Challenges to Blending Theory
© Mark Turner 2013
Challenge (1): Isn’t blending just epiphenomenal, a kind of linear sum over many other processes (counterfactual thinking, metonymy, categorization, metaphor, other “rhetorical” forms of thought, etc.) that already have names? Why lump them together? Answer: Scientific generalization consists of locating substantive and systematic patterns that run across many apparently different products and effects. Science seeks generalizations and investigates whether hypothesized generalizations are solid and useful. We should focus on what we think is actually happening mentally. What labels we apply when we agree on hypotheses for which there is good evidence is only a matter of shorthand. It is easy to fall prey to a nominalist bias—if a language has a noun for something, it must somehow exist; for example, we have the word “metonymy,” and can feel that there must be an objective category of “metonymy” and “metonymies.” It is even easier in the study of language and mind to fall prey to a bias about mental causes, such that, since there are “metonymies,” there must be a dedicated process, a “metonymic thinking” that produces those products. But such words are only efficient labels, not analyses. Where we find commonality of mental process, we should model it; where we find distinctions of process, we should model them. What words we use to label parts of our analyses is a matter of rhetorical convenience. Blending theory hypothesizes that blending is not epiphenomenal; that there are important patterns of process and intricate mechanics and patterns of compression that run over these apparently many different kinds of products. Such a view is quite common elsewhere in cognitive science. For example, the classical computational view that neurons spike or do not, generating a code (all those 1s and 0s), sees the process as operating over very different domains and phenomena, and in different neural real estate, but models that process uniformly and demarks exceptions and nuances here and there.
Challenge (2): Everyone has known about blending—the mental combination of old things to get new things—for a long time. So what is new? Answer: If we agree on the power of blending and the need to study it, then, united, we can plant the flag and turn to the minor parts of the challenge. Gilles Fauconnier and I have surveyed the work of invaluable thinkers dating from classical antiquity, including Aristotle, who analyzed particular products of blending quite insightfully, and who sometimes even commented on the general mental power of combining ideas. Yet, my view is that these excellent forerunners typically thought of blending as an exotic, exceptional, cognitively-expensive event, used exclusively in rare moments of high creativity, rather than as a basic mental operation, non-costly, constantly deployed in everyday cognition by every cognitively modern human being. Additionally, modern blending theory has proposed that there are overarching systematic principles of blending, generic blending templates, intricate mechanisms, and constraints that run across swaths of mental work whose products look quite different. Support for the view that blending theory’s original assertions were new and non-obvious is available from the fact that they were roundly resisted in cognitive science and cognitive linguistics at the outset. If they look old and obvious now, that is progress, and we can move forward with specific analyses. Priority of claims is not an interesting scientific topic, but, still, I find no awareness of the intricate mechanics and pervasiveness of blending and compression prior to the original Fauconnier & Turner work.
Challenge (3): If blending exists, shouldn't we be able to program it algorithmically? Isn't that how we specify a mental process in science? Answer: I have been lucky to participate in projects dedicated to exploring the ways in which blending theory might serve the valuable field of computational creativity and inference. Computational modeling has much to teach us. Yet, the mental process itself does not appear to me to be algorithmic. It is an important part of the flexibility of blending that outputs are not determined algorithmically from inputs, or, at a minimum, computer science, artificial intelligence, and psychology have not been able so far to characterize “inputs” to blending in such a way as to make computational models of blending more than suggestive and illustrative.
Challenge (4): Where are the quick psychological tests to show us how blending works? Why don’t you put people in the fMRI machine so we can see where and when blending happens and how it works? Answer: The empirical evidence for blending is at the moment mostly in the classic form of making generalizations over in-sample data (grammatical patterns, for example) and then testing those hypotheses against out-of-sample data to determine whether they in fact apply. But experimental tests are also worthy. Turner (2010) surveys possibilities for such experiments and the great obstacles to experiment design. McCubbins & Turner (2013) discusses a set of experiments we designed and ran to begin to locate which patterns of blending are more or less tractable for subjects. Turner (2014, appendix, “The Academic Workbench”) reviews possibilities for future experiments and tests. But the tension between blending theory and the expectations of popular psychology runs deep. A way does not yet seem to have been found to get past the gross and indirect nature of current techniques for “reading the mind” to pick blending directly out. In Gilles Fauconnier's phrase, blending is often deemed to have neural correlates, “hardly a surprising assumption. But the correlation is complex: blending creates networks of connected mental spaces, a 'higher level' of organization, if you like. It is presumably not itself a primitive neural process. It is however a capacity of some brains, and perhaps an exclusively human capacity in its double-scope form.”
Challenge (5): But we have no awareness except in rare moments that we are doing any blending, so what is the evidence that it is so? Answer: Awareness is immaterial. We have no awareness during vision that our brains are doing fabulous work, that 50% of neocortex is implicated in this work. In awareness, we just open our eyes and the visual field comes flooding in. That is utterly wrong, as any physiologist of vision will attest. Vision takes spectacular work, and the only moment we are likely to notice that anything is going on is when it fails, as when we have motion sickness and the visual field starts to “do” utterly inexplicable things. It's just so with blending.
Challenge (6): Isn't blending theory incomplete? Answer: Absolutely. In fact, “blending theory” is more a framework for a research program on conceptual integration, compression, mapping, and so on than it is a theory. Many new insights have been generated inside this framework since its proposal. Its original architects eagerly look forward to future developments, which are sure to be surprising.