related texts

Comments on the MIT Encyclopedia of Cognitive Science (MITECS)

In this text I give comments on the texts in the MITECS. The advantage of this is that it represents the current 'paradigm' in the field of cognitive science (at least tries to), and is online.

[ 17 Jul 2010 ] By now it is not online anymore. If you find here something that is interetsing and you want to see the actual text in MITECS, you can try to google part of the quote that I give. With some luck you can find the text online. Note also that these commenst are pretty old by now.

In general, while the texts in MITECS do not contain many explicit errors, in many cases they implicitly lead in the wrong way. The target of this page is to highlight the mistakes that are implied in the texts. That presents a problem, because implied meaning is dependent on the interpretation, which differs between readers. In particular, experts in the field interpret the text differently from non-experts, because they use their additional knowledge. In my interpretation, I assume low-level of expertise with cognitive science, and hence try to answer the implied message that a non-expert would read in the text.

Note from the MITECS Executive Editor:

Keep in mind that the current mitecs site is a developmental, unedited site. The final site will be posted this spring.

The text below is intended to be read side by side with the MITECS text. Indented text is always an exact quote from MITECS, and the user can use search to find the quote in the MITECS text. Where the quote ends with several dots, the comment is about all the paragraph starting with this text. This page contains comments on the neuroscience domain. Other pages contain comments on the other domains:

================================================================================
================================================================================

General: While MRI and PET are useful clinicaly, current cognitive studies are totally irreplicable, and hence useless.

The most common misconception about the cortex seems to be the myth of the two separate VISUAL PROCESSING STREAMS. In reality, these refer to different tendencies in the different regions in the visual cortex, and the regions are not separated at all. This is probably a manifestation of the Modularity error and simplicity error.

Introduction: Neuroscience

Cognitive neuroscience is thus a science of information processing.
There is nothing wrong with this statement as it is, but it should be noted that many cognitive scientists use 'information processing' to mean 'symbolic processing', in which case this is wrong.
Viewed as such, one can identify key experimental questions and classical areas of study: How is information acquired (sensation), interpreted to confer meaning (perception/recognition), stored or modified (learning/memory), used to ruminate (thinking/consciousness), to predict the future state of the environment and the consequences of action (decision making), to guide behavior (motor control), and to communicate (language)?
This is already going wrong. The way the question is posed, it is already assumed that information is separately acquired, interpreted, stored or modified etc., but the separation is not necessarily true. In some cases we know from experimental evidence of the separation (e.g. we have quite a good idea how sensation happens) but not in all cases, and we cannot assume the separation.

2.2 Neuron doctrine
That is misleading, as it is not a doctrine (a set of principles to guide behavior). It is an experimental fact, which has been confirmed beyond reasonable doubt. Calling it a 'doctrine' gives the impression that there may be any doubt about it.
The pioneering studies of V1 by Hubel and Wiesel (1977) established both the form in which visual information is represented by the activity of single neurons and the spatial arrangement of these representations within the cortical mantle ("functional architecture").
That is simply false. Hubel and Wiesel found some properties of the cortex, and these properties may have functional significance. The statement gives the impression that Hubel and Wiesel found all the important functional properties, which is very far from the truth.
More generally, the receptive field is a characterization of the filter properties of a sensory neuron, which are commonly multidimensional and include selectivity for parameters such as spatial position, intensity, and frequency of the physical stimulus.
This is very misleading. If the 'sensory neuron' here refers to the primary sensory neurons (those that perform the transduction), then it is OK. However, the concept of receptive field is normally applied to neurons in the cortex, and that seems to be the case here as well. In this case it is simply false. It is not the properties of the neuron that determine its selectivity, but its connectivity, i.e. which neurons synapse on its dendrites, and their activities. In other words, it is a property of the network the neuron is in. The sentence is an example of the 'intelligent neuron' misconception (see in myths and and misconceptions for a discussion).

Experts in neuroscience may object that it is clear that the 'properties' are the properties of the network, but it is clear only to somebody that already knows that. For a non-expert there is no way to understand this from the text.

Modular and Columnar Organization The proposal that ...
It needs a careful reader to figure out that this paragraph actually says that while people looked for modularity in the monkeys visual cortex, there isn't any evidence for it. I doubt if any reader without expertise in neuroscience can read this paragraph and not come with the impression that there is evidence for modularity in the monkey visual cortex, and this impression is strengthen by the next paragraph.

Modular organization of sensory cortex also occurs at a finer spatial scale, in the form of regional variations in neuronal response properties and/or anatomical connections, which are commonly referred to as columns, patches, blobs, and stripes.
First, this sentence strengthen the impression that the previous paragraph showed evidence for modularity (by using 'also'). Secondly, the term modularity is redefine with a completely new meaning ('regional variations'). Redefinition of terms is a standard way of confusing the discussion (see in Reasoning errors, and specific to modularity in Reasoning errors)
The concept of modular functional organization was later expanded upon by the physiologist Vernon B. Mountcastle, who obtained the first evidence for columnar function through his investigations of the primate somatosensory system, and offered this as a general principle of cortical organization (Mountcastle 1957).
Mountcastle did not obtain any evidence for 'columnar function'. He obtained evidence for 'regional variation' on a small scale. There is still no evidence that these columns have any functional significance, and that is true for all the evidence that is quoted in the next few sentences.
Other investigations have demonstrated that modular representations are not limited to strict columnar forms (Born and Tootell 1993; Livingstone and Hubel 1984), and can exist as relatively large cortical zones in which there is a common feature to the neuronal representation of sensory information (such as clusters of cells that exhibit a greater degree of selectivity for color, for example).
This sentence uses again the trick of implying that the previous text showed evidence for modularity even though it didn't. Here 'modularity' is actually 'clustering', increasing the confusion further.
Hierarchical Processing A consistent organizational feature of sensory systems is the presence of multiple hierarchically organized processing stages, through which incoming sensory information is represented in increasingly complex or abstract forms.
The normal usage of the word 'stages' implies that each stage is distinct from the other stages, and the word 'stages' is quite meaningless if they are not distinct. There isn't any evidence for distinct stages once the information reached the cortex, and the text does not even try to support this notion. Thus the idea that the processing is divided to distinct parts (stages) is sneaked in by a linguistic trick.

The term 'hierarchical' implies well defined organization, and again there is no evidence for this notion, and it is simply sneaked in.

Beyond V1, the ascending anatomical projections fall into two distinct streams, one of which descends ventrally into the temporal lobe, while the other courses dorsally to the parietal lobe.
That is simply false. There isn't anything like 'distinct streams' inside the cortex. What the experimental evidence shows is that the dorsal regions tend to be more sensitive to motion (and therefore to spatial relations, because these are based on information from movement) than the ventral regions.

The concept of 'stream', which refers to what in reality is a smooth variation in a continuous sheet, is one of the most misleading concepts in neuroscience today.

Although the suggestion that perceptual experience may be rooted in the activity of single neurons or small neuronal ensembles has been decried, in part, on the grounds that the number of possible percepts greatly exceeds the number of available neurons, and is often ridiculed as the "grandmother-cell" hypothesis, the evidence supporting neuronal representations for visual patterns of paramount behavioral significance, such as faces, is now considerable (Desimone 1991; Rolls 1992).
The way this sentence is structured, it gives the impression that there is evidence to support the "grandmother-cell" hypothesis, but actually all it claims is that there is evidence for 'neuronal representations'. Since everything in the brain is neuronal, this is obvious.

The "grandmother-cell" hypothesis is stupid not because the number of cells, but because the effect of a single cell on the system is too small to be significant. Neuroscientists find it difficult to figure this out, because they are constrained by the 'intelligent neuron' misconception (see in Myths and misconceptions).

A classic and familiar example is the Necker Cube, in which the 3-dimensional interpretation -- the observer's inference about visual scene structure -- periodically reverses despite the fact that the retinal image remains unchanged.
The assertion that the 'retinal image remains unchanged' is ridiculous, because it does change as a result of movements of the eye. A very minute movement can already generate a change. Ignoring eye movements is quite common error in cognitive science. In the context of this text, though, this error is not significant.
Several lines of evidence suggest that there may be multiple steps along the path to extracting meaning from sensory signals.
Since this sentence is qualified by 'suggest' and 'may be', it is not definitely wrong, but it is definitely not right either. There isn't any evidence for distinct steps in the processing of sensory signals inside the cortex. All the evidence for these steps is based on demonstrating the humans perform operations that would be perform by these distinct steps, if they exist, but this does not show that there are distinct steps. On the other hand, the cortex is continuous sheet, which suggests that the processing inside it is unlikely to be discontinuous. This is completely ignored.
Mid-Level Vision This step involves a reconstruction of the spatial relationships between environmental surfaces.
To repeat, there is no evidence to support a separate step involving 'reconstruction of the spatial relationships'. It is done in an integrated fashion with the rest of processing. The text refers the reader to the section about MID-LEVEL VISION, but this takes it for granted that this level exist, and does not bother with any evidence for it.
Generally speaking, neuronal gain control is the process by which the sensitivity of a neuron (or neural system) to its inputs is dynamically controlled.
This sentence is confusing. The term 'its inputs' may mean 'The stimulus to which it responds', in which case the sentence is OK. However, most people would understand 'its inputs' to mean the neurons that synapse on its dendrites and their activity. In which case the definition is wrong, because the 'gain control' can be (and mostly is) a result of changes in the inputs themselves, rather than the sensitivity of the neuron to them.

The following few paragraphs continue this confusion, and a reader without a good understanding of neurobiology would get the impression that neurons are more intelligent than they really are, and can adjust their activity autonomously.

The stimulus and mnemonic factors that influence attentional allocation have been studied for over a century (James 1890), and the underlying brain structures and events are beginning to be understood (Desimone and Duncan 1995).
The second part of the sentence is simply false. We still don't have a clue about the structures and events that underlie attention.
Much of our understanding comes from analysis of ATTENTION IN THE HUMAN BRAIN -- particularly the effects of cortical lesions, which can selectively interfere with attentional allocation (VISUAL NEGLECT), and through electrical and magnetic recording (ERP, MEG) and imaging studies -- POSITRON EMISSION TOMOGRAPHY (PET) and FUNCTIONAL MAGNETIC RESONANCE IMAGING (FMRI).
This is completely spurious. The section about ATTENTION IN THE HUMAN BRAIN contains a confused discussion of volition control of attention and movement and other stuff (like reading), and completely ignores lesions and neglect. PET and fMRI are totally irreplicable, and ERP and MEG hardly tell us anything about structures and events in the brain. The evidence from VISUAL NEGLECT does not show anything specific to attention.
In addition, studies of ATTENTION IN THE ANIMAL BRAIN have revealed that attentional shifts are correlated with changes in the sensitivity of single neurons to sensory stimuli (Moran and Desimone 1985; Bushnell, Goldberg, and Robinson 1981; see also AUDITORY ATTENTION).
There is no doubt that attentional shifts, like any change in thinking and perceiving, must correlate with changes in the activity of some neurons, because all the action in the brain is by neuronal activity. Thus there is no need for evidence for it. That the authors bring this evidence, implies that the authors think that there is another possibility, and it would be interesting to figure out what this is.
The neuronal mechanisms that enable information about the world to be stored and retrieved for later use have a long and rich history
This repeats the standard error of talking about storing and retrieving, terms which are inapplicable to the brain. The correct terms are 'forming memories' and 'recalling' or 'recollecting'. See various points in Reasoning errors.
One of the first cognitive functions to be characterized from a biological perspective was language.
That is simply false, as language has not been 'characterized from a biological perspective'. The following evidence is about finding some regions in the cortex where damage causes linguistic deficit, which is far from being 'characterization from a biological perspective'.
Investigators since then have discovered that different aspects of language, including the PHONOLOGY, SYNTAX, and LEXICON, each rely on different and specific neural structures (see PHONOLOGY, NEURAL BASIS OF; GRAMMAR, NEURAL BASIS OF; LEXICON, NEURAL BASIS OF).
That is a blunt lie. While PHONOLOGY is clearly associated with Broca and Wernicke areas, there isn't any consistent evidence about the structures that underly SYNTAX and LEXICON. The sections that it refers to do not report any consistent evidence.
Modern neuroimaging techniques, including ERPs, PET and fMRI, have confirmed the role of the classically defined language areas and point to the contribution of several other areas as well.
The PET and fMRI are totally irreplicable. ERPs don't actually give us much.
Such studies have also identified "modality neutral" areas that are active when language is processed through any modality: auditory, written, and even sign language (see SIGN LANGUAGE AND THE BRAIN
Another lie. The sole consistent finding is preference for the left hemi-sphere.

Cerebral Cortex

General: The important thing about this section is that it does not contain any discussion of the most important property of the cortex, i.e. its internal connectivity.

Its most distinctive anatomical features are i) the very extensive internal connections between one part and another part, and ii) its arrangement as a six-layered sheet of cells, many of which are typical pyramidal cells.
1) The very extensive internal connections are not only between 'parts', but also inside each 'part'. In addition, apart from the division to hemi-spheres, the cortex is not actually divided to parts.

2) It is not clear that the six-layered organization has any functional significance as far as information processing is concerned.

It seems most likely that animals with a dominant neocortex have a combination of the two -- they have the best of both worlds by combining genetic and individual acquisition of knowledge about their environments (Barlow 1994).
Because of the stochastic connectivity in the cortex, it cannot contain any genetic knowledge. Because the author completely ignores the connectivity, he cannot figure this out.
Neuropsychologists who have studied the defects that result from damage and disease to the cerebral cortex emphasize localization of function (Phillips, Zeki and Barlow 1984): cognitive functions are disrupted in a host of different ways that can, to some extent, be correlated with the locus of the damage and the known connections of the damaged part.
The Neuropsychologists need to emphasize the localization because the data itself does not support it, as the second part of the sentence effectively says. Apart from Input/Output localization and the Broca and Wernicke area, there is no localization in the cortex, and that is shown by the fact that the author does not bring evidence for localization.
But the cells everywhere have an unusual and similar form, which suggests they have a common function; an attractive hypothesis is that this is prediction.
The first half of the sentence is correct. The second half is simply stupid. 'Prediction' is a very complex operation that cannot be done by individual cells, and it is not useful on its own. The function of the cortex is clearly done by integrated action of many neurons, and it is to support everything that is required for generating appropriate behavior, not only subsection of it.

face recognition

General comment: The main thrust of this section is the "specialness" of faces. The exact meaning of this term is not made explicit, and in particular it does not say if this "specialness" here means "innate" or not. In some places the "specialness" seems to include innateness. It is also seems odd to look for evidence of "specialness" of face recognition, when it is obvious that behaviorally it is special, unless the "specialness" also means innateness. On the other hand, In some parts of the text the "specialness" seems not to include innateness. The confusing usage of "specialness" gives the impression that the text bring evidence for innateness, even though it doesn't.

Analysis and retention of facial images is a crucial skill for primates
That is simply false. What is crucial is the appropriate responses to different faces and facial expressions. If you use 'Analysis' in its widest meaning, you can say it is required by definition for appropriate response. You cannot say this about retention of images, and saying this is a 'crucial skill' is an experimental statement, which has to be supported by evidence. None of the evidence that is brought later supports it.

readers tend to accept this sentence as obviously true either because they already believe that the brain works like a computer, or because of the blatant nonsense effect. The latter is enhanced by the fact that readers do not expect nonsense in the first sentence of a text.

These indicate that face recognition in primates is a specialized capacity consisting of a discrete set of component processes with neural substrates in ventral portions of occipito-temporal and frontal cortices and in the medial temporal lobes.
The text does not actually says whether the 'specialized capacity' is innate or not, but strongly implies it by discussing primates and talking about 'discrete set of ...' . Note that the areas of the cortex that are listed here cover very large fraction of the cortex.
One early milestone was Yin's (1969) demonstration of the inversion effect, the tendency for recognition of faces to be differentially impaired (relative to that of other "mono-oriented" stimuli such as houses) by turning the stimulus upside-down. This finding has been widely interpreted to mean that face recognition depends on specialized mechanisms for configural processing (i.e., analysis of small differences in details and spatial relations of features within a prototypical organization).
This is reasonable provided the 'specialized mechanism' is not confused with 'specialized built-in mechanisms'. Recognizing faces requires 'analysis' (in the wider sense of the word) of small and many differences, and hence is going to be different (specialized) whether it is learned or built-in. While the text does not actually say it, for many readers it implies that the mechanisms are built-in, and the way the text continues supports this.
Face "specialness" is also supported by data showing that infants preferentially look at or track face-like arrangements of features relative to jumbles of features or control stimuli.
That is as close as you can get to a lie without actually lying. Infants are totally uninterested in faces and face-like objects. All the experiments that 'show' that babies prefer 'face-like' features are based on showing the baby extremely simplified drawings. In most of the experiments, the control that were used were bad ones, for example not controlling for symmetry, smoothness of lines and balance of features. The experiments that did control for these features were mostly irreproducible.

Readers which are not familiar with the literature would tend to assume that 'face-like' really means like a face, and that the experiments are mostly reproducible. In addition, they would not know that infants are not interested in the real faces. The latter fact is carefully hidden from the reader, since it completely undermines the argument about recognizing faces being 'crucial'. Obviously, if infants response only to extremely simplified face drawing but not to real faces, it does not help them at all in forming a bond with their parents or learning to recognize faces.

Nonetheless, the protracted development of adult levels of performance indicates that face recognition additionally involves either a long period of NEURAL DEVELOPMENT and/or cognitive processing capacity, specific experience with faces, or both.
That is the closest this text gets to admitting that the infants data does not support what it claims to support. This sentence implies the possibility that humans can acquire face recognition without experience with faces, which is clear nonsense.

The degree of DOMAIN SPECIFICITY present in prosopagnosia is relevant to whether face recognition is best viewed as a unique capacity, or merely as an example of general mechanisms of OBJECT RECOGNITION (Farah 1996).
This is doing the reasoning error about dissociation (Reasoning errors).

In both humans and monkeys, faces are analyzed in subregions of the visual cortical object recognition pathway. In particular, temporal neocortex in nonhuman primates (notably inferior temporal cortex or "area TE") contains neurons which fire selectively to face stimuli (Gross and Sergent 1992).
This invites the reader to deduce that the subregions are the temporal neocortex. However, the evidence clearly does not support this conclusion. To reach it, you need to show that there are no cells which respond differentially to faces in other areas of the cortex. It worth noting that the text does not actually state the wrong conclusion, but tempts the reader to make it.
The question arises as to whether such cells truly respond to visual information unique to faces, or whether their selectivity is more parsimoniously explained as responsiveness to features shared by faces and other object classes; the bulk of the evidence supports the former description.
depends on what the term 'features' means here, this is either and example of getting the 'right answer' by asking the wrong question, or a simple lie. Note that it is possible that some cells do respond only to faces (whether genetically determined, result of learning or simply by chance), but it needs more than the evidence that is discussed in the text to show this.
First, although face cells make up only a tiny fraction of neurons (1-5%) within TE and adjacent areas as a whole, their concentration is much higher in irregular localized clumps.
While this is true, the implication that it is a special feature of face recognition is wrong. Neurons that deal with some concept have to be connected to each other, and closer neurons have higher probability to connect to each other, so the neurons that deal with any concept would tend to cluster. For example, see above in the introduction for neuroscience (sentence starting with "Other investigations have demonstrated ").
Secondly, different types of face-selective cells are found in different regions, such that cells sensitive to facial expression and gaze direction tend to be found within the superior temporal sulcus, whereas cells more generally selective for faces and, purportedly, for individuals tend to be located in TE on the inferior temporal gyrus.
It is worth noting that this sentence starts with an absolute statement, which tells the reader that different types of neurons are separated from each other. The rest of the sentence is more realistic, and uses the relative term like 'tend' and the qualification 'purportedly'. Thus the absolute statement is actually wrong, and it is here just because it fits better with the ideas of the text, as expressed in the first paragraph.
Longer-latency face-specific potentials were also recorded from the anterior portions of ventral temporal cortex activated by face recognition in POSITRON-EMISSION TOMOGRAPHY (PET) studies.
[ Also later paragraphs ] PET studies are completely irreplicable, See the survey of the literature.
Neuropathological data are generally consistent with results of imaging and evoked potential studies regarding anatomical substrates for face recognition.
Note that the only point of 'consistency' is that the right hemisphere seems to be more important, and the data is not consistent on more specific details. It is also consistent about it being in the posterior cortex, but since face recognition is based on visual input, and visual input is processed in the posterior cortex, that is already obvious.
Many explanations have been given for the apparent "specialness" of faces.....
The discussion in this paragraph is realistic. The point to note is that the idea that we have innate specialized features for face recognition does not appear explicitly in this paragraph. It is possible that the 'unique behavioral capacities and reflect dedicated neural circuits' supposed to mean innate, but it is not clear from the text. This realistic discussion is in contrast to the first paragraph, which strongly implies innate and discrete components.

Columns and modules

General: This chapter plays very much on the difference between the way expert neuroscientists and other people read the text, and is full of statements that may not look wrong to the neuroscientist, but draw a very misleading picture for other people.
Though Area 17 seems to be a consistent functional unit, other areas prove to contain a half-dozen distinct physiological subdivisions ("maps") on the centimeter scale.
It is not obvious what 'distinct physiological subdivisions' supposed to mean here. The cortex of each hemi-sphere is continuous, without any borders anywhere, and without distinct functional divisions except when it is associated with input or output into the cortex.
Columns are usually subdivisions at the sub-millimeter scale, and modules are thought to occupy the intermediate millimeter scale, between maps and columns.
This already assumes that there are modules in the cortex, without anything like an evidence. Maybe it is using it in a different way than the common usage, but it does not say this.
The CEREBRAL CORTEX sits atop the white matter,
Funny way of putting it, as if the white matter and the cortex are separate entities. Neuroscientists know that the white matter is simply the axons of the neurons in the cortex itself, but other people may get confused.
Both may be regarded as the outcomes of a self-organizing tendency during development, patterns that emerge as surely as the hexagons of a honeycomb arise from the pounding of so many hemispherical bee's heads on the soft wax of tunnel walls. The hexagonal shape is an emergent property of such competition for territory; similar competition in cortex may continue throughout adult life, maintaining a shifting mosaic of cortical columns.
Gives the reader the impression that there are hexagons in the cortex, which is obviously false, as nobody has ever seen anything like it in the cortex, on any scale.
A column functionally ties together all six layers.
That is nonsense, as the layers are not functionally separated to start with. Neurons do not respect layers boundary, and a typical pyramidal neuron (more than half of the neurons in the cortex) traverses most of the layers with its dendrites, and to some extent with its axon processes.
Layer IV neurons send most of their outputs up to II and III.
Saying that the output is sent 'up' is confusing, as it is ignores the fact that in most of cases, the axons first goes horizontally a distance which is large to very large compare to the size of the neuron, and only then goes up. The result is that most of the outputs are outside the current 'column', which undermines all the concept, and therefore it is important to hide it from the reader.
Some superficial neurons send messages down to V and VI, though their most prominent connections (either laterally in the layers or via U-fibers in white matter) are within their own layers.
Repeats the same misleading picture of local communication, and in additional now says that neurons send 'messages', i.e. doing the 'intelligent neuron' misconception (see in Myths and misconceptions). The phrase in parenthesis tells experts that it is not local, but people that do not already know it cannot understand it.
So for any column of cortex, the bottom layers are like a subcortical outgoing-mail box, the middle layer like an in-box, and the superficial layers somewhat like an interoffice-mail box spanning the columns and reaching out to other cortical areas (Calvin and Ojemann 1994).
Another 'intelligent neuron' allegory, to give naive readers the impression that neurons are much more 'intelligent' than they are.
Indeed, Diamond (1979) argues that the "motor cortex" isn't restricted to the motor strip but is the fifth layer of the entire cerebral cortex.
Since brain damage studies show clearly that some regions in the cortex are more involved in motor control than the rest of the cortex, (by definition, these are the motor cortex), diamond idea is simply stupid. In addition, as I said above, layers are not separated functionally.
It now appears that a column is like a stalk of celery, a vertical bundle containing axons and apical dendrites from about 100 neurons (Peters and Yilmaz 1993) and their internal microcircuitry.
'Internal circuitry'? what he is talking about? That is simply an invention. Neuroscientists know that, and may assume that he means the dendrites and axons, but other readers may be confused that there is some 'internal microcircuitry' which is not the dendrites or axons.
Macrocolumns may, in contrast, reflect an organization of the input wiring, for example, corticocortical terminations from different areas often terminate in interdigitating zones about the width of a thin pencil lead.
The second half of this sentence is simply false. All the evidence for 'macrocolumns' is concerned extracortical input into the cortex. There is no evidence for corticocortical connections organized this way, and the author does not try to claim this in the following text. Again, it is obvious to neuroscientists, not to more naive readers.
The superficial pyramids send myelinated axons out of the cortical layers into the white matter; their eventual targets are typically the superficial layers of other cortical areas when of the "feedback" type; when "feedforward" they terminate in IV and deep III.
What is "feedforward" and "feedback" in the brain, and what is the evidence for this distinction? That is introducing computer science terms without any justification.
The axon may continue for many mm, repeating such clusters about every 0.43 mm in primary visual cortex, 0.65 mm in the secondary visual areas, 0.73 mm in sensory strip, and 0.85 mm in motor cortex of monkeys (Lund et al. 1993).
The numbers are averages, in a quite large range, and in most of the cases the distance is different from the average. This is important for the idea in the next sentence, which is based on the assumption that the distances are 'standard', and most of the neurons conform to it.
Because of this local standard for axon length, mutual re-excitation becomes probable among some cell pairs.
This is based on the confusion in the previous sentence between average and standard.
Macrocolumns of similar emphasis are seen to be connected by such synchronizing excitation.
Plain lie. There is no evidence for 'such' (i.e. based on standard distance) synchronizing excitation.
Though COMPUTATIONAL NEUROANATOMY has proved more complex,....
Surprisingly sober concluding paragraph, compared to the rest of the text.
Two adjacent ocular dominance columns, each containing a full set of orientation columns, suggested similar internal wiring, whatever the patch of visual field being represented.
... Except to anybody that knows anything about the wiring in the cortex, because there isn't similar wiring in the cortex anywhere. It is mostly stochastic (see for discussion).

computational neuroanatomy

General: This chapter contains 6 references to other researchers, all of them background information, and 13 references to the work of the author himself. It is hard to see any significance in any of the mathematical models that are discussed, and they all look like mathematical juggling rather than investigation of the properties of the cortex.
The results of this experiment (Schwartz et al. 1989; Schwartz 1994) confirmed that cortical topography is in strong agreement with the conformal mapping hypothesis, up to an error that was estimated to be roughly 20%.
With this kind of error, there are infinite number functions that agree with the observation. It is specially ridiculous to say 'strong agreement' in this case, and the only reason is that the author refers to his own research.
One recent result of this analysis is that the zero-crossings of the cortical orientation map were predicted, on topological grounds, to provide a coordinate system in which left and right handed orientation vortices should alternate in handedness (i.e. clockwise or counter-clockwise orientation change).
'the handness of the vortices' is not a feature of the cortex, and even the 'vortices' do not correspond to any functional entity in the cortex. They are interpretation of the data, which can be made to fit the model.

Computation in Single Neuron

General: Some neurons can do what can be reasonably called 'computation', but not neurons in the cortex. The word 'computation' implies a well defined function, and the neurons in the cortex has stochastic connectivity, which means none of them can compute a well-defined function. Claiming that they do 'computation' is based on the 'intelligent neuron' misconception (see in Myths and misconceptions). The argument in this text shows that the processing inside a single neuron is much more complex than linear integration, but it still does not make sense to call it computation.
If spikes can be generated locally under physiological conditions, they could implement powerful logical operations far away from the cell body (Softky 1994).
... But since their input (i.e. which neurons are connected to the dendrite) is stochastic, the result is also stochastic. It is meaningful only as part of the whole network, which as a result of learning and global tendencies can have non-stochastic activity.
Regarding the first question, some animals can discriminate intervals of the order of a microsecond (for instance to localize sounds), implying that the timing of sensory stimuli must be represented with similar precision in the brain. This is probably based on the average timing of spikes in a population of cells.
That is nonsense. All that is require is a 'transducer' that convert difference in time to some pattern of neural activity, and from this point onwards the difference can be represent by pattern of neural activity. All the stimuli are converted to patterns of activity, and there is no reason to believes that timing difference is different in this respect.
For instance, certain cells in the monkey VISUAL CORTEX are preferentially stimulated by moving stimuli, and these cells can modulate their firing rate with a precision of less than 10 ms (Bair and Koch 1996).
Another example of the 'intelligent neuron' misconception. The cells do not modulate their firing rate, they are getting different input from other neurons, and as a result fire in different rate. In this case, it seems that the author himself believes that the cell modulate their firing rate directly.
If neurons care so much about precise timing of spikes -- that is, if information is indeed embodied in a temporal code -- how, if at all, is it decoded by the target neurons? Do neurons act as coincidence detectors, able to detect the arrival time of incoming spikes at the millisecond or better resolution? Or do they integrate more than a hundred or so relatively small inputs over many tens of milliseconds until the threshold for spike initiation is reached (Softky 1995; Fig. 1)?
That is very misleading. There is no question that neurons are sensitive to differences in their input in the range of milliseconds, but this does not make them coincidence detectors. For that, their output must contain information about the coincidence that has been detected (see Myths and misconceptions).
The functional resolution of these pulses is in the millisecond range, with temporal synchrony across neurons likely to contribute to coding.
See the discussion in Myths and misconceptions for the reason that this is wrong.
Reliability could be achieved by pooling the responses of a small number (20-200) of neurons.
And who is doing the pooling? Since the connectivity is stochastic, it is not going to happen on its own.

Visual processing streams

General: The idea of having 'streams', which implies a well defined route of flow (of information, in this case), in the cortex is clear nonsense, because the cortex has an extremely complex connectivity, with information flowing in almost every direction from every point. This is simply because from any point there are axons going out in almost any direction.

The 'evidence' for 'streams' just shows different tendencies in different areas in the cortex, but nothing like a well defined flow. The sole reason for introducing streams is that they are simpler and easier to model on computer than realistic models.

Despite the complexity of the interconnections between these different areas, two broad "streams" of projections from V1 have been identified in the macaque monkey brain: a ventral stream projecting eventually to the inferotemporal cortex and a dorsal stream projecting to the posterior parietal cortex (Ungerleider and Mishkin 1982).
This introduces the "streams" concept, without explaining what it means, so the reader must understand it as having the typical characteristics of streams, i.e. well defined flow. The following evidence doesn't show anything like a well defined flow, but the reader is left with the wrong impression.
Consider, for example, the patient DF, ...
Using a single case study is always a mistake, because it assumes that the symptoms in one patient are general to the whole population (The 'sameness error', Reasoning errors ). The author does not quote any evidence that the symptoms seen in the case of DF are general.
Goodale and Milner (1992) have suggested that one way to understand what is happening in these patients is to think about the dorsal stream not as a system for spatial vision per se, but rather as a system for the visual control of skilled action.
It is worth noting what Goodale & Milner do here: while attributing completely different properties to the "streams" than Mishkin & Ungerleider did, and not bringing any evidence for well-defined flow, they still take the "streams" concept for granted.

Working Memory, neural basis of

General: the most interesting thing about this chapter is what is missing from it: any evidence that 'working memory' is a separate system.
Working Memory, as defined by cognitive psychologists, refers to "a system for the temporary holding and manipulation of information during the performance of a range of cognitive tasks such as comprehension, learning and reasoning" (Baddeley 1986).
Note that the definition already assumes the existence of the separate system.
A cellular basis for working memory has emerged from the study of activity in single neurons recorded from the prefrontal cortex of monkeys that have been trained to perform delayed response tasks (Fuster and Alexander 1971; Kubota and Niki 1971; Goldman-Rakic et al. 1990)......
This paragraph is a gross overinterpretation of the data. The data shows that in the monkeys the prefrontal cortex has important part in the task of remembering spatial positions. It does not tell us that the prefrontal cortex is the only region that takes part in this task, it does not tell us that this is the only important function of the prefrontal cortex, and, most importantly, it does not tell us anything about working memory in general. The belief that it does is based on the baseless assumption that there is 'working memory system', and it is mediating all short-term remembering, including spatial locations.
This functional distinction provides a valuable clue to how the neural circuitry subserving working memory might be organized. In particular it points to the role of neural inhibition in sculpting the memory field of these neurons.
A piece of nonsense. That neural inhibition has an important role in all mental activities is obvious to anybody that bother to learn about the brain, as inhibitory neurons are 10-20% of the neurons in the cortex, and obviously they are not there for decoration purposes. The author, however, missed this point, and hence regard evidence for the importance of inhibition as a 'valuable clue'. See in myths and misconceptions for discussion.
Working memory is considered a major component of the machinery of executive function and it is not surprising that POSITRON EMISSION TOMOGRAPHY (PET) and functional MAGNETIC RESONANCE IMAGING (fMRI) studies in human subjects have focused on this function.
Currently, PET and fMRI studies are irreplicable, so this evidence is useless.
All of these results speak to the modular organization of working memory systems in the prefrontal cortex, with each working memory "domain" associated with an informationally constrained sensory processing stream.
Another gross overinterpretation. Even if the studies that this statement is based on were replicable, they did not show modular organization.
As might be expected if working memory were essential to executive function, working memory deficits and correlated prefrontal dysfunction have been demonstrated in schizophrenics
That is ridiculous, because working memory is clearly at least useful for 'executive function', so there is no need for evidence for it. This 'need' arises only because of the baseless assumption that 'working memory' is a separate system from the rest of the cognitive system.

Memory, Human neuropsychology

The brain is highly specialized and differentiated, organized so that different regions of neocortex simultaneously carry out computations on separate features of the external world (for example, the analysis of form, color, and movement).
That is false. There are tendencies in the cortex, but not separation. The brain is not highly specialized, and most of the mental activity is done in the cortex, where apart from sensory, motor and Broca/Wernicke areas there is no specialization.
Memory for a specific event, or even memory for something so apparently simple as a single object, is thought to be stored in a distributed fashion, essentially in component parts.
There is no evidence for 'storage' in component parts.
In one sense, memory is the persistence of perception.
... and thoughts, emotions etc. Somehow the author forgot about these.
It has long been appreciated that severe memory impairment can occur against a background of otherwise normal intellectual function. ....
The author uses in this paragraph the word 'memory' to actually mean 'formation of memory'.
The same brain lesions that cause difficulties in new learning also cause retrograde amnesia, difficulty in recollecting events that occurred prior to the onset of amnesia. ......
I wasn't intended to put compliments in this text, but this paragraph deserves one. That is the first time I see any researcher explicitly makes the point that the HIPPOCAMPUS formation function to form memories, but have little role, if any, in using them. I suspect the reason that this is normally 'hidden' is that this fact does not sit well with the notion of separate memory systems.
Memory is not a unitary mental faculty but depends on the operation of several separate systems, which operate in parallel to record the effects of experience.
There isn't any evidence for separate system working in parallel. That is simply confabulation.
The major distinction is between the capacity for conscious recollection about facts and events (so-called declarative or explicit memory) and a collection of nonconscious memory abilities (so-called nondeclarative or implicit memory), whereby memory is expressed through performance without any necessary conscious memory content or even the experience that memory is being used.
Nobody ever shown any correlation between different brain systems and implicit or explicit memory. It is totally clear that both are together inside the cortex.
Declarative memory is a brain-systems construct.
I don't understand what that actually means. Do you? let me know.
Nondeclarative memory is not itself a brain-systems construct, but rather an umbrella term for several kinds of memory, each of which has its own brain organization.
Obviously there are no different 'brain organizations' for each kind of memory. Maybe the author meant different structures, in which case that is simply false. All memories are in the cortex. Different responses and processing of different stimuli require different parts of the system, but not the memory itself.
Among the prominent kinds of nondeclarative memory are procedural memory (memory for skills and habits), simple classical CONDITIONING, and the phenomenon of priming.
Including simple CONDITIONING (later it mention eye blinking) in memory is just adding confusion. This does not depend on knowledge in anyway, so is a different thing.
Emotional learning, including fear conditioning, depends on the amygdaloid complex.
That is misleading. Some emotional responses are dependent on the amygdaloid complex, and hence emotional effects on cognition, including learning, are dependent to some extent on the amygdaloid complex.

Visual cortex, cell types and connections

General: A positive point about this chapter is that it does mention the most important attribute of neurons in the cortex, i.e. their connectivity. However, it does it only a very superficial way, and does not try to tie the connectivity of a neuron to its activity. Like most of texts, it gives far too much importance to layers, and refer to them as if they were separate entities.
Thus, the laminar position of a neuron's cell body tends to be highly correlated with its connectivity.
That is simply false for pyramidal cells, i.e. for most of the neurons in the cortex (it is clear from the figure). The author himself mentions that in the next two sentences, but does his best to mislead the reader to get the impression that this statement is indeed true.
This is particularly true of aspinous, inhibitory neurons and spiny stellate neurons since their dendrites, which receive connections from the axons of other neurons, are usually confined to a single layer.
Note that these neurons do not actually respect layer boundaries, as the term 'confined' implies. The reason that they are 'confined' is that they are smaller than the width of a layer, so only those of them that are on the border span two layers.
But dendritic arbors of pyramidal neurons can span several layers, with apical dendritic branches also being highly stratified.
Here the author actually mentions the fact that makes laminar organization functionally irrelevant, but he avoids actually mentioning connectivity in this sentence. The next sentence and following paragraphs emphasize the importance of laminar organization. Thus, while alert reader can figure out the irrelevance of laminar organization, most of readers would be left with wrong impression.
Forward connections, from lower to higher areas in the hierarchy, originate in superficial layers (layers 2+3) and terminate in the middle layer (layer 4). (In the case of V1 the forward input to layer 4 originates from the THALAMUS.) Feedback connections originate from deep layer (layers 5 and/or 6) neurons and the axons of these cells terminate in superficial and deep layers.
Except the connections from the THALAMUS (which is not part of the cortex), there is no evidence that the 'laminarity' (i.e. which layer they synapse on) of the connections from lower areas to higher areas is different from that of the connections from higher areas to lower areas. A recent review from the same author ( Callaway, E.M. (1998). Annu. Rev. Neurosci 21: 47-74) does not cite any such evidence. Considering the fact that people do look for this kind of features, and the technology available today, that probably means that there is no such difference.

A confusing factor is that the definition of forward connections vs. feedback connections is not clear. Here forward connections mean connections from lower areas to higher areas, but in the next paragraph this term is applied to connections inside the same area as well, and is defined to mean 'dominant'.

The forward connections are strong and their organization has a dominant influence on the visual responses of recipient neurons (see SINGLE-NEURON RECORDING).
If the term 'forward connections' here means 'connections from lower areas to higher areas', than the first part of the statement (that they are strong) is wrong. If 'forward connections' means connections to level 4, then it may be true. Certainly, SINGLE-NEURON RECORDING does not yet gives any evidence about the strength of the connections. The first part is true in the case of input from the THALAMUS to V1.

The second part of the sentence is trivially true: Neurons in higher levels have very little input that carries visual information except the 'forward connections', so their visual response are necessarily dominated by the by these connections. Since this domination is not dependent on the strength of the connections, this does not support the assertion in the first part of the sentence.

Local excitatory connections, intrinsic to a single visual cortical area, are also highly layer-specific and can be classified as forward/dominant and feedback/modulatory.....
This paragraph is very misleading, because it uses the organization of V1 as an example for the organization of visual cortex regions. That is a bad idea, because V1 is clearly a unique area, mainly because most of the visual input to the cortex is in it, by projections from the LGN, which are dominant in V1. In addition, the author grossly simplify the connectivity of V1, which contains many connections which do not fit the circuit he describes.

Note also the change in the definition of forward from 'from lower to higher' to 'dominant'.

Binding by neural synchrony

This requires binding mechanisms that can cope efficiently with combinatorial complexity.
Why it requires 'binding mechanisms' rather than other mechanisms? The author does not explain, and it seems that he takes it for granted that the mechanisms are binding mechanisms.
However, this strategy of binding features together by recombining input connections in ever changing variations and representing relations explicitly by responses of specialized cells results in a combinatorial explosion of the number of required binding units.
The author did not tell us what 'binding unit' is, but he probably means a single-unit coding (see in BINDING PROBLEM chapter which is discussed in mitecs-psychology).
Here, a particular constellation of features is thought to be represented by the joint and coordinated activity of a dynamically associated ensemble of cells each of which represents explicitly only one of the more elementary features that characterize a particular perceptual object.
The basic idea is correct, but this sentence adds a lot of excess baggage. For distributed representation all that is required is that the neurons in the representation fire at the same time, where the timescale of the 'sameness' is shorter than the time it takes to do a simple mental operation. That maybe what 'joint an coordinated' means, though it is not obvious why the two adjectives are required. Distributed representation does not require any of: These requirements are completely spurious, and have no relevance to the brain. They probably come from 'computational considerations', which stands for 'it is easier to code this in computer program'. It may also be a kind of 'intelligent neuron' misconception (myths and misconception), of ignoring the effects of neurons. The author does not realize that if the neurons in the representation are active at the same time, they will tend to activate a specific set of neurons, and that all that is required for representation (we know that because that is the most that any neural representation can ever do).

The rest of this paragraph is irrelevant, because it is based on these spurious requirements.

Both the ambiguities and the temporal constraints can be overcome if the selection and labeling of responses is achieved through synchronization of individual discharges (Gray et al. 1989; Singer and Gray 1995).
As explained in myths and misconceptions, this does not work because of the problem of propagation. In this text, the author simply ignores this problem.
If synchronization serves as a selection and binding mechanism neurons must be sensitive to coincident input. Moreover, synchronization must occur rapidly and show a relation to perceptual phenomena.
1) Ignoring the question of propagation.
2) The test of 'relation to perceptual phenomena' is so weak that it is meaningless, because there are many possible explanations for correlational firing, e.g. simply a noise, or a way for the system to make firing in general more synchronous, and hence make the activity of neurons more effective in activating other neurons.

Aging, Memory, and the Brain

Memory is not a unitary function but instead encompasses a variety of dissociable processes mediated by distinct brain systems.
Declares the existence of distinct systems as an obvious fact, even though all memory processes are together in the cortex.
Explicit or declarative memory refers to the conscious recollection of facts and events, and is known to critically depend on a system of anatomically related structures that includes the HIPPOCAMPUS and adjacent cortical regions in the medial temporal lobe.
That is simply false. Recollection is not dependent on the HIPPOCAMPUS (see above in the comments about the section Memory, Human neuropsychology ). The HIPPOCAMPUS is important for formation of memories.

Defining declarative memory as associate with conscious recollection is confusing, because we use our declarative knowledge unconciously all the time.

Animal Navigation, Neural Networks

One type of cell, known as the place cell, was originally discovered in the rat HIPPOCAMPUS (O'Keefe and Dostrovsky 1971).
It should be noted that this 'type' of cell is not really a type of cell, i.e. this cells cannot be distinguished from other cells by any property, except their activity. Expert neuroscientists know this, but less experts will be confused by the language to believe that these cells are really of a distinct type.
Remarkably, these cells also seem to utilize the two general processes mentioned above; dead reckoning and landmark-based orientation. Evidence that these cells use landmarks, .....
Here, and in the following paragraphs, the language that the author uses implies that the cells themselves are 'utilizing', 'using landmarks', etc. Obviously, for neuroscientists, it is the connectivity of the cells that determine their activity, and the models that are described later are explicitly based on connectivity. For a non-expert reader, however, the 'intelligent neuron' language is very confusing.

Attention in the Animal Brain

In most contexts ATTENTION refers to our ability to concentrate our perceptual experience on a selected portion of the available sensory information, and in doing so, to achieve a clear and vivid impression of the environment. To evaluate something that seems as fundamentally introspective as attention,....
Attention has a simple explanation: it is what you think about (see in Myths and misconceptions).
Attention is a dynamic process added on top of the passive elements of selection provided by the architecture of the visual system.
There is no evidence that there is any separate process associated with attention.
Visual information is dispersed from primary visual cortex through extrastriate cortex along two main routes.
This repeats the misconcepts of two 'routes' (or 'streams'), when the actual observations are just tendencies in a continuous sheet. See the comments above about the introduction (sentence starts with 'beyond V1').
The different emphasis in information processing within parietal and temporal areas is also apparent with respect to the influences of attentional states.
Interestingly, after using the misleading terms of 'route', 'stream' and 'devoted to' two sentences earlier, the author now uses the much more realistic terms of 'areas' and 'differential emphasis'.

Blindsight

See the discussion in Myths and misconcepts.

Computational neuroscience

General: Underlying assumption of this section is that when doing a specific task, all humans perform the same computation. Because of the stochastic connectivity in the cortex, this is clearly false for processes in the cortex, and hence false for all the high- and medium-levels of information processing.
Most neural systems are specialized for particular tasks, such as the RETINA which is dedicated to visual transduction and image processing.
Outside the brain, yes, Inside the brain, this is less obvious, and the main structure of the human brain (Cerebral Cortex) is not specialized.
Brains are complex, dynamical systems, and brain models provide intuition about the possible behaviors of such systems, especially when they are nonlinear and have feedback loops.
.. But it is not obvious that these intuitions are correct, or even useful. This needs to be supported by some evidence, and it isn't.
The predictions of a model make explicit the consequences of the underlying assumptions, and comparison with experimental results can lead to new insights and discoveries.
Maybe they can, but the fact is that the current computational models don't give us new insights about the brain, and do not lead to new discoveries. This observation is easily explained by the stochastic connectivity, and based on this it can also be predicted that the current models are not going to give any new insights in the future. Only when models that take the stochastic connectivity in account are used, they have a chance to be useful.
Perhaps the most successful model at the level of the NEURON has been the classic Hodgkin-Huxley model of the action potential in the giant axon of the squid (Koch and Segev 1998).
Nice example, but this is not a model of the brain.
Realistic models with several thousand cortical neurons can be explored on the current generation of workstation.
That is ridiculous, because a model with several thousand cortical neurons is far from realistic, because it is far too small compared to the real system. There is no reason to believe that there is any correlation (above the obvious ones) between the properties of systems that are so different in their size.

Consciousness, Neurobiology of

General: as usual, the definition of 'Consciousness' is left open, to make the discussion more confused. The authors equate it with the existence of 'subjective content associated with conscious sensation' (qualia), but in the rest of the discussion they discuss functional features of perception.
Consciousness is one of the principle properties of the human brain, a highly evolved system. It therefore must have a useful function to perform.
Because the authors consider only functional features in their discussion of conciousness, that may be obviously true, but with the most common definition (having 'subjective content'), this is clearly nonsense, because Consciousness can be a side effect.
Crick and Koch (1995) assume that the function of visual consciousness is to produce the best current interpretation of the visual scene -- in the light of past experiences -- and to make it available, for a sufficient time, to the parts of the brain which contemplate, plan and execute voluntary motor outputs (including language).
This 'function of the visual consciousness' is (obviously) a function of the visual system. It is not obvious what 'visual consciousness' means here, and hence what this sentence actually claims.
In the visual modality, Milner and Goodale (1995) have made a masterful case for the existence of so-called on-line systems that bypass consciousness.
Nonsense. Milner and Goodale (1995) haven't establish the existence of separate systems. All the observation they made are to do with overlearned processes, which are therefore fast, efficient (use small amount of resources), and accurate. Because of their efficiency, these processes have negligible effect on the activity of the system, and hence are 'unconscious' and 'don't access working memory'.
An alternative hypothesis is that there are special sets of "consciousness" neurons distributed throughout cortex (and associated systems, such as the THALAMUS and the BASAL GANGLIA).
This is definitely an hypothesis, but it does not have any supportive evidence, and it is difficult to reconcile with the stochastic connectivity in the cortex, because the latter means that the activity of "consciousness" neurons and other neurons cannot be sensibly separated. The authors, however, take for granted that it is true in the rest of the discussion.
It is also possible, of course, that all cortical neurons may be capable of participating in the representation of one percept or another, though not necessarily doing so for all percepts.
So why making the more complex assumption of special "consciousness" neurons? The authors do not bother to give any argument.
The secret of consciousness would then be the type of activity of a temporary subset of them, consisting of all those cortical neurons which represent that particular percept at that moment (BINDING BY NEURONAL SYNCHRONY).
The reference to BINDING BY NEURONAL SYNCHRONY is misleading, because this kind of representation does not require NEURONAL SYNCHRONY in the usual sense of the term. See the comments about BINDING BY NEURONAL SYNCHRONY.
Since in the macaque monkey, no neurons in primary VISUAL CORTEX project to any area anterior to the central sulcus, Crick and Koch (1995) proposed that neurons in V1 do not directly give rise to consciousness (although V1 is necessary for most forms of vision, just as the retina is).
Note that this discussion is all in functional terms, and has nothing to do with the 'subjective content' of perception.
A promising experimental approach to locate the NCC has been the use of bistable percepts, that is a constant visual stimulus that gives rise to two percepts, alternating in time, as in a Necker cube (Crick and Koch 1992).
A side point: As usual these authors make the mistake of assuming the that visual stimulus is constant, just because the person (animal) looks at a fixed image. That is never true, because of eyes movements.
Finding the NCC would only be the first, albeit critical, step in understanding consciousness.
Here the authors make explicit their belief in the NCC, even though they did not give any evidence for their existence. By making finding these NCC a 'critical step', they encourage researchers to take them as an established fact, and hence to research imaginary features of the brain. The prestige of one of the authors (Crick) makes the effect specially strong.

Lexicon, Neural Basis

Evidence from neuropsychological and neuroimaging studies have converged on one widely-shared conclusion: the mental LEXICON is organized into relatively autonomous neural subsystems in the left hemisphere, each dedicated to processing a different aspect of lexical knowledge.
(1) Neuroimaging studies are irreplicable, except some general tendencies. One of these is that Broca's and Wernicke's areas are important in language processing. Apart from these, their data is unreliable.

(2) Apart from the concentration in the left-hemisphere, the conclusion is trivial. Different aspects of lexical knowledge require different operations, and hence must be handled by different sets of neurons, i.e. they are necessarily 'autonomous'. The interesting is whether they are completely separated or innate, and the data does not really tells us anything about these properties.

Converging evidence for this conclusion comes from functional neuroimaging studies which have shown that distinct brain regions are activated when neurologically intact participants are engaged in processing the phonological (frontal-temporal) versus the orthographic (parietal-occipital) forms of words.
That is clearly modality effect, because auditory input is in the temporal lobe and visual input is in the occipital lobe.
These results, and the fact that the reliable categories of category-specific deficits are those of animals, plant life, artifacts, and conspecifics, have led to the proposal that conceptual knowledge is organized into broad, evolutionarily-determined domains of knowledge.
This is doing the 'dissociation myth' error (Reasoning errors). The author seems to believe that the fact that the deficits tend to dissociate animals, plant life, artifacts and conspecifics suggests it is evolutionary, which is nonsense. This categories are distinct, so would be handled by partly separated sets of neurons, and hence tend to dissociate.

Magnetic Resonance Imaging

General: While MRI is useful clinicaly, current cognitive studies are totally irreplicable, and hence useless.

Positron Emission Tomography

General: While PET is useful clinicaly, current cognitive studies are totally irreplicable, and hence useless.

Conditioning and the Brain

How the brain codes, stores, and retrieves memories is among the most important and baffling questions in science.
The brain does not 'store' or 'retrieve' memories, it form memories and uses (recalls) them. By using the wrong terms, the author, like many others, already implies mechanisms (moving units of data around) before considering the evidence.
It is clear that various forms and aspects of learning and memory involve particular systems, networks, and circuits in the brain, and it appears now that it will be possible to identify these circuits, localize the sites of memory storage, and ultimately to analyze the cellular and molecular mechanism of memory.
All the evidence that we have (mainly from brain damage data) shows that memories are distributed in the cerebral cortex, with tendencies (memories associated with sensory input near the entry of the input, assymetry between the two hemi-spheres), but no specific locations.
In general, in basic associative learning and memory, the structures most involved in generating the appropriate responses seem also to be the most likely sites of memory storage (see below).
This is not based on any evidence. In general, memories are in the cerebral cortex. There may be some plasticity outside the cortex, but there is still no evidence for anything like memory outside the cerebral cortex.
Importantly, higher brain structures also become critically engaged in learned fear under certain circumstances.......
The amazing thing about this discussion is that it totally ignores the cerebral cortex (note that is different from the Cerebellar cortex). Considering that this is clearly the most important part of the brain (in mammals) for mental operations (including perception and control of movements), that is quite ridiculous. The author does not give us any reason for this omission.
The cerebellum has long been a favored structure for modeling a neuronal learning system.
... but not because there is any evidence for learning in the cerebellum. The actual reason for favouring the cerebellum, as the author says in the next sentence, is that it is more organized than the cerebral cortex, so fits better the computational approach to brain studies. See for a different explanation of the function of the cerebellum.
The CS (e.g., tone) pathway projects to the forebrain and also, via mossy fibers, to the cerebellum...
This and the rest of the text is mostly confabulations. We still don't have a way to follow the projection of CS in details, but we know for sure that the picture is much more complex than this description.
A wide range of evidence including electrophysiological recording, lesions, electrical stimulation and reversible inactivation during training has demonstrated conclusively that the cerebellum is necessary for this form of learning (both delay and trace), that the cerebellum and its associated circuitry form the essential (necessary and sufficient) circuitry for this learning, and strongly support the hypothesis that the essential memory traces are formed and stored in the localized regions in the cerebellum (see Thompson and Krupa 1994; Lavond et al. 1993; Yeo 1991).
The evidence certainly cannot show 'conclusively' that the cerebellum and its associated circuitry are sufficient for learning, unless the 'associated circuitry' includes most of the rest of the brain, (including the cerebral cortex). None of the 'wide range of evidence' show that the learning is not done elsewhere, most likely in the cerebral cortex.

Memory storage, modulation of

General: Interestingly, this text does not contain any reference to the cerebral cortex, where the memory is 'stored' (more accurately, formed).

Motion, Perception of

These V1 neurons give rise to a larger subsystem for motion processing that involves several interconnected regions of the dorsal (or "parietal") VISUAL PROCESSING STREAM (Felleman and Van Essen 1991).
What does 'give rise' means? Neuroscientists would know that it at most means 'send their axons to', but novices don't, and are likely to interpret it in a different way, i.e. be confused.

The 'subsystem' is not separated from other 'subsystems', so it is not a subsystem. it is not a STREAM either (see the comments about VISUAL PROCESSING STREAMS).

This COMPUTATION is thought to be achieved neuronally via convergence of temporally staggered outputs from receptors with luminance sensitivity profiles that are spatially displaced.
The implication that the COMPUTATION is the same in all cases is not based on any evidence. Considering the stochastic connectivity in the CORTEX, which is true for inputs from the LGN (though these are more organized), this is extremely unlikely.
In concert, the results of these experiments indicate that MT neurons provide representations of image motion upon which perceptual decisions can be made.
Overinterpretation. The results show that neurons in the MT tend to be sensitive to motion information, and that their response is important in processing motion information. It does not tell us what the processing is, so saying they provide 'representations of image motion' (whatever that actually mean) is unjustified.
Computational steps and corresponding neural substrates have been identified for many of these perceptual/motor functions.
The computational steps have been identified only in computational models. We still don't have evidence that this computational steps are done in the brain, and the stochastic connectivity in the cortex makes this extremely unlikely.
In concert, these studies demonstrate that cortical motion processing areas -- particularly MT and MST -- forward precise measurements of object direction and speed to the oculomotor system to be used for pursuit generation.
The term 'precise measurements' is an overinterpretation, because it implies mathematical description of the direction and speed. We know that they send information that the oculomotor system can use for precise movements, but it is not necessarily 'precise measurements'.
Indeed, motion processing is now arguably the most well-understood sensory sub-system in the primate brain.
Ridiculous statement, as we haven't got a clue yet how motion processing is actually done. We have some idea of the which parts of the cortex are relatively important in it, but that is all.

Motor Control

Thus, the CNS must transform information about a small number of variables (direction, amplitude and velocity) into a large number of signals to many muscles.
There is no evidence whatsoever for the assumption that the CNS convert the input to small number of variables, and then use the latter to compute movements. In fact, it seems that cognitive scientists haven't even figure out that this is not obvious, and require some evidence. This is an example of 'relying on common sense' error (Reasoning errors).

From what we know about the cerebral cortex, and from human performance (which takes into account many variables), it seems more reasonable to assume that the information is kept all the time in very large number of "variables". These variables correspond to the activity of a neuron or a small group of neurons, and cannot be grouped in larger groups in any meaningful way. The latter is mainly based on the stochastic connectivity in the cortex.

According to this view, during planning the brain is mainly concerned with establishing movement kinematics, a sequence of positions that the hand is expected to occupy at different times within the extra-personal space.
A more reasonable view is that the brain is concerned with the movement kinematics and other aspects of the movements, rather than only kinematics, and the evidence cannot distinguish between these options.
Later, during execution, the dynamics of the musculoskeletal system are controlled in such a way as to enforce the plan of movement within different environmental conditions.
This fits with my hypothesis about the function of the cerebellum, which is to give fast corrections during the movement.
There is evidence indicating that the planning of arm trajectories is specified by the CNS in extrinsic coordinates....
This paragraph is an example of the 'matching optimal behaviour' error (Reasoning errors). Arm movements can be most easily controlled when they are smooth in fixed coordinated (in other word, don't involve large acceleration). Hence we should expect all efficient movers (i.e. any healthy adult) to move smoothly in fixed coordinates, and the observation that they do tell us nothing.
In this context, a representation in the CNS of the inertial, viscous, and gravitational parameters contained in the equations of motion is no longer necessary.
You can ignore these 'parameters' (more accurately, information) only if you don't care about the speed of the motion. Since normally we do, we cannot ignore this information.

Phonology, Neural Basis of

In order to produce a word or group of words, a speaker must select the word(s) from the set of words in long-term memory; encode its phonological form in a short-term buffer in order to plan the phonetic shape which will vary as a function of the context (articulatory phonological planning); and convert this phonetic string into a set of motor commands or motor programs to the vocal tract (articulatory implementation).
That is one way to implement speech generation, but not necessarily the correct one for the brain, and there is no evidence for it, and it is unrealistic. In particular, the idea of a buffer into which the word is moved is incompatible with the stochastic connectivity in the cortex.

Structure from Visual Information Sources

This development was driven by psychophysical, physiological and anatomical determination of two VISUAL PROCESSING STREAMS in VISUAL CORTEX, one specialized for object shape and one for SPATIAL PERCEPTION.
Repeat the usual nonsense about two separate streams. See the comments about VISUAL PROCESSING STREAMS.
However the question as to whether the primate brain implements mathematically based approaches remains open.
At least this author has doubts, while most authors take it for granted that the brain does implement mathematically based approaches. the stochastic connectivity in the cortex rules this out.

Surface Perception

More generally, the visual system must be able to decompose a variation in luminance caused by intrinsic changes in surface properties (such as surface reflectance), from those caused by the variations that are extrinsic to a surface. There are two ways in which this decomposition seems to be accomplished.
This skips what is probably the most important way, which is information from the effects of eyes movements. The eyes moves essentially all the time, so the retinal image continuously changes. When this does not happen (this can be achieved by an image that moves with the eyeball), humans have serious problem in perceiving not only 3D structure, but anything at all.