COGSYS 274 1 September 2008 Disk Used

ARTICLE IN PRESS
Available online at www.sciencedirect.com

No. of Pages 11, Model 5+

Cognitive Systems Research xxx (2008) xxxxxx www.elsevier.com/locate/cogsys

2 3 4 5 6 7 8

Qi Zhang *

Sensor System, 8406 Blackwolf Drive, Madison, WI 53717, USA Received 19 February 2008; accepted 20 June 2008

9 10 11 12 13 14 15 16 17 18 19 20 21

Abstract

Keywords: Computational model; Dream; Random activation; Learning; Memory consolidation

RR E

22 23 24 25 26 27 28 29 30 31 32 33 34

1. Introduction

CT

A number of studies have concluded that dreaming is mostly caused by random signals because dream contents are random impulses, and argued that dream sleep is unlikely to play an important part in our intellectual capacity. On the other hand, numerous functional studies have suggested that dream sleep does play an important role in our learning and other intellectual functions. Specifically, recent studies have suggested the importance of dream sleep in memory consolidation, following the ndings of neural replaying of recent waking patterns in the hippocampus. This study presents a cognitive and computational model of dream process that involves episodic learning and random activation of stored experiences. This model is simulated to perform the functions of learning and memory consolidation, which are two most popular dream functions that have been proposed. The simulations demonstrate that random signals may result in learning and memory consolidation. The characteristics of the model are discussed and found in agreement with many characteristics concluded from various empirical studies. 2008 Published by Elsevier B.V.

ED

Dreaming refers to the subjective conscious experiences we have during sleep. The experience is vivid, intense, bizarre, and is hard to recall. Various studies have concluded that dream sleep may help us in learning (e.g., Greenberg & Pearlman, 1974; Hennevin, Hars, Maho, & Bloch, 1995; LaBerge, 1985; Smith, 1995), and may be a perceptible embodiment of a dreamers conceptions (Hall, 1953). Findings of the correlation between REM (rapid eye movement) sleep and waking learning have suggested that dream sleep may play an important role in learning and memory consolidation (e.g., Bloch, Hennevinm, & Leconte, 1979; Fishbein, 1970; Pearlman, 1971; Winson,

Tel.: +1 608 217 0742; fax: +1 608 821 0068. E-mail address: qzhang_sensor@yahoo.com

1985). Similar ndings have been also concluded in more recent psychophysiological studies, although only these initiating studies are cited. Long-term memory can be fractionated into declarative (explicit) memory and nondeclarative (implicit) memory, and declarative memory can be further divided into episodic memory and semantic memory. Episodic memory is the memory of past experiences, while semantic memory is about factual and generic knowledge (Tulving, 1972). The hippocampal complex, including the hippocampus and its surrounding areas, is considered a critical region in retaining recent episodic memory or its traces. On the other hand, the general neocortex is considered the place where semantic memory is stored. Memory consolidation is considered a neural process by which episodic memory becomes independent of the hippocampal complex and is consolidated into the neocortex (Squire & Alvarez, 1995). Findings from neural recording, which reveal the replaying

PR

OO

Action editor: Rajiv Khosla

A computational account of dreaming: Learning and memory consolidation

35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52

1389-0417/$ - see front matter 2008 Published by Elsevier B.V. doi:10.1016/j.cogsys.2008.06.002

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

CO

COGSYS 274 1 September 2008 Disk Used 2 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx

No. of Pages 11, Model 5+

of recent waking patterns of neuronal activity within the hippocampus during sleep, have further reinforced the view that dreaming may play an important part in memory consolidation. This reactivation of hippocampal cells has been recorded in rats (e.g., Pavlides & Winson, 1989; Wilson & McNaughton, 1994) and in humans (Staba, Wilson, Fried, & Engel, 2002); in SWS (slow wave sleep) that dominates Non-REM sleep (Pavlides & Winson, 1989; Wilson & McNaughton, 1994) and in REM sleep (Louie & Wilson, 2001; Poe, Nitz, McNaughton, & Barnes, 2000). Furthermore, synchronized activity in the hippocampus and neocortex in sleep is also reported and attributed to memory consolidation into the neocortex (Battaglia1, Sutherland, & McNaughton, 2004). The signicance of dreaming on learning and memory consolidation, suggested by psychophysiological studies, is also supported by the ndings of hippocampal ring in sleep. In some studies (e.g., Fosse, Fosse, Hobson, & Stickgold, 2003; Pavlides & Winson, 1989; Schwartz, 2003), dream sleep has been directly associated with the hippocampal rings for the possible link between the cognitive activity of brain and the activation of stored episodic memory. This association is reasonable when we consider the fact that both vivid dreams and thought-like experiences can be recalled in >70% of REM awakenings (Hobson, 1988), and >48% of Non-REM awakenings (Nielsen, 2000), respectively. However, studies in dream reports may lead to a dierent conclusion. Carefully looking into dream contents, it is often concluded that dreams are more or less random thoughts (Foulkes, 1985; Hobson & McCarley, 1977; Wolf, 1994). A recent study by Fosse et al. (2003), which is focused on the correlation between daily experiences and dream contents, again conrms the randomness nature of dreams. This study found that daily experiences are replayed in the form of segments, rather than entire episodes, during REM sleep. In other words, daily experience is replayed more in random fashion and less in sequential fashion in dreams. The randomness has led to the proposal of the activation-synthesis model (Hobson, 1988; Hobson & McCarley, 1977; Hobson, Pace-Schott, & Stickgold, 2000). The model states: dreams are caused by random signals arising from the pontine brainstem during REM sleep; the forebrain then synthesizes the dream and tries its best to make sense (i.e., dream images) out of the nonsense (i.e., random impulses) it is presented with. In short, the dream randomness has been used against some proposed intellectual functions of dreaming, and has divided dream theories into functional and functionless. Therefore, whether or not dreaming has intellectual functions depends on whether random impulses can lead to intellectual consequences, e.g., learning and memory consolidation. Compared to numerous psychophysiological and neurobiological studies that are associated with dream mechanisms and functions, few computational studies have been reported on the same aspects. These reported studies are based on connectionist modelings. In the most cited

PR

study by Crick and Mitchison (1983), it is concluded that dreaming is the reverse-learning process to remove socalled spurious memories (i.e., useless and old memories) in order to avoid overload of the brain. A similar simulation, however, indicates that the useless and old memory is actually increased after the reverse-learning of the simulated dreaming process, and suggests that we dream to roughen up our memory space (Christos, 2003). In either case, it is said (Botman & Crovitz, 1989; Domho, 1996) that the conclusions are generally disassociated with what has been found about dreaming, and are typically interpreted to contradict the psychoanalytic account of dreams. In this study, a cognitive and computational model of dreaming is presented. This model is developed from a previous construct (Zhang, 2005) of a learning system. In this present study, dreams of the computational system are performed. The outcomes of dreaming are examined in terms of naming and picture drawing, which are typical tasks in the tests for semantic memory. The characteristics of the model are discussed and found in agreement with many empirical ndings from dream studies.

110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

CO

RR E

CT

ED

2. The construct of an articial intelligence dreamer 2.1. A brief revisit of the learning system Knowledge can be learned from experience. How learning occurs and how new knowledge is associated with prior knowledge, are questions yet to be answered. A previous study (Zhang, 2005) presents a cognitive learning system, namely AI counter, which can learn to count. That system is constructed based on two rules: (1) a concept (represented by a common feature) is learned when the common feature is abstracted and generalized; and (2) new learning has to rely on prior learning if the newer knowledge is an extension of the prior knowledge. The learning system was built with a multi-level structure of information processing, which is in fact the system shown in Fig. 1, except for the hippocampal memory. In the system, the base level of cognition is called single memory that stores and reacts to one piece of an entire external input. A single memory has three inputs (excitation input Iexc, signal input Isig, and interlock input Iint) and four outputs (excitation output Oexc, signal output Osig, coordination output Ocor, and interlock output Oint). The function states of a single memory on how these signals interact are summarized in Table 1. These states reect the two functions of a single memory. One function is to store an Io/Iexco pair in two steps. In step 1, when Isig = Io and Iint = yes, a single memory res an Ocor (=yes), which is then transported to the bundle of intersubsystem signals (see Fig. 1). Only when the bundle receives a coordination signal from both the symbol and representation subsystems, it generates a unique excitation signal (Iexco) and sends it back to all single memories in both subsystems. In step 2, the single memory that red Ocor stores both Io and Iexco together after receiving

OO

COGSYS 274 1 September 2008 Disk Used

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx

No. of Pages 11, Model 5+

CT

Table 1 Three important states of a single memory State Learning Input Step 1: ring Ocor signal Output Isig = Io Iexc = null Iint = yes Isig = Io

ED

Fig. 1. The cognitive framework of a computational dreamer. This system has two cognitive sections: the learning center (i.e., the combination of the symbol and representation subsystems) that acquires conceptual knowledge, and the hippocampal memory that stores past experiences. The Sin and Rin are input ports of a symbol and representation, respectively; while Sout and Rout are output ports. The mode selector can set the system to either waking mode (for the learning center to receive external signals) or dreaming mode (for the learning center to receive randomly activated past experiences coming from the hippocampal memory). The pathways and kinds of signals are indicated under the Signal keys. For clarity, coordination and excitation signals are combined. Detailed signal kinds are given in Table 1.

RR E

Step 2: storing Io and Iexco permanently

Osig = null Oexc = null Oint = no Ocor = yes Osig = null

Iexc = Iexco Iint = yes Firing stored Io upon receiving Iexco after the single memory has learned

CO

Isig = any Iexc = Iexco

Iint = yes

UN

Firing stored Iexco upon receiving Io after the single memory has learned

Isig = Io

Iexc = any Iint = yes

Oexc = null Oint = yes Ocor = yes Osig = Io Oexc = Iexco or null* Oint = yes Ocor = no Osig = Io or null** Oexc = Iexco Oint = yes Ocor = no

* **

Depending on Isig. Depending on Iexc.

164 165 166 167 168 169 170

the unique Iexco. The other function is to re the stored information accordingly, i.e., an Iexco can excite the stored Io, and an Io can excite the stored Iexco. The cognitive level above the single memory is called memory triangle, which consists of three single memories, as shown in Fig. 2. In a triangle, the Oint of a single memory is the Iint of next single memory. As a result, all

three single memories form a loop with the interlock signals. The purpose of a triangle is to learn a unique data point (Io) three times, based on an assumption that, if a data point can represent a common feature of external stimuli, it should appear in all external stimuli. After a memory triangle has stored the common data point for three times into each of its single memories in the order from 1 through 3, a common feature is considered learned and generalized because of the existence of the loop. A subsystem with several interlocked memory triangles (e.g., the MTr1 and MTr2 in Fig. 1) is constructed to learn more common features. Since a memory triangle is the base for a common feature, knowledge is locally stored. The localized characteristic has the advantage of self-assembling a logically interrelated structure among acquired knowledge. A concept is an abstract idea or a common feature, and a word is the symbol for a concept. A conceptual knowledge is considered learned only when a symbol is associated to a common feature. This association is realized by pairing the two subsystems into a semantic system. Such pairing is inspired by the nding of split-brain (Myers & Sperry, 1953; Sperry, 1982) that indicates each brain half appears to have its own learning processes and its own separate chain of memories. In his Nobel lecture, Sperry (1982) further noted that our left hemisphere is capable of comprehending printed and spoken word, and our right hemisphere is word-deaf and word-blind, but capable of comprehending spatial and imagistic information. Based on these characteristics, the symbol subsystem (dedicated for symbolic learning) and representation subsystem (ded-

PR

OO

171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

COGSYS 274 1 September 2008 Disk Used 4

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx Table 2 Sequenced input pairs

No. of Pages 11, Model 5+

201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244

icated for common feature learning) can be correspondent to our left and right hemispheres, respectively. And, the subsystem pair can be correspondent to the neocortex since both have the functions of the learning and storage of semantic knowledge. The underlying mechanism of subsystem pairing is done via the unique excitation signals generated by the bundle of inter-subsystem signals whenever learning occurs. As mentioned earlier, the bundle generates a unique excitation signal, Iexco, only when it receives a coordination signal (Ocor) from each subsystem, and the Iexco is stored together with Io in the two single memories that red the Ocor in both subsystems (see also Table 1). Thus, a known common feature can excite a stored Iexco in the representation subsystem, which can excite the stored symbol in the symbol subsystem. Similarly, a known symbol can excite a stored Iexco in the symbol subsystem, which can excite the stored common feature in the representation subsystem. According to Immanuel Kant, a concept is a common feature or characteristic, and concepts are abstracts in that they omit the dierences of the things in their extension, treating them as if they were identical. In order for the AI counter to be able to tally, it has to learn three concepts that are associated with three common features. The three features are about the meaning of zero, one and tally, and are carried by those eight pairs of examples in either Table 2 or Table 3. For example, the 4th example, in Table 2, is an input pair of a symbol I, and a singlepeaked representation. This example tells the learning system that the symbol I is the sign for a one-peaked representation. Further, the 6th example is an input pair of the symbol I, and another single-peaked representation. However, this single peak looks dierent from that of the 4th representation input. Thus, this example tells the system that the symbol I is also the sign for a dierent onepeaked representation. The learning system has to omit the dierences of the things in their extension and to abstract what is in common from given examples. In here, the dierences are the widths of dierent single-peaks, while the common feature is one peak. In Table 2, the input pairs of example 1 through 3 are about the concept of zero or nothing, whose representation input is a at line with no peak in it. The purpose of

CO

RR E

CT
the three examples is to teach the concept that a at line is a background and it is like the surface of a desk with no object on it. The pairs of example 4 through 6 are about the concept of one, and whose presentation input has a single peak. The pairs of example 6 through 8 are about the pattern of increment in both peaks and symbols. After the system has learned these three concepts, it can recognize external stimuli and count them, even if the stimuli have never been experienced before, as long as they are consistent with the concepts that have been learned. When receiving a symbol input, the system answers it with a representation of the symbol; when receiving a representation input, it responds with the symbol of the representation. The former operation is equivalent to object drawing and the latter to object naming that are terms of typical tasks for semantic memory.

ED

Table 3 Non-sequenced input pairs

PR

OO

Fig. 2. A memory triangle that is made up of three single memories (after Zhang, 2005). All signals are correspondent to those in Table 1.

245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

COGSYS 274 1 September 2008 Disk Used

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx

No. of Pages 11, Model 5+

5 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370

261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315

2.2. An AI dreamer

In humans and almost all mammals, there is such a memory system, namely the hippocampal complex, which has been revealed to have the function of immediate storage (e.g., McClelland, McNaughton, & OReilly, 1995; Squire, 1992; Tulving, 1972). When the complex is damaged, the capacity of retaining episodic memory is impaired (Scoville & Milner, 1957) and patients become amnesic (e.g., Squire & Alvarez, 1995). Deterioration of this complex has been commonly found in patients with Alzheimers disease for whom poor retention of episodic memory is the rst sign. Based on the function of the biological hippocampal complex, a hippocampal memory subsystem is added to the learning system of AI counter. As shown in Fig. 1, the subsystem has a number of memory cells, and each cell is designated to store a segment of an external experience. All of the cells are interlocked so that the sequenced segments of an external experience are stored in their original order of occurrence. The interlocking mechanism has similar functions as the hippocampus in terms of sequential learning (e.g., Levy, 1996) when the variable is time, and spatial navigation (e.g., Muller, Kubie, & Ranck, 1987) when the variable is distance. Thereby, all stored segments of an external experience can be re-accessed and replayed in their originally occurring sequences (to be demonstrated). The segments can also be red randomly, which becomes the source of random signals in dream simulations of this study.

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

CO

RR E

CT

ED

However, the learning rate of the system is dependant of training examples. It only needs one round of training with the examples given in Table 2, which has been demonstrated in the previous study. However, it has to be trained for six rounds with the examples given in Table 3, or for eight rounds if examples in Table 2 were in reverse order. Both Tables 2 and 3 contain the same eight examples. The only dierence is that the examples in Table 2 are listed in sequence for the meanings of zero, one and tally, and are listed from simple to complex. Comparably, the examples in Table 3 are not listed in the sequence from simple to complex. Based on the proposed rule of learning (Zhang, 2005), new learning has to rely on prior learning if the newer knowledge is an extension of the prior knowledge. Thus, the learning system cannot learn extended knowledge if its base knowledge has not already been learned, which is the reason why it learns slower with the examples in Table 3. When it cannot learn from an example, it simply let the external experience pass through like nothing has happened. Therefore, it has to experience those examples many more times before learning is nally realized. What if the learning system had a subsystem that could store external experience as its episodic memory for later reprocessing? Comparably, humans, and most mammals, do have the capability to remember rapidly what has happened and to retain this as episodic memory after a single experience.

The system, shown in Fig. 1, is named AI dreamer, which is made up of two cognitive sections. One section is the AI counter of the previous study, which is also called the learning center in later discussions for convenience. And, the other is the hippocampal memory that stores its experience and allows the experience to be re-accessed. Here, the conceptual knowledge and stored experience are compared with the concepts of semantic and episodic memories dened by Tulving (1972), respectively. This two-sectioned construct also agrees with an understanding that the brain has a complementary memory system of knowledge and experience, each of which uses dierent computational strategies for storing information (e.g., McClelland, McNaughton, & OReilly, 1995; OKeefe & Nadel, 1978). Besides the added hippocampal memory, a mode selector is also added that can set the system either to waking mode in order to receive external stimuli, or set the system to dreaming mode to process internal signals that are red randomly from the hippocampal memory. Similar as for the learning system of the AI counter, the cognitive system (namely AI dreamer) shown in Fig. 1 is also built in the LabView language (e.g., Bishop, 2001), although the same cognitive construct can be built in any commonly used computer language. 3. Simulations of the AI dreamer The external experience for the system to learn is a series of input pairs given in Table 3. Each pair of the input consists of a symbol and its representation. When the system learns, it takes in an input pair via separated channels, Sin and Rin, shown in Fig. 1. The representation input carries the meaning of the symbol input, while the symbol is a name that symbolizes the representation. The term meaning in here means that the input carries the conceptual knowledge that can be perceived and understood in a real world. We can see from Table 3, that a representation input may be a at line or may come with a number of peaks. Every peak represents one object, while a at line means no object. When we see an input has three peaks, the meaning of three peaks is to be perceived by the learning center. Those input pairs in either Tables 2 or 3 carry the same knowledge (i.e., zero, one and tally) necessary for the system to learn how to tally. The only dierence between the two tables is in the given sequence of the input pairs. Firstly, in training phase, the system is set to waking mode to receive all given pairs in Table 3 in their given sequence. This system only receives those pairs once, but it must remember the entire experience thereafter. This is comparable with the fact that human can remember what has happened after a single experience, and this property of rapid remembering is demonstrated through the serial recall (that is also a standard test in memory study) shown in Table 4. When an input pair of symbol and representation arrives during the training, the symbol is forwarded to

PR

OO

COGSYS 274 1 September 2008 Disk Used 6

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx

No. of Pages 11, Model 5+

Table 4 Recorded streams of rings from the hippocampal memory

371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403

the symbol subsystem, while the representation is forwarded to the representation subsystem. In turn, the inputs are delivered to the memory triangles and single memory at lower cognitive levels. The learning center may learn from, react to, or reject the input pair. What the learning center eventually does is a result of interactions among all its enclosed single memories, each of which it also has the potential of learning from, reacting to or rejecting. If the center cannot learn from an input pair, it rejects and forwards the pair to the hippocampal memory via the two pathways as indicated in Fig. 1. The pair is then stored in one of the memory cells as a part of episodic memory. After the training phase, the learning center is only able to learn from the 6th pair, and it has to store the rest of the pairs into the hippocampal memory. This is because the 6th pair is the starting point of the most basic knowledge of tally a at baseline that means there is no object. Learning from the 6th pair is only the beginning to learn the baseline. Thereafter, the learning center has to learn from the 5th pair and then the 4th pair before it can fully generalize the meaning of a baseline (i.e., the meaning of zero). Only when it knows what is a baseline, it is able to subtract one object (i.e., one peak) from the baselines of the 3rd, 8th, and 2nd pairs, and to learn the property of the subtracted one object. After it has learned the meaning of one, it then can subtract more peaks and abstract the meaning of tally in subsequent learnings of the 7th and 1st pairs. This progressive learning is the result of the mechanism proposed in the previous study that new learning has to rely on prior learning when the newer knowledge is an extension of the prior knowledge. All these subsequent learning steps, however, does not occur in the training phase, and only occur in the dream phase in which the

ED

stored examples are randomly red and processed by the learning center. After the initial external experience, the system is set to dream mode to allow the learning center to reprocess those stored experiences red from the hippocampal memory. During dreaming, the learning center acts in the same way as it is awake. It learns from an input pair if the inputs prior knowledge has already been learned; it responds to an input if the meaning of the input has already been learned; and it rejects an input pair if the pairs prior knowledge has not been learned. During dreaming, the stored segments of experience are randomly red, which is intended to simulate the random nature of dreaming that has been concluded by a number of dream studies. The randomness has strongly challenged the view that dreaming has any intellectual function. For example, dreaming is thought only a neural process by which our brain produces dream imagery from noisy signals (Hobson, 1988; Hobson & McCarley, 1977). The central point of the challenge is based on the assumption that randomly assembled information hardly contributes to cognition and knowledge. But, the assumption may not be valid. From Russells viewpoint (1913), knowledge can be learned through experience of a direct causal (experience-based) interaction between a person and the object that the person is perceiving. On the other hand, it has also been argued that true knowledge (or at least the most important knowledge) is essentially independent of sensory experience (Locke, 1690). If both statements are true, it may be said that knowledge can be acquired through experience, but the acquisition process is not necessarily dependent on an exact experience. Random ring may be sucient enough to provide an

PR

OO

404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

CO

RR E

CT

COGSYS 274 1 September 2008 Disk Used

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx

No. of Pages 11, Model 5+

7 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547

437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490

4. Discussion

4.1. Phenomenal and neurobiological properties of dreaming As we go to sleep, we slowly sink down into deeper stages of sleep (i.e., Non-REM or NREM sleep). After an hour or two, the rst REM period begins and lasts a few minutes. Then, we sink back into NREM sleep again. This cycle occurs about every 90 min. Towards the end of the night, the REM periods get longer. NREM sleep alternates with REM sleep and includes all sleep apart from REM. Each NREM sleep has four stages corresponding to increasing depth of sleep as indicated by progressive

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

CO

alternative to the exact replay of an actual experience in knowledge acquisition. Table 4 lists some recorded streams of representation ows that are intercepted along the pathway from the mode selector to the representation subsystem. The rst simulation is a serial recall recorded in waking mode. If we simply attach all representation signals, given in Table 3 except the 6th input, into a long representation signal, we will see that the long representation signal is same as the rst recorded stream in Table 4. For example, one may notice that the combination of the rst three peaks in this serial recall is the rst representation input in Table 3, and the last peak is the eighth representation input. Other streams are recorded in six dierent dream simulations of randomly activated experiences. Again, each stream is made up of activated experiences of the eight examples in Table 3. The only dierence is that the experiences are randomly activated. The streams are compressed in order to ll in the limited space in Table 4, thus the coordination between the streams and the eight examples are not clearly shown in Table 4. But some of them may still be seen, e.g., the rst two peaks in the last stream are the two peaks of the seventh example in Table 3. The results of learning, after the learning center has experienced the recorded dream streams, have also been indicated in the same table. As noted before, the input pairs in Table 3 carry all conceptual knowledge (including zero, one and tally) that are sucient for the system to perform tallying after full learning. These concepts may be partially or fully learned in dreams under a given number of total rings. The eciency of learning in a specic dream varies depending on how soon newer knowledge arrives after its required prior knowledge has been learned. The learning center may learn all of the conceptual knowledge in one dream (e.g., the rst and second dreams in Table 4) or in many dreams (e.g., the series of four dreams in Table 4). Furthermore, with additional cognitive structure, this system can be extended to learn other counting systems, e.g., Roman numerals or Arabic numerals. The latter has been demonstrated in the previous study (Zhang, 2005). However, since the present study is centered on semantic learning from stored experiences, the extension is not discussed.

dominance of the electroencephalography (EEG) by highvoltage, low frequency wave activity (Rechtschaen & Kales, 1968). This low frequency wave is also called slow wave, which dominates the deepest stages of NREM. Almost all mammals have the NREMREM cyclic alternation in sleep, which suggests not only a shared mechanism across species but also a universal functional signicance (Hobson et al., 2000). Dreams can occur in both REM and NREM sleep. In 7095% of awakenings from the REM state, normal subjects report that they have been dreaming whereas only 510% of NREM awakenings produce equivalent reports (Hobson, 1988). If there is no clear dream reported, 43 50% of NREM awakenings can elicit reports of thoughtlike or mentation recall (Foulkes, 1962; Nielsen, 2000). If the mentation recall can also be accounted as the result of dreaming, it means the mind is mostly in dream states during sleep. There are more subtle dierences between REM and NREM dreams. Reports from REM sleep awakenings are relatively longer, more perceptually vivid, more motorically animated, more emotionally charged, and less related to waking life than NREM reports (e.g., Antrobus, 1987; Foulkes, 1962). In contrast to REM reports, NREM reports are more thought-like and contain representations of current concerns more often than do REM sleep reports (Foulkes, 1962). It has been noted earlier in the simulation of dreams that, in dream mode, the learning center is as active as it is in awaking mode. A similar characteristic has been concluded from neurobiological studies. It was stated (Hobson et al., 2000) that, in REM sleep, the brain is almost as active as when awake, except for the sensory input and motor output, which are blocked, and for certain areas of the dorsolateral prefrontal cortex and the primary visual centers which are selectively deactivated. A number of functional imaging studies in humans have revealed a fairly consistent pattern of activities in REM sleep that appear to reect dream processes (Braun et al., 1997; Maquet et al., 1996; Nofzinger, Mintun, Wiseman, Kupfer, & Moore, 1997). Some important ndings from the studies are: in REM sleep, the brainstem reticular formation is highly active; primary sensory areas (e.g., striate cortex for the visual system) are inactive; by contrast, extrastriate (visual) regions (as well as other sensory association sites) are very active; limbic and paralimbic regions (including the hippocampus, amygdala and anterior cingulate) are intensely activated; and widespread regions of the frontal cortex including the lateral orbital and dorsolateral prefrontal cortices show marked reductions in activity. These studies also show that the following areas are relatively less active in NREM than in REM sleep; they are the brainstem, midbrain, anterior hypothalamus, hippocampus, caudate, and medial prefrontal, caudal orbital, anterior cingulate, parahippocampal and inferior temporal cortices (Braun et al., 1997). Besides the fact that the general neocortex in dream sleep is almost as active as when awake, the highly acti-

RR E

CT

ED

PR

OO

COGSYS 274 1 September 2008 Disk Used 8 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx

No. of Pages 11, Model 5+

CO

vated limbic system, especially the hippocampus, is of special interest. The activated hippocampus coincides with the reactivation of recent waking patterns of neuronal activity within the hippocampus. The reactivation has been revealed in human and other mammals, and in REM and NREM sleeps, while the patterns of neuronal activity are generally thought to be associated with daily experience or episodic memory (Louie & Wilson, 2001; Pavlides & Winson, 1989; Poe et al., 2000; Staba et al., 2002; Wilson & McNaughton, 1994). What could be the common functioning areas in both REM and NREM sleep? Functional imaging studies are able to indicate the activity intensities of various brain areas, but are not adequate to rule out the involvements of the less active areas. For example, bursts of rings have been recorded from neurons in the hippocampus during NREM sleep (Wilson & McNaughton, 1994), although brain imaging has indicated that this area is less active in NREM sleep. On the other hand, the ndings that patients with large pontine lesions can still dream suggest that the brainstem is not the critical component for all dreams to occur, but rather it may only be important for REM dreaming (Solms, 1997). This study of pontine lesions is consistent with the nding from functional brain imaging that the pontine brainstem and anterior cingulate cortex is deactivated during NREM sleep (Braun et al., 1997). If only considering those activated areas in both REM and NREM sleep, it may be concluded that the commonly activated areas necessary for dreams to occur, at least, include: most of the general neocortex and of the limbic system (especially the hippocampus). This simplied activation pattern is compatible with the cognitive structure of the computational dreamer, which has a learning center (mimicking the conceptual learning function of the general neocortex) and a hippocampal memory (mimicking the storage and reactivation functions of episodic memory in the hippocampus). However, it must be pointed out that this system is only a simplied approach to explain some common properties of dreams in both REM and NREM sleep. By no means, is it thought to be able to enclose all dream properties of a brain.

RR E

present, dreams often involve weak, post-hoc, and logically awed explanations of improbable or impossible plots; and (3) dreams lack orientational stability; persons, times, and places are fused, plastic, incongruous and discontinuous. In Hobsons view (e.g., 2000), the bizarreness or impossible plots in dreams are simply resulted from the randomness. The second one is the repetitive characteristic of dreams repeated themes and repeated dreams (including recurrent dreams) are common among dreams (Domho, 1993). The proposed mechanism of conceptual learning states that new knowledge cannot be learned if its prior knowledge is not already learned. As a result, a segment of stored experience may by chance have to be red multiple times before the learning center is able to learn from it. This multi-appearance of the same segment of stored experience can be commonly observed in those dream-source streams shown in Table 4. When the learning center is not able to fully abstract and generalize the knowledge from the random signals in one dream, more dreams become necessary before dream learning is thoroughly realized. That is the scenario of the multi-dream learning, shown also in Table 4, in which repeated themes and similar dreams are expected. The lack of self-reection and self-control are also the typical characteristics of dreaming. For example, selfreection in dreams is generally found to be absent (Rechtschaen, 1978) or greatly reduced (Bradley, Hollield, & Foulkes, 1992); the dreamer rarely considers the possibility of actually controlling the ow of dream events (Purcell, Mullington, Mott, Homan, & Pigea, 1986); volitional control is greatly attenuated in dreams (Hartmann, 1966). Similarly, this computational model of AI dreamer does not have a necessity of a central controller in regulating its learning process.

603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657

The characteristics of dreaming that are concluded from many empirical studies can be employed to examine the proposed theory of dreaming and its simulations. The rst is the characteristic of the randomness. This characteristic is concluded from or supported by empirical ndings conducted by, e.g., Hobson and McCarley (1977), McCarley and Homan (1981), Hobson (1988), Reinsel, Antrobus, and Wollman (1992), Williams, Merritt, Rittenhouse, and Hobson (1992), Revonsuo and Salmivalli (1995). These ndings have been summarized by Hobson et al. (2000) as followings: (1) dream imagery can change rapidly, and is often bizarre in nature; (2) relative to waking and, when

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

4.2. Psychophysiological characteristics of dreaming and dream mechanism

CT

ED

4.3. The dream function of memory consolidation and learning The dream function of memory consolidation has long been suggested in various studies. Newer and relatively direct evidence comes from the neuronal activity of seemingly replay of waking-experience in the hippocampus in both REM and NREM sleep. Many researchers agree that such reactivation is experience dependent, and conclude that the reactivation is associated with the consolidation of episodic memory stored in the hippocampus (Pavlides & Winson, 1989; Poe et al., 2000; Skaggs & McNaughton, 1996; Staba et al., 2002; Wilson & McNaughton, 1994). Others who studied amnesia and hippocampal lesion have also expected the consolidation function in dream sleep. For example, Squire and Alvarez (1995) have argued that if episodic memory is revived constantly, dreaming may be the best answer to explain why the memory consolidation process does not regularly intrude into our consciousness. Neuronal recordings have suggested that daily experiences stored in the hippocampal complex can be reacti-

PR

OO

COGSYS 274 1 September 2008 Disk Used

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx

No. of Pages 11, Model 5+

9 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752

658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695

Table 5 Naming and picture drawing tests

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

CO

RR E

vated or replayed during REM and NREM sleeps, but there are no indications about whether the replaying is sequential or random. Study of dream contents and their connections with daily experience may provide this detail. A recent study (Fosse et al., 2003) concluded that only 1.4% of dream reports contained the reproductions of experience ensembles; while 65% of dream elements are linked to fragments and features of waking events. This apparent contradiction between dreamings lack of a fully episodic structure and its proposed role of the consolidation, has led to the suggestion (Fosse et al., 2003; Stickgold, Hobson, Fosse, & Fosse, 2001) that sleep only has its role in the consolidations of associated semantic memory and procedural memory, not episodic memory. In interpreting the data, however, some researchers have realized and suggested that the key is to understand why memories appear in dreams as fragments or partial episodes but also occasionally as complete replays (Nielsen & Stenstrom, 2005; Schwartz, 2003). The understanding oered by this study is that, when random ring is the only fashion of the reactivation, daily experiences should only appear in the form of segments rather than the sequential replay of entire past experience. Memory consolidation is dened as the process by which memory becomes independent of the hippocampal region (Squire & Alvarez, 1995). The presented computational system can demonstrate that past experiences stored in the hippocampal memory have been consolidated into the learning center in tests called naming. Naming test is a standard test in memory studies in which the tested subject calls out the name of a presented item. During the naming tests, the hippocampal memory is removed after dream consolidation, in order to demonstrate that the memory has become independent of the hippocampal memory. The rst three rows in Table 5 are summaries of three naming tests. In these tests, dierent objects (or pictures) are shown and the system answers with the correct names to each of the given objects. When compared

with Table 3, one can see that the objects tested and the names answered are part of the inputs given in the table that have been experienced by the system previously. Since the tests are done after the hippocampal memory was removed, it indicates that, after dreaming, the episodic memory has been consolidated into the learning center and become free of the hippocampal memory. It may be noticed, from those examples in Table 5, that the memory consolidation in the computational model is not a simple relocation of stored experiences, rather a learning process that gradually incorporates facts and events into an already existing framework of knowledge (e.g., McClelland, McNaughton, & OReilly, 1995). The learning by consolidation is the natural outcome of the computational system that coincides with the understanding that mechanisms of the hippocampus and neocortex involved in the consolidation process, are dierent (e.g., OKeefe & Nadel, 1978) however complementary (e.g., McClelland, McNaughton, & OReilly, 1995). In other words, learning and memory consolidation can be seen as dierent sides of the same coin. The function of dream learning is the most popular dream function that has been observed and concluded. The impact of dream learning is broad and can be often observed after dreams. For example, unprepared learning, more slowly mastered and dicult, is especially dependent on the quality of REM sleep (Greenberg & Pearlman, 1974); learning tasks requiring signicant concentration or acquisition of unfamiliar skills are followed by increased REM sleep (LaBerge, 1985); material learned during the day and consolidated over a night of sleep is recalled better the next morning (Hennevin et al., 1995). The dream learning function can also be demonstrated with the presented system through both naming tests and picture drawing tests (also one of the standard memory tests) after the removal of the hippocampal memory. The last ve rows in Table 5 are summaries of two naming tests and three drawing tests. One can see that the inputs in four out of the ve tests are new to the system because they were not included in Table 3. In other words, the system has never had the chance of knowing them. The system would have made mistakes or made no response to the inputs if memory consolidation was a simple relocation of experience. Instead, it names those presented pictures without mistake, and counts the peaks of the representation inputs, and draws pictures correctly by tallying the symbol and drawing the peaks out accordingly. Although those four inputs seem new to the system, they are not new at all in meaning, or in the sense of common features and conceptual knowledge. The same common features (and conceptual knowledge) have already been carried by those input pairs in Table 3 and have already been abstracted and generalized into the learning center during the systems dreams. This demonstrated property of acquired knowledge is called exibility, which is considered one of the key characteristics of semantic memory that is associated with the general neocortex.

CT

ED

PR

OO

COGSYS 274 1 September 2008 Disk Used 10 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx

No. of Pages 11, Model 5+

5. Conclusion Few computational models (Christos, 2003; Crick & Mitchison, 1983) have been reported in exploring the underlying mechanisms and functions of dreaming. And, the conclusions from these models are generally not associated with what have been revealed in empirical studies of dreams (Domho, 1996). Memory consolidation has been simulated in a number of studies based on connectionist modeling (Alvarez & Squire, 1994; McClelland et al., 1995; Murre, 1996) to understand the underlying mechanisms of memory consolidation and retrograde amnesia. In these studies, the consolidation processes were not directly associated with dreaming process and its characteristics, although the authors suggested a connection between dreaming and memory consolidation. Due to the nature of parallel distributed processing of connectionist modeling, one neuron in one subsystem may have to connect with many neurons in a dierent subsystem, thus a single stream of information ow that is compatible with dreams would not exist during the simulations of consolidation process. Furthermore, these studies were unable to demonstrate the exibility of the acquired knowledge from memory consolidation, whereas exibility is one of the key characteristics of semantic memory that is generally considered as a result of memory consolidation. The presented cognitive model of dreaming conrms computationally that learning and memory consolidation of episodic memory can be realized with randomly arriving signals that are stored segments of experience or episodic memory. This model oers a positive answer to Freuds inquiry (1900): Is the dream capable of teaching us something new concerning our internal psychic processes and can its content correct opinions which we have held during the day? It also demonstrates that dreaming is a cognitive process that deals with the daily residual (Freud, 1900), unnished business (Hall & Van De Castle, 1966) and correspondences with waking thoughts (Domho, 1996; Foulkes, 1985). Although this system is only able to demonstrate the dream functions of learning and memory consolidation, there is no doubt that dreaming functions should cover many other cognitive aspects and psychological properties as many studies have concluded. References

795 796 797 798 799 800 801 802 803 804 805 806 807

Antrobus, J. (1987). Cortical hemisphere asymmetry and sleep mentation. Physiological Review, 94, 359368. Alvarez, P., & Squire, L. (1994). Memory consolidation and the medial temporal lobe: A simple network model. Proceedings of the National Academy of Sciences of USA, 91, 70417045. Battaglia1, F. P., Sutherland, G. R., & McNaughton, B. L. (2004). Hippocampal sharp wave bursts coincide with neocortical up-state transitions. Learning and Memory, 11, 697704. Bishop, R. H. (2001). Learning with LabVIEW 6i. New York: Addison Wesley. Bloch, V., Hennevinm, E., & Leconte, P. (1979). Relationship between paradoxical sleep and memory processes. In M. A. Brazier (Ed.), Brain

mechanisms in memory and learning: From the single neuron to man. New York: Raven Press (pp. 329343). Botman, H. I., & Crovitz, H. F. (1989). Dream reports and autobiographical memory. Imagination. Cognition and Personality, 9, 213214. Bradley, L., Hollield, M., & Foulkes, D. (1992). Reection during REM dreaming. Dreaming, 2, 161166. Braun, A. R., Balkin, T. J., Wesensten, N. J., Carson, R. E., Varga, M., Baldwin, P., et al. (1997). Regional cerebral blood ow throughout the sleepwake cycle. Brain, 120, 11731197. Christos, G. (2003). Memory and dreams: The creative human mind. New Jersey: Rutgers University Press. Crick, F., & Mitchison, G. (1983). The function of dream sleep. Nature, 304, 111115. Domho, G. W. (1996). Finding meaning in dreams: A quantitative approach. New York: Plenum. Domho, G. W. (1993). The repetition of dreams and dream elements. In A. Mott, M. Kramer, & R. Homann (Eds.), The functions of dreaming up (pp. 293320). Albany: SUNY Press. Fishbein, W. (1970). Interference with conversion of memory from shortterm to long-term storage by partial sleep deprivation. Communications in Behavioral Biology, 5, 171175. Fosse, M. J., Fosse, R., Hobson, J. A., & Stickgold, R. J. (2003). Dreaming and episodic memory: A functional dissociation? Journal of Cognition Neuroscience, 15, 19. Foulkes, D. (1985). Dreaming: A cognitive-psychological analysis. Hillsdale: Erlbaum. Foulkes, D. (1962). Dream reports from dierent states of sleep. Journal of Abnormal and Social Psychology, 65, 1425. Freud, S. (1900). In J. Strachey (Ed.), The interpretation of dreams. New York: Basic Books. Greenberg, R., & Pearlman, C. (1974). Cutting the REM nerve: An approach to the adaptive role of REM sleep. Perspectives in Biology and Medicine, 17, 513521. Hall, C. S. (1953). A cognitive theory of dream symbols. Journal of General Psychology, 48, 169186. Hall, C. S., & Van de Castle, R. L. (1966). The content analysis of dreams. New York: Appleton-Century-Crofts. Hartmann, E. (1966). The psychophysiology of free will. In E. Lowenstein (Ed.), Psychoanalysis: A general psychology (pp. 521536). New York: International University Press. Hennevin, E., Hars, B., Maho, C., & Bloch, V. (1995). Processing of learned information in paradoxical sleep: Relevance for memory. Behavioral Brain Research, 69, 125135. Hobson, J. A. (1988). The dreaming brain. New York: Basic Books. Hobson, J. A., & McCarley, R. W. (1977). The brain as a dream-state generator: An activation-synthesis hypothesis of the dream process. American Journal of Psychiatry, 134, 13351348. Hobson, J. A., Pace-Schott, E., & Stickgold, R. (2000). Dreaming and the brain: Towards a cognitive neuroscience of conscious states. Behavioral and Brain Sciences, 23, 793842. LaBerge, S. (1985). Lucid dreaming. New York: Ballantine. Levy, W. B. (1996). A sequence predicting CA3 is a exible associator that learns and uses context to solve hippocampal-like tasks. Hippocampus, 6, 579590. Locke, J. (1690). An essay on human understanding. London: Penguin Books, Republished in 1997. Louie, K., & Wilson, M. A. (2001). Temporally structured replay of awake hippocampal ensemble activity during rapid eye movement sleep. Neuron, 29, 145156. Maquet, P., Peters, J. M., Aerts, J., Delore, G., Degueldre, C., Luxen, A., et al. (1996). Functional neuroanatomy of human rapid-eye-movement sleep and dreaming. Nature, 383, 163166. McClelland, J. L., McNaughton, B. L., & OReilly, R. C. (1995). Why there are complementary learning systems in the hippocampus and neocortex: Insights from the successes and failures of connectionist models of learning and memory. Psychological Review, 102, 419457.

808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

CO

RR E

CT

ED

PR

OO

COGSYS 274 1 September 2008 Disk Used

ARTICLE IN PRESS
Q. Zhang / Cognitive Systems Research xxx (2008) xxxxxx

No. of Pages 11, Model 5+

11 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959

Please cite this article in press as: Zhang, Q., A computational account of dreaming: Learning ..., Cognitive Systems Research (2008), doi:10.1016/j.cogsys.2008.06.002

UN

CO

RR E

876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917

McCarley, R. W., & Homan, E. (1981). REM sleep dreams and the activation-synthesis hypothesis. American Journal of Psychiatry, 138, 904912. Muller, R. U., Kubie, J. L., & Ranck, J. B. (1987). Spatial ring patterns of hippocampal complex spike cells in a xed environment. Journal of Neuroscience, 7, 19351950. Murre, J. M. (1996). TraceLink: A model of amnesia and consolidation of memory. Hippocampus, 6, 675684. Myers, R. E., & Sperry, R. W. (1953). Interocular transfer of a visual forma discrimination habit in cats after section of the optic chiasm and corpus callosum. Anatomical Record, 115, 351352. Nielsen, T. A. (2000). A review of mentation in REM and NREM sleep: Covert REM sleep as a possible reconciliation of two models. Behavioral and Brain Sciences, 23, 851866. Nielsen, T. A., & Stenstrom, P. (2005). What are the memory sources of dreaming? Nature, 437, 12861289. Nofzinger, E. A., Mintun, M. A., Wiseman, M. B., Kupfer, D. J., & Moore, R. Y. (1997). Forebrain activation in REM sleep: An FDG PET study. Brain Research, 770, 192201. OKeefe, J., & Nadel, L. (1978). The hippocampus as a cognitive map. Oxford: The Clarendon Press. Pavlides, C., & Winson, J. (1989). Inuences of hippocampal place cell ring in the awake state on the activity of these cells during subsequent sleep episodes. Journal of Neuroscience, 9, 29072918. Pearlman, C. (1971). Latent learning impaired by REM sleep deprivation. Psychonomic Science, 25, 135136. Poe, G. R., Nitz, D. A., McNaughton, B. L., & Barnes, C. A. (2000). Experience dependent phase-reversal of hippocampal neuron ring during REM sleep. Brain Research, 855, 176180. Purcell, S., Mullington, J., Mott, A., Homan, R., & Pigea, R. (1986). Dream self-reectiveness as a learned cognitive skill. Sleep, 9, 423437. Rechtschaen, A. (1978). The single-mindedness and isolation of dreams. Sleep, 1, 97109. Rechtschaen, A., & Kales, A. (1968). A manual of standardized terminology, techniques and scoring system for sleep stages of human subjects. Washington: Public Health Service. Reinsel, R., Antrobus, J., & Wollman, M. (1992). Bizarreness in dreams and waking fantasy. In J. S. Antrobus & M. Bertini (Eds.), The neuropsychology of sleep and dreaming (pp. 157184). Hillsdale: Lawrence Erlbaum Associates. Revonsuo, A., & Salmivalli, C. (1995). A content analysis of bizarre elements in dreams. Dreaming, 5, 169187.

Russell, B. (1913). Theory of knowledge: The 1913 manuscript. London: Allen and Unwin, Republished in 1984. Schwartz, S. (2003). Are life episodes replayed during dreaming? Trends in Cognitive Sciences, 7, 325327. Scoville, W. B., & Milner, B. (1957). Loss of recent memory after bilateral hippocampal lesions. Journal of Neurology. Neurosurgery and Psychiatry, 20, 1121. Skaggs, W. E., & McNaughton, B. L. (1996). Replay of neuronal ring sequences in rat hippocampus during sleep following spatial experience. Science, 271, 18701873. Smith, C. (1995). Sleep states and memory processes. Behavioral Brain Research, 69, 137145. Solms, M. (1997). The neuropsychology of dreams. Mahwah: Lawrence Erlbaum. Sperry, R. (1982). Some eects of disconnecting the cerebral hemispheres. Science, 217, 12231226. Squire, L. R. (1992). Memory and the hippocampus: A synthesis from ndings with rats, monkeys, and humans. Psychological Review, 99, 195231. Squire, L. R., & Alvarez, P. (1995). Retrograde amnesia and memory consolidation: A neurobiological perspective. Current Opinion in Neurobiology, 5, 169177. Staba, R. J., Wilson, C. L., Fried, I., & Engel, J. J. (2002). Single neuron burst ring in the human hippocampus during sleep. Hippocampus, 12, 724734. Stickgold, R., Hobson, J. A., Fosse, R., & Fosse, M. (2001). Sleep, learning, and dreams: O-line memory reprocessing. Science, 294, 10521057. Tulving, E. (1972). Episodic and semantic memory. In E. Tulving & W. Donaldson (Eds.), Organization of memory (pp. 381403). New York: Academic Press. Williams, J., Merritt, J., Rittenhouse, C., & Hobson, J. A. (1992). Bizarreness in dreams and fantasies: Implications for the activationsynthesis hypothesis. Consciousness and Cognition, 1, 172185. Wilson, M. A., & McNaughton, B. L. (1994). Reactivation of hippocampal ensemble memories during sleep. Science, 265, 676679. Winson, J. (1985). Brain and psyche: The biology of the unconscious. New York: Doubleday/Anchor Press. Wolf, F. A. (1994). The dreaming Universe. New York: Simon & Schuster. Zhang, Q. (2005). An articial intelligent counter. Cognitive Systems Research, 6, 320332.

CT

ED

PR

OO