Relations Principle: other things being equal, intensify vital relations. f. The Web Principle: 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. g. The Unpacking Principle: other things being equal, the blend all by itself should prompt for the reconstruction of the entire network. h. The Relevance Principle: other things being equal, an element in the blend should have relevance, including relevance for establishing links to other spaces and for running the blend. Conversely, an outer-space relation between the inputs that is important for the purpose of the network should have a corresponding compression in the blend. i. The Compression Principle: achieve compressed blended spaces. There are several routine strategies for satisfying the compression principle: borrow compression from a compressed input; compress a vital relation (such as time or space or change) by scaling it to human scale in the blend; compress a vital relation by syncopating it as it is projected to the blend; compress one vital relation into another that fits human scale better; achieve scalability of one kind of vital relation by compressing it into a different kind of vital relation; create vital rela- tions in the blend that are not otherwise available in the network by compress- ing one vital relation into another; or create compressed blends by projecting only highlights from other mental spaces in the network that are connected by a struc- ture of story or sequential action. Figures 15.2 and 15.3 illustrate two of the compression hierarchies for which we currently have evidence. The constitutive and governing principles of conceptual integration, operat- ing over Vital Relations, are driven by an overarching goal, Achieve Human Scale, and have the effect of creating blended spaces at human scale. The most obvious human-scale situations have direct perception and action in familiar frames that are easily apprehended by human beings: an object falls, someone lifts an object, 382 mark turner two people converse, one person goes somewhere, and so on. They typically have direct intentionality, very few participants, and immediate bodily effect. They are immediately apprehended as coherent. Once blending achieves a human-scale blend, the blend also counts as human- scale, and so can participate in producing other human-scale blends, in a boot- strapping pattern that characterizes much of cultural evolution. To achieve a human-scale blend often requires imaginative transformations of elements and structure in an integration network as they are projected to the blend. There are several subgoals: Compress what is diffuse Obtain global insight Strengthen vital relations Come up with a story Go from many to one Conceptual integration operates indispensably in all the areas of thought and action that distinguish human beings from members of other species. Language is prominent among them and our concern here. As we surveyed in Fauconnier and Turner (2002), even very simple con- structions in language depend upon complex blending. It is natural to think that adjectives assign fixed properties to nouns, so that The cow is brown assigns the fixed property ‘brown’ to ‘cow’. By the same token, there should be a fixed property associated with the adjective safe that is assigned to any noun it modifies. Yet consider the following unremarkable uses of safe in the context of a child playing at the beach with a shovel: The child is safe, The beach is safe, The shovel is safe. There is no fixed property that safe assigns to ‘child’, ‘beach’, and ‘shovel’. The first statement means that the child will not be harmed, but so do the second and third—they do not mean that the beach or the shovel will not Figure 15.2. Compression hierarchy for Analogy/Disanalogy conceptual integration 383 be harmed (although they could in some other context). Safe does not assign a property but rather prompts us to evoke scenarios of danger appropriate for the noun and the context. We worry about whether the child will be harmed by being on the beach or by using the shovel. Technically, the word safe evokes an abstract frame of danger with roles like ‘victim’, ‘location’, and ‘instrument’. Modifying the noun with the adjective prompts us to integrate that abstract frame of danger and the specific situation of the child on the beach into a counterfactual event of ‘harm’ to the child. We build a specific imaginary scenario of ‘harm’ in which ‘child’, ‘beach’, and ‘shovel’ are assigned to roles in the danger frame. Instead of assigning a simple property, the adjective is prompting us to blend a frame of danger with the specific situation of the child on the beach with a shovel. This blend is the imaginary scenario in which the child is harmed. The word safe implies a disanalogy between this counterfactual blend and the real situation, with respect to the entity designated by the noun. If the shovel is safe, it is because in the counterfactual blend it is too sharp, but in the specific situ- ation it is too dull to cut. The disanalogy between the actual situation and the counterfactual blend is a relation between two mental spaces. One of them is already a blend. They are both inputs to yet a second blend, the one in which we have the actual situation, but now the disanalogy between the input spaces is compressed into a property of an element in the blend: in the blend, we have a safe child or a safe beach or a safe shovel. The word safe is a prompt to construct an elaborate conceptual integration network with the actual situation and a harm scenario as inputs to a counterfactual blend, and the actual situation and the counterfactual blend as inputs to a final blend in which there is an element with the property ‘safe’. To understand safe requires constructing and using the entire network. Figure 15.3. Compression hierarchy for Cause-Effect 384 mark turner We can create many different blends out of the same inputs. The process is the same in all of them, but the results are different. In The shovel is safe, the child is the victim in the blend if we are concerned about the shovel’s injuring the child, but the shovel is the victim in the blend if we are concerned about the child’s breaking the shovel. Furthermore, any number of roles can be recruited for the ‘danger’ input. In the counterfactual blend for The jewels are safe, the jewels are neither victim nor instrument; they are ‘possessions’ and their ‘owner’ is the victim. If we ship the jewels in packaging, then in the counterfactual blend for The packaging is safe, the jewels are the ‘victim’, external forces are the ‘cause of harm’, and pack- aging is the ‘barrier to external forces’. Other examples showing the variety of pos- sible roles would be Drive at a safe speed, Have a safe trip, This is a safe bet, and He stayed a safe distance away. The beach is safe shows that the ‘‘matches’’ between inputs are not achieved independently of blending and that there is nothing simple about ‘‘matching.’’ The beach in the real situation is matched to the role ‘doer of harm’ in the ‘harm’ scenario because we have achieved an imaginary blend that counts as counterfac- tual to the real situation. That match, however, is a match between a role in a frame and a specific element that is in fact not an instance of the role. The real ‘safe beach’ is not a ‘doer of harm’. That is the point of the utterance. The role ‘doer of harm’ in the ‘harm’ input is matched to a ‘beach’ in a blend that is imaginatively a ‘doer of harm’. And the ‘beach’ in the specific situation is matched to the ‘beach’ in the counterfactual blend because they are opposites in the way that counts for this situation: one is a ‘doer of harm’, and the other is not. Safe is not an exceptional adjective with special semantic properties that set it apart from ordinary adjectives. The principles of conceptual integration are general. Even color adjectives, which at first blush look as if they must assign fixed features, turn out to require noncompositional conceptual integration. Red pen- cil can be taken to 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, or a pencil used only for recording deficits (see also Travis 1981 on black kettle; in addition, see the sections on ‘‘active zones’’ in Langacker 1987, 1990, 1991). Theories of semantics typically prefer to work with examples like black bird or brown cow, since these examples are supposed to be the principal examples of compositionality of meaning, but even these examples illustrate complicated processes of conceptual integration. Many expressions prompt directly for blends. The exact words used in the news report in Latitude 38 to describe Northern Light and Great American II were: (3) 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, conceptual integration 385 The word ghost points explicitly to the blend. Its effect is to indicate how to build connections over three separate spaces: someone in the temporally later input (in this case, the crew of Great American II, the reporters, and everyone following the event) remembers an element in a temporally earlier input (here, Northern Light in 1853); that element is not in the temporally later input, but it has a counterpart in the blend (here, the ‘‘ghost’’ ship). This use of ghost to indicate how to build an integration network is quite conventional. It cannot be explained as just predicating a feature of a single ele- ment; it also tells us something important about the web of that element’s con- nections. We saw a similar lexical phenomenon with safe, which again could not be explained as just predicating a feature of a single element; instead, it told us something important about the web of connections across spaces in an integration network involving a specific counterfactual scenario of harm. In addition, ghost signals that events involving the ‘ghost’ in the blend are constrained by the events involving its ancestor counterpart. In the Northern Light example, the run of the ghost must be the same as that of its ancestor. It cannot go faster than it went in 1853, benefit from 1993 weather, collide with Great American II, and so on. So, again, ghost is not telling us about specific features of the events in the two spaces, but only that those events have a particular cross-space relationship: this kind of ghost, at least, must copy its ancestor counterpart. Nobody confuses the blend with reality. There is no inference that the sailors actually saw a ghost ship or imagined one. The construction and operation of the blend is creative, but also conventional in the sense that readers know immediately and without conscious effort how to interpret it. Ghost is specialized as a prompt for elaborate conceptual blending that pro- vides thorough compression to human scale. It signals various vital relations be- tween the input spaces and also signals a category in the blend that is a compres- sion of those vital relations. The sailors in the 1993 space know about the history of Northern Light. That is a link of intentionality through memory. Ghost prompts for the construction of that outer-space vital relation, which is often supplemented with an outer-space representation link, making the content of the memory a representation of the past event. The particular ghost in the blend compresses that outer-space representation link into a directly perceived instance of the category ‘ghost’. There are three pairwise outer-space Identity links connecting that instance in the blend, an element in the input historical space, and an element in the content of the memory that is in the later input space. There is also an outer-space relation of counterfactuality between the blend and the inputs; for example, in the blend there are two boats racing, one of them a ghost, but that structure is incompatible with the structure in the inputs for 1853 and 1993. The ghost in the blend is natu- rally decompressed into a time link between the two inputs. These features of ghost are quite general. The Northern Light example illustrates, therefore, not only the constitutive and governing principles, the overarching goals, and the Vital Rela- tions, but also a much more specific template of conceptual integration for ghost that seems to be available in every culture. 386 mark turner For ghost, one of the inputs has an element that the other input does not, and this disanalogy on existence of an element is compressed into an element in the blend that has special properties. Once this strategy of compression of disanalogy to property is recognized, we can see a vast range of similar constructions in the language. Coming home, I drove into the wrong house and collided with a tree I don’t have is counterfactual because it depends upon the evoked but counterfactual scenario of driving into the right house and therefore not colliding with a tree. The grammatical trigger here is not if then but rather the adjective wrong. In one input, the driver drives into the parking place at his or her home. In the other, he or she drives onto the property of some different house and collides with a tree. These inputs share the frame of parking a car at a house, and there are identity con- nectors between the cars and the drivers, but there are disanalogies between the two inputs, having to do with the value of the role ‘house’ and the existence of a tree at a particular location. In the blend, we have, from the space of what actually happened, the house where the driver did drive, the tree, and the collision. The disanalogy between the houses is compressed in the blend into a property of the house: it is now the ‘wrong’ house. And the disanalogy having to do with the tree is compressed into a property of the tree: ‘a tree I don’t have’. It is tempting to think that this is a property of the tree independent of the blend, but note what happens if our companion on a walk through some public woods says, pointing to a tree, ‘‘That is a tree I don’t have.’’ We are likely to interpret the speaker as meaning that he or she does not own a tree of that type. It would be quite strange if he or she actually meant to point out that he or she did not own that particular tree. In the statement we are looking at, a tree I don’t have is not interpreted to mean that the driver does not own that particular tree, but rather that there is a counterfactual relation between the blend and the input with the driver parking his or her car at his or her home: there is no tree in the corresponding spot at his or her home, and no collision when he or she drives through that spot, either. Very gener- ally, when disanalogy operates on existence of a value for a role, that disanalogy is a good candidate for compression into nonpossession, as in That car does not have air conditioning, Arkansas has no coastline, Africa does not have bears, and My house doesn’t have that porch. Caffeine headache, money problem,andnicotine fit are straightforward phrases— for a headache that comes from lack of coffee, a problem that consists of a lack of money, a fit brought on by lack of nicotine, presumably from not smoking enough—all set up an integration involving a counterfactual link between spaces. Caffeine headache brings up two situations, one in which you have your coffee and one in which you have a headache. There is evident identity, analogy, and disanal- ogy between these two situations: in both, it’s late morning, and you are at work. But there is the coffee only in the first and the headache only in the second. A blended network is constructed in the following way: there are input spaces cor- responding to the two contrasting situations, links of analogy, disanalogy, and identity between them, and projection of the frame of morning activities from both inputs to the blend. From the input with the headache, we project the headache. conceptual integration 387 From the desired input, we project the causal relation and the causal element. In the blend, the headache is now the effect of something (see figure 15.4). The blend is the new construal of the situation. The input with coffee is coun- terfactual with respect to the blend. In the blend, there is a counterpart for coffee that causes the headache. It is what we refer to by means of the expression absence of coffee. The expression caffeine headache brings in the label ‘caffeine’ from the coffee element in the counterfactual input and applies it to its counterpart in the blended space (see figure 15.5). In the linguistic construction shared by caffeine headache, money problem, and nicotine fit, the first noun picks out the element in the desired input whose absence in the blend is causal for the unwanted state, and the second noun picks out the bad state that obtains in one of the inputs and in the blend. So we have, for ex- ample, security problem, arousal problem, insulin coma and insulin death (in the case of hyperglycemia, which results from absence of adequate insulin), food emergency, honesty crisis, and rice famine. These examples demonstrate the way in which blending has multiple possi- bilities. For example, we could read caffeine headache as referring to a headache ‘caused’ by the caffeine. For both networks, there is a cause-effect relationship in the blend, in the first case between ‘absence of caffeine’ and ‘headache’, in the second between ‘presence of caffeine’ and ‘headache’. In both, the Cause-Effect Vi- tal Relation is further compressed into property. There can now be caffeine head- aches, whisky headaches, and sex headaches. In just the same way, we have missing tooth, absent students, and a gap in the fence. Figure 15.4. The caused headache network 388 mark turner Blending thus creates important elements that have the property of being non- things. Similarly, non-events and non-actions are nearly everywhere in our cogni- tion. Physical reality is a material anchor for conceptual blends that typically carry many projections from counterfactual spaces. The jar-lid won’t come off, The stack of books has not fallen, The stack of books will fall, The jar-lid refuses to come off, and The stack of books wants to fall over all present networks in which one input has nothing happening and the other input has something happening. In the blend, the nothing happening becomes an event that is contrasted with the other event: the stack of books stays upright versus the stack of books falls. Missing a shot evokes a blend that contains a non-event: both inputs have the shot, one input has the ball going somewhere other than into the goal, the counterfactual input has the ball going into the goal. In the blend, the ball’s not going into the goal becomes a ‘missed shot’, a non-event. The notion of ‘absence’ is not explicitly indicated by any part of the expression caffeine headache. It emerges from the entire network, as prompted by this gram- matical construction. But there are linguistic expressions for indicating this com- pression explicitly: absence of, lack of, want of, even no,asinI have a no-caffeine headache. These counterfactual networks are frequently very hard to notice, since we con- struct them so effortlessly as part of backstage cognition. Consider, for example, the report in USA Today for January 31, 2000, of that year’s Super Bowl. In the last play of the game, the ball carrier for the Titans was tackled one yard from the goal line. We unavoidably construct the contrasting space in which the runner advances Figure 15.5. The caffeine headache network conceptual integration 389 . blends out of the same inputs. The process is the same in all of them, but the results are different. In The shovel is safe, the child is the victim in the blend if we are concerned about the shovel’s. space, we have the event of 1853; in a second, we have the event of 1993. These are the input spaces to the blend. There is a partial cross- space mapping that connects counterparts in the input mental. events involving the ‘ghost’ in the blend are constrained by the events involving its ancestor counterpart. In the Northern Light example, the run of the ghost must be the same as that of its ancestor.