Narrative and Abstraction: Some Problems with Cognitive Metaphor

I’ve had problems with cognitive metaphor theory (CMT) since Lakoff and Johnson published Metaphors We Live By (1981) – well, not since then, because I didn’t read the book until a couple of years after original publication. It’s not that I didn’t believe that language and cognition where thick with metaphor, much of it flying below the radar screen of explicit awareness. I had no trouble with that, nor with the idea that metaphor is an important mechanism for abstract thinking.

But it’s not the only mechanism.

During the 1970s I had studied with David Hays in the Linguistics Department of the State University of New York at Buffalo. He had developed a somewhat different account of abstract thought in which abstract ideas are derived from narrative – which I’ll explain below. I was reminded of this yesterday when Per Aage Brandt made the following remark in response to my critique of Lakoff and Turner on “To a Solitary Disciple”:

Instead, the text sketches out a little narrative. The lines run upwards, the ornament tries to stop them, they converge and now guard, contain and protect the flower/moon. This little story can then become a larger story of cult and divinity in the interpretation by a sort of allegorical projection. All narratives can project allegorically in a similar way.

Precisely so, a little narrative. Narratives too support abstraction.

My basic problem with cognitive metaphor theory, then, is that it claims too much. There’s more than one mechanism for constructing abstract concepts. David Hays and I outlined four in The Evolution of Cognition (1990): metaphor, metalingual definition and rationalization, theorization, and model building. There’s no reason to believe that those are the only existing or the only possible mechanisms for constructing abstract concepts.

In the rest of this note I want to sketch out Hays’s old notion of abstraction, point out how it somewhat resembles CMT and then I dig up some old notes that express further reservations about CMT.

Narrative and Metalingual Definition

The fact that various episodes can exhibit highly similar patterns of events and participants is the basis of Hays’s (1973) original approach to abstraction. He called it metalingual definition, after Roman Jakobson’s notion of language’s metalingual function. While Hays’ notion is different from CMT of Lakoff and Johnson, I do not see it as an alternative except in the sense that perhaps some of the cases they handle with conceptual metaphor might better be explicated by Hay’s metalingual account. But that is a secondary matter. Both mechanisms are needed, and, as I’ve indicated above, a few others as well.

Let us start with an example of the metalingual function. As Jakobson defined it, the metalingual function is the capacity for language to talk about language. When, for example, you ask for clarification – “What’d you say?” – you are employing the metalingual function to direct your addressee’s attention to their language. Now consider this diagram, which depicts two episodes (indicated by the square nodes):


In one episode (at the left) Mrs. Poyser is saying something; her mouth is moving and sound is coming from the direction of her head. What she’s saying is that Tottie wants a cookie (at the right). The metalingual arc (MTL) connects the content of her assertion to a node that, in effect, represents her speech stream.

In metalingual definition, an abstract and generalized story is used to define the meaning of a word. That is to say, an episodic pattern is used to define the meaning of a systemic node, with the head of the episodic structure being connected to the systemic node with a metalingual arc. Thus one might say that charity is when someone does something nice for someone else without thought of reward. That is depicted in diagram below (without a lot of messy detail).

MTL def

The point is that charity inheres in the entire pattern of relationships over entities and events rather than being some specific participant in the act. Any specific event that conforms to the overall pattern will be considered an act of charity. Note also that, because abstractly defined concepts can participate in language episodes just as concretely defined concepts can, that it is possible to have metalingual definitions nested within one another. In this case, reward would seem to be abstract. Rewards can take many physical forms; apples and oranges, pieces of candy, a piece of jewelry, a vacation in Bermuda, various forms of currency, a kiss on the cheek, etc. – all these and more can serve as rewards. There is no way this variety can be characterized by a common set of physical characteristics, hence the notion of reward needs to be given an abstract characterization. The abstract concept of charity thus has the abstract concept of reward embedded within it. If the definition of reward is through the mechanism of metalingual definition, then we have one metalingual definition (reward) embedded within another (charity). If the definition of reward proceeds through some other mechanism (such as conceptual metaphor) then we have one kind of abstraction embedded within another.

Narrative and CMT

Now consider the following pair of diagrams, which is derived from my work on Shakespeare’s Sonnet 129:


On the left we have a diagram for the little narrative that Shakespeare packed into lines seven and eight: “Past reason hated as a swallowed bait/ On purpose laid to make the taker mad”. On the right we have a diagram for the lust sequence that underpins the poem’s main narrative: someone sees a suitable sexual partner, is so consumed by lust as to do anything to possess that object; has sex; and afterward is consumed by guilt and regret. The two diagrams are the same, point for point.

CMT would have us see one of those as the source domain and the other as the target. As in the journey mapping:


In that mapping, the concrete business of taking a journey from one place to another (on the left) is mapped onto the rather more abstract business of living one’s life (on the right). The journey domain is the source while the life domain is the target.

But that’s not what Shakespeare is doing. The lust domain is well established in the poem before Shakespeare introduces the animal analogy. And the animal analogy can stand on its own. The two narratives have the same (abstract) form, but neither can reasonably be said to the ground of the other. Rather both are instances of some third still more abstract pattern, one that is unnamed.

What, then, is the point of introducing the animal narrative? I suggest – though here we’re beginning to wander away from my main argument – he introduces it because it has a term that isn’t in the lust narrative. Someone has laid the animal bait on purpose. What is the agent that has laid the lust bait, and what is it’s purpose? Who or what is that secret agent? Does the concluding couplet shed any light on that question?

All this the world well knows; yet none knows well
To shun the heaven that leads men to this hell.

I’ll leave those questions as an exercise for the reader. I want to return to CMT and, in particular, the journey metaphor.

The Journey Metaphor: What's a Domain?

I've had problems with the use of journey as a source domain for some time, but it's only recently that I've had anything useful to say about it. I've been working on music and ended up reading some of the literature on navigation (human and animal) and spatial cognition (see citations below). I now have a few comments that folks might find useful.

In thinking about language, cognition, mind, etc. I invariably end up with the brain. So, one of my concerns about the journey domain is: Where do we find it in the brain? It is not at all obvious just how to go about locating the various source domains in the brain. It could be that at least some of them correspond to more or less functionally distinct neocortical regions. It could also be that a bunch of them are overlapped and intermingled in the same (possibly distributed) volume of neural tissue. And there's no a priori reason why we shouldn't expect both.

However, consider what you have to do to actually execute a journey. You've got to move the body from one place to another. All by itself that requires quite a large chunk of cortical and subcortical tissue. You also have to navigate from your starting point to your finishing point. And you have to eat, drink, sleep, and be merry on the way. It thus seems that, to execute a real journey, you have to use the whole brain.

That's not a very useful conclusion – hence one of my problems with the journey domain. However, maybe our primary interest is in the control structure that pulls it all together. That would seem to be a more limited set of neural structures. The navigation system would seem to be in limbic cortex, the hippocampus and associated structures. Of course, the navigation system is not the top-level control structure for journeying. The top-level control system is the one that tells as when to perform a navigational computation (fix current position, establish next line of travel), when to move, and when to eat, drink, sleep, and be merry. But the general idea is that we're interested in how all this stuff is controlled.

This line of thought occurred to me as I was thinking about how a brain would improvise jazz. Of course, there could be a special music module, with jazz attachment, but I happen to think the whole mental modules line is, at best, not helpful, but more likely it's dangerous nonsense. I'd rather think about how to make used of existing equipment. So, the scheme that occurred to me is that the brain could treat the chord changes (e.g. the standard I IV I V IV I blues progression) as a landscape. The act of improvising then becomes one of navigating through the landscape.

Thus, during a real journey navigational control is normally mapped to visual, auditory, olfactory, sensors tracking the external world. Similarly, the locomotor control is mapped to the legs. But, during a jazz journey, we map navigational control to both the internalized memory of the chord progression and to the auditory world of sound actually being created. And we map locomotor control to the muscles playing the instrument.

Note that, in suggesting this, I'm NOT talking about the CONCEPTS we use to talk and think about jazz improvisation. That's a whole other discussion, one that doesn't interest me at the moment. I'm talking about how we actually perform an improvisation, or, by implication, how we understand an improvisation someone else is performing. So, we have the neural configuration required to execute a real journey in the external world, and we have the neural configuration required to execute a (virtual) journey in the musical world.

It’s not clear to me just how these two configurations are related to that required to use and understand the LIFE IS A JOURNEY or LOVE IS A JOURNEY mappings. I observe, however, that the Berkeley group's neural modeling makes use of a model of human locomotor control. And it’s the control structure of that model that is, e.g. determining verb aspect.

One could, however, argue that this whole approach to domains is simply wrong. Domains aren’t in the brain. They’re in the world.

OK, so the journey domain is ‘out there’ in the world. Doesn’t that effectively make the whole physical world into the journey domain? Where in the physical world can’t we journey to? Well, we can’t go to the center of the earth or to Alpha Centauri. But that’s not a matter of principle, that’s simply a matter of contingent fact. Excluding those as potential destinations doesn’t much help us with the problem: Is there any way to specify a journey domain that doesn’t include everything one can see, hear, and do in the world?

On to the next problem:

What's Concrete And What's Abstract?

It seems that one of the primary things we want from those two mappings is the capacity to see life/love problems and achievements/goals as analogues to physical obstacles and goals in a real journey.

So, what's the image schema for the prototypical obstacle? There isn't one. There's a whole mess of them. For lots of things can be an obstacle in a physical journey – a log across the road, a pack of wolves, an avalanche, a fallen bridge, etc. Each has it’s own prototypical form and image schema, but there is no “covering schema” for the lot. If that is the case, then how can the notion of an obstacle be a concrete notion? And if it isn't concrete, then it’s useless as the ground for an abstraction. You can run through the same drill for goals – landmarks and destinations come in all shapes and sizes, and even for whole journeys.

Sure, real journeys are physical events in the physical world. You can see and hear and smell them. Etc. But the notion of a journey itself, its goals and obstacles, that would seem to be abstract.

Unless, once again, we focus on control structure and control events. What does an obstacle mean to the control structure? No matter what the obstacle is, you have to stop and you have to do something. Similarly, landmarks, no matter what they are, require that you fix your position, check it against your itinerary, and plot the next leg (if there is one). Thus it would see that what all journeys have in common is a control structure.

So, the journey domain that serves as the target domain in metaphor mappings is best conceived of as a control structure. And, when you consider evolutionary matters, it’s obvious that this control structure must be very flexible and powerful, and ancient. All animals have to make their way around and about in the world. Primates in particular are foragers. We must know our territory well in order to survive. And my colleague Valerius Geist maintains that what turned a bunch of clever apes into human beings was a long march across the ecologically demanding African steppes.

This gets at one of my general difficulties with cognitive metaphor. The whole thing seems like a giant passive data structure that takes the human body as its template rather than set theory or predicate calculus. Well, that’s a difference. But a data structure is a data structure. It’s passive, it’s static.

Once you start storing control structures and processes in your database – as you have to do in order to make the journey mapping work – you've got a different ball game. You have to start thinking about processes. There is no doubt a pile of recent literature here and there that speaks to this issue, but I'm not up on it. So I suggest two ancient texts, Norman and Rumelhart, 1976 (pp. 35 - 65), which I'm sure some of you read back in the Jurasic era, and Hays, 1981 (pp. 5 ff., 84 ff).

Now for a bonus:

Cognitive Domains and Mental Spaces

As I understand it, cognitive domains are more or less permanent mental structures. Mental spaces are temporary structures, constructed on-the-fly during thought and speech. We've got metaphor theory constructed in terms of permanent cognitive domains and blending theory constructing in terms of temporary spaces. It would be nice to put them together in a uniform framework.

That's certainly more than I can do, much less do in a brief and informal email. But at least some of the neural machinery involved overlaps with that involved in controlling journeys.

The hippocampus is part of the limbic cortex; that is to say, it is phylogenetically old tissue, older than the neocortex. The rat hippocampus has been classically associated with spatial cognition. There are hippocampal cells called place cells that are active only when the rat is at a certain position in its world. Different place cells are sensitive to different locales. So, there's been a great deal of experimental work on the role of the rat hippocampus in spatial cognition and navigation.

The human hippocampus, at least superficially, seems to be a different beast. When it is destroyed the person is unable to learn anything new (at least in so-called episodic memory). They have normal recall for events that happened before the injury, but no recall for events that happened after the injury. Patients can carry on a normal intelligent conversation, and 30 seconds later they've forgotten it. They thus seem unable to transfer the content of a temporary mental space to long-term memory. Long-term memory is not necessarily, of course, the world of all those cognitive domains, which would be part of so-called semantic memory (rather than episodic).

Now, what, if anything, do spatial cognition and episode retention have in common? That's a tricky question, though I observe that journeys typically have many episodes, etc. David Redish (1999) has recently reviewed the hippocampal literature and reported on some computer simulations of his own. He thinks the common link is control of context and the ability to segment the experiential stream into discrete chunks. I can't summarize his argument here, but it's worth thinking about.

My general point is that it now seems to me that an inquiry into the neural underpinnings of journeying will also be an inquiry into the neural underpinnings of the relationship between mental spaces and permanent cognitive structures.


William Benzon. Cognitive Networks and Literary Semantics. MLN, Vol. 91, pp. 952-982, 1976.

Willam Benzon and David Hays. The Evolution of Cognition. Journal of Social and Biological Structures 13(4): 297-320, 1990.

C. R. Gallistel, ed., Animal Cognition, MIT Press 1992. This is a reprint of Cognition, 37 (1990). See Gallistel's introduction and papers by Gibbon and Church, Church and Broadbent.

Reginald G. Golledge, Wayfinding Behavior, Johns Hopkins Press, 1999.

David G. Hays. The Meaning of a Term is a Function of the Theory in Which It Occurs. SIGLASH Newsletter 6: 8-11, 1973.

David G. Hays. Cognitive Structures. HRAF Press. 1981.

Donald A. Norman and David E. Rumelhart, Explorations in Cognition, WH Freeman, 1976.

A. David Redish, Beyond the Cognitive Map, MIT Press, 1999.