Neural basis of semantic-like memory :
representation, activation and cognitive control

(modified from: Ryoko Fujimichi, Yuji Naya, and Yasushi Miyashita, in “The Cognitive Neurosciences III, ed by Gazzaniga, M.S., MIT Press, 2005”)



Imagination is a wonderful gift that improves with use. We now attempt to stimulate your imagination to help you recall a whole portrait of this chapter in an image (Fig.1).

A few days ago, I found a piece of paper with a strange scribble in my suitcase.

Fig. 1 The house on the hill. See the poetry in the first section.

Suppose we think of the monkey as an agent of “the Society of Mind” (Minsky, 1986) or a neural assembly (Hebb, 1949). Each egg appears to have two possible destinies, to be broken or to be hatched. In either case, a butterfly representing the worm’s associative component is born. After you read this chapter, your imagination will help you to understand the structure of the memory systems and to recall the somewhat complicated visual associative processes described in this chapter by using the image of the house on the hill.

Various forms of memory can be classified as declarative (explicit) or nondeclarative (implicit) on the basis of how information is stored and recalled (see,Squire in this volume ; see Schacter in this volume). Declarative memory underlies the learning of facts and personal experiences.Nondeclarative memory includes forms of perceptual and motor memory. Declarative memory can be further classified as episodic (a memory for events and personal experience) or semantic (Tulving,1972). According to Tulving (1972), semantic memory is “a mental thesaurus, organized knowledge a person possesses about words and other verbal symbols, their meaning and referents, about relations among them, and about rules, formulas, and algorithms for the manipulation of these symbols, concepts, and relations...” According to Lindsay and Norman (1977), such knowledge “contained within human memory forms an interrelated web of concepts and actions. Knowledge of one topic is related to knowledge of others. The human memory system makes possible the tracing of the relationships among the knowledge within the database”. They also noted that “much of our knowledge is probably encoded in combinations of network representations,sensory images, and motor control images.”

Thinking of semantic memory as above and as reflecting relationships among symbols, a prototypical form of semantic memory system in the brain would be substantiated as a cumulative collection of cell assembly for objects or their neuronal associative networks. In this chapter, we will provide evidence obtained by others and ourselves of possible neural mechanisms for such an associative network. We will first discuss how a memory engram is organized and introduce the neural representation of semantically linked symbols. We will then discuss how memory engrams are activated, highlighting two different retrieval processes, ‘active’ and ‘automatic’. Finally, we will make further reference to a metamemory system, which supervises the retrieval process and exerts cognitive control.

Representation of visual objects: Organizing a memory engram

Neuronal correlates of associative long-term memory were first reported in the monkey inferior temporal (IT) cortex by Miyashita (1988) and Sakai and Miyashita (1991). Their single-unit recordings identified two mnemonic properties of IT neurons: 1) that IT neurons can acquire stimulus selectivity through learning in adulthood; and 2) that their activity can link representations of temporally associated but geometrically unrelated stimuli. How these memory neurons function as basic elements of semantic networks, and how semantic networks are created through interaction among multiple representations in temporal cortical areas, is now firmly established (Miyashita and Chang, 1988; Miyashita and Hayashi, 2000). In that regard, investigation of the neural basis of semantic networks has been greatly facilitated by reducing complex associative networks to elementary associative links between two objects, and then asking what are the neural mechanisms underlying such elementary associative links (Miyashita, 1993). The pair-association (PA) memory task is the best-known neuropsychological test with which to tap the memory of such an elementary pair-wise associative relation (e.g. WMS-R, Wechsler, 1987). In the following section, we discuss how investigations using PA task revealed the neuronal machinery of associative memory in the IT cortex.

Fig. 2 A. Lateral view of a macaque brain. TE is located at the final processing stage of the ventral visual pathway. A36 is thought to be a part of the medial temporal lobe system. V4, visual area 4; TEO, area TEO.B. Coronal cross section indicated by the horizontal line on the ventral view in panel A. The black and gray areas indicate the locations of the recoding sites in TE and A36, respectively.C. Sequence of events in one trial of the PA task. Fixation points and cue stimuli were presented at the center of a video monitor. Choice stimuli were presented randomly in two of four positions on the video monitor. (Modified from Naya et al., 2000)

Forward processing of pair-association memory

The IT cortex contains two cytoarchitectonically distinct but mutually interconnected areas: area TE (TE) and area 36 (A36) (Suzuki and Amaral, 1994; Saleem and Tanaka, 1996) (Fig.2A, 2B). TE is a unimodal neocortex and located at the final stage of the ventral visual pathway, which processes object vision (Logothetis et al., 1995; Tanaka, 1996; Sheinberg and Logothetis, 1997; Janssen et al., 2000). A36, on the other hand, is a limbic polymodal association area and a component of the medial temporal lobe memory system, which is involved in the formation of declarative memory (Zola-Morgan and Squire, 1990; Murray et al., 1993; Higuchi and Miyashita, 1996; Yakovlev, 1998; Murray and Bussey, 1999; Liu and Richmond, 2000; Brown, 2001). Naya et al. (2003) found that association between the representations of different but semantically linked objects proceeds from TE to A36. To do this, they trained monkeys to perform the PA memory task in which meaningless computer-generated pictures were sorted randomly into pairs. We refer to each member of a pair as a paired associate. The monkeys were trained to memorize combinations of paired associates, which could not be predicted otherwise. In each trial of the task, a cue stimulus was presented, and the monkey was rewarded when he/she chose the paired associate of the cue (Fig.2C). After training, extracellular spike discharges were recorded from single neurons in TE and A36.

A total of 2368 neurons were recorded from A36 (510 neurons) and TE (1858 neurons) in the three monkeys performing the PA task. Of those, 475 neurons (85 neurons in A36 and 390 neurons in TE) showed responses to the cue presentation for at least one stimulus among the 24 learned stimuli (visually responsive cells). Out of them, 423 neurons (76 neurons in A36 and 347 neurons in TE) showed significant (p < 0.01, ANOVA) stimulus selectivity during the cue period (60-320 ms from cue onset)(cue-selective cells). The responses of a representative cue-selective neuron in A36 are shown in Figures 3A and 3B. Note that one stimulus elicited the strongest response from this neuron during the cue period (Fig.3A, thick black; Fig.3B, filled bar in pair 4). This neuron was also activated when the paired associate of the preferred stimulus was presented (Fig.3A, thick gray; Fig.3B open bar in pair 4). In contrast to the robust responses to this stimulus pair, this neuron responded only negligibly when stimulus from any of the other pairs were presented as cue stimuli. This type of neuron, which is found in the IT cortex and is referred to as a ‘pair-coding neuron’, selectively responds to both paired associates (pair-coding response). This property indicates that memory storage is organized such that single neurons can code both paired associates in the PA task.

The pair-coding responses in A36 with those in TE were compared by examining the distributions of the response amplitude for the pair trials. It was found that the distribution were significantly shifted toward positive values in both areas (A36, median = 0.27; TE, median = 0.03; p < 0.001 in both areas, Wilcoxon’s signed-rank test), with the distribution for the A36 neurons shifted to significantly higher values than that for the TE neurons (p < 0.001, Kolmogorov-Smirnov test). Thus, in addition to the preferred stimulus, neurons in both A36 and TE responded selectively to the paired associate of the preferred stimulus, and the response was more prominent in A36 than in TE. In addition, the percentage of pair-coding neurons among the cue-selective neurons was significantly higher in A36 (33 %) than in TE (4.9 %) (p < 0.001, x2 test) (Naya et al., 2003). This means that although neurons in both areas acquire stimulus selectivity through associative learning, the effect is engraved more intensely on the neuronal representation in A36 than in TE. The dramatic increase in the percentage of pair-coding neurons that one sees by going from TE to A36 indicates that the association between representations of paired associates proceeds forward through this anatomical hierarchy of the IT cortex.

Circuit reorganization during formation of the pair-association memory

It has been long hypothesized that memory engrams of declarative knowledge, as exemplified by the emergence of pair-coding neurons, develop with a structural and functional reorganization of neural circuits in the cerebral association cortices. (Squire and Zola-Morgan, 1991; Miyashita Y, 1993; Mishkin et al., 1997; Jones, 2000; Martin et al., 2000). This reorganization of neural circuits would be accomplished through a cellular program of gene expression leading to increased protein synthesis and then to alteration of synaptic connections (Bailey and Kandel, 1993). To date, this hypothetical framework has been primarily investigated in invertebrates and lower mammals, in which it is difficult to examine the organization of semantic memory. Still, the hypothesis as applied to semantic memory has been tested in a series of molecular biological studies carried out in monkeys (Okuno et al., 1996; Tokuyama et al., 2000, 2002), showing that upregulation of mRNA encoding proteins thought to be involved in structural reorganization occurred during formation of the pair-association memory. In this series of studies, the RT-PCR mRNA quantitation was combined with three unique experimental strategies. The first strategy entailed the use of split-brain monkeys, which were prepared by transecting the anterior commissure and the entire extent of the corpus callosum (Hasegawa et al., 1998). The fact that there was no interhemispheric transfer of mnemonic engrams in this preparation (Hasegawa et al., 1998; Gazzaniga, 1995) enabled us to compare mRNA expression within individual monkeys, thereby eliminating genetic and cognitive variations between individuals. The second strategy entailed the use of a visual discrimination (VD) task, rather than a no-task condition, as the control. This enabled the motivational and attentional states in the two hemispheres, as well as the input of visual stimuli, to be appropriately balanced. The third strategy entailed training monkeys to first learn a 'rule' or 'strategy' component of the tasks using training stimulus sets, after which a test stimulus set was introduced for new learning of the declarative components of the task. Before the learning process with the test stimulus was complete, the animals were perfused, and expression of mRNA in the brains was evaluated. This enabled investigation of gene expression during formation of associative memory but not during formation of procedural memory.

Using the approach described above, it was found that expression of mRNA encoding Brain Derived Neurotrophic Factor (BDNF) was significantly higher in A36 of the PA hemisphere than in the VD hemisphere (p < 0.05). In the early visual cortex (e.g., V1 or V4), by contrast, expression of BDNF mRNA did not differ in the two hemispheres (V1, p > 0.60; V4, p > 0.87), indicating that the increased expression of BDNF mRNA level in A36 did not reflect a difference in the amount of visual input. The RT-PCR analysis also showed that expression of the mRNA encoding trkB, a specific receptor for BDNF (Bonhoeffer, 1996; McAllister, 1999), was slightly increased in A36 of the PA hemisphere, though the increase did not reach statistical significance. The expression of the mRNA encoding the immediate-early gene zif268 was also selectively upregulated in A36 during formation of PA memory (Tokuyama et al., 2002). On the other hand, expression of a 'housekeeping gene,' β-actin, did not differ between the two hemispheres in any of the cortical areas examined.

Fig. 3 In situ hybridization of BDNF mRNA. A-D. Distribution of BDNF mRNA in the IT gyrus of the PA (A) and VD (B) hemispheres. BDNF mRNA accumulated in a patch in A36 of the PA hemisphere (framed area), but not in A36 of the VD hemisphere. The framed areas in (A)and (B) are enlarged in (C) and (D), respectively. BDNF mRNA-positive cells were observed in layers V/VI and in layers II/III of the PA hemisphere (C); the image of the cell marked by the arrow are defined in (E-G). En, entorhinal cortex; 35, area 35; 36, area 36; TE, area TE; rs, rhinal sulcus. Arrowheads mark the boundaries between different cortical areas.E-G. BDNF mRNA-positive cells in layers II/III of the PA hemisphere. The cell marked by an arrow in (C) is enlarged and shown in darkfield (E), bringht field (F), and brightfield with epi-illumination (G). Silver grains were concentrated around lightly Nissl-stained neuronal nuclei. Cortical layers of A36 are indicated along margin of (D). Scale bars, 1 mm (A, B), 250 μm (C, D), 50 μm (E-G).(Modified from Tokuyama et al., 2000)

The spatial distribution of the BDNF mRNA was visualized using in situ hybridization (Fig.3). Notably, BDNF mRNA-positive cells accumulated as a “patchy” cluster in A36 of the PA hemisphere, but not in the same area of the VD hemisphere, which suggests that upregulation of BDNF expression is associated with neurons located within the patches in A36. These patches were most prominent in layers V/VI, but were also observed in layers II/III (Fig.3C), extending for at least 0.4 mm along the anteriorposterior axis. In contrast to the PA hemisphere, the VD hemisphere contained only scattered BDNF mRNA-positive cells in layers V/VI of A36 (Fig.3B and D). And when the magnitude of the local increase of BDNF mRNA expression was estimated by grain-counting analysis (framed areas, Fig.5A and B), it was found that significantly more neurons in the PA hemisphere expressed detectable levels of BDNF mRNA than in the VD hemisphere [9.1±0.7 % in the PA hemisphere vs. 3.7±0.6 % in the VD hemisphere (x2 = 72.4, p < 0.001)]. In area 35 of the PA hemisphere, expression of BDNF mRNA also seemed slightly stronger than in the VD hemisphere, but there were no differences in the patterns of BDNF mRNA expression in any other regions of the PA and VD hemispheres.

BDNF is thought to mediate activity-dependent synaptic plasticity, even in mature nervous systems (Bonhoeffer, 1996; Thoenen, 1995; McAllister et al., 1999). Consistent with that idea, its expression is regulated by changes in neuronal activity. Moreover, since zif268 encodes a transcription factor, its expression could function as a trigger for a cascade of gene activation that leads to the cellular events underlying neuronal reorganization. Thus, analysis of the formation of PA memory has provided the first evidence supporting the hypothesis that BDNF contributes to the reorganization of neural networks, and that perhaps this reorganization is initiated by zif268, which triggers a cascade of gene activation.

The location of the focal patch expressing BDNF approximates the location of aggregates of pair-coding neurons detected by single-unit recording (referred to as a ‘hotspot’)(Naya et al., 2003, Yoshida et al., 2003). A combined anatomical-physiological analysis recently showed that structural reorganization does indeed occur at hotspots and that the fiber terminals of picture-selective neurons in TE are retracted out of the hotspot in A36 but remain to project within the hotspot (Yoshida et al., 2003). We therefore suggest that BDNF expression may induce axonal and synaptic reorganization in the hotspot in A36, and that such reorganization of local networks is detected electrophysiologically as a change in neuronal stimulus selectivity, typically as the emergence of pair-coding neurons.

Activation of memory engrams: ‘active’ vs. ‘automatic’ retrieval

There appears to be two types of memory retrieval processes, automatic and active; their differences are illustrated in the following example. Whoever loved adventure stories when they were boys and girls, a beautiful fairy who called TinkerBell strongly associates with the boy who would not grow up, Peter Pan. It would then be easy to recall TinkerBell’s name from that of Peter Pan - i.e., it would be automatic. On the other hand, if you were asked the name of the author who wrote “Peter Pan and Wendy”, you would likely find it more of an effort to recall the author’s name - e.g., James M. Barrie.

Now you know that sometimes one needs no effort to recall and at other times one has to strive toward a successful recall. We refer to the former as automatic retrieval and to the latter as active retrieval (Petrides, 2000; Fletcher and Henson, 2001). In the previous section, we discussed visual associative memory stored in the IT cortex. It follows that such long-term memory could then be retrieved from the IT cortex by either of these two processes. In this section, we will suggest that whether retrieval is automatic or active depends on whether the retrieval signal is created within the network of the IT cortex or runs from the prefrontal (PF) cortex to the IT cortex. We begin by introducing a study that showed that the automatic memory retrieval signal flows backward through the IT cortex, from A36 to TE.

Automatic retrieval signal: Backward spread of memory retrieval signal in the inferior temporal cortex

The theory of semantic network visualizes a retrieval of an item as activation of a corresponding node at the network. The neural correlate of such a node-activation was first reported in the pair-association task by Sakai and Miyashita (1991). They found a group of IT neurons that showed an activation related to the retrieval of the paired associates from a cue stimulus. The response is referred to as pair-recall response. Using a modified PA task (PA with a color switch task), Naya et al. (1996) showed that this pair-recall response indeed corresponded to the recall of the visual image in subject’s mind, since IT neurons started to fire just after a color switch that signaled the necessity and timing of memory retrieval during its delay period and the IT neurons also stopped to fire just after another color switch that signaled retrieval of other memorized items. Such change of neuronal discharge did not occur when a color switch signaled a simple maintenance of short-term memory. Therefore, this type of delay activity in the IT cortex represents the internal target that is retrieved from long-term memory. Recently, it was reported that this target-related activity is transmitted backward, from A36 to TE, as illustrated below (Naya et al., 2001).

The responses of representative A36 neurons showing the target-related activity specified by a cue stimulus have been well documented (Sakai and Miyashita, 1991; Naya et al., 1996). One stimulus elicited the strongest response during the cue period and the response continued into the delay period. In the trial when the paired associate of this preferred stimulus was presented as a cue, the neurons in both A36 and TE exhibited the highest delay activity among the stimuli. This type of activity is referred to as target-related. The onset of the target-related activity of the TE neuron was later than that of the A36 neuron.

The time course of the target-related delay activity of each neuron was examined by considering the responses to all cue stimuli. The partial correlation coefficients of instantaneous firing rates at time t for each cue stimulus were calculated with the visual responses to its paired associate (pair-recall index, PRI). The time courses of the average PRI(t ) across the population of stimulus-selective neurons were found to significantly differ in A36 and TE (repeated-measures ANOVA, p < 0.0001). The PRI(t ) for the A36 neurons began to increase during the cue period and developed with a rapid time course. The PRI(t ) for the TE neurons, by contrast, increased slowly and reached a plateau in the middle of the delay period. To summarize this section, memory-retrieval signals appeared first in A36, after which TE neurons were gradually recruited to represent the sought target. Thus mnemonic information that was extracted from long-term storage, spread backward, from A36 to TE.

Top-down signaling appears when active retrieval is required.

A clinical case study helps us to highlight the active retrieval in humans and provides a clue to an experimental model with which to investigate active retrieval (Sidtis et al., 1981). In that study, an epileptic patient who had undergone posterior callosotomy was given a word in his left visual field. He could never read the name of it directly, although he claimed to ‘see’ its image in his mind. He was nevertheless able to eventually answer the name using inferential strategies based on his mental image. His limited ability suggests that his right hemisphere was transmitting to his left hemisphere semantic information about the stimulus but not the actual stimulus. After the callosum was completely sectioned, semantic information was no longer transferred from his right hemisphere to left. Hasegawa et al. (1998) combined this posterior-split-brain paradigm with the associative memory task in monkeys. In the posterior-split-brain monkey, in which the posterior corpus callosum and the anterior commissure are sectioned, the cortex receives bottom-up visual information only from the contralateral visual field. With this paradigm, long-term memory acquired through stimulus-stimulus association does not transfer interhemispherically via the anterior corpus callosum; nonetheless, when the visual cue was presented to one hemisphere, the anterior callosum could instruct the other hemisphere to retrieve the correct stimulus specified by the cue. Thus, although visual long-term memory is stored in the temporal cortex, memory retrieval is under the executive control of the PF cortex.

Direct proof of the existence of top-down signal from the PF cortex to the temporal cortex and of its contribution to the active retrieval process was provided by single-unit recordings from the posterior-split-brain preparation (Tomita et al., 1999). In this protocol, inferior temporal neurons in one hemisphere are to be activated by bottom-up visual inputs when an object is presented in the visual hemifield contralateral to the recording site. When the object is presented in the ipsilateral hemifield, however, these neurons should not be able to receive bottom-up visual inputs. Any neural activation should therefore reflect top-down inputs from the PF cortex. It was found that a considerable number of IT neurons did indeed receive top-down signals from the PF cortex; the activity of one such IT neuron is shown in Figure 8. This neuron was not only activated by contralateral presentation of stimuli, but was also activated by ipsilateral presentation of stimuli. In these neurons, which showed stimulus-selectivity in both top-down and bottom-up responses, it was found that the latency was significantly longer in the top-down condition (p < 0.001). These top-down responses were abolished after transection of the remaining anterior corpus callosum. The partial split-brain studies in human and monkeys reveal the events occurring during the active retrieval process, which are indicative of purely top-down signaling.

Fig.4  A. Experimental procedures of the Feeling-of-Knowing (FOK) trials. Subjects were required to recall word answers to general-information questions during fMRI scans. By pressing buttons, they indicated whether they recalled the target words or not. Then, outside the scanner, they judged their degree of FOK to the nonrecalled questions on a scale of three, or they wrote their answers to the recalled questions. Each trial was sorted into trial type (Recalled, FOK3, FOK2, and FOK1) according to the participant’s judgment and was subjected to event-related fMRI analysis.B. Regions involved in FOK and/or successful recall. A subset of the FOK regions was also activated in successful recall (c, d, and e). A subset of the FOK regions in the bilateral IFGs was not activated in successful recall (a and b). A small subset of the FOK region in the left IFG was also activated in successful recall, but its cluster size did not reach the significant, given the correction for the whole brain multiple comparisons. (c), bilateral caudate nuclei/thalami; (d), left middle frontal gyrus; (e), ACC/SMA; (f), left superior parietal lobule/inferior parietal lobule; (g), precuneus.(Modified from Kikyo et al., 2002)

Metamemory: Cognitive control of memory system

During an exam, you might have experienced a situation in which, although you were unable to answer a question in a closed test, you were sure you could have answered correctly if it was a multiple-choice question. In fact, you accessed related items in your semantic network, since there is a positive correlation between the objective score in the recognition test and the degree of a subjective feeling whether you knew the answer or not. Metamemory refers to knowledge about one’s memory capabilities and knowledge about strategies that can aid memory (Shimamura AP, 1995). Metamemory requires execution of extensive retrieval process and, at the same time, supervises the retrieval process. Kikyo et al. (2002) successfully identified brain areas related to a metamemory system in humans using a “feeling-of-knowing” (FOK) paradigm, which is a well-established tool for assessing metamemory system (Fig.4A). The FOK is a subjective sense of knowing a word before recalling it - i.e., a sense that “I know that I know it”. Based on the recall-judgment-recognition (RJR) paradigm (Hart, 1965), subjects in this experiment were asked to recall word answers for general-information questions during fMRI scans. After the scans, the subjects wrote their answers to the recalled questions and were instructed to judge their degree of FOK to the non-recalled questions on a three-point scale, where 3 = < I definitely could recall the answer if given hints or more time>; 2 = < I probably would recognize the answer>; 1 = < I definitely did not know>. Event-related fMRI with a parametric analysis showed stronger activity in the bilateral IFGs (BA 47), left MFG (BA 46/9, BA 10) and ACC/SMA (BA 32/24/6) when people have a greater FOK (Fig.4B)(See also, Maril et al., 2003). These activation areas are referred to here as the FOK regions. Among these FOK regions, subregions in the bilateral IFGs (BA 47) were not recruited for successful recall processes, suggesting a specific role of these regions in human metamemory system. One of the FOK regions was located in the anterior portion of the left MFG, BA10. This area was regarded as a part of the memory areas in the anterior prefrontal cortex (AFC) in some literatures and was related to retrieval strategy and/or “third level of executive control” (Fletcher and Henson, 2001).

Fig. 5  Comparison of shift-related activation in the PF cortex in monkeys and humans. Lateral views of 3D-rendered brain image in which the activation shown was superimposed (A, monkeys; B, humans). In the human data, a prominent activation focus in the posterior part of the inferior frontal sulcus is shown. Other activations in the precentral gyrus and in the anterior insula were also observed. For reference, cytoarchitectonic maps of macaque monkeys by Walker (C, 1940) and that of humans by Brodmann (D, 1910) are presented. These maps correspond approximately to areas in the white squares in A and B. Green arrowhead, the principal sulcus; blue arrowhead, the inferior ramus of the arcuate sulcus; yellow arrowhead, the inferior frontal sulcus. Scale bar, 30 mm.(Modified from Nakahara et al., 2002)

Future perspective

The extensive investigations described in this chapter have been revealing a whole picture of the semantic memory system. There are, however, still pieces missing that will be needed to make the picture complete. For example, in the metamemory system, how are the active retrieval subsystem and its cognitive control subsystem integrated? ; How does each component of the identified distributed metamemory network in the PF cortex differentially support those subsystems? ; Is there any specific molecular/cellular basis for those differential functional subsystems? It is anticipated that a number of missing pieces will be found when the gap between the information provided by invasive studies carried out with monkeys and that provided by non-invasive human imaging studies is filled (Logothetis et al., 1999; Miyashita and Hayashi, 2000). Most of the detailed knowledge of the anatomy, function and cellular basis of the cortex has come from studies in monkeys (Felleman and Van Essen, 1991; Miyashita, 1993; Gaffan, 1994; Goldman-Rakic, 1995; Fuster, 1995; Desimone, 1996; Miller, 2000; Logothetis, 2002). With that as background, using the same methods to study humans and monkeys would advance our understanding of the neural organization of higher order cognitive function. For example, fMRI is a method that may bridge this gap by enabling direct comparison of the functional organization of the brains of monkeys and humans (Logothetis et al, 1999; Hayashi et al., 1999; Logothetis et al., 2001; Vanduffel et al., 2001; Logothetis et al., in this volume). Using that approach, Nakahara et al. (2001) observed that, when subjects performed a high-level cognitive task, transient activation related to cognitive set shifting occurred in focal regions of the prefrontal cortex in both monkeys and humans (Fig.5), and that these functional homologues were located in cytoarchitectonically equivalent regions in the posterior part of ventrolateral prefrontal cortex. Such comparative imaging also has the potential to provide significant new insight into the evolution of cognition in primates.

We began this chapter by describing the image of the house on the hill. Visual object information (a worm) reaches the final processing stage of the ventral visual pathway (climbing up the hill). There are cases in which the eggs hatch naturally (automatic retrieval). On the other hand, if the cell assemblies in the PF cortex (fairy monkeys on the second floor) and those in the IT cortex (monkeys at first floor) interact, a top-down signal triggers active retrieval from long-term memory storage (eggs are broken by monkeys). Any case, butterflies (its paired associates) would flutter into the backyard. It is interesting to further understand what a monkey in the garret actually does. Our future perspective is to put the missing pieces in their appropriate places, one by one, and in that way to make the image of the house completely clear.