This document is also available as a postscript file.
In this paper I review some of the theoretical issues surrounding metaphor, and trace them through the context of the cognitive neuroscience debate. Metaphor, like all figurative language, has been usually explained as a secondary linguistic process which takes place as a function taking place on literal language. However this explanation does not fit well with some of the recent work on right hemisphere processing of language or recent cognitive studies, both of which suggest that the figurative and literal language are processed simultaneously and share much structure. In seeking ways to operationalize the Lakoff and Johnson view of metaphor as a constitutive cognitive phenomenon, I begin to spell out what kinds of theoretical predictions the Lakoff-Johnson model would make on the neurophysiological levels cognitive investigation. I conclude by offering some rudimentary thoughts on possible proposals for further investigation using these methods.
Whatever the skill employed in thought-that of logic, mathematics, language, spatial or musical symbols-we must not forget that it is driven by the Jamesian processes, undergoes flights and perchings, is susceptible to great variations in attention, and in general, is fueled by metaphorical and metonymic processes. (Gerald Edelman, Bright Air, Brilliant Fire, p. 174)
Though it has long been recognized that most of our everyday uses of language involve metaphor, with a few notable exceptions (such as Winner & Garner, 1977) the enterprise of cognitive science has largely ignored the investigation of figurative language in favor of investigating literal language until recently. Of the many factors which contributed to the paucity of research on figurative language comprehension, the instantiation hypothesis is perhaps the most onerous. The instantiation hypothesis argued that reason, intelligence and minds were substrate neutral, that is, independent of any specific embodiment, so long as the body was a algorithmic device. The computational device of choice was a serial processor driven digital computer. Since the mathematician Alan Turing proved that all digital computers were in principle reducible to recursive elaboration of a finite state algorithm (a 'Turing machine'), minds (and mental 'processes') were in principle reducible to finite state algorithms. Finite state algorithms have a peculiar literal quality in that their variables are either true or false with no admixture of truth or falsity permitted. Upon this view the fundamental problem of language comprehension was determining how the brain's representations and the world literally matched up: the world outside the brain was thought to be represented symbolically inside the brain by a series of finite states. If the mind was the kind of software program running on the that kind of hardware, the lack of attention given to figurative language comprehension results from an obvious source: it would be a mere afterthought to solving the problem of literal language comprehension, since Turing had proven that all computational processes must ultimately decompose into finite states-that is, into literality. The mantra of this dogma was clear: solve how language represents and the problems posed by figurative language will inevitably solve themselves.
Unfortunately for proponents of the instantiation hypothesis, many of the recent findings in cognitive science are motivated by its antithesis: the embodiment hypothesis. The embodiment hypothesis argues that minds are fundamentally not disembodied algorithmic processes like a computer program, but are instead constituted and constrained by the kinds of organization reflected in the biological, anatomical, biochemical, and neurophysiological characteristics of the body and the brain. While both hypotheses share the materialist assumption that higher-level processes, such as abstract thought, language comprehension and the like, are built up out of lower-level processes, the embodiment hypothesis explicitly denies the substrate neutrality claim of the instantiation hypothesis proponents. Rather than processes which manipulate symbols in a finite-state fashion, the study of mind as a biological and neurophysiological enterprise requires thinking about the mind in a series of interwoven levels of investigation, many of which appear to behave analogically and frequently exhibit a kind of adaptive plasticity not found in digital computers. A good cognitive theory requires a kind of vertical convergence across a number of different levels of investigation ranging from philosophy and cognitive psychology through neuropsychology to neuroanatomy and neurochemistry.
Figurative language comprehension has robust connections with the embodiment hypothesis, especially in the area of metaphor. As George Lakoff (1987) and Mark Johnson (1987) have argued, our ordinary use of language is largely structured by metaphoric and metonymic principles which exhibit a directionality. Human beings systematically characterize abstract ideas-thoughts, religious beliefs, political and ethical situations-in terms of bodily movements and bodily functions, for example. The primary claim of their position is that these metaphors and the directionality are not arbitrary, but instead are a natural outgrowth of the manner in which our minds are constituted. In this paper I seek to tie together threads from the philosophical, cognitive, and neurophysiological levels. In operationalizing the Lakoff-Johnson hypothesis I have made use of Posner's schematization of cognitive science (table I) as a framework for seeking a vertically integrated cognitive theory.
Conceptual Metaphor Theory Level of Investigation Methods of Study
Tasks
Understanding a metaphor Cognitive Systems Verbal report,
textual analysis,
discourse analysis
Cuing and retrieval of a Mental Operations Cognitive studies
conceptual metaphor examining RT
Facilitation of related Performance Domain RT in priming
information, inference experiments with
generalizations polysemic words, ERP
studies of the time
course involved
Activation in Neural Systems lesion, FMRI, PET,
somatosensory, auditory, sometims in
and visual processing combination with ERP
areas when processing
metaphorically constituted
abstract concepts
Intraneural connections Cellular Electrocellular
from visual, auditory, stimulation, and
somatosensory region to anatomical methods,
language areas supplemented by
computer simulations
Table I-Overview of the proposed operationalization of the conceptual mapping hypothesis on metaphor in relation to Posner's conceptual framework for characterizing the levels of investigation in cognitive science
Most philosophers have generally argued that figurative language involves tricks or plays on the literal. John Searle (1979), for example, argued that metaphor is a simply a roundabout way to express literal semantics. I have schematized Searle's view of metaphor in figure 1. According to Searle's model, all utterances would be processed as literal utterances first. Only once the mind was unable to find a literal meaning for an expression would the utterance be sent to a special non-literal processing center for decoding. Searle's view necessarily entails that the comprehension of a metaphor assumes the metaphorical expression will eventually be decomposed into a literal paraphrase.
Figure 1-A schematization of Searle's sequential model (1979) of metaphor comprehension
On the face of it Searle's model has much going for it, for it fits with our common sense experience of understanding language and speech. However that common sense understanding fundamentally confounds the distinction between literal and figurative speech with another distinction: that between attended and automatic processes. In our common sense understanding we think of metaphors as unusual utterances which require our attention, such as we encounter in poetry or theater. We think of a metaphor as something jarring which interrupts our ordinary way of thinking about the world and challenges us to enter into another way of thinking. However, Lakoff and Johnson have shown that much of our everyday language, including what we would ordinarily call literal language, is structured by conventional metaphors. In other words, the bulk of our metaphoric processing is automatic-only some metaphors require attentional processing. Couple their observations with the fact that some forms of language which we would ordinarily consider literal (such as scientific reports) also requires attended processing even though most forms of literal language (such as social interactions, reading a newspaper and the like) are also processed fairly automatically, and Searle's model of how we understand metaphor begins to pale. This realization suggests two different kinds of theoretical models: one in which both literal and metaphorical processing are done in parallel and another in which both literal and metaphorical processes are largely the same process.
A number of cognitive methods can be applied to help adjudicate between these three theoretical models. As Searle's model is a sequential hypothesis, it requires that a metaphorical utterance first be processed as if it were literal, judged non-literal and then alternate strategies are employed to decode the utterance into literal. Presumably all these actions take place in real time, which would imply that metaphorical utterances take longer to understand than literal utterances. However, many studies of figurative language-on idioms, proverbs and metaphor-show this is not the case (see Gibbs 1994, pp. 92-108 for a useful overview), provided there is sufficient context. Ortony, Schallert, Reynolds & Antos (1978) measured the time it took subjects to comprehend literal v. metaphorical sentences at the end of long and short contexts, and reported no difference for long contexts but that metaphorical sentences took significantly longer to read in short contexts. Janus & Bever (1985) tracked eye movements and compared the amount of time people focused on the target sentences. Subjects again responded as quickly in the long context condition. These results contraindicate the Searlean sequential model of metaphor processing; additionally, the possibility that metaphorical contexts are 'chunked' and processed as semantic units is contraindicated by the differing results in the long v. short context conditions. (Presumably if metaphors were chunked in a fashion similar to lexemes they would have been retrieved in nearly equal times for either the short or long context conditions--this was clearly not the case.) Finally, it should be noted as a caveat that other studies (Blasko & Connine, 1993) show that highly familiar metaphors are understood more quickly than novel metaphors-in other words, these processing time studies must be taken cautiously as they show practice effects.
While these type of cognitive studies adequately contraindicate the Serlean model of metaphorical processing, they have not yet proved as revealing in determining whether literal and figurative language processing are parallel processes or are largely the same process. With respect to idiomatic language Gibbs (1980, 1986) has shown that subjects take less time to read idiomatic phrases when the context supports an idiomatic interpretation than the same phrases in contexts supporting a literal interpretation. The differing results with respect to context suggest that the literal and figurative language comprehension processes operate in parallel and the context primes one process or the other. Similar results have been obtained with regard to metaphor (summarized in Gibbs, 1994, p. 101-4). Yet subjects seem unable to ignore a figurative meaning even when instructed to focus exclusively on the literal context (Keysar, 1989). Keysar investigated whether metaphor and literal interpretations of a context would produce a Stroop-like interference effect when a target was, for example, both metaphorically true and literally false. Reaction times increased in the invalid condition but decreased in the valid condition, suggesting that these processes may well at least share component subsystems. Ultimately cognitive methods alone may not be able to decisively resolve the parallel and same process debate because that question may be better posed at other levels of investigation than the cognitive systems level.
Finally, there has recently been an interesting exchange between the Lakoff-Johnson-Gibbs approach and the largely complementary Glucksberg-Keysar-McGlone approach. Along with Lakoff and Johnson, Gibbs (1992) claims that long-term memory is structurally organized by prototypes extended by metaphoric and metonymic principles called conceptual metaphors or conceptual mappings. A conceptual metaphor, such as LOVE IS A JOURNEY, is constituted by a mapping between areas of the brain, such as between affect and sensorimotor areas. In the appropriate context, most conventional metaphoric expressions, such as we're at a crossroads or our marriage is on the rocks, access these conceptual metaphors from long-term memory. Glucksberg, Keysar and McGlone (1990) argue for a class inclusion view in which some metaphoric expressions build up an ad-hoc category in working memory rather than accessing conceptual metaphors from long-term memory, even when it would be expected that they would draw on a conceptual metaphor. Gibbs (1992, 1994) argues that the class-inclusion view requires an understanding of metaphor in which each metaphorical expression creates a unique or novel mapping in working memory. In their reply Glucksberg, Keysar and McGlone (1992) suggest that only some cases require the development of ad-hoc category, citing a brief initial experiment in which subjects were given metaphorical with minimal context and asked to paraphrase them.
Some of the strongest evidence against a purely parallel processing model for figurative and literal language comprehension comes from a number of studies on right hemisphere-damaged (RHD) patient populations. One possible localization of these parallel processes would suggest that they could be lateralized with respect to brain hemispheres. In an early study often cited as establishing that metaphoric processing is right hemisphere dependent, Winner & Gardner (1977) compared left hemisphere-damaged (LHD) aphasic patients, RHD patients, bilaterally damaged patients and a non-neurological group. Participants were presented with a figurative sentence such as "he has a heavy heart" and asked to perform two tasks. One task involved matching the sentence to one of four pictures, one of which literally represented the sentence (i.e., a man carrying a large heart), an appropriate metaphoric interpretation (a crying man), a salient quality depicted by the metaphoric adjective (a 500 lb weight) and one which illustrated the noun (a large heart). In the second task patients were asked to verbally explain their choices. While RHD patients selected the metaphoric picture much less frequently than LHD patients or the control group in the first task, in the second task RHD patients were able to verbalize their choices using figurative language whereas LHD patients were unable to explain their 'correct' choices verbally. Winner & Gardner argue that these results show that the effective interaction of the hemispheres was important in appreciating figurative meaning. Though this study clearly shows that the figurative and literal language comprehension processes are not parallel processes in differing hemispheres, the study also indicates that at least some subcomponents of language comprehension concerned with metaphoric processing are RH dependent. In short, the RH makes a positive contribution to figurative language processing. This kind of lesion evidence further suggests that parallel processing of literal and figurative language comprehension is too crude a level of analysis to be a useful description at the level of localization, suggesting that the figurative and literal language processing share at least some but not all components.
This conclusion was underscored in a related study by Brownell et al. (1990) which also showed that RHD patients were more impaired in pairing a word with a metaphoric synonym than LHD patients. In addition to providing further support for the idea that the right hemisphere makes a unique contribution to figurative language processing, they also report that LHD patients were affected by the degree of semantic similarity between the target and its synonym in the metaphoric-but not the literal-condition. This finding is intriguing because it suggests that the right and left hemispheres may code semantic information in different ways. Work on RH contributions to semantic priming (summarized in Beeman et al.1993) generally suggests that semantic memory is either stored differently in the RH or processed differently in the RH. Beeman argues for the latter, suggesting that the RH and LH do operate in parallel with respect to semantic processing, with the RH processing the information in a coarser fashion and the LH in a fine fashion. The coarse-fine distinction is drawn from neural network modelling of vision which suggests that networks which use larger receptive fields code information more coarsely, thereby maximizing responsivity to precisely localizing a source of continuous input. According to this hypothesis the coarser semantic fields of the RH would operate in parallel with the more finely grained semantic fields of the left hemisphere. Figurative language comprehension would then entail a RH activation strong enough to influence LH hemisphere processing.
If Beeman's theory can be adapted to understanding the phenomenon of conceptual metaphor, the context effect observed in the cognitive studies would presumably be instantiated as a build-up of strong patterns of activation in the RH. This explanation is also consonant with the Stroop-like interference effect that Keysar (1989) noted in which the availability of metaphoric interpretations slowed the RT of literal interpretations. By positing a shared semantic network which differs only in the scope and weighting of the neural nets, this theoretical model nicely captures both the shared componentry evidenced by the interference between literal and figurative processing and the parallel processing suggested by the decreased reaction times in the long context condition. Rather than architecturally separate modules for literal and figurative processing, this theoretical model posits a literal-metaphorical continuum instantiated by a varying the organization of neural networks. However, the conceptual metaphor model seems to argue against a strict localization of metaphor in th RH as it requires activation in related sensorimotor areas (i.e. visual metaphors should activate visual areas, etc.). This may prove to be a significant incompatibility with Beeman's hypothesis.
There remain, however, unanswered questions about the role of conceptual metaphors as largely automatic (unattended) processes in the brain. Lakoff (1994) argues that "the system of conventional conceptual metaphor is mostly unconscious, automatic and available to users with no noticeable effort." If conceptual metaphors are largely automatic and metaphoric processing is largely instantiated by coarse semantic fields in the RH, the RH should show a strong and growing activation pattern in the long context condition of the cognitive studies on metaphor. This hypothesis could be investigated using FMRI methodology by reproducing the cognitive studies on metaphor and rather than measuring the RT needed to comprehend a sentence, look for a growing pattern of activation in the RH. Imaging techniques could compare early, middle and late activations involved in the comprehension of the passage. I believe the appropriate subtractions would be the event images minus an image of the subject at rest (either reading nonesense text or 'literal' text). Presumably these subtractions would show greater RH activation late in the passage. If ERPs were used in conjunction with a neuroimaging method, a clearer picture of the time course might provide us with insight about the rate of the expected build-up in the RH.
Another line of inquiry to investigate would focus on the notion of a conceptual metaphor. The conceptual metaphor hypothesis proposes that there will be interconnections between semantic areas and areas tightly related to sensory input. A passage which uses visual metaphors, for example, should activate pathways connected to the visual system and to some degree the areas themselves. A FMRI study could be designed which checked for activation in the visual, auditory, somatosensory and motor cortices after reading or hearing a passage with strong visual, auditory, tactile or bodily movement metaphors. The relevant subtraction would change in each modality, but the basic approach would be to compare a loaded condition minus an unloaded condition; e.g., for the visual modality an image after hearing a passage with visual metaphors minus an image of passive listening to a visual-metaphor neutral text. One would expect to see activations. If there was no such activation, it would be substantial evidence contraindicating the conceptual metaphor view. However, all of these suggestions are subject to the caveat that the activation should be of large enough scale to be picked up by the resolution of the neuroimaging technique. On the neuroanatomical level, pathways should be observable between associated areas.
Finally it strikes me that the differences between the class inclusion view and the conceptual metaphor view may also lend themselves to investigations at the neurophysiological level, but I am not clear on how the distinction between long term and working memory can be made testable at this time, and that appears to be the key bone of contention for the experimental design.
I have provided an initial attempt at linking the conceptual metaphor hypothesis to part of the recent neurophysiological literature on metaphor. In tracing theoretical issues through this debate it has become clear that the differing levels of investigation have historically had differing concerns which are only beginning to be translatable into a vertically integrated cognitive theory which encompasses levels from biology and anatomy through neurophysiology and cognitive studies. I have focused on the relationship between neurophysiology, cognitive studies and philosophy in this paper. Finally, I've proposed some avenues of neurophysiological investigation which are theoretically significant given recent philosophical and cognitive work.
copyright (c) Tim Rohrer---draft--comments are welcomed at rohrer@darkwing.uoregon.edu
Beeman, Mark et al. (1993). "Summation Priming and coarse semantic coding int he right hemisphere." Journal of Cognitive Neuroscience, 6, 26-45.
Blasko, D., & Connine, C. (1993). "Effects of familiarity and aptness on metaphor processing." Journal of Experimental Psychology: Learning Memory, and Cognition, 19, 295-308.
Brownell, Hiram H. et al. (1990). "Appreciation of metaphoric alternative word meanings by left and right brain-damaged patients." Neuropsychologia, 28, 375-383.
Edelman, Gerald M. (1992). Bright Air, Brilliant Fire: on the matter of mind. New York: Basic books.
Gibbs, Raymond. (1980). "Spilling the beans on understanding and memory for idioms in conversation." Memory & Cognition, 8, 449-456.
Gibbs, Raymond. (1986). "Skating on thin ice:literral meaning and understanding idioms in conversation." Discourse Processes, 9, 17-30.
Gibbs, Raymond. (1992). "Categorization and Metaphor Understanding." Psychological Review, 99, 572-577.
Gibbs, Raymond. (1994). The Poetics of Mind: Figurative Thought, Language and Understanding. Cambridge, Eng.: Cambridge University Press.
Glucksberg, Sam, B. Keysar and M. McGlone. (1990). "Metaphor understanding and accessing conceptual schema: Reply to Gibbs (1992)." Psychological Review, 99, 578-581.
Janus, R., & Bever, T. (1985). "Processing metaphoric language: An investigation of the three stage model of metaphor comprehension." Journal of Psycholinguistic Reseach, 14, 473-487.
Johnson, Mark. (1987). The Body in the Mind: The Bodily Basis of Meaning, Imagination and Reason. Chicago: University of Chicago Press.
Keysar, Boaz. (1989). "On the functional equivalence of literal and metaphorical interpretations in discourse." Journal of Memory and Language, 28, 375-385.
Lakoff, George. (1987). Women, Fire and Dangerous Things. Chicago: University of Chicago Press.
Lakoff George. (1994). "The Contemporary Theory of Metaphor." In Ortony, Andrew (ed.) Metaphor and Thought, 2nd. ed., Cambridge University Press.
Ortony, Andrew, D. Schallert, R. Reynolds and S. Antos. (1978). "Interpreting metaphors and idions: Some effects of context on comprehension." Journal of Verbal Learning and Verbal Behavior, 17, 465-477.
Searle, John. (1979). "Metaphor"." In Ortony, Andrew (ed.) Metaphor and Thought, Cambridge University Press.
Winner, Ellen & Gardner, H. (1977). "The Comprehension of Metaphor in brain-damaged patients." Brain, 100, 717-729.