Tuesday, July 14, 2009

Concepts, Cognition, and Anthropomorphism

Ever since Plato, western philosophers have looked at words, and the objects they represent, as a sort of fixed ideal:  the word "couch", for instance, represents an object, or rather one of a class of objects.  Lately, in the last few centuries, scientists and/or philosophers have recognized that the same word might represent different, but widely overlapping, classes under different circumstances (contexts), a fact that often leads to confusion, sometimes deliberate.  All of this tends to focus attention on the relationship between words and symbols, strengthening the notion that humans are the only "Symbolic Species", since the evidence is overwhelming that no other species (with a few possible cetacion exceptions) uses language in anything like the way we do. 

But is that true?  How closely are symbols, and the concepts they represent, tied to language and the use of words as symbols?  For that matter, what is a symbol, in terms of cognition, or the mechanics of our brains? 

Before we explore these questions, let's back up and take a look at the context:  the paradigm within which we are asking these questions, understanding what they mean, and judging the answers. 

It began with Plato, as mentioned above.  From his work we get the concept of the Platonic Ideal:  the perfect representation, the archetype, behind a concept represented by a word.  As he explained it, we are chained in a cave as it were, looking at the back wall, able to see only shadows cast by the ultimate, ideal, reality cast by the light of "Truth", because we are unable to turn and look out the entrance. 

This paradigm was pretty much universal by Hellenistic times (a few centuries after Plato) in the west, and also, I suspect, in the east where it had spread via Persion/Greek influences to India, and from there via Chan/Zen Buddhism (and possibly other schools) across North Asia to China.  (My suspician is partly founded on the fact that Zen Koans seem well designed to break the tyranny of words/concepts as absolutes, forcing the seeker to recognize that these things are only limited tools, like a set of shelves which are fine for holding small solid objects, but limited for very large objects and useless for liquids.) 

The reality, as recent research is showing, is that our brains typically identify words heuristically:  we learn of a number of items that belong to a class represented by a word, our minds (often subconsciously) identify common features or characteristics of these items, and we tentatively identify new members of the class based on those same shared features/characteristics. ...

Backing up again to Plato, we can see that there's no reason that such classes require an "ideal" at all:  they are just groupings of items that are all similar enough to justify attaching a word, a symbol, to the class.  A concrete example may be helpful:  consider the word "couch".  This may be loosely defined as a bench with a back and armrests (at least in modern usage), long/wide enough for at least two people to sit comfortably.  Can we define a Platonically ideal couch?  Not without context.  For instance, to Victorians such a couch would have had a reasonably vertical back and stiff cushions, helping to support people who were expected to sit up with good posture while socially interacting.  But for modern laid-back music lovers, the ideal couch will be much more relaxed, allowing occupants to lean back and totally relax while losing themselves in musical reproductions not available even live to Victorians. 

Thus, the ideal, singular to Plato, splits into many depending on context, observer, and time.  The problem is that our thought patterns, our paradigms, are constrained by over two thousand years of Plato's influence, and they often interfere with efforts to understand how language, symbols, and concepts work. 

It's certain that the human brain has a mechanism for representing words:  my best guess is that they're normally represented (in at least one area of the brain) as abbreviated "shorthands" for the sequence of tongue, lip, mouth, throat, and diaphragm movements that go into pronouncing them.  But is this true of concepts?  Did recent human ancestors invent the concept, or the symbol that stands for it?  Are even our closest relatives lacking this feature? 

I think not.  We know that many primates possess "mirror neurons", pyramidal cells in parts of the neocortex that tend to fire in the same groups when the animal is performing (or preparing to perform) some action and when it's observing some other animal performing that action.  I would say that actions of this sort represent concepts, and that the processes around the firing of these "mirror neurons" represent neurological symbols (or the use of them).  At a sufficiently abstract level, there's no real difference between a specific pattern of lines on a piece of paper (or clay) and a specific pattern of neurons in the neocortex (or a specific pattern of sounds). 

Before we go farther, we need to examine how the mammalian neocortex operates.  I'm going to use Francis Crick's Astonishing Hypothesis as a base, although it's highly simplistic and probably wrong in some details.  As he does, I'm going to focus on the V1 area in the visual cortex, which receives input from the retina (via the lateral geniculate nucleus: a part of the thalamus), performs a complex transformation on it, and sends the results on to further areas of the brain. 

The large majority of axons coming from the retina carry a single type of message:  each neuron is watching a specific location in the retina, looking for a specific type of spot against an opposite background:  some look for big spots, some for small.  Some for dark spots (against a light background), some for light.  some for red spots (against a presumably green background), some for blue, green, purple, etc.  Now all of these things these different neurons are looking for can be mapped onto an n-dimensional space:  in this case with n probably equal to five:  two dimensions for the location within the retinal field, one for size, one for color, and one for light/dark.  The light/dark axis may be discrete rather than continuous; it may even be two-valued.  The color axis also may be somewhat more complex than my simple description makes it sound:  there are some complications regarding the distribution of cone types within the retina that time and space don't permit discussing.  These points don't affect the basic picture:  the visual field of the retina can be mapped as an n-dimensional space of "spots" with each neuron sampling a specific location in that space and responding accordingly. 

Of course, most of the visual field isn't made up of spots (although in the world of the earliest chordates it may well have been:  our evolutionary history is often built into our design in ways we wouldn't suspect).  What each neuron in the retinal population I'm discussing does is look to see how much the visual input resembles the type of spot it's looking for, and the more it does, the larger its firing rate (the faster it fires action potentials).  Notice, then, that what we have here is a continuous function over a continuous space (disregarding complications involving a discrete light/dark axis) varying continuously over time, but sampled at discrete points in space and time.  Each neuron occupies a specific point in the n-space and samples that one point (although since it's actually examining the vicinity looking for a spot, one close to its location would generate a partial signal).  Each neuron intermittently sends a signal with its instantaneous firing rate, thus the closer it is to seeing what it's looking for, the more frequently its signal is being sampled. 

What does the V1 area do with this information?  The major populations of pyramidal cells look for bars:  wide bars, thin bars, vertical bars, horizontal bars, dark bars, light bars, red, blue, green, purple, etc.  And everything in-between.  The information brought from the retina (relayed via the lateral geniculate nucleus) is integrated and messaged by the cortical neuropil and the stellate cells of layer IV, eventually allowing the pyramidal cells to sample an n-dimensional space of bars and send their signal accordingly.  (Actually, the situation is more complex because these cells are also looking for time-dependent patterns as well, but if they can support the more complex situation, they can certainly support the simpler one.) 

Are we beginning to see a pattern here?  Nerve cells in at least two parts of the brain can, in effect, occupy a discrete location in some n-dimensional space, sample a scalar signal at that location, and output the results to their axons, whence it's carried to other parts of the brain.  (The vertebrate retina is part of the brain developmentally:  it starts out as part of the neural tube and is then drawn away into the eye structure.  It's not part of the neocortex, however, the retina had differentiated and specialized a long time before the ancestors of mammals invented the neocortex.[1]

We don't need, here, to go into how this process is achieved:  it's enough to know it can be done.  And what one area of the brain can do, so can any other:  whatever the mechanism, it's almost certainly reusable.  (Note, however, that just because this mechanism is available to every part of the brain doesn't mean every part of the brain uses it.  OTOH when we're looking for a mechanism to explain some observation, it is an obvious candidate.) 

Let's get back to our "concepts" within the brain represented by specific groups of neurons (such as "mirror neurons") firing when a specific "concept" is called to mind in an animal, either by seeing some other animal performing an action or by performing, or preparing to perform, that action.  Based on the description above, it's a good guess that those neurons occupy a specific location in some n-dimensional "concept space" that corresponds to the specific act involved.  Of course, since at this level of consciousness humans, and probably other primates, normally pay attention to only one general subject at a time, it may be that each individual concept is represented by a unique pattern:  sort of like an ideogram but in n dimensions rather than just two. 

Figure 1:  Cursive hieroglyphs (a type of ideogram) from the Papyrus of Ani, an example of the Egyptian Book of the Dead.  (From Wiki)

At this point I want to discuss some very recent research:  Mirror Neurons Differentially Encode the Peripersonal and Extrapersonal Space of Monkeys ( by Vittorio Caggiano, Leonardo Fogassi, Giacomo Rizzolatti, Peter Thier, Antonino Casile), which is unfortunately behind a paywall, but let me blockquote the abstract:[3]
Actions performed by others may have different relevance for the observer, and thus lead to different behavioral responses, depending on the regions of space in which they are executed.  We found that in rhesus monkeys, the premotor cortex neurons activated by both the execution and the observation of motor acts (mirror neurons) are differentially modulated by the location in space of the observed motor acts relative to the monkey, with about half of them preferring either the monkey's peripersonal or extrapersonal space.  A portion of these spatially selective mirror neurons encode space according to a metric representation, whereas other neurons encode space in operational terms, changing their properties according to the possibility that the monkey will interact with the object.  These results suggest that a set of mirror neurons encodes the observed motor acts not only for action understanding, but also to analyze such acts in terms of features that are relevant to generating appropriate behaviors.
Unlike many abstracts, this one is pretty easy to understand, and its implications are extremely relevant to the subject here.  The "mirror neurons" under examination are associated with motor activity, and may well be associated with a system of representing "concepts" with shorthands of motor sequences. 

I mentioned above that in my view the representation for words in the human brain is some sort of "shorthand" for the sequence of muscle actions involved in pronouncing it; similarly it makes sense that in monkeys, apes, and other primates with "mirror neurons" the representation in at least one area of the brain for "concepts" involving actions is a similar shorthand for the muscle actions involved.  Or rather, the shorthand in primates came first, and the process was adapted for use with language by adding specific tongue, lip, mouth, throat, and diaphragm movements to a system already adapted for arm, leg, hand, and foot movements. 

In trying to understand how "concepts" are represented in the brain, we need to consider how they're used.  As Caggiano et al. point out, the relevance of an action undertaken by another depends (in kind as well as quantity) on the distance of that other, and it makes sense that the specific "concept" activated in the brain is slightly different depending on that distance.  Rather than trying to separate these different reactions into completely separate concepts, perhaps we should think of this as more like adverbs modifying a verb:  the same action stimulates the same concept, but the distance, both absolute and operational, stimulates distinct modifying adverbs.  Of course, we must also keep in mind that this research looked at only one area of the brain, and that one being associated with motor activity, specifically the planning and execution of such activity.

We need to keep in mind that the brain is made up of many areas, all talking to one another:  in rhesus monkeys there are 52 areas just on each side of the neocortex, this may also be the number for humans, although the detailed research needed to prove it hasn't been done yet for ethical reasons.  There may be, probably are, more than one area involved in expressing these "concepts", and each area probably uses a different system of representation.  Exactly how these areas express their concepts, and the precise relationships among expressions in different areas, is a subject for future research.  However there are some things we can be fairly sure of from what we know already. 

For one thing, the brain appears to operate on a very associational basis:  if one area is stimulated to "fire" a certain concept it will likely stimulate other areas to fire the same concept.  Some of these areas may act to relate concepts to one another, others may act to related observations of other animals to various concepts that classify their actions, yet others may act to relate concepts for proposed actions to specific conditions such as food lying on the ground nearby or hanging from a nearby branch, yet others may serve to react to various concepts with responses from memory regarding past incidents where similar actions took place in similar circumstances.  The relationship among these various areas may then allow a decision regarding whether, and how, to undertake the action this concept represents. 

Again quoting from Caggiano et al., this time from the conclusion:[3]
Our results suggest a cognitive role for mirror neurons as a system that not only encodes the meaning of observed actions but also contributes to choosing appropriate behavioral responses to those actions.  In particular, a stimulating (although admittedly speculative) interpretation of our results is that mirror neurons not only may represent a neuronal substrate for understanding "what others are doing," but also may contribute toward selecting "how I might interact with them."
Notice that in all this I've essentially expanded the use of "concepts" to all primates possessing "mirror neurons".  This makes sense, since if the firing of a certain pattern of neurons in even one area of the brain represents an attentional focus on one particular idea, such as a class of actions, it makes a valid symbol, and symbols can be reasonably considered to represent concepts.  Language doesn't come into it.  Or rather, language came into it very late in the game, when one lineage of apes came to represent a well-established system of symbolism within the brain with something that could be easily communicated to other members of the local group. 

Thus we can see that while language is very likely (but not certain) to be a human invention, the use of symbols, and the concepts they represent, is probably much older:  language certainly added a dimension or more to the utility and effectiveness of symbols and concepts, but the original system was in place at least tens, and possibly hundreds, of millions of years earlier. 

Caggiano, V., Fogassi, L., Rizzolatti, G., Thier, P., & Casile, A. (2009). Mirror Neurons Differentially Encode the Peripersonal and Extrapersonal Space of Monkeys Science, 324 (5925), 403-406 DOI: 10.1126/science.1166818

Links: Not all of these have been called out in the text.  Use the back key if you came via clicking a footnote. 

1.  The Early Differentiation of the Neocortex:  a Hypothesis on Neocortical Evolution

2.  Genetic and epigenetic contributions to the cortical phenotype in mammals

3.  Mirror Neurons Differentially Encode the Peripersonal and Extrapersonal Space of Monkeys paywall

1 comment:

  1. Brilliant post! Have linked to it from my blog.