Download Attention: Some Theoretical Considerations and more Study Guides, Projects, Research Psychology in PDF only on Docsity!
Psychological Review 1963, 70, 80-
ATTENTION: SOME THEORETICAL CONSIDERATIONS^1
J. A. D EUTSCH AND D. D EUTSCH
Stanford University
The selection of wanted from unwanted messages requires discriminatory mechanisms of as great a complexity as those in normal perception, as is indicated by behavioral evidence. The results of neurophysiology experiments on selective attention are compatible with this supposition. This presents a difficulty for Filter theory. Another mechanism is proposed, which assumes the existence of a shifting reference standard, which takes up the level of the most important arriving signal. The way such importance is determined in the system is further described. Neurophysiological evi- dence relative to this postulation is discussed.
(^1) This work was wholly supported by Grant M-4563 from the National Institute of Mental Health, National Institutes of Health, United States Public Health Service and Grant G 21376 from the National Science Foundation. We are grateful to D. E. Broadbent, K. L. Chow, D. A. Hamburg, and F. Morrell for helpful comment and discussion.
There has, in the last few years, been an increase in the amount of research devoted to the problem of attention, which has been sum- marized in Broadbent’s (1958) important work. Whilst psychologists have been investigating the behavioral aspects of attention, suggestive evidence has also been found by neurophysiol- ogists. We feel that it would be useful at this time to consider the theoretical implications of some of this research. Our paper is divided into three parts. In the first we consider some of the behavioral find- ings on attention. In the second a system is pro- posed to account for various features of this behavior. Although we do not consider it nec- essary to identify a system of this type with par- ticular neural structures (see Deutsch, 1960) since a machine embodying such a system would also display the behavior we wish to explain, we do, however, venture some tenta- tive hypotheses concerning the neural identifi- cation of the proposed system.
BEHAVIORAL CONSIDERATIONS
However alert or responsive we may be, there is a limit to the number of things to which we can attend at any one time. We cannot, for instance, listen effectively to the conversation of a friend on the telephone if someone else in the room is simultaneously giving us complex instructions as to what to say to him. And this difficulty in processing information from two different sources at the same time occurs even if no overt response is required. This phenom- enon of selective attention has been investigat- ed in a number of experiments. The most important of these deals with the processing of information emitted simultaneously by two separate sound sources (Broadbent, 1954; Cherry, 1953; Spieth, Curtis, & Webster, 1954). Two problems arise from the results of such experiments. The first is how different streams of information are kept distinct by the nervous system, and how a resultant babel is thereby avoided. The second is why only one of the messages (once it has been kept distinct and separate) is dealt with at any one time. A pro- posed solution to the first problem, based on experiments in which two messages were fed simultaneously one to each ear, was that the messages were kept distinct by proceeding 1
down separate channels (such as different neu- ral pathways). Nor was it difficult for Broadbent (1958) to extend such a notion to other cases. It had been shown in numerous experiments that we are enabled to listen to one of the two simultaneous speech sequences while ignoring the other, by selecting items for attention which have some feature or features in common, such as their frequency spectra (Egan, Carterette, & Thwing, 1954; Spieth et al., 1954) and their spatial localization (Hirsch, 1950; Poulton, 1953; Webster & Thomson, 1954). It was supposed that relatively simple mecha- nisms were responsible for segregation accord- ing to these categories, though the principles of their operation were not made clear. Broadbent’s (1958) answer to the second problem, of how one message is admitted to the exclusion of others, followed from the notions we have already considered. It was proposed that there was a filter which would select a mes- sage on the basis of characteristics toward which it had been biased and allow this mes- sage alone to proceed to the central analyzing mechanisms. In this way, messages with other characteristics would be excluded and so the total amount of discrimination which would have to be performed by the nervous system would be greatly decreased. Whole complex messages could be rejected on the sole basis of possessing some simple quality, and no further analysis of them would occur. However, it seems that selection of wanted from unwanted speech can be performed on the basis of highly complex characteristics. For instance, Peters (1954) found that if an unwant- ed message is similar in content to the wanted one, it produces more interference with the ade- quate reception of the latter than if it is dissim- ilar to it. This shows that the content of the two messages is analyzed prior to the acceptance of one and rejection of the other. Gray and Wedderburn (1960) have also found that when speech was delivered to subjects in both ears simultaneously, such that a meaningful sequence could be formed by choosing sylla-
bles or words alternately from each ear, the sub- jects reported back the meaningful sequence rather than the series of words or syllables pre- sented to one ear or the other. Treisman (1960) presented two messages, one to each ear, and subjects were asked to repeat what they heard on one ear. The messages were switched from one ear to the other in the middle and it was found that subjects tended to repeat words from the wrong ear just after the switch. “The high- er the transition probabilities in the passage the more likely they were to do this” (Treisman, 1960). Other evidence, indicating that complex discriminations would be required of the filter, has been produced by experiments concerning the selection of novel stimuli, for which func- tion Broadbent (1958) assumes the filter to be responsible. Sharpless and Jasper (1956), studying habituation to auditory stimuli in cats, found that habituation, both behavioral and EEG, was specific not only to the frequency of sound presented, but also to the pattern in which a combination of frequencies was pre- sented. Evidence for human subjects is pre- sented by Sokolov (1960) and Voronin and Sokolov (1960), who report that when habitua- tion has been established to a group of words similar in meaning but different in sound, then arousal occurred to words with a different meaning. Behavioral data on the arousal of curiosity in rats upon the presentation of novel visual patterns are reported by Thomson and Solomon (1954). Such evidence as the above would require us, on filter theory, to postulate an additional discriminative system below or at the level of the filter, perhaps as complex as that of the cen- tral mechanism, to which information was assumed to be filtered. Howarth and Ellis (1961) have presented an ingenious experimental argument to show that the same discriminatory mechanism functions in normal perception and when, on filter theory, the discrimination would have to be performed at the level of the filter. The case they put forward
2 DEUTSCH & DEUTSCH
which is the most important by seeing which signal has no other signal which exceeds it in the physical dimension by which “importance” is represented. But a small amount of reflection will suffice to show that such a system is very uneconomical. Each possible incoming signal must have a provision in the shape of numerous comparing mechanisms, through each of which it will be connected to all other possible signals. So that as the number of possible signals increases, the number of mechanisms to com- pare them all against each other will increase at an enormous rate. If the same comparing mechanisms are to be shared by pairs of signals then the time to reach a decision will increase out of all bounds. However, there is a simpler and more eco- nomical way to decide that one out of a group of entities is the largest. Suppose we collect a group of boys and we wish to decide which is the tallest. We can measure them individually against each other and then select the boy in whom this comparison procedure never yielded the answer “smaller.” This is like the system outlined above. The decision smaller will be made in this case when we lower a horizontal plane or ruler down on the heads of two boys. The boy whose head is touched by this instru- ment is declared to be larger and the other boy smaller. But such a procedure is cumbersome because there are many pairs of boys and we must scan through many records of individual boys before we can select the tallest. We could, of course, argue that a simpler solution would be to use an absolute measure of height, such as a ruler with feet and inches inscribed on it. But this procedure is not really simpler. Each boy must be compared against the ruler, and then the measurements themselves must be com- pared against each other in much the same way as the boys were to decide on the larger and smaller in each couple. If we are simply interested in finding the tallest boy, then an alternative procedure may be used. Suppose we collect our group below our board which is horizontal and travels light- ly up and down, and then ask all our group to
stand up below it. Then the boy whose head touches the board when the whole group is standing up will be the tallest boy in the group. If then we call him out, the board will sink until it meets the head of the next tallest individual. If we introduce some other boys into the group, then if there is a taller boy in this group the boards will be raised until it corresponds to his height. In such a system only the tallest indi- vidual will make contact with the board, and so he will himself have an immediate signal that he is the tallest boy. Now suppose that instead of boys, we have signals, not varying in height, but in some other dimension (which we may continue to call “height”) which corresponds to their impor- tance to the organism. Suppose that each signal as it arrives is capable of pushing up some “level” up to its own “height” (the height deter- mined by its importance), then the most impor- tant signal arriving at any particular time will determine this level, analogous to the horizon- tal board in our example. It will then be the case that any signals which arrive then or after and are of lesser importance and so of smaller height will be below this level. However, if the signal of greater height ceases to be present, then the level will sink to the height reflecting the importance of one of the other signals which is arriving. If we suppose that only signals whose height corresponds to the height of the level switch in further processes, such as motor out- put, memory storage, and whatever else it may be that leads to awareness, we have the outline of a system which will display the type of behavior we associate with attention. Only the most important signals coming in will be acted on or remembered. On the other hand, more important signals than those present at an immediately preceding time will be able to break in, for these will raise the height of the level and so displace the previously most important signals as the highest. So far we have omitted any discussion of the role of general arousal in selective attention. Without such arousal, usually (but not invari-
4 DEUTSCH & DEUTSCH
ATTENTION 5
ably, Bradley & Elkes, 1953; Gastaut, 1954) indicated by characteristic patterns on the elec- troencephalogram, awareness of and behavioral responsiveness to peripheral stimulation are absent. Some degree of general arousal is thus necessary for attention to operate. Furthermore, individuals when aroused will attend to any incoming message, provided that it is not concomitant with a more important one, whereas when asleep they will only respond to very “important” messages, such as a person’s own name (Oswald et al., 1960) or, in the case of a mother, the sound of her infant crying. And when drowsy, though responsive to a larger range of stimuli than when asleep, subjects will tend to “miss” signals which they would notice when fully awake. The system which takes this into considera- tion is schematically represented in the diagram below (Figure 1). Any given message will only be heeded if the horizontal line (Y) representing the degree of general arousal meets or crosses the vertical line, the height of which represents the “importance” of the message. Whether or not alerting will take place then depends both on the level of general arousal and on the importance of the message. Attention will not be paid to Message b though it is the most important of all incoming signals, when the level of general arousal is low (Position X). When the level of general arousal is at Z, which is very high, attention could be paid to all the signals a, b, c, d, and e. In fact, attention is paid
only to b as a result of the operation of the spe- cific alerting mechanism. Further, it is supposed that a message will increase the level of general arousal in propor- tion to its importance and for various lengths of time in proportion to its importance, so that messages which would not have been heeded before will command attention if they follow in the wake of a more important message. The mechanism whereby the weighting of importance of messages is carried out is given by Deutsch’s (1953, 1956, 1960) theory of learn- ing and motivation, and will be only briefly sum- marized here, since it is not the main point of the paper. It is assumed that on exposure to a suc- cession of stimuli, link-analyzer units responsive to these stimuli will be connected together. Certain primary links, when stimulated by phys- iological factors, generate excitation, and this is passed on from link to link along the connec- tions established by experience. Each link-ana- lyzer unit will receive excitation depending first, on the state of the primary links to which it is connected, either directly or indirectly, and sec- ond, on the “resistance” of such a connection, which is determined by past learning. It is assumed that the amount of such excitation arriving at a link-analyzer unit determines both its threshold of excitability by incoming stimuli (leading to an increased readiness to perceive a stimulus whether it is there or not) and the rank- ing of importance of such a stimulus (e.g., Lawrence 1949, 1950). We should predict from this theory an inverse correlation between the attention-getting or distracting value of a stimu- lus when attention is being paid to another, and its threshold (regarded as the likelihood of its being reported by a subject when he is asked to say what he perceives). We should also expect that stimuli which have a high importance weighting should more often be mistakenly per- ceived when similar stimuli are present.
N EUROPHYSIOLOGICAL CORRELATES
We may ask how the suggested system would fit what is known of the physiological
Figure 1. Diagram to illustrate operation of proposed system. (The interrupted horizontal line – 1 – represents the “level” of importance in the specific alerting system which is raised and lowered according to the incoming messages. The solid hori- zontal lines represent levels of general arousal. At X, the organ- ism is asleep and none of the actual messages produce alerting. At Z, the organism is awake. All messages could be alerted to, but the specific alerting system allows only b to be heeded.)
ATTENTION 7
greater sensitivity in the appropriate region, rather than signifying a reduction in incoming information. However, other recent experi- menters continue to maintain that evoked responses diminish in amplitude when attention is not being paid to the test stimulus. Garcia-Austt, Bogacz, and Vanzulli (1961) recorded scalp visu- al evoked responses in human subjects (who were able to give introspective reports) during presen- tation of flash stimuli. They report,
When the stimulus is significant and therefore atten- tion is paid to it, the response is relatively simple and widespread. When, on the other hand, the stimulus is not significant and no great attention is paid to it, the response is reduced, complex, and localized.
It would seem that changes in the evoked poten- tial at the cortex do indeed take place during habituation and attention shifts; but that what those changes exactly are, and what they repre- sent, is not yet clear. We should indeed expect, on the above the- ory of attention, changes in the cortical evoked potential when attention is being paid to a stim- ulus, reflecting the activation of various processes, such as motor output and memory storage. Pertinent to this assumption is the dis- covery by Hubel, Henson, Rupert, and Galambos (1959) of what they term “attention” units in the auditory cortex. By the use of microelectrodes implanted in unanesthetized and unrestrained cats, they obtained records from units which responded only when the ani- mal was “paying attention” to the sound source. These attention units appeared to be both inter- spersed amongst the others and segregated from them. We may venture to interpret these results by supposing that the units in question formed part of the systems, discussed above, responsi- ble for the appropriate motor response to stimu- lation or the committing of items to memory, and so forth, or that they lay on the pathway to these systems. Thus they would be inactive even if impulses evoked by auditory stimulation were reaching the cortex, provided that the ani- mal was not also attending to the stimuli.
There is another theoretical assumption for which we might reasonably seek a neurophysi- ological counterpart. We suppose that a selec- tion of inputs from a variety of sources takes place by comparison with a fluctuating stan- dard. This implies the existence of an undiffer- entiated structure with widespread connections with the rest of the central nervous system. We are tempted, on account of the evidence for the diffuseness of its input, to identity the brain stem reticular formation as this particular struc- ture. Potentials may be evoked throughout this structure by excitation of various sensory sys- tems (French, Amerongen, & Magoun, 1952; Starzl, Taylor, & Magoun, 1951), and various cortical structures (Bremer & Terzuolo, 1952, 1954; French, Hernández-Peón, & Livingston, 1955). Occlusive and facilitatory interaction between responses evoked in the reticular for- mation from very different sources have further been observed (Bremer & Terzuolo, 1952, 1954; French et al., 1955). Single unit studies demonstrating a convergence of input from sev- eral sources have also been reported (Amassian, 1952; Amassian & De Vito, 1954; Hernández- Peón & Hagbarth, 1955; Scheibel, Scheibel, Mollica, & Moruzzi, 1955). A similar conclu- sion, that the reticular formation is capable of acting as a nonspecific system, can be based on neuroanatomical evidence. Scheibel and Scheibel (1958) state on the basis of their exten- sive histological study:
the degree of overlap of the collateral afferent plexuses is so great that it is difficult to see how any specificity of input can be maintained, rather it seems to integrate and vector a number of inputs.
We have also postulated that the fluctuating level correlates with states or arousal. Again the brain stem reticular formation seems well suited to fulfill this function. Its importance in the regulation of states of arousal has been demonstrated both through work involving lesions (Bremer, 1935; French, 1952; French & Magoun, 1952; Lindsley, Schreiner, Knowles, & Magoun, 1949) and stimulation of this struc-
ture (Moruzzi & Magoun, 1949; Segundo, Arana, & French, 1955). Recently Moruzzi (1960) has shown that the lower brain stem may play an important role in the initiation of sleep. It also seems likely that the thalamic reticular system is involved in the regulation of states of arousal. Large bilateral lesions of the anterior portion of this system may produce coma anal- ogous to that produced by lesions of the mid- brain (French et at., 1952) although the depth of coma so produced is less profound. Simulation of portions of this system has also been shown to produce either sleep or arousal depending on the parameters of stimulation (Akimoto, Yamaguchi, Okabe, Nakagawa, Nakamura, Abe, Torii, & Masahashi, 1956; Hess, 1954). The work of Adametz (1959), Chow and Randell (1960), and Doty, Beck, and Kooi (1959), who demonstrated that with different operational techniques and with assiduous nursing care massive lesions of the mid-brain- reticular formation need not produce coma, should, however, be considered. Chow, Demenet, and Mitchell (1959) found also that massive lesions in the thalamic reticular system need not produce coma. Until reasons for these discrepant results are found we must regard our conclusions as to the role of the reticular system in attention as tentative. Whatever the explanation of the findings on lesions in the reticular formation may turn out to be, it seems that, if we are right, some diffuse and nonspecific system is necessary as a part of the mechanism subserving selective attention. Such a system should be found to have afferent connections from all discriminatory and per- ceptual systems. Through these connections it should be influenced to take up a variety of lev- els; the level at any one time corresponding with the level of the “highest” afferent message from the discriminatory mechanisms. On its efferent side such a nonspecific system should again be connected with all discriminatory and perceptual mechanisms. Through such connec- tions it would signal to them its own level. If this level of the nonspecific system was above
that of a particular discriminatory mechanism, no registration in memory or motor adjustment would take place, if such a discriminatory mechanism was stimulated. Consequently, only that discriminatory mechanism being acti- vated whose level was equal to that of the dif- fuse system would not be affected. In this way the most important message to the organism will have been selected.
REFERENCES
ADAMETZ, J. H. Rate of recovery of functioning in cats with rostral reticular lesions. J. Neurosurg., 1959. 16, 85-98. AKIMOTO, H., YAMAGUCHI, M., OKABE, K., NAK- AGAWA, T., NAKAMURA, I., ABE K., TORII, H., & MASAHASHI, I. On the sleep induced through electrical stimulation of dog thalamus. Folia psychi- at. neurol. Jap., 1956, 10 , 117-146. AMASSIAN, V. E. Interaction in the somatovisceral pro- jecton system. Proc. Ass. Res. Newv. Ment. Dis., 1952, 30, 371-402. AMASSIAN, V. E., & DEVITO, R. V. Unit activity in reticular formation and nearby structures. J. Neurophysiol., 1954, 17 , 575-603. BRADLEY, P. B., & ELKES, J. The effect of atropine, hyoscyamine, physostigmine and neostigmine on the electrical activity of the brain of the conscious cat. J. Physiol., 1953, 120 , 14-15. BREMER, F. Cerveau isolé et physiologie du sommeil. CR Soc. Biol., Paris, 1935, 118 , 1235-1242. BREMER, F., & TERZUOLO, C. Rôle de l’écorce cérébrale dans le processus physiologique du réveil. Arch. int. Physiol., 1952, 60 , 228-231. BREMER, F., & TERZUOLO, C. Contribution à l’étude des mécanismes physiologiques du maintien de l’ac- tivité vigile du cerveau. Interactión de la formation réticulée et de l’écorce cérébrale dans le processus des réveil. Arch. inter. Physiol., 1954, 62 , 157-178. BROADBENT, D. E. The role of auditory localization in attention and memory span. J. exp. Psychol., 1954, 47 , 191-196. BROADBENT, D. E. Perception and communication. London: Pergamon, 1958. CHERRY, E. C. Some experiments on the recognition of speech with one and with two ears. J. Acoust. Soc. Amer., 1953, 25 , 975-979. CHOW, K. L., DEMENT, W. C., & MITCHELL, S. A., JR. Effects of lesions of the rostral thalamus on brain waves and behavior in cats. EEG clin. Neurophysiol., 1959, 11 , 107-120.
8 DEUTSCH & DEUTSCH
production of sleep by sensory stimulation. Int. J. EEG clin. Neurophysiol., 1960, Suppl. No. 13 , 231-
MORUZZI, G., & MAGOUN, H. W. Brain stem reticu- lar formation and activation of the EEG. EEG clin. Neurophysiol., 1949, 1 , 455-473. MOUSHEGIAN, G., RUPERT, A., MARSH, J. T., & GALAMBOS, R. Evoked cortical potentials in absence of middle ear muscles. Science, 1961, 133 , 582-583. NAQUET, R., REGIS, H., FISCHER-WILLIAMS, M., & FERNANDEZ-GUARDIOLA, A. Variation in the responses evoked by light along the specific path- ways. Brain, 1960, 83 , 52-56. OSWALD, I., TAYLOR, A., & TREISMAN, M. Discrimination responses to simulation during human sleep. Brain, 1960, 83 , 440-453. PALESTINI, M., DAVIDOVICH, A., & HERNÁNDEZ- PEÓN, R. Functional significance of centrifugal influences upon the retina. Acta neurol. Lat.-Amer., 1959, 5 , 113-131. PETERS, R. W. Competing messages: The effect of interfering messages upon the reception of primary messages. USN Sch. Aviat. Med. res. Rep., 1954, Project No. NM 001 064.01.27. POULTON, E. C. Two-channel listening. J. exp. Psychol., 1953, 46 , 91-96. SCHEIBEL, M.E., & SCHEIBEL, A. Structural sub- strates for integrative patterns in the brain stem retic- ular core. In H. H. Jasper, L. D. Proctor, R. S. Knighton, and S. Roberts, W. C. Noshay, C. William, and R. T. Costello (Eds.), Reticular formation of the brain. Boston: Little, Brown, 1958. Pp. 31-55. SCHEIBEL, M. E., SCHEIBEL, A., MOLLICA, A., & MORUZZI, G. Convergence and interaction of afferent impulses on single units of reticular forma- tion. J. Neurophysiol., 1955, 18 , 309-331. SEGUNDO, J. P., ARANA, R., & FRENCH, J. D. Behavioral arousal by stimulation of the brain in monkey. J. Neurosurg., 1955, 12 , 601-613. SHARPLESS, S., & JASPER, H. Habituation of the arousal reaction. Brain, 1956, 79 , 655-678. SOKOLOV, E. N. Neuronal models and the orienting reflex. In Mary A. B. Brazier (Ed.), The central nervous system and behavior: Transactions of the third conference. New York: Josiah Macy, Jr. Foundation, 1960. Pp. 187-276. SPIETH, W., CURTIS, J. F., & WEBSTER, J. C. Responding to one of two simultaneous messages. J. Acoust. Soc. Amer., 1954, 26 , 391-396. STARZL, T. E., TAYLOR, C. W., & MAGOUN, H. W. Collateral afferent excitation of reticular formation of brain stem. J. Neurophysiol., 1951, 14 , 479-496.
THOMPSON, W. R., & SOLOMON, L. M. Spontaneous pattern discrimination in the rat. J. comp. Physiol. Psychol., 1954, 46 , 281-287. TREISMAN, A. M. Contexual cues in selective listen- ing. Quart. J. exp. Psychol., 1960, 12 , 242-248. VORONIN, L. G., & SOKOLOV, E. N. Cortical mech- anisms of the orienting reflex and its relation to the conditioned reflex. Int. J. EEG clin. Neurophysiol., 1960, Suppl. No. 13, 335-346. WEBSTER, J. C., & THOMPSON, P. O. Responding to both of two overlapping messages. J. Acoust. Soc. Amer., 1954, 26 , 396-402.
10 DEUTSCH & DEUTSCH
