related texts

Various bits about various CNS systems


1. Why humans are sensitive to mirror symmetry

Humans are specially sensitive to mirror symmetry, particularly when the axis of symmetry is vertical. What causes this sensitivity?

Part of the special sensitivity can be explained by the special features of symmetry. In general, symmetrical objects/pictures have 'internal' control, which gives the observer an impression of internal consistency. That would make them special for any observer, not only for humans.

Mirror symmetry has an additional advantage over other point symmetries. This is the fact that there is a relatively large area in which the distance d between a point and the symmetry related point is small. For d < D (for some small D), the area is L*D, where L is the length of the object along the symmetry axis. For other point symmetries, the area is pi * D * D, which is very small for small D. Again, that would make mirror symmetry special for any observer. However, that does not explain the preference for vertical axis, and does not seem to be enough to explain the strength of the preference.

The preference for vertical axis can probably be explained based on the symmetry of the CNS, but how? Here I suggest the following hypothesis:

The sensitivity to vertical mirror symmetry is a result of a small portion (order of magnitude 10 ** -3) of the neurons in the optical nerves going to the wrong side in the optical chasm, but connect correctly (in the topological sense) to the LGN on the wrong side.

We already know that all the neurons from the retina converge in the optical chasm, and from there neurons from the left of the retina of both eyes continue to the left side of the brain (mostly the left LGN), and from the right side of the retina they go to the right side of the brain. In more technical terms, nasal neurons cross to the contralateral optical tract, while temporal neurons continue to the ipsilateral tract. As a result, information from the left of the retina, which corresponds to the right visual field (the image is inverted when it passes the lens), reaches the left hemisphere of the cortex. The information is mapped in a topological way from the retina to the LGN and from the LGN to the cortex.

Neurons that 'miss the turn' in the optical chasm (i.e. either nasal neurons that continue to the ipsilateral tract or temporal neurons that do cross), but still connect correctly to the LGN, end up delivering the information to the wrong hemisphere, but in the right position. That means that the information is delivered to the mirror position of the correct place.

In most of the cases, this information is just noise, which blurs the picture. However, when the person focuses on the axis of a mirror symmetric object/picture with a vertical axis, this wrong information is actually the same as the correct information. As a result, mirror objects/pictures, with vertical axis, are clearer than any other objects/pictures, and hence easier to perceive.

Obviously, it also requires the person to focus on the axis of symmetry, which adults have already learned to do, while babies do it by chance. Once they happen to fixate close enough to the axis of symmetry, they have a clearer picture (i.e. a more coherent and stronger activity in the visual areas), which apparently causes them to keep their gaze longer (*).

The number of 'stray neurons' must be quite small, because otherwise non-vertical-mirror-symmetrical object/pictures would be very blurred. That, and the fact that researchers did not look for them, probably account for the fact these stray neurons are not documented in the literature.

These 'stray neurons' may also explain other phenomena. For example, people with non-functional visual cortex in one hemisphere (or even with a complete hemisphere non-functional) can sometime still show sensitivity to visual input in the affected (contralateral) visual field. This sensitivity is probably mediated by the stray neurons.

The 'stray neurons' can explain special sensitivity to mirror symmetry only when the axis is vertical and in the middle of the visual field. However, as explained above mirror symmetry is also inheritedly more attractive, and in addition the fact that it is very specal in one angle will make it more special in other angles. Thus we should expect special sensitivity to mirror symmtery even when it is not in the right place(vertical, central), though weaker than when it is in the rigth place.

While the 'stray neurons' are 'stray', the sensitivity to mirror symmetry may have evolutionary advantage, because it makes detecting animals (which are in general mirror symmteric) easier. It also may be useful in mate selection, because asymmteric mate probably had some problem during development. Thus the existence of these neurons may be 'intentional', in the sense that natural selection does not favour individualls with better accuracy of sorting in the optic chasm.


(*) I take it for granted that babies look longer at objects/pictures that look to them clearer.


2 .The main function of the Cerebellum

For accurate movements, it is needed to continuously apply small corrections to the basic movement program. This is required to correct for inaccuracies in the program itself, for small effects that where not taken into account, and for inaccuracies in the performance of the muscles. These corrections are required even to keep a limb in one place. They are based mainly on input from proprioceptors in the muscles, but also on other sensory input.

These corrections have the following characteristics:

An effective way to get the required speed is to have a dynamically configurable switching board, which can be dynamically changed such that it will generate the right output from the input. This switchboard must have a way to connect any input to any output, and be able to dynamically switch these connections on and off.

Minor hypothesis : The Cerebellum is a dynamically configurable switchboard for controlling accurate movements. When the cerebral cortex 'decides' on a movement, it activate the appropriate muscles to do the movement directly, and in the same time activates/inhibits the right neurons in the Cerebellum, such that the Cerebellum will give the appropriate corrections in respond to proprioceptionic (and other) input. The cerebral cortex knows what to activate/inhibit in the Cerebellum (in other words, which signals to send to the cerebellum) by learning it, by the same learning mechanism which causes learning of mental operations.(For a theory of this learning mechanisms see here).

The important difference between this hypothesis and current theories is that in my hypothesis the precise connectivity in the cerebellum is not important, and the cerebral cortex learns to use whatever connectivity there is in the cerebellum. The structure of the cerebellum needs only to ensure that it is possible to connect any (or at least almost any) combination of inputs with any combination of outputs, and that the input-to-output transformation is fast and reliable.

A related difference is that other theories tend to hypothesize that the Cerebellum takes part in planning the movement. Here, the Cerebellum only gives corrections while the movement is being performed.

The main control of the switching is probably through climbing fibers, which switch the appropriate Purkinje cells on. However, signals from the cerebral cortex also go through the granule cells. The output of the Cerebellum is also not completely restricted to motor control. That means that the cerebral cortex can use the Cerebellum for other uses, though it main usage is for controlling accurate movements.

Becuase the cerebellum acts as a switchboard, there is no stable correlation between activity of each neuron and the neurons it activates, because this is dependent on input from the cortex, and keeps switching around. As a result, in an adult Hebbian learning mechanism would not be useful (during development, there should be some pruning of the space of all possible connections to the space of all useful connections, which can be done by Hebbian mechanism). Thus, if the hypothesis is correct, it is unlikely that learning is taking place in the adult Cerebellum, and learning of movements in adult is based on changes in the Cerebral Cortex, like anything else. Note that this does not mean that there is no plasticity in the adult Cerebellum. It just means that it is not significant for learning.


After writing this, I found this article. Below the heap of waffle, the basic idea, of the cerebellum as a switching board mapping combination of inputs to combination of outputs, is similar. However, these authors still believe that the precise connectivity of the cerebellum is defining its detailed operation, and that significant learning happens in the cerebellum.


[17Mar2005] There is by now a large literature showing "learning" in the cerebellum, but this literature is not really relevant to motor learning. The major problem is that it is all about acquiring association between a specific stimulus and a specific reflex (condioning). While this is an interesting area, this kind of acquisition is a very different thing, in particular becuase what is learned is several orders of magnitude simpler than motor learning, and because it is not involved learning to perform new movements or improve existing movements. It is an acquisition of a simple association that causes the stimulus to activate an existing reflex.

There are also studies that show that the cerebellum is esential for learning more complex motor actions, but these studies don't show that the learning occurred in the cerebellum.


3. What is innately coded in the Wernicke area?

From brain damage studies, it is known that Wernicke area is important in understanding language. How?

Notwithstanding the hallucinations of Chomsky et al, there is no evidence that humans are innately sensitive to any language structures like words and sentences. On the other hand, infants are clearly sensitive to the differences between phonemes, and to other basic attributes of the sound, like rhythm and pitch. The other basic attributes sensitivity does not seem to be coded in the Wernicke. Thus the Wernicke area is probably innately coded to distinguish between phonemes {1}.

What is needed to make the Wernicke area good in distinguishing between phonemes ? Clearly, we need neurons that response significantly differently to different phonemes. So the next question is what are 'phonemes' as far as the brain is concerned.

Phonemes are distinct patterns of air vibrations, which are converted to neural signals by the inner ear (cochlea). The cochlea functions as mechanical device for converting vibrations (originally vibration of air, converted by the external and middle ear mechanism to vibrations of the oval window) to location. Due to the design of the cochlea, different frequencies cause vibrations, and hence neural signals by the hair cells, at different locations along the cochlea. Thus different phonemes are, as far is the brain is concerned, different patterns of activity in the hair cells along the cochlea.

Phonemes also have temporal component, so a complete description of a phoneme is actually not a single pattern, but a continuous change in the pattern of activity of the hair cells. This is what the neurons in the Wernicke area have to be sensitive to.

There is no problem explaining the sensitivity of neurons in the Wernicke area to temporal changes, because any network is sensitive to temporal changes in its input. Thus the question is how neurons in the Wernicke area are sensitive to patterns in activity along the cochlea.

For that, some of the neurons in the Wernicke should be getting input from several patches along the cochlea{2}. Different neurons should be getting input from different patches, so they respond differently to different phonemes. Note what we do not need to assume:

  • Each neuron has to get input from a specific set of patches in the cochlea.
    Since there is no fixed mapping (across individuals) between phonemes and any specific neuron groups in the wernicke area, the distribution of input may be random between individuals, as long as it is stable once it is formed.

  • The pattern of activity along the cochlea from which a neuron gets input must match exactly the pattern of activity evoked by some phoneme.
    The match can be approximate, because that will be enough to make the neuron respond strongly to this phoneme, and much weaker to other phonemes, and that is all that is required.

  • All the neurons in the Wernicke should be getting input from the cochlea.
    It should be enough if only part of the neurons respond specifically to different phonemes.
  • Once we have in the Wernicke area neurons that stably respond differently to different phonemes, any phoneme that the infant hears would cause a stable subset of neurons to become active. A simple Hebbian process can cause the neurons in this subset to strengthen the connections between them, and thus to improve the respond to the phoneme (unsupervised learning). Thus, assuming Hebbian process is in action in the Wernicke area, just being exposed to language (sequences of phonemes) would cause the infant to increase his sensitivity to the phonemes that shhe hears.

    As the child grows and becomes more adapt in associating patterns of activity in his/her brain, shhe can start to associate temporal patterns of activity in the Wernicke area (which are the result of temporal sequences of phonemes, i.e. spoken words), with other patterns of activity in the cortex. In other words, shhe learns the meaning of words. Note that this does not require any special mechanism above a generic learning mechanism that associates two pattern of activity in the cortex that tend to predict each other (See in Cognition for a model of how this general learning mechanism works). Learning more complex structures in language (sentences etc.) follows, and naturally tends to concentrate in and around the Wernicke area, because the basic information about words is there.

    Thus to explain the importance of the Wernicke area in understanding language, all that is needed is that some the neurons in the Wernicke area will be getting input from patches along the cochlea. As far as I know, the connectivity of the Wernicke area is not mapped well enough to tell if this is true, but it is a reasonable assumption .

    The important point of the analysis above is that it shows that to explains the observations about the Wernicke area there is no need for elaborate models with explicit design. In addition, these models are incompatible with the stochastic connectivity of the Wernicke area (and the rest of the cortex, see connectivity in the cortex and brain-symbols, section 4) and therefore are extremely unlikely to be correct.


    {1} Most cognitive psychologists would probably say that it is coding to 'recognize phonemes'. However, there is no reason to believe that the infant 'recognizes' the phonemes in any sense except that shhe can distinguish between them, so using 'recognize' is misleading.

    {2} By 'a neuron getting input from X' I mean that there is some pathway from X to the neuron, and that this pathway has strong and stable (over time) effect on the activity of the neuron.

    Yehouda Harpaz