You are in: Home » History of the Unit
ATTENTION AND COGNITIVE CONTROL
General Notes. This material has been scanned from the original typescript. While we have done our best to remove errors, some may well remain. You can access other parts of this particular Progress Report either from the menu at the bottom of this entry, by using your browser's back function, by navigating back to the Unit History Timeline, or by accessing the relevant section of the electronic archive. Reference for this report are indexed by number and these can be found in a dedicated section also accessible from the menu at the bottom of this entry.
Barnard 4.25, Blandford 2.92, Brown 4.58, Chapman 4.42, Duncan 5.0, Emslie 3.92, Godden 0.67, Goodrich 3.50, Groeger 4.42, Howes 5.0, Lavie 1.0, Marcel 5.0, May 2.25, Nimmo-Smith 1.0, Sellen 0.73, Shallice 1.0, Ward 2.9, Whittington 2.58, Wilkinson 1.0, R Young 5.0, Allison (SO) 1.0, Bright (SO) 0.37, Clegg (SO) 1.58, Duff (SO) 0.34, Gourdol (CO) 0.17, Grande (SO) 3.5, A Green (SO) 2.08, P Houghton (Student) 0.21, van Houten (SO) 0.16, Kolodny (SO) 0.75, Newell (SO) 0.42, Robinson (SO) 0.18, Salber (SO) 0.17, Scott (SO) 0.85, Taylor (SO) 0.08, Tweedie (SO) 1.25.
Total Person Years: Scientists 61.1; Research Support 13.1
Abstract
Objectives
In this programme we investigate the coordination of cognitive activity. Even a simple cognitive event, such as directing attention to a relevant object in the visual world, requires coordinated activity in multiple brain systems. The problem is brought even more to the fore in tasks of realistic complexity, such as learning a novel computer system or controlling a vehicle through traffic. Our strategy is to investigate examples at multiple levels of complexity. The approach is broadly multidisciplinary: for simple cases, the approach combines cognitive psychology and basic neurobiology; in more complex cases, there is major input from artificial intelligence and computer science. We seek to understand basic processes of attention, intelligence, acquisition and use of knowledge, and consciousness.
Scientific progress and achievements over the past five years
Good progress has been made towards a cognitive-neurobiological model of selective attention in vision. This case provides a simple, concrete example of the general problem of coordination. According to the model, attention develops as an emergent state in which multiple brain systems work concurrently on the same object. Collaborative projects have allowed tests of the model in single unit recording from the macaque, and in studies of brain lesions, regional cerebral activation and normal behaviour in the human. Clinical applications are also beginning through collaborative work with the programme addressing rehabilitation of unilateral neglect.
At the next level of complexity are studies of executive functions, intelligence and consciousness. Cognitive and neuropsychological studies suggest a central role for frontal lobe functions, in particular goal management and selection, in traditionally-defined intelligence. Our studies of consciousness have dealt especially with visual, spatial and bodily awareness. Dissociations of consciousness have been demonstrated through neuropsychological studies of mode of report, cueing conditions for action, and awareness of deficit. Again a key question is how dissociable systems combine to produce unitary behavioural phenomena such as goal selection and awareness.
Besides their immediate practical application, studies of real-world tasks allow us to consider the full problem of cognitive models sufficiently powerful and integrated to explain real behaviour. We have worked with models such as Soar, a general architecture for integrated control of cognition, and Barnard's Interacting Cognitive Subsystems, in which content-specific cognitive modules combine to produce coherent behaviour. Continuing our strong tradition in the field, a major test bed for such approaches has been human-computer interaction (HCI), supplemented by studies of driving and of action slips by anaesthetists. Our studies have shown contrasts between multiple forms of learning and knowledge, including instructed and exploratory learning in both HCI and driving, people's knowledge of their own proficiency, and action slips from familiar automatisms in the setting of anaesthesia.
Future plans for the next five years
The joint cognitive-neurobiological approach to visual attention, frontal lobe functions, intelligence and consciousness will be extended over the next five years. Behavioural studies will address the conditions, time-course and nature of multiple forms of competition between candidate cognitive activities, the limits of voluntary control, and the contribution of executive functions to intelligence. In the area of consciousness, a major emphasis will be on bodily sensation, including studies of deafferentation, body image and pain, including pain under general anaesthetic. These cognitive studies provide the basis for studies of neural implementation. Through collaborations in Cambridge and elsewhere, we have a developing programme in functional brain imaging. As one example, we shall investigate the role of frontal functions in various tests of general intelligence. We are also developing (i) a panel of patients with well-characterized, focal brain lesions for joint behavioural and functional activation work; (ii) our collaborations on rehabilitation after stroke and head injury; and (iii) our collaborations on single cell electrophysiology. Our aim is to give the APU a central role in a multidisciplinary team addressing the neurobiology of well-specified higher cognitive functions.
With the departure of Brown and Groeger, the APU's distinguished 40-year programme of research in driver behaviour has come to an end. This makes the HCI work even more central to our study of cognition in tasks of realistic complexity. New topics to be addressed include the structuring and selection of dynamic visual input, control of behaviour by input from the device being utilized, and the development of formalisms for communication between experts in cognitive psychology, AI and design. The topics of complex visual selection, goal representation and action choice are all points of contact with the more basic, neurobiological parts of the Attention and Cognitive Control programme. By moving between simpler and more complex cases, we hope to maintain the balance of tractability and applicability to real behaviour.
Implications for improving health, health care and wealth creation
The tremendous economic significance of both driving and HCI research hardly needs stating. On British roads there are around 70,000 serious injuries each year; the European research to which we have contributed, dealing with development of new driving technology, has potential application to hundreds of millions of vehicles. As witnessed for example by the overwhelming success of icon-based systems, development of usable computer interfaces now has direct impact on almost every field of activity in industrialized countries. In our work we have given explicit attention to the problem of moving from scientific analysis to real device design. In 1991, Brown's distinguished contributions to British driver safety were recognised by award of the OBE.
Our neurobiological work too has increasing clinical and economic significance. Disabling executive impairments are common, for example, following closed head injury. The most common victims of head injury are young men at the start of their work lives. Attentional impairments, especially unilateral neglect, are also a common consequence of strokes and other brain insults. Management of these conditions is now the subject of active collaboration with the Neuropsychological Rehabilitation team.
INTEGRATED COMPETITION IN DISTRIBUTED BRAIN SYSTEMS (Duncan, Emslie, Houghton, Ward)
Introduction
A major theme guiding research into attention is control of behaviour by competition. One perceptual input, action or line of processing gains precedence at the expense of others. Different forms of competition are reflected, for example, in different sources of interference between concurrent tasks.
Two different aspects of competition are emphasised here. The first and best developed concerns vision: At any given time, we have detailed awareness of only a few objects in the visual field, the remainder being disregarded or ignored. Experiments are concerned especially with interference between one part of the visual input and another. A second project concerns relationships between frontal lobe functions, individual differences in general intelligence or Spearman's g, and interference between tasks sharing little in the way of input modality, output modality or common content. We relate these problems to competition between concurrent action requirements or goals.
A second major theme is integration between brain systems. Multiple brain systems are involved in the competitive control of even simple behaviour. As well as defining their separate roles, we must understand how systems are integrated to work concurrently on the same object or task.
Over the past 5 years the work of this group has become increasingly interdisciplinary. As the central element, we have continued to develop functional models from research on normal human cognition. Within the APU, we have begun to work closely with Robertson's group on the consequences of attentional models for clinical practice and rehabilitation (this work is described largely in Robertson's section of the report, in the Rehabilitation programme; see also the related proposals by Goodrich in this programme). At the same time, we have turned, through a series of outside collaborations, to investigations of how attentional functions are implemented in the human and animal brain. Single unit recording in awake monkeys has allowed us to study attentional modulation of neural responses at different levels of the visual system. To initiate this project, Duncan spent the year 1989-90 working with Robert Desimone at NIMH in Bethesda; the work has since developed with grant support from the U.S. Office of Naval Research and the Human Frontier Science Program. The latter grant has also allowed collaborations with groups working on human lesion studies (Glyn Humphreys, Birmingham), functional activation using PET (Guy Orban, Leuven), and modelling (Claus Bundesen, Copenhagen).
A. Visual Attention (Duncan, Ward)
In the last Progress Report, we outlined a general account of the functions involved in visual selective attention (2.25, 2.26, 2.95). At any given time, objects in the visual field compete for representation in visual short-term memory. Subjectively, this competition is reflected in limited attentional capacity. Competition introduces the problem of selectivity: we attend to objects that are relevant to current behaviour, and ignore the remainder. To bias competition towards relevant objects, candidate objects in the visual input are compared against a flexible attentional template, or advance description of information currently required (2.26, 2.95). According to the task, this template can specify many different object properties, including location, motion, object features etc. Objects matching the template gain a competitive advantage, and are more likely to be selected for control of behaviour. The model has been applied to a range of problems, including selective and divided attention (2.95), visual search and visual grouping (2.25, 2.26).
In the present reporting period, the model has been developed through a series of further studies, both behavioural and physiological:
1. One set of behavioural studies has dealt with the role of objects in attentional competition. When concurrent discriminations concern different objects, competition results in reduced accuracy. When discriminations concern different attributes of the same object, however, all interference disappears (2.22). This general conclusion holds true even for attributes such as colour and motion, whose processing is separated into quite distinct cortical systems (2.21, 2.95). The implication is that, when an object is selected for control of behaviour, all of its attributes are selected together.
A second set of behavioural studies has dealt with the time-course of attentional competition. Conventional serial models suggest that each object consumes processing capacity for only a few dozen msec; in contrast, our single cell results (see below) suggest that between-object competition and nontarget suppression develop over hundreds of msec. To resolve the issue we have measured the time-course of competition directly in behaviour. Following identification of one object, interference with another lasts up to half a second, then rapidly resolves (2.27). This time-course is exactly the same whether the task requires identification of one or two attributes of the first object, even though making two identification responses of course takes substantially longer. Increasing the difficulty of the first discrimination also has no effect though, as we should expect, interference is substantially increased when the first display contains two objects rather than one. The results show that high-speed serial models of visual attention are incorrect. Like nontarget suppression in the physiological data, interference in behaviour lasts for hundreds of msec.
2. Our single unit studies have dealt with both attentional templates and attentional competition in the high-level visual cortex of the monkey. Animals were trained to direct attention to different objects in a multi-object display, depending on a prior instruction cue. Recordings were made from single cortical neurons in each phase of the task. While some recordings have been made in both V4, an early input area, and the inferior convexity of the frontal lobe, the main results come from inferotemporal (IT) cortex, a visual area involved in object recognition.
In IT, we observed preactivation or priming of neurons responsive to the current target. Such priming began with the cue indicating which object was the target, and continued until presentation of the display. On arrival of the display, we observed initial responses to both target and nontarget objects, followed by rapid nontarget suppression (2.15, 2.92, 2.93).
Such results suggest the following model (2.17). In line with behaviour, there is competition between populations of extrastriate neurons responsive to different objects in a display. When a particular object is defined as behaviourally relevant, the corresponding neural population is given a competitive advantage by preactivation. The result is domination of the IT response by the target, with suppression of responses to nontargets.
Different regions of extrastriate cortex are specialised to code different attributes (shape, location, etc.) of visual objects. In our experiments, attentional selection was based on nonspatial object characteristics. The cue indicated what sort of object the target would be, not where it would appear. For the case of spatial cues, preliminary results in Desimone's laboratory suggest priming of cells with a corresponding spatial receptive field; this priming occurs earlier in the visual system than IT. The general scheme (2.24) suggested by these results is that -- for different selection cues -- local priming in different regions of extrastriate cortex can give the target object a competitive advantage. Such a scheme implies integration between visual areas: a local bias in one area leads eventually to a competitive advantage for the whole, distributed description of the selected object (2.21, 2.24).
3. Our lesion studies deal with attentional deficits consequent on parietal, occipitotemporal and frontal lesions in the human. In collaboration with Bundesen and Humphreys, a method has been developed to differentiate deficits including competitive bias against certain regions of the visual field, and poor task-dependent selectivity. One of our most general and surprising findings concerns the phenomenon of extinction. Spatial extinction consequent on right parietal lesions is well-known: objects in the left (contralesional) field can be identified when they occur alone, but are extinguished by an accompanying object on the right (see related contributions of Robertson, and Goodrich and Ward, in this report). Our results suggest this to be a special case of a much more general effect. Spatial extinction also occurs following occipitotemporal lesions, a result confirmed by Desimone in the monkey. In a case of bilateral parietal lesions, we have also observed extinction based on nonspatial characteristics; the patient identifies either a picture or a word presented alone, but extinguishes the word when both are presented together (Humphreys et al., submitted). Again, these results show integration between brain systems (2.17). A lesion which affects one object more than another, whether for spatial or other reasons, produces a competitive bias against that object. Though originally arising in one brain system, this bias is generalised to others. The result is extinction of the whole object and its associated behaviour.
The multiple forms of competitive bias produced by different brain lesions correspond to multiple strategies for rehabilitation. Robertson's work shows how neglect of visual targets in contralesional space can be relieved by irrelevant motor activity on that side. Others have obtained similar results using unilateral vestibular stimulation, or a unilateral eye patch to bias input to the colliculus (e.g. Rubens, 1985). Again, an initially local bias produces generalised results.
4. Though our PET data are preliminary, one initial result may be mentioned. When subjects make concurrent discriminations of orientation and location, we see activation of both occipital area 19 (expected from other studies of orientation discrimination) and parietal area 7 (expected for location discrimination). Other activations seem largely to be motor (cortical areas 4 and 6, caudate, cerebellum), and dividing attention between objects produces no new positive foci. A tentative conclusion is that, in this type of task, attentional competition is largely a property of the visuomotor network itself. We see no activation of "higher-order" areas such as the anterior cingulate.
We believe that these findings make a strong case for interdisciplinary research into attentional functions. Functional models define questions for physiological studies, which in turn feed back to the model level. To our knowledge, few other groups in the world have been able to assemble such an integrated cognitive, neurophysiological and clinical programme of research on high-level brain function. The findings suggest the following conclusions:
(a) Competition between objects occurs in multiple brain systems, visual and motor, cortical and subcortical etc. (2.17, 2.24).
(b) Competition however is integrated across systems. Thus a bias towards one object or region of space, initially arising in any system, is generalised to others (2.24); resulting in a tendency for multiple brain systems to deal concurrently with the same object.
(c) A particular example is integration across visual areas dealing with an object's different attributes (shape, colour, location, motion etc.) (2.95). Indeed this is needed to allow object selection based on different attentional templates. Templates for location, colour and identity are reflected in very different patterns of pre-activation or priming in the extrastriate network; yet all must finally result in selection of a whole object description.
(d) The outcome of competition in the visuomotor network is a long-lasting selective state, in which the description of a whole object gains control of response systems.
B. Selection of Goals: Frontal Functions and Intelligence (Duncan, Emslie, Bourke)
The integrated competition approach is well established for the case of visual attention; in this case hypotheses are relatively well worked out, and suitable lines of attack relatively clear. In this section, the approach is extended to deal with higher-order aspects of attention and control. The work has developed from a prior project dealing with general intelligence or Spearman's g, frontal lobe functions, and interference between dissimilar, concurrent tasks. We suggest that these problems are related to competition, not between perceptual objects, but between action requirements or goals. Though in comparison to visual attention these problems are less well understood, they are of broad significance both theoretically and clinically. Executive functions are emphasised in accounts of memory, planning, attention, ageing etc.; executive disorders are currently implicated in a broad range of clinical problems, including head injury, degenerative disorders, schizophrenia, depression and autism. Again, related clinical work is described by Robertson.
The concept of Spearman's g is central to the study of individual differences. Roughly speaking, g reflects a person's tendency to perform relatively well or not so well in all manner of diverse real-life and experimental activities. In the last Progress Report, we proposed that g might be largely a reflection of frontal lobe functions (2.20, 2.95, 2.96). Following frontal damage, there can be a widespread disorganisation of behaviour, manifested in activity that seems impulsive, stereotyped, irrelevant or otherwise inappropriate to task demands. In the current reporting period, we have carried out a number of studies supporting this general hypothesis. The work has been supported in part by funding from the U.S. Air Force (held jointly with Alan Baddeley), and has involved extensive collaboration with both Rehabilitation (Roger Johnson, Michaela Swales) and MRIS (Charles Freer) Units at Addenbrooke's Hospital. Major findings are as follows:
1. Conventional belief is that frontal deficits are rather unrelated to conventional intelligence. This view is based largely on the existence of patients with obvious cognitive deficits following frontal lesions, but preserved superior IQs on conventional tests such as the Wechsler Adult Intelligence Scale (WAIS). On psychometric grounds, however, changes in g after brain damage are more appropriately measured by tests of "fluid intelligence", typically involving novel problem-solving (2.96). In collaboration with Paul Burgess from University College London, we have shown massive impairments in fluid intelligence (up to 60 IQ points) in frontal patients with preserved WAIS IQs (2.73, 2.96).
2. A series of studies has dealt with a phenomenon we term goal neglect, or disregard of a task requirement even though it has been understood (2.74, 2.96). Subjectively, it is as though a task requirement "slips the subject's mind". Occasionally mentioned in the literature on frontal patients, goal neglect has also been shown to occur in normal people in our studies. Important conditions for its occurrence include novelty, weak verbal feedback on errors, and multiple concurrent task requirements. In the normal population, goal neglect is very closely related to g, though the task is simple and has no apparent element of problem-solving. Under the same conditions, goal neglect is extremely common in frontal patients.
At any given time, behaviour is shaped by a set of requirements or constraints, specified at multiple levels of abstraction (2.74, 2.96). An example is a conventional goal-subgoal hierarchy. We suggest that goal neglect provides a simple behavioural model of a general frontal process of shaping behaviour by activation and competition of appropriate action constraints; and that variability in this process is largely reflected in Spearman's g.
3. In a third project, we attempted both to segregate frontal impairments into distinct components, and to separate them from impairments less sensitive to frontal lesions. A large battery of conventionally "frontal" tests (e.g. Wisconsin card-sorting, verbal fluency) and conventionally "nonfrontal" tests (e.g. recognition memory, object recognition from unusual views) was administered to a series of 90 head injured patients. Head injury was chosen because of its highly variable damage, often with a strong frontal element. From each test we obtained both overall scores, and scores related to putative frontal subfunctions such as switching cognitive set or inhibiting inappropriate behaviour.
Four results are noteworthy. First, correlations between different measures were generally low, in the range of 0.3. In particular, the conventionally "frontal" tests did not correlate more strongly with one another than with the "nonfrontal" tests. Second, extracting more detailed measures from each test -- e.g., measures of inhibition from two different tests -- produced even lower correlations. Third, average performance in a range of 6 "frontal" tests was correlated around 0.6 with fluid intelligence. Both were weakly related to frontal damage as assessed on MRI; the relationship was strengthened if an estimate of premorbid IQ was partialled out. Fourth, the same average was even more closely related to our standard measure of goal neglect, showing that even patients with apparently different profiles of "frontal" deficits share a common impairment. Our hypothesis is that each separate test simply provides a weak opportunity for the common deficit to be revealed.
These findings notwithstanding, it is obvious from animal work that frontal cortex is in some respects heterogeneous, both anatomically and functionally. A clear question for the future is how an integrated behavioural function such as goal selection or activation can arise from the activity of separate frontal subfunctions.
4. Our final studies concern a possible "general factor" in dual task interference, and its relationship to the g factor of psychometrics. It is well known that concurrent tasks are especially prone to interference when they are obviously similar, for example when they share input or output modality. The competition between one visual object and another described in the previous section is one clear example. Some interference remains, however, even when tasks are as dissimilar as possible.
In one series of studies (2.71, 2.72, 2.157), Bourke asked whether a single, general source of dual task interference accounted well for the data when superficial task similarities were avoided. In this case, tasks should be ordered in demand on the general factor, and the order obtained should be consistent no matter what concurrent task was used to measure it. Results confirmed this prediction, suggesting the possibility of a principled scale for measuring the general "demand" of different tasks (2.71).
One hypothesis is that interference between dissimilar, concurrent tasks might reflect conflict within those same action control functions that we have related to g. In this case, a concurrent task might act like a simple reduction in g; across any set of tasks, profiles of g correlation and dual task decrement should be similar. In an initial study, this prediction was confirmed for impairments in 12 driving skills, produced by concurrent generation of random letters (2.29). Baddeley's development of the random generation task is described elsewhere in the Report; for present purposes random generation provides a convenient way to avoid stereotypy in a task with minimal input and simple output. The prediction was confirmed in a second study using 15 conventional psychometric tests; when the concurrent task was changed from random generation to tone discrimination, however, results were rather less clear. A more fine-grained analysis of dual task conflicts will be needed to relate them clearly to g.
C. Inhibitory Control (Houghton)
A variety of phenomena in visual attention (negative priming, inhibition of return, interference) have been taken to reflect direct inhibition of irrelevant inputs. In negative priming (Tipper, 1985), for example, responses to a current target stimulus are slowed down if this stimulus has previously been ignored. In a collaboration with Tipper (2.113), Houghton produced a detailed simulation of negative priming and other phenomena based on inhibition of ignored representations. The sources of inhibition were related to specific brain systems; on this basis, the model has since been extended to data from impaired populations including Parkinson's disease and neglect patients, and to other forms of cognitive inhibition.
FUTURE PROPOSALS
Our general interdisciplinary approach is already proving itself for the case of visual attention. In our proposals this approach is continued and strengthened. At the same time, we have made moves to establish a similar, broad approach to the problem of frontal lobe functions.
The collaborations we have already established outside the APU are continuing, and new ones developing. For functional activation studies of frontal lobe functions, we have established a collaboration with the PET group at Heinrich-Heine University in Dusseldorf (Rudiger Seitz). As functional MRI (fMRI) develops over the next 5 years, we should also expect to gain the increased temporal resolution it allows. Later we shall describe the steps already taken in this direction.
In fact, the opportunity exists to establish the general research programme much more centrally in Cambridge itself. This would involve collaboration between a number of Cambridge laboratories, with a central role for the APU. Later we shall describe the basis for such an development, the support it would need and the progress made so far towards it.
A. Visual Attention (Duncan)
We have suggested that the phenomena of visual attention reflect competition between objects in a distributed network of visual and motor systems. We propose studies to amplify this model at the levels of both behaviour and brain systems, and in collaboration with Robertson (see his section of this report, in Neuropsychological Rehabilitation), to develop useful clinical techniques for the treatment of attentional disorders.
A1. Scope of Interference: What is the basis for the extended interference we have observed between visual targets presented at temporal separations up to 500 msec? One hypothesis is that interference reflects mainly competition within visual processing structures, but supramodal and other, broader aspects of attentional competition have also often been proposed. More generally, it has often been suggested that interference between concurrent activities reflects brain systems they require in common (e.g. Allport, 1980). This hypothesis will be tested using converging behavioural and functional activation studies.
In fact, we have pilot data already to suggest that interference is all but eliminated when one target is visual and the other auditory. If confirmed, such results will lend strong support to the hypothesis of specifically visual competition. In a parallel functional activation study, pairs of visual and auditory stimuli would be presented. In different conditions, subjects would identify both stimuli in one modality, or stimuli in different modalities. We predict competition (e.g. reduced activation for both right and left visual stimuli) in the within-modality case, but parallel activation of modality-specific systems. Following our previous PET results, we predict little activation of supramodal systems for these simple, well-practised discriminations.
A second interesting case would be concurrent, lateralised motor activity. Robertson's work on rehabilitation of visual neglect has suggested integration between spatial biases in visual and motor systems. To examine such integration, we should present a visual object to left or right, accompanied by a distractor in the contralateral field. At varying intervals surrounding this display, a central auditory stimulus would call for a lateralised manual response. Integration would be indicated by a visual bias initiated by the manual requirement. Following on from this study, we should examine visual biases consequent on other aspects of movement. The parietal lobe of the macaque, for example, contains neurons active both during a particular kind of action (e.g. pulling) in the dark, and in response to a visual input suitable for such an action. In general, does performance of some action bias the visual system towards inputs "affording" that action?
Finally we shall ask what factors influence the duration of interference between one visual object and another. When does one object's demand on the visual system resolve? Our findings to date suggest that the duration of response selection operations is immaterial, but this needs to be confirmed. An interesting hypothesis is that each eye movement clears the previous selective state, setting the stage for the next.
A2. The Attentional Template: We have suggested that, at least in part, the attentional template is implemented by priming of extrastriate neural populations. Issues arising include the distinct representation of templates coding different object properties such as location or shape; the coordination implied by using a local template to guide selection of a whole object description; and interaction between extrastriate and frontal systems.
A continuing collaboration with Desimone at NIMH addresses coordination between visual areas. When monkeys search for a target object, we find template-like activity only late in the visual system. Neurons selective for the target are pre-primed in inferotemporal (IT) cortex, but not earlier in IT's main input area V4. What of the subsequent suppression of nontarget responses seen in IT when the visual array is presented? Our preliminary data suggest that, in the earlier area V4, such suppression develops later in the visual response, as if fed back from the higher-order system. Such a pattern of interaction between areas is exactly what we would need to coordinate whole-object selection and suppression. To support such a picture, the data need confirmation and extension to other selection tasks with other kinds of template activity.
For the human, a broader set of hypotheses may be tested. Specialisation of an occipitoparietal system for spatial vision and an occipitotemporal system for object vision is well known. We propose that priming of parietal neurons will be involved in location-based templates, and thus in controlling selection by a spatial cue, while occipitotemporal priming is involved in object-based templates, and thus in controlling selection by object features. Close connections are also known, however, between parietal and dorsal prefrontal cortex, and between inferotemporal and more ventral prefrontal cortex; and there are good reasons from others' work to suppose that these prefrontal regions are involved in holding cues to guide future behaviour (e.g. Funahashi, Bruce & Goldman-Rakic, 1989). In general, frontal systems seem especially important when behaviour must be altered; in the present case, we expect greater prefrontal involvement when target locations or objects must switch from trial to trial. Related and broader experiments dealing with frontal functions and practice are presented in a later section.
Ideally, these hypotheses will be tested in both lesion and functional activation studies. For the lesion studies, we shall compare selection deficits following parietal, occipitotemporal and frontal lesions. Encouraging pilot data for the first two groups have already been obtained. We shall compare spatial and nonspatial selection cues, either fixed or varying between trials. With fMRI, it may be possible to image activity in the period between presentation of a cue and the subsequent visual array. In this way we should specifically target the kind of sustained priming activity we suggest is important in the attentional template. We predict parietal activation for spatial cues, and occipitotemporal activation for object cues. A particularly interesting result would be differentiation of more dorsal and more ventral frontal activations during spatial and object tasks with varying cues.
A further important case is long practice with search for a particular target. In this case, the familiar target seems automatically to draw attention to itself. A possible study in both human and monkey is to investigate changes in template-related activity when practice is prolonged.
A3. Extinction and Integration: Contralateral neglect and extinction are common in the acute phase following a unilateral cerebral insult. With time the most severe problems may resolve, leaving subtle disorders manifest on detailed examination. According to our hypothesis, a competitive imbalance in one brain system can be transferred to others, producing widespread consequences of even a restricted lesion.
Using the brief exposure techniques we have already developed to measure subtle forms of visual extinction, we shall pursue our observation that this disorder is common to many forms of unilateral lesion. Of particular interest will be the visual consequences of lesions to various parts of the motor system, including posterior frontal cortex and the basal ganglia. At the same time we shall develop our work on nonspatial extinction. Our hypothesis is that, whenever a brain lesion affects the representation of one object more than another, a consequence is competitive bias against the impaired object when both are presented together. For example, it has been reported (Robertson & Lamb, 1991) that left and right temporoparietal regions are specialised for respectively more local and more global objects; we shall explore the consequences for extinction when both kinds of object occur together.
In parallel with these studies are those described by Robertson in the Neuropsychological Rehabilitation section of this report, dealing with clinical implications and tests of the general competitive model.
B. Selection of Goals: Frontal Functions and Intelligence (Duncan, Bright)
In this section we describe behavioural, lesion and functional activation studies of dual task interference, frontal functions and g. Linked clinical studies are described by Robertson (see his section of this report, in Neuropsychological Rehabilitation).
B1. Dual Task Interference: As we have said, there is generally interference when two tasks are performed at once, even when those tasks are maximally dissimilar. According to Bourke's work, such interference suggests competition for a single, common system or resource, and we have suggested it may be closely related to g. In fact, little is known about this rather general form of dual task interference. We plan to investigate it in detail.
Following the strategy we have used for visual interference, we shall begin with a study of time-course. The literature suggests two factors that might contribute to interference between very dissimilar activities. First, there is the "psychological refractory period", reflected in interference when two responses must be selected or two facts retrieved simultaneously. This form of interference takes the form of a real-time processing bottleneck. Second, there is a much longer-lasting difficulty in keeping the rules or control structures of two tasks active at once. This difficulty is reflected in interference even when stimuli for two tasks are separated by hundreds of msec, and subjects can prepare for one or the other. By presenting stimuli at varying temporal separations, we shall ask which form of interference acts like Bourke's "general factor", and which is related to g correlations.
In the context of vision, we have already proposed one test of the general hypothesis that dual task interference reflects demands on common brain systems. As a converging test of the same hypothesis, we should like to conduct a functional activation study of dual tasks and practice. Practice is a major factor reducing dual task interference; subjectively, practised behaviour becomes "automatic". Among the effects of practice, the PET literature suggests reduction in frontal activation. To investigate this link, we should measure local cerebral activity in both single and dual tasks, early and late in practice; mapping changes in cerebral activation to changes in dual task conflict.
B2. Frontal Functions and g: In our collaboration with the Dusseldorf PET group, we have begun work on a functional activation study designed to test the hypothesis of a close link between frontal lobe functions and g. It is well known that fluid intelligence tests can be based on very different materials, e.g. verbal and spatial. These tests have similar, high g correlations; our question is what they share in terms of cerebral activation.
For use in PET, we have developed three tests with high g correlations, one verbal, one spatial, and one perceptual-motor. Each is paired with a control, based on similar materials and operations but with a lower g component. The three control tests will of course show very different patterns of regional cerebral activation. Our question is how each pattern is modified in the high-g case. Three broad hypotheses may be distinguished. First, different high-g tests may activate common frontal regions. For example, broad activations of both dorsolateral prefrontal and anterior cingulate regions have previously been described for a range of relatively complex tasks.
Second, high-g tests may activate largely content-specific cortical areas. Based on known content-specificity of monkey frontal cortex, this may be the picture for both frontal and posterior regions. Third, increasing a test's g correlation may simply increase involvement of a broad range of cortical systems; in line with the idea that g reflects some overall average of many different functions or skills. In general, our question is how a unified construct such as g relates to modular cerebral organisation.
In a linked study, we shall compare the deficits of patient groups with lesions in different major subdivisions of the frontal lobe (see below). The literature shows frontal deficits in an enormous variety of different tests, with little in the way of principles predicting which tests will be most sensitive. We predict that a major factor should be a test's g correlation: the higher the correlation, the greater should be the frontal deficit. This prediction will be tested both for specific frontal groups, and as a baseline against which to assess more specific impairments.
B3. Goal Neglect: Based on our work to date, we have suggested that goal neglect may provide a simple behavioural model of action control functions that are basic to both frontal impairment and Spearman's g. In general, these functions concern the shaping of behaviour by activation of appropriate task constraints or requirements. To address these functions in more detail, our analyses of goal neglect will be extended and developed.
A first step is generalisation. Our hypothesis is that, whenever a task is novel and has multiple concurrent requirements, one requirement can lose activation and hence control over behaviour. To date, however, the work has been based on one specific test, involving switches of attention between one high-speed stream of visual input and another. An additional problem is sensitivity: In the current test, goal neglect is restricted to around a quarter of the normal population. To address these problems, we are developing a second test which is quite unlike the first. There is no speeded input, and no call for a switch of attention; we simply give subjects a central task (finding groups of words with related meanings), and add additional, independent requirements. With this test we plan parametric studies of the number of concurrent requirements, the order in which they are presented, and individual differences in g. Pilot data confirm that goal neglect is commonly observed, at least in elderly subjects.
When this test has been developed, a number of extensions will be undertaken. First, the head injury project we have described suggested that goal neglect is a difficulty common to patients with very different profiles of "frontal" impairment. A possible hypothesis is that a general goal activation function - central to Spearman's g - arises from integrated activity in separate frontal systems. To test this hypothesis we shall assess neglect following lesions of distinct frontal regions, and compare with posterior lesion controls. Second, neglect is sensitive to task complexity or number of requirements, but we should also like to investigate other aspects of difficulty, less directly related to requirement activation and frontal function. To link to our visual studies, we shall investigate simple discriminability - which has no effect on an object's demands on visual attention - and whether requirements to be concurrently activated concern the same or different objects. Third, the phenomenon of goal neglect has much in common with the problem of "prospective memory" (see contributions by Maylor and Sellen in the Memory section of this report), or remembering to satisfy an intention. To investigate this link, we shall study the effects of delays and interfering activity between initial goal specification and subsequent requirement to perform.
C. Inhibitory Control
With Houghton's departure this topic is continued elsewhere.
Centralising the Programme in Cambridge
Though the interdisciplinary programme we have described has been based largely on outside collaborations, this approach has obvious disadvantages. Collaborators in other laboratories have their own priorities, day-to-day involvement in the research is lost, and the work depends on continued success in attracting outside funding. In fact, the opportunity exists to centralise much of the existing programme in Cambridge, based around a number of new developments in other institutions. This is a possibility which, in consultation with the Visiting Committee and new Director, Duncan is keen to explore. We briefly outline the opportunities that exist and some potential costs.
Funding already secured for establishment of a Functional Brain Imaging Centre, based at Addenbrooke's Hospital under the direction of Professor John Pickard (Neurosurgery), will make both PET and fMRI available in Cambridge within the next few years. An additional request for core funding to support clinical research into closed head injury is under consideration by the MRC; Duncan, Robertson and Baddeley have all been closely involved in this development. Meanwhile, Duncan has also been establishing pilot collaborations with Professor Laurie Hall (Medicinal Chemistry), who will take primary responsibility for establishing fMRI in the new Centre, and in the interim is conducting pilot studies in his own department. The expertise we have already developed through outside collaborations will be of undoubted benefit to the Centre's work, and places us well for studies of normal volunteers and clinical groups other than head injury to complement the Centre's own programme. To support such studies, however, and to give the APU a viable role in the Centre's work, would probably require commitment of at least two scientific posts.
Investigating the functional consequences of human brain lesions has of course been a major part of the APU's programme for a substantial period. When lesions are well characterised, the technique is complementary to functional activation. To optimise such studies, as well as functional activation studies of patients themselves, structural brain lesions must be characterised with the same precision as, and in comparable coordinates to, functional activations. In collaboration with John Pickard (Neurosurgery), Charles Freer (MRIS Unit), Laurie Hall (Medicinal Chemistry) and Trevor Robbins (Experimental Psychology), Duncan has begun to establish a panel of stable patient volunteers for neuropsychological research, with CT- or MRI-imaged lesions mapped to standard coordinates. We are establishing a number of focal lesion groups, including groups with lesions in distinct subdivisions of the frontal lobe as well as posterior groups. Again, this endeavour is of broad benefit to APU research. A full-time scientific officer would be able to support both this development and Duncan's work with normal subjects.
More remotely, the potential of animal work in Cambridge should also be considered. At the new Innes Centre, the MRC is currently establishing a major programme of neuropsychological work with the marmoset. This programme is under the direction of Rosalind Ridley. Though a major collaboration with animal research would be a novel departure for the APU, our work with Desimone's group shows the potential of this approach. With its combination of fine temporal and single-neuron spatial resolution, primate electrophysiology offers a detailed, direct view of brain activity. An attractive approach is to follow gross spatial studies of human brain activation with detailed investigation in the primate. Preliminary discussions with Dr. Ridley suggest a clear possibility of collaborative work on frontal lobe functions.
There are obvious attractions to linking together these different developments in Cambridge, many already supported by the MRC. At the same time, the momentum for such an enterprise needs to come from a specific, integrated research programme of the sort Duncan has developed. We should expect broad benefits not only for the APU's programme as a whole, but for the MRC's other investments in Cambridge as well. We are keen to discuss the desirability of such an approach, and how it might best be integrated into the APU's programme.
SELECTIVE ATTENTION
(Lavie)
Introduction
Three primary questions have been the focus of selective attention research.
1. What causes selection to occur?
A major issue is the extent to which attentional instructions can result in selective perception of task-relevant stimuli. The work detailed in section A attempted to resolve this issue by proposing a model for selective attention that specifies the conditions under which attention would result in selective perception.
2. When selection does occur, what is it based on?
Both response selection and perceptual selection require some degree of physical distinction between the stimuli that are selected for further processing or response, and those that are not selected. Section B describes work investigating the nature of the various stimulus dimensions available to control selection (e.g. colour, shape, location). The studies detailed in B have specifically focused on describing the nature of space-based selection and its contrast with object-based selection.
3. Which perceptual processes do not require selection, and operate independently of attention?
The work detailed in section C has centred on designing a new test for the prevailing attentional theory of Feature Integration (e.g. Treisman, 1988), which claims that the perception of separate features (e.g. colour, orientation) is preattentive, i.e., can proceed without requiring focused attention. On this theory, focused attention is only needed when perception of the particular conjunctions of features into objects is required.
Since Lavie arrived at the APU in October 1994, her report of past studies mainly describes research on programs that began before arrival at the APU. More emphasis has been given to future plans.
A. The Role of Information Load in Selective Processing
A1. Studies of Visual Attention: One of the basic questions in the study of selective attention is whether the perception of irrelevant information can be avoided. This issue led to the long-standing debate on early versus late selection. The early selection approach claims that selective attention results in selective perception. By contrast, the late-selection approach claims that perception proceeds automatically, regardless of selective attention. A definitive resolution of this issue has proved elusive over three decades of research, which have produced some support for both sides of the argument. Lavie has proposed a new approach to this debate which seems capable of resolving previous discrepancies (2.41; Lavie, 1992). She suggested that a major determinant of whether selective perception takes place is the information load imposed by processing task-relevant information. Identification may proceed automatically until perceptual capacity is exhausted. On this view, whether or not irrelevant stimuli are identified should depend on the capacity demands imposed by relevant processing.
A review paper which re-examines previous findings from this perspective provides clear support for the load hypothesis. Results of early selection have usually been obtained in high load tasks; late selection results have been obtained when the relevant processing carried a low load (2.41). This review has also considered the role of physical distinction between relevant and irrelevant stimuli in the task in determining the efficiency of selective processing. This factor was previously held to be the major determinant of selection; but the review points out that, on its own, physical distinction between relevant and irrelevant stimuli leads to apparent discrepancies in results, whereas a combination of physical distinction plus relevant information load provides a resolution. The conclusion is that physical distinction between relevant and irrelevant stimuli is a necessary condition for correct selective processing, but it is not a sufficient condition for the occurrence of selectivity, as long as task-load is low.
Lavie has continued with empirical studies, testing this hypothesis directly. New experiments varying the load of the relevant processing have shown that this variable determined the level of processing for distracting information (2.42). This result was obtained with a number of converging operational definitions of load in relevant processing. These included varying the number of items relevant for processing, and varying the processing requirements for identical displays (e.g., simple detection versus difficult discrimination or feature versus conjunction tasks).
A2. The Load Hypothesis and Studies of Neuropsychological Impairment: Since the studies on normal subjects show that increasing the load of relevant processing can help in the rejection of irrelevant information, implications are raised for patients with lateralized damage (typically stroke) to the parietal lobe. These patients suffer from 'neglect' of visual information presented to their contralesional side, especially when ipsilesional information captures their attention (see sections by Duncan, Goodrich and Robertson in this report). On the load theory for selection, increasing the demand of relevant processing (e.g., by requiring patients to process two relevant letters rather than one) should help eliminate the distraction from irrelevant information in the ipsilesional field. A study that tests this hypothesis is currently being conducted (in collaboration with Ian Robertson), with initial results supporting the prediction.
B. Objects and Space
Lavie's model for selective attention (detailed in section A) attempted to clarify the interactive role of load and physical distinctiveness in determining selective attention. There are however important questions regarding the definition of physical distinctiveness. In her past work with Tsal (Tsal & Lavie, 1988; Tsal & Lavie, 1993), Lavie has argued that not all dimensions of stimuli are equally efficient for selection (c.f. Duncan, 1981). They suggested that space-based selection underlies selection by other attributes such as colour or shape.
A major controversy in recent studies concerns the role of empty space per se, versus the perceptual objects that occupy the space, in determining attentional selection (e.g., Duncan, 1984). Location-based selection studies have usually shown the importance of space effects when distance is manipulated between separate objects (e.g., Tsal & Lavie, 1988). By contrast the theory of object-based selection has been supported in studies which manipulated the separation or integration of attributes into objects, while holding distance between the attributes constant and usually near (e.g., Duncan, 1984 ).
It remains unknown what selection will be based on when the two selection-rules are set against one another. In other words, would object-based selection break down when the attributes to be selected are grouped into one object but over a wide spatial area? Conversely, should distance between the attributes matter once an object is selected? A study in which both spatial separation and object integrality are manipulated orthogonally within the same widely-spaced situation is required to answer these questions. Lavie (in collaboration with J Driver at the University of Cambridge) has designed such a situation in a recent study, comparing performance in judging two attributes which are close or distant either within the same large object or in two separate objects. Preliminary results show that object effects can be obtained even when the object links attributes over a very large space, provided that spatial attention is diffuse. On the other hand, when attention is spatially focused on a narrow region (following a precue), object segmentation over larger spaces no longer has an effect.
C. FIT and Response Competition Effects from Unattended Distractors
In contrast to the view of perception as a unitary process, the prevailing Feature Integration Theory (FIT) of attention (e.g., Treisman, 1988) proposes a major distinction between perception of separate features and of conjunctions of those features to form multidimensional objects. According to FIT, unattended visual features are registered on separate feature maps (e.g., colour maps, orientation maps, etc.). One role of focused attention is to conjoin the separate features into consciously seen objects. Thus, for unattended objects, only the separate features are known (e.g. blue, green, circle, triangle), but not their specific combinations (e.g. a green triangle, and a blue circle). Although this theory has been tested extensively in the paradigms of visual search and of illusory conjunctions, those lines of evidence remain equivocal (see for example: Duncan & Humphreys, 1989; Tsal, Meiran & Lavie, 1993).
Lavie has designed a new test for FIT, using the response competition paradigm. The response competition paradigm provides an indirect procedure for assessing the perception of unattended items, by measuring their effects on the response times to attended items. The unattended items may either be compatible or incompatible with the relevant response. Perceiving the identity of the unattended distractors should lead to longer RTs for targets that appear with incompatible rather than compatible distractors.
According to FIT, the compatibility effects of irrelevant distractors which include target features should not differ according to whether the target features are conjoined within one distractor object or disjoint between two distractors. If distractors are unattended, their features remain unconjoined. In Lavie's study subjects were requested to respond to targets which were defined by conjunctions of colours and shapes (e.g., respond to a purple cross or a green circle, and withhold response to a purple circle or to a green cross). They were also requested to ignore irrelevant distractors that flanked the target on both sides. The features of the distractors could either be compatible or incompatible with the target, but more importantly, the same incompatible (or compatible) distractor features were either conjoined in one of the distractor items (e.g. a green circle) while the other distractor had neutral features (e.g., a blue triangle); or the distractor features were disjoint across two of the distractor items (e.g. a blue circle and a green triangle). Hence, the same distractor features appeared in both types of trials (conjunctive or disjunctive). Only their combinations into objects varied.
The results of this study supported the Feature Integration theory. Both conjunctive and disjunctive distractors had significant compatibility effects of equivalent magnitude, suggesting that the unattended features were not conjoined into individual objects.
FUTURE PROPOSALS
A. The Role of Information Load in Selective Processing
A1. Load and the Efficiency of Visual Search: The load hypothesis has interesting implications for previous accounts of the efficiency of visual search. On the load account, as the relevant search task becomes easier, prevention of irrelevant processing becomes harder. A study in progress has manipulated determinants of efficient search, such as the degree of similarity between target and non-target items in the search task (as discussed by Duncan & Humphreys, 1989), and measured their effect on the prevention of irrelevant distraction. Preliminary results support the contention that more efficient search among the relevant items results in a less efficient rejection of the distractor.
A2. Implications of the load hypothesis for neuropsychological patients and elderly subjects: In addition to studying effects of relevant load on the distraction potency of ipsilesional items in neglect patients, there may be more general implications of the load hypothesis with regard to patients and elderly subjects who have abnormally restrictive capacity limitations (see Baddeley, Bressi, Della Sala, Logie & Spinnler, 1991). These populations should perform better than normal on selective attention tasks, in the following restricted sense. They should be less susceptible to interference from distractors under low load conditions, since less spare capacity should be available for processing these distractors. Studies on this issue will be conducted in collaboration with Ian Robertson and Elizabeth Maylor.
A3. What is the Nature of Load that is Relevant for Selective Processing? Thus far Lavie's studies have shown that loading a task makes rejection of irrelevant information more efficient. Should we conclude that difficult tasks in general lead to more selectivity? This issue will be studied in several ways:
A3.1 The role of data- versus resource-limitations in the processing of irrelevant information: The concept of perceptual load encompasses more than just delayed RT or increased difficulty. It implies that the system must carry out further operations, or must apply operations to additional units. It is these additional demands on capacity, rather than a mere delay in processing, which should block low-priority, irrelevant items from consuming spare capacity. By contrast, data limitations on perception, while increasing task difficulty, should not impose additional demands on capacity, as increases in resource allocation cannot compensate for data degradation (Norman & Bobrow, 1975). Thus, the process of perception for degraded relevant information should be merely prolonged. Indeed the chance of distractor intrusion during this extra time may be even be increased (e.g., Eriksen & Schultz, 1979; Navon, 1988).
Thus, according to the load hypothesis, data limitations in relevant information should actually have the opposite effect on irrelevant information processing to increases in relevant load (i.e., resource limitations), even though both factors make any task harder. Studies that test this account are currently being conducted, in which load is manipulated by the number of relevant items. Data limits are induced by either reduced acuity for more peripheral targets, smaller and lower contrast targets, or masking characters superimposed on the target position.
A3.2 Types of load: Does loading post-perceptual routines (e.g., response selection, memory) result in a similar effect to the perceptual load manipulations discussed above? There are grounds to expect that loading a more central aspect of control, responsible for setting up and maintaining the priorities in a task, should lead to confusion, i.e., a reduced distinction between high- and low-priority items, resulting in impairment of selectivity. It should be interesting therefore to contrast the effect of loading early perceptual processes versus late response-selection processes within comparable tasks.
A3.3 Auditory load and selective attention: There is evidence for general limitations in capacity which hold across modality (see Duncan's section in this programme). Thus, one might extend the load hypothesis across modality to predict that selectivity in vision would improve under increased load in the auditory modality. In a preliminary study conducted in collaboration with Richard Ivry (UC Berkeley), we are investigating whether load in semantic processing for auditory stimuli can eliminate Stroop-interference from an irrelevant visual word on responses to colour patches.
A4. The Relationship Between Load and Other Factors Influencing Irrelevant Processing
A4.1 The relationship between perceptual load and spatial cueing: In a study by Johnston and Yantis (1990), cueing the location of a target in advance resulted in successful rejection of information in irrelevant locations. The displays used in their study were characterised by high perceptual load, and thus it is difficult to determine the roles played by either cueing or load. In a collaborative study with James Johnston (of NASA-Ames, California), both cueing and load are manipulated in a factorial design in order to study their contributions in determining selective processing. Initial results suggest that load and spatial cueing interact such that effectiveness of cueing is greater with high loads.
A4.2 Attention capture: The load hypothesis suggests that spare capacity inevitably 'spills over' to distracting information once the relevant processing required by the explicit task has been completed. However, it may be possible to prevent this without any increase in load. It is known that some factors (e.g. abrupt onsets) are particularly effective in attracting attention. This raises the possibility that interference from distractors may be prevented, even under conditions of low load, provided the relevant stimuli are sufficiently attention capturing (e.g., they flicker) to continue to attract any spare capacity.
A5. Testing the Extent of Voluntary Control in Selective Attention Tasks: One of the implications of Lavie's previous load studies is that selective perception is involuntary in the sense that it is more intimately related to the load of the relevant task than to the subject's intentions to be selective. When the load in the relevant task is low, subjects apparently cannot ignore the distractors, even though this is explicitly required and failure to comply with the requirement impairs performance. The extent to which attending to/ignoring distraction is involuntary seems to have significant applied consequences. In education, for example, there is little point in emphasising instructions to ignore distraction if this is not under voluntary control. It therefore seems important to investigate whether distractor processing is truly involuntary by directly manipulating subjects' motivation and expectations in the task. (for example, increasing motivation by offering some reward for selective processing, or changing expectations by manipulating the relative frequency of different kinds of distractor).
B. Objects and Space
The popular "spotlight" metaphor for visual attention implies that attention is primarily a spatial phenomenon. Tsal and Lavie (1988, 1993), suggested a more specific role for location in controlling visual selection. Although subjects can use various dimensions as cues for selection (e.g. colour, shape), Tsal and Lavie proposed that this proceeds by means of allocating attention to the location of the colour or shape cue. In several studies they demonstrated that when a target was selected by colour (i.e., subjects were requested to report a red item) or by shape (i.e., report an angular letter), neighbouring stimuli with a different colour or shape were enhanced, whereas nonadjacent items with the target colour or shape were not. In addition, a 'spatial carry-over' effect was established: subjects were subsequently faster if the next target appeared at the same location, even though location was completely irrelevant to the explicit task. There was no analogous colour or shape carry-over if the target was specified by location and colours or shapes were repeated from trial to trial.
A number of theorists have recently claimed that "perceptual groups", rather than positions per se, are the targets of visual attention. It is suggested here that these studies can be reconciled with the special role for location proposed above, if we assume that perceptual groups are selected by means of their location, analogously to the coloured stimuli discussed above. This proposal will be tested in several ways, for example by looking for spatial carry over effects analogous to those observed by Tsal and Lavie (1993) in the situations which have previously produced evidence for perceptual grouping effects.
C. FIT and Response Competition Effects from Unattended Distractors
Further investigation of compatibility effects for conjunctive versus disjunctive distractors is planned along several lines. One direction will be to measure whether factors that are known to affect feature integration in visual search paradigms have a similar effect on the extent to which response competition is based on separate feature processing. For example, will the distance between the two distractors affect the extent to which disjunctive distractors produce interference? Another direction would investigate the relationships between processes of feature integration for distractors and the load of relevant target processing. Given that the task used in Lavie's previous FIT study seemed to carry an intermediate level of load, a few questions arise. (i) Would increasing load in the central target task eliminate the distractor effects, given that they may be produced by an early stage of separate feature registration? (ii) Would reducing the load in the present task provide an advantage for the conjunctive distractors over the disjunctive ones, by allowing sufficient spare capacity to produce (involuntary) feature conjunction for the distractors?
ATTENTIONAL IMPAIRMENTS (Goodrich, Ward)
Introduction
At any moment in time, we are aware of, or can attend to, only a limited number of visual events from the hundreds potentially available to us. Theories of normal attentional function commonly consider objects in the visual world to be competing for limited attentional resources. It is the winners that are selected and reach awareness; the losers might be disregarded or simply fail to reach awareness. The following studies explore factors which can influence this competition for attention.
The subjects were patients with attentional dysfunction following neurological impairment. Disabling deficits in selective attention are relatively common following unilateral brain damage, especially to the parietal lobe. Patients ignore information on the side of space contralateral to their lesion, even though it can be demonstrated that this information is processed perceptually, leading to the belief that the neglected information suffers from a 'lack of attention' rather than a lack of perceptual processing (other work on neglect is discussed in the sections of this report by Duncan, Marcel and Robertson). A second group comprised people with Parkinson's disease. Within this category exists a subgroup with 'frontal' signs who appear to have difficulty on performance of tasks with specific attentional requirements.
The goal of these studies is to understand attentional dysfunction subsequent to brain damage in terms of damage to the normal components of selective attention, and so contribute to the development of theories of normal attentional mechanisms. In addition, such theories contribute to the development of rehabilitation interventions.
A. Visual Attention and Extinction (Goodrich, Ward)
We began a new series of studies aimed at relating the phenomenon of visual extinction, frequently found following unilateral parietal damage, to models of normal visual attention. We applied the notion of competition between target items for limited capacity processing to the phenomenon of visual extinction (2.61). By this account, the effect of unilateral parietal damage is to weaken the competitive strength of contralesional items. When two items are simultaneously presented, the more ipsilesional item wins the competition for selection at the expense of the contralesional item. However, when the contralesional item appears alone, it can still draw attention to itself, since there are no other competitors within the display. Extinction may therefore simply be a spatially specific exaggeration of a normal attentional limitation, arising from competition between multiple targets.
A1. Perceptual Grouping and Selective Attention: Within this framework, we have employed visual extinction as a technique for investigating factors underlying the competition for attention. Using subjects with unilateral parietal lesions, we find that under conditions of brief exposures, detection of contralesional items is improved when contra- and ipsilesional items form a good perceptual group (2.61). We argue that compe¬tition for selection is sensitive to the representations given to possible targets: when targets are represented as elements of separate groups, they compete for limited attentional resources; when they are represented as elements within a group, they become allies, rather than competitors, for selection.
We have utilised the extinction paradigm to explore the competition for selection between different types of visual representation. For instance, do object representations compete as integrated wholes, or as lower-level attribute values? We now have initial evidence comparing the rate of extinction for line drawings of objects versus scrambled versions of these same line drawings. The results suggest an object superiority effect in extinction: whole drawings show signifi¬cantly less extinction than scrambled drawings -- that is, whole drawings are stronger competi¬tors for selection than scrambled collections of the same features.
In a further extinction study, we looked at competition between items in terms of their spatial organisation. Our results show that extinction is greater when right and left items are horizontally, rather than diagonally, aligned about a central fixation point. Further work is required to determine whether this advantage for diagonal items reflects a dissociation between items represented in upper versus lower visual fields, or whether competition for selection is increased when items are spatially mirrored in right and left space.
A2. Non-perceptual Task Variables and Selective Attention: In a second series of studies, we explored the relationship between visual competition and non-perceptual task variables. We report a patient who, following right parietal damage, showed the opposite of the normal extinction pattern ('anti-extinction'). On brief visual presentations, he consistently showed improved detection and identification of contralesional items when presented with an accompanying ipsilesional target.
Interestingly, this benefit for contralesional items on simultaneous presentation only holds when the same task is performed with both contra- and ipsilesional items. In other words, a common task specification for contra- and ipsilesional items appears to reduce the severity of his attentional deficit. This beneficial effect of common task variables between target items provides a clear demonstration that attentional deficits resulting from parietal damage can be mitigated by non-perceptual task variables.
B. Identifying Factors which Reduce Unilateral Neglect (Goodrich)
B1. Intermodality Interactions: Unilateral neglect can be found in each sensory modality, and in some cases may be restricted to a single modality. Work with normal subjects has shown strong links between attentional mechanisms across modalities (Driver & Spence, 1994). If subjects attend to a specific location in one modality, it is difficult for them to attend elsewhere in a different modality. We (with Jon Driver, University of Cambridge) explored whether these intermodality links could be exploited in order to alleviate neglect (i.e., as a way of biasing selection).
We began with the effect of touch on visual neglect in a line bisection task. Subjects had to bisect a wooden rod with their ipsilesional index finger after feeling the full extent of the rod with the same finger, which was moved either actively or passively. Active tactile information reduced neglect (the degree of rightward displacement was reduced) compared to the passive condition. The advantage we have found for active over passive touch is consistent with 'premotor' theories of attention which suggests that attending to a location in space requires programming a movement towards it (Rizzolatti & Gallese, 1988).
B2. Ipsilesional Capture in Unilateral Neglect: We (with Driver) have also begun a series of studies investigating the notion of ipsilateral capture in unilateral neglect and prospects for rehabilitation. The idea here is that in neglect, ipsilesional events invariably 'win' the competition for selection at the expense of contralesional events. Using a shape comparison task, we have shown that when ipsilesional stimulation is reduced, neglect is attenuated.
B3. Axis Based Visual Neglect: In a final study, we (with Driver) investigated what defines the contra-/ipsi-lesional divide determining the subset of information that is neglected in unilateral neglect. We report three patients with unilateral neglect who showed a deficit defined by axis-based, object-centred coordinates (2.18).
C. The Simple Reaction Time Deficit in Parkinson's Disease (Goodrich)
It has been widely held that performance of simple RT utilises only a subset of those processes underlying choice RT performance. Following on from the work of Frith and Done (1986), we have shown that the traditional subtractive view of RT performance is not tenable and that simple and choice RT are doubly dissociable (2.30). Using a series of RT tasks with different stimulus and response modalities, we showed that, in normal subjects, simple RT is more affected than choice RT by the addition of a secondary task. This suggests that at least part of the normal simple-RT advantage is attributable to an attentional-demanding process peculiar to the simple RT task.
Previous studies suggest that the same process that speeds simple RT in normal subjects is impaired in at least some patients with Parkinson's disease. In a series of studies (with John Harrison, Charing Cross Hospital and Leslie Henderson, University of Hertfordshire) we attempted to define the nature of this attention- demanding process. In one study, we showed that the Parkinsonian disadvantage for simple RT did not appear to be due to an inability to maintain focused attention on a single, expected stimulus. In a further study, in response to inconsistencies in the literature, we used a model-driven approach in order to isolate a subtype of Parkinsonian subjects with neurological signs of frontal lobe dysfunction who reliably exhibit a selective simple RT prolongation. In a further paper, we refuted the widely held notion that in Parkinson's disease, the simple RT deficit and the impairment shown on predictable tracking tasks stem from the same underlying disorder of preparation (2.37).
FUTURE PROPOSALS
A. Visual Attention and Extinction (Goodrich)
A1. Perceptual Grouping and Selective Attention: We will continue with experiments examining the relationship between (i) perceptual organisation and selection, and (ii) non-perceptual task organisation and selection. In the first category of studies, we wish to explore precisely which object properties produce an advantage for selection, and therefore, presumably, which object properties are being analysed before and/or during the process of selection. This would involve comparing rates of extinction while manipulating lower-level variables (such as whether the items formed good gestalts) as well as higher-level variables (such as the meaningful nature of the objects).
A2. Non-Perceptual Task Variables and Selective Attention: In the second category of studies, we will explore the nature of the task variables which produces the 'anti-extinction' effect. For example, in one study (with Rob Rogers, University of Cambridge) we are comparing the effects of internal versus external task priming on the 'anti-extinction' effect.
B. Identifying Factors which Reduce Unilateral Neglect (Goodrich)
Links have been established with the physiotherapy department, Hinchingbrooke Hospital, Huntingdon. We (with Driver) will offer supervision and ideas for research projects which would be carried out by the physiotherapists. Our proposed interventions are all driven by theories of normal attentional function. This collaboration has great potential since a direct contact will be established between experimental psychology (and laboratory based studies) and rehabilitation studies in hospital settings. Among a range of ideas for studies are the following:
B1. Intermodality Interactions: We will continue exploring the rehabilitative implications of intermodality interactions by setting up series of studies in which we attract the patient's attention to the impaired side in one modality by stimulation in another. Tactile stimulation appears to be promising; since the patient is generally surrounded by sights and sounds on all sides, vision and audition might be less efficacious. By contrast, an abrupt tactile stimulus would rarely have to compete with comparable stimulation on the ipsilesional side.
B2. Ipsilesional Capture in Unilateral Neglect: Further work will be carried out exploring the notion of ipsilateral capture and rehabilitation in a more realistic setting than in the shape studies described above. For instance, on hospital wards, the general trend is to place the patients' possessions to the contralesional side in the hope of facilitating leftward exploration. We shall not only place key information on the contralesional side of space, but also reduce the degree of ipsilesional stimulation. We will test patients performing everyday activities with and without blank screens placed on their ipsilesional side.
B3. Axis Based Visual Neglect: No future plans at present.
B4. Attentional Load and Unilateral Neglect: The degree of neglect shown by patients appears to vary with their activity. For example, neglect appears more pronounced in walking than in sitting. One interpretation is that when 'attentional load' is greater, less capacity remains to utilise strategies designed to overcome neglect (e.g. consciously scanning to the left). We will ask patients to perform a standard test of neglect in sitting and standing (and walking if possible). If neglect is greater in standing, a possible rehabilitation strategy is "WALK-STOP-LOOK-WALK-STOP-LOOK".
CONSCIOUSNESS (Marcel)
Introduction
Two main themes occupy the current investigations: (i) the unity and fractionation of consciousness; (ii) the extension of the study of consciousness to bodily experience and sensation (and to emotional experience). The approach taken in many of the studies combines theoretical and applied issues and interrelates clinical and normal populations. The applied issues include the diagnosis and rehabilitation of neurological patients and the understanding of effects of general anaesthetic in surgery. The clinical populations, while mostly neurological, also include and extend to psychiatric and affective disorders.
A. Unity of Consciousness
Most theoretical and empirical studies assume a unity of consciousness. Unity of consciousness can mean three things: (i) that there is no real distinction between first-order phenomenal experience and second-order reflexive or introspective consciousness (the source of report), or at least that the former is transparent to the latter; (ii) that each is singular and indivisible; (iii) that the contents of consciousness are integrated by mutual co-reference at any one time, and therefore that there cannot be (unnoticed) contradictions of experience. The following studies cast doubt on these assumptions and suggest (a) a real division between first- and second-order consciousness, (b) that one or both levels may be non-unitary, (c) that there can be failure of access between first- and second-order levels, and (d) that simultaneous contradictions in experience which go unnoticed do occur.
A1. Influence of Mode of Report (Marcel): Psychophysical studies with a blindsight subject and normal subjects showed essentially the same phenomenon (2.147). Subjects were asked either to guess the presence/absence of a threshold visual stimulus or to report their awareness of such presence/absence. In each condition they responded concurrently by three means: blinking, button press with finger, vocal. Accuracy was higher when guessing than when reporting experience, which suggests (according to current conventions) that report data reflect consciousness; and the guessing condition showed no dissociation between the response modes. When reporting, however, the different response modes were often in contradiction. Especially when responding fast, subjects were unaware of these self-contradictions. Conditions with each response mode tested singly and with different response delays show the same pattern. The account that is most parsimonious and yet applicable to other phenomena is that second-order reflexive consciousness may be split, in this case according to response mode, and that separate divisions may differ in their access to first-order phenomenal experience. This conceptualisation has major implications both for the analysis of psychological data, clinical and normal, and for long-standing philosophical assumptions. It is backed up by and applicable to the other studies reported in this section. Further studies, some started, will look at other stimulus modalities and the role of "stimulus-response compatibility" (see Proposals for Future Work).
A2. Unawareness of Sensory and Motor Loss: The single most predictive index of neuropsychological recovery and rehabilitative efficacy is awareness of the deficit, which is often lacking (anosognosia) in the acute phase after trauma. In addition, lack of awareness of certain types of deficit has important theoretical implications for aspects of self consciousness. Despite these two tributes to its importance, this is a largely unexplored area. In a joint British-Swedish study, we have attempted to study anosognosia for hemiplegia and hemianaesthesia in stroke patients. We have developed a successful combination of structured interview and tests for an exploratory group study, and the data indicate how it can be refined as a clinical and research tool (2.147). Results indicate that anosognosia is (i) unrelated to confusion, (ii) can be independent of degree of awareness of other deficits, sensory or psychological, (iii) is related to but not accounted for by sensory loss, (iv) is related to time since lesion (acute-chronic), (v) is unrelated to unilateral personal neglect. In addition, the study has revealed several intriguing phenomena. (i) Importantly for clinical purposes, patients may be aware of having a plegia but quite unaware of its specific everyday consequences. (ii) A number of patients show evidence of being simultaneously aware and unaware of their specific disabilities. First, a number of patients are unaware of their inability on bimanual tasks if asked "In your present state how well could you ...?", but are accurate if asked in the form "If I was in your present state, how well could I ...?". Second, some patients are unaware if asked "Is this arm (the completely plegic arm) weak?" but are fully aware if asked in the form "Is this arm ever naughty? Does it ever not do what you want ?". Apart from the interpretation that self-knowledge is sensitive to the description under which it is addressed, this suggests a kind of split in reflexive introspective consciousness. Indeed, an extension of this characterisation may be made to spatial neglect in general (see discussions of neglect in this report by Duncan, Robertson, Goodrich), where a number of studies suggest that, rather than being unaware of neglected stimuli, the patient is aware but "dissociated" from them.
A3. Unawareness of Blindness with Blindsight: One of the most intriguing, though very rare, instances of unawareness of a deficit is Anton's Syndrome, unawareness of blindness. I have been able to study such a patient in whom a head injury bilaterally impacted the occipital poles. Although he showed some mild frontal effects, the patient was not clinically confused. Also he was fully aware of his slight memory impairments and of his physical state (broken legs). Further, while he was aware, from failures in visually based behaviour, that he had some visual problems, and acknowledged frustration at this, he was not aware of being blind. Therefore his unawareness is unlikely to be an intentional denial to preserve self-esteem. His adequate visual imagery performance suggests no problem with internally generated visual experience. The impairment thus appears to be a selective failure of introspective access to lack of visual experience of the world. A highly important aspect of this case is that, while being blind on conventional direct or explicit testing, the patient appears to have Blindsight when tested appropriately. He can accurately point at and grasp objects when forced to do so rapidly. While performing at chance in making conventional confident binary judgements of extremes of luminance, he performs nearly perfectly in the same situation when asked to guess luminance level. The presence of blindsight in Anton's syndrome has never been tested until now. This double lack of awareness (loss of visual experience with preserved aspects of visual processing, and unawareness of that loss) once again suggests two levels of consciousness, and suggests that the way that we know many of our sensory and cognitive capacities is by reflexive access to specific types of phenomenal experience (seeing, feeling, recognising). In addition, this is one of the first cases (if not the first) of bilateral blindsight to be studied, which I propose to continue with the patient's willingness.
A4. Measures of Nonconscious Perception: Two approaches to measuring nonconscious perception of masked visual stimuli were investigated. According to the first, nonconscious perception is demonstrated when indirect measures (priming) are more sensitive than direct measures (forced choice discrimination accuracy). Various direct and indirect measures were compared to investigate their potential nonequivalence, and the effect of this on experimental results. Indirect measures were not consistently more sensitive than direct ones. This does not, however, imply absence of nonconscious perception, since two direct measures, forced-choice discrimination accuracy and subjective awareness of the relevant information, may be dissociated. The second approach was therefore to investigate this dissociation. Traditional criticisms of introspective report as an index of awareness were overcome by progressively more detailed questions addressing awareness of different aspects of masked stimuli. This employed trial-by-trial ratings and post-block questionnaires, but did not depend on absolute accuracy. Evidence for nonconscious perception was found in some tasks (semantic categorisation) but not in others (stimulus detection). "Passivity" of the perceiver seems an important determinant of obtaining effects of nonconscious perception. The approach provides an interpretative tool for analysing the relationship between subjective experience and performance in a variety of situations (2.176).
A5. Emotional Experience and Consciousness: An area relatively neglected by both students of consciousness and students of emotion is emotional experience. Lambie has extensively reviewed data on emotional experience and its constitutive, phenomenal and formal properties from normal, clinical and cross-cultural literatures, and has placed this in the context of theories of emotion and of consciousness. Empirical dissociations of emotional experience from awareness of other emotion components (e.g. bodily sensations) cast doubt on most existing theories. The account emerging from this study is compatible with the distinction drawn above between non-reflexive and reflexive consciousness, and implies certain kinds of pre-conscious mechanisms, e.g., attention and attribution.
A6. Summary: The above-mentioned studies, together with preliminary work on general anaesthesia (section A4, Future Proposals), can be unified in their implications. They suggest, in addition to nonconscious processing, not only a separation of first-order awareness from second-order reflexive awareness, but also the existence of dissociative states or splits in the latter, usually discussed in the context of hypnosis and post-traumatic stress. Traditionally, the condition most often suggested to reflect dual consciousness is that of split-brain (commissurotomised) patients. In a discussion paper (2.148) I have criticised some of the criteria currently used to adduce disunified or partially unified consciousness in these patients. The distinction between personal and subpersonal levels is not co-extensive with conscious and nonconscious, but is rather a distinction between levels of explanatory discourse. However the personal level is linked to consciousness in its recourse to concepts such as intention, self-reference and subjective experience. In a review paper (2.116) I have discussed how the personal level manifests itself in neuropsychological syndromes. The conclusion is that, both for explanation and rehabilitation, neuropsychology cannot successfully reduce to the level of sub-personal mechanism but must also take account of personal-level structures.
B. Bodily Experience and Sensation
Theoretical and empirical work on consciousness have paid little attention to bodily experience. Yet it is arguably central to any (non-dualist) sense of self, as well as to understanding our emotional and spatial experience. Several issues are addressed in the current work. (i) What underlies our bodily experience? While a Helmholtzian approach (nonconscious inference) is more easily seen to be applicable to perceptual consciousness of the external world, it would also appear that it is applicable to the experience of bodily sensation, i.e., the result of nonconscious inference on the basis of a spatial representation. (ii) What is the nature of the deficit in neurologically based losses of awareness of the body? (iii) What is the relationship of bodily representation in sensory experience and in action?
In addition to the empirical work referred to below, some of the context for that research is captured in theoretical work and reviews. One approach focusses on the nonconscious basis of conscious experience of our bodies and of sensation, reviewing experimental psychology and physiology, neurological syndromes, clinical (affective) research and cross-cultural studies. Our bodily experience is suggested to be the result of (i) a nonconscious structural representation (of the body as distinct from the rest of the spatial world), (ii) inference and attribution, (iii) culturally relative somatic symbolisation of self, (iv) spatially selective attention. In addition, a distinction is drawn between (a) the experienced space of sensations and of the egocentric disposition of body parts (Body Image), versus (b) the space of body parts as targets and obstacles for movements of other body parts (Body Schema). A second paper (2.117) argues that bodily experience and sensations are (explanatorily) secondary to spatial cognition and self. (i) Bodily sensations experienced as such (e.g., pressure on object from finger experienced as pain in finger as opposed to experience of mass and location of the object) depend on attention to oneself as a spatial entity, and therefore depend on a spatial self-world distinction. (ii) In different cultures, whether certain physical and emotional situations are experienced as bodily sensation or as purely mental is the result of the degree of somatization of self. This existential sense of "self" is a social construct. Much of this theoretical work has taken place in the context of an Interdisciplinary Project on Spatial Representation based at King's College Cambridge. Two edited volumes have resulted from this project. In the second, "The Body and the Self" (2.2), a jointly written introductory chapter (2.97) examines the importance of the body in the light of developmental, neurological, cognitive and philosophical work and issues.
B1. Role of Body Image and Nonconscious Inference in Bodily Experience - Intersensory Illusions and Dissociations
B1.1. Touch and Vision: The sensation of single or double tactile stimulation (the 2-point threshold) has usually been thought of as due only to, and the direct reflection of, density of innervation at the skin. In a series of experiments (2.77) double, disparate retinal images of a stimulated finger were produced by subjects focussing visual attention on a display two metres beyond the finger. In this situation, although requested to attend primarily to touch, subjects experience double tactile sensation to a single touch, the effect diminishing in subjects with greater ocular dominance. It is suggested that one's conscious access to a modality (touch) is not direct but via a level of representation (body image) affected by other modalities (e.g. vision). This is paralleled by other sensory interactions, e.g. visual-vestibular in experience of self-motion. This work is now completed.
B1.2. Location and Shape of Body Parts: Replications and extensions have been carried out of the effects of vibrotactile muscular stimulation (biceps or triceps). (a) If the forearm is held in a rest during stimulation, it feels to have moved in a direction opposite to that in which it would have moved if left free. If the finger is in contact with another part of the body, then that part of the body is felt to have grown or receded appropriately, e.g. the nose feels to be growing. This latter effect is difficult to account for without the notion of nonconscious inference. (b) If subjects cannot see the stimulated arm, their verbal reports of the position of the hand of the stimulated arm reflect the primary vibratory illusion. However, if subjects are asked to quickly grasp the hand with their other hand, they do so correctly. If the grasp is delayed, subjects are more likely to reach to the illusory position. This is interpreted in terms of the distinction between a nonconscious Body Schema, whose spatial representation of body parts modulates immediate spatial coordination of the body, and a conscious proprioceptive Body Image of the space of sensations and body part disposition. Body space is coded quite differently in these two representations. If a body-directed movement is delayed, the specification of target location becomes increasingly subject to its coding in the conscious Body Image.
B2. Somatosensory and Motor Imagery and Experience of Manual Posture (Marcel, Bisiach, Johns, Berti, Ladavàs, Pizzamiglio): The context for these studies is that (a) patients with unilateral personal neglect are sometimes thought to have lost part of the representation of their bodies; (b) patients with hemiplegia or hemianaesthesia (unilateral loss of movement or sensation) due to stroke might be thought to be unable to have somatosensory or kinaesthetic imagery for the affected body parts (on the basis of the idea that perception and imagery share common representational bases). In the first part of our studies, latency patterns suggested that normal subjects identify rotated photographs of left and right hands via rotation of perceptual and motor images respectively for dorsal and palmar views (cf. Shepard & Cooper, 1982). This was confirmed by the effect on latency patterns of selective interference tasks (e.g. maintenance of pressure by one hand in a fixed position). The same patterns of data were obtained when the task was given to right-brain-damaged patients with and without hemiplegia, with and without hemianaesthesia, with and without unilateral personal/bodily neglect. The patients showed no difference in data patterns for photos of left and right hands, but when the same interference task as in normals was applied to their unaffected hand, they showed the same selective interference effects as normal subjects, i.e. patterns of identification latency for the affected hand were uninfluenced. The results suggest (i) that patients who seem phenomenally unaware of a limb, or who lack movement or sensation in it, can nonetheless manipulate a motor or perceptual image of that limb; (ii) that in somatic and motor imagery, just as in visual imagery, spatial representation and processing should be distinguished from phenomenal experience; (iii) that bodily neglect does not involve loss of the representation of the neglected body parts. At the time of writing, publication awaits completion of testing of control groups of brain-damaged patients. The data on normal subjects have been written up by Marcel and Johns.
B3. Unawareness of Sensory and Motor Loss (Marcel, Tegnèr): The work reported in section A2 as bearing on the fractionation of consciousness is clearly also highly relevant to bodily experience. Taken together with the study of pain in general anaesthesia (section A4. of Future Proposals), the apparent "split consciousness" suggests that bodily sensations (or their lack) may have phenomenal status even if we are not attending to or aware of them. The immediate practical implication of this work is that the actions that may be attempted by plegic or anaesthetic stroke patients are not constrained by their reflexive awareness of their sensory or motor capabilities, i.e. general awareness of deficit does not prevent patients attempting impossible actions. Therefore extra care is required in rehabilitation before early discharge.
C. Motor Control - Influence of Intention on Accessibility and Fluency of Action (Marcel)
In Ideomotor Apraxia, actions which cannot be produced on request or by imitation are often produced or improved when performed as a component of normal activities. We have explored the extent of occurrence of such phenomena in a range of neurological patients, one application being identification of appropriate eliciting conditions for therapy. Patients showing the phenomenon were filmed in a variety of situations. Dual-task conditions (appropriate to each patient's impaired behaviour) suggested that, in general, the performance difference is not due to "over-attention". When an activity was induced of which the relevant behaviour was a component, improvement was obtained in all cases. Unexpectedly, a further significant improvement was observed when the action had social or personal significance (e.g. (i) lifting and moving a cylinder; (ii) drinking a tumbler of liquid; (iii) offering mugs of tea to guests). These findings suggest that motor control is strongly influenced by the nature of the intention. This should be exploited where possible in choosing appropriate situations for eliciting and retraining of motor acts. This work (2.116) is completed and is not being continued.
PROPOSALS FOR FUTURE WORK
A. Unity of Consciousness
A1. Influence of Mode of Report - Psychophysical Studies in Normal Subjects: The first set of studies on the influence of Mode of Report (see A1. above) all used visual stimuli. It is important to examine the effect of e.g. tactile and auditory-verbal stimuli, in order to see if the observed differential sensitivity of response modes is in any way due to stimulus-response compatibility. Since selective response-mode preparation may also play a part in differential sensitivity, it is important to examine conditions where the required response-mode or order of response-modes is indicated by a post-stimulus signal.
A2. Unawareness of Sensory and Motor Loss: There are two extensions planned for the work on anosognosia for hemiplegia and hemianaesthesia.
A2.1 The degree of interest shown in this work means that it would be useful to develop our battery of tests into a clinical and research tool. For the former purpose it requires shortening, since at present it takes at least 40 minutes to administer. Our current analyses are already aimed at selecting the most interesting tests and questions. For the latter purpose, in addition to selecting the most significant tests and questions, we need to standardise and validate to some degree. Since our control groups consisted of normal subjects of different age-ranges, some of this work is already done.
A2.2 We plan to carry out detailed single case and experimental studies of stroke patients anosognosic for hemiplegia and hemianaesthesia who show dissociations of awareness of their state. The phenomena indicated in section A2 suggest that such dissociations are linked to aspects of selfhood and to social and emotional aspects of questioning. We envisage enlisting clinical psychologists to aid design of studies to illuminate these phenomena and to provide rehabilitative heuristics.
A3. Bilateral Blindsight: The patient with both unawareness for blindness and blindsight has now recovered awareness of his phenomenal blindness. However, he is now left with bilateral blindsight. This is an important case to study since no study of blindsight has yet been carried out on a case with full visual field loss. Scattered light can no longer be appealed to for explanation. More importantly, several well-established blindsight phenomena may depend on the visual experience from remaining fully intact parts of the visual field. For example my own (2.78) and other researchers' studies have shown awareness of static stimuli in the blind field provided that appropriate stimuli are presented simultaneously to the sighted field (usually an intact hemi-field). Would the same effects be obtained if both fields are phenomenally blind but with blindsight?
A4. Awareness of Pain in General Anaesthesia: Certain techniques used in short gynaecological operations provide a rare opportunity to study consciousness under general anaesthetic. When the fallopian tubes are clipped by laproscopy, occasionally three conditions obtain: (a) the patient is tilted slightly head-down so that the viscera can be pushed to rest on the diaphragm while the abdomen is inflated; (b) the musculature is neither paralysed nor fully relaxed, reflex movements being used by the operating team; (c) when possible, the patient is not artificially ventilated. Under such conditions it is sometimes possible (physically) for the patient to talk. In pilot studies at the John Radcliffe Hospital, Oxford, we have attempted to ask such patients during the operation if they experience pain, and, if so, where. Three important phenomena have emerged. Even when patients show no general signs of awareness, while many patients make no response, some (c. 40-50%) do respond vocally. While some of those responding report no pain, others "correctly" report pain in the chest. Since very few know how the operation is conducted they would hardly expect pain in the chest. Most intriguingly, a few patients who at first report no pain, do so when asked a further question derived from studies of hypnotic analgesia : "Does any part of you experience any pain ?". A small number of women have responded "She has a pain in her chest (and/or groin/stomach)". This response, especially use of the third-person (she), suggests a dissociative state of split consciousness, and that the functional effect of certain types and levels of general anaesthetic is similar to that of hypnosis. It must be emphasised that these are as yet pilot studies. Parametric studies of anaesthetic level and type as well as of behavioural and physiological indices are clearly needed.
A5. Theoretical Work on Self-Consciousness: Following the success of the Interdisciplinary Project on Spatial Representation (1989-93), a group of philosophers and psychologists intends to collaborate on a similar venture on Foundations of Self-Consciousness. Among other themes, this project will focus on the relation between bodily awareness and self-consciousness, on foundations of autobiography as transtemporal selfhood, and on social and emotional bases of self-consciousness. The British Academy have shown interest in supporting such a project. At the moment the core collaborators include Marcel, Naomi Eilan, John Campbell, William Brewer, John O'Keefe and Josef Perner. However a number of other scientists and philosophers have shown a keen interest.
B. Bodily Experience and Sensation
B1. Role of Body Image
B1.1 Location and Shape of Body Parts: The different effects of vibrotactile stimulation on the experienced location of a body part and on its location as a target for movement were interpreted in section B1.2 (current work) to suggest different spatial coding for sensation and motor control. This can also be seen as a difference between the conscious and nonconscious spatial aspects of intentions for action, e.g. location versus movement paths. When the current work is completed, I propose to investigate this as follows. The stimulated arm is visually occluded by a horizontal board on which light-emitting diodes are distributed. After initial vibrotactile stimulation, illumination of one L.E.D. indicates a target location. The subject either (a) has to move hand of the the stimulated arm to a position beneath the target or (b) has to draw the direction and extent of required movement of that hand with the other hand. Degree of influence of the sensory illusion and dissociation of the two measures should distinguish the sensory-versus-motor hypothesis from the intention-coding hypothesis.
B1.2 Role of Body Image - Structural Divisions: A new line of work is planned on this topic. Several types of evidence suggest that a nonconscious structural description of the body underlies the spatiality of experienced bodily sensations. It is of relevance to explaining certain phenomena whether such a description explicitly represents structural segmentations of the body, e.g. divisions at mechanical articulations. Two types of study are proposed. In the first, it is predicted that sensory discriminations will be greater across putative segment boundaries than within segments (equal separations of tactile stimuli across the wrist versus within the forearm or hand). In the second type of study, it is predicted that tactile priming or tactile aftereffects will be more effective within a segment than for equal separations across segment boundaries. Such effects would cast doubt on certain theories of spatial attention in bodily sensation (e.g. Kinsbourne, in press) which postulate no representation of body space as distinct from external space. They would also illuminate certain topological phenomena in peripherally demyelinized patients (see B2.2 below).
B2. Demyelinization and Proprioception (new work): Patients who, through peripheral neuropathy, are deprived of almost all afferent information carried by fast, myelinated nerves provide a unique opportunity to study several issues applicable to normal people. The main patient to be studied is IW (already partly studied) who from the neck down is assumed to have no kinesthetic or proprioceptive information and has lost all corresponding (conscious) sensation except deep pain and temperature at the surface. Several other patients have been traced abroad and one (GL, Montreal) has been partially studied.
B2.1 Tactile Sensation and Kinaesthesis: Absent, Reduced, or Nonconscious?: At present the only psychophysical studies of sensation carried out on I.W. have been by report. We intend to use forced-choice techniques. We have started this and already found evidence of some somatic "blindsight". This is important since otherwise it remains a mystery how IW has retaught himself to walk and reach without constant visual guidance. It also suggests additional different psychological roles for fast and slow afferent pathways similar to those for cortical and tectal pathways in vision.
B2.2 Location of Sensation in Body Parts: IW knows proprioceptively in which part of his body a pain or temperature stimulus is but does not know where the body-part is in egocentric space. However, preliminary studies suggest that this experienced location of sensation is only topographic (in the forearm versus hand) but not determinate (where in the forearm). It is possible that it is dynamic information (which he lacks) that provides such determinate information, a view suggested by pre-existing studies by Brouchon and Paillard. This work will involve experiments on IW and on normal subjects involving location of sensation at different time delays after passive and active movement.
B2.3 Coding of Location of Stimuli in Peri-Personal Space: Since IW is sensitive to temperature, we can produce non-visible somatic stimuli (heat spots) on a surface. Comparison of his learning of (a) the egocentric locations and (b) configural relationship of such stimuli when he can see his arm movements and when he cannot should help to illuminate the relationship between topokinetic and morphokinetic spatial coding.
B2.4 Distinguishing Self-Movement from World-Movement: IW's pain receptors provide a non-painful sensitivity to certain types of texture (e.g. stiff bristles of a scrubbing brush). We can thus assess the role of reafference in distinguishing one's own from the world's movements by comparing three conditions (blindfolded): (i) self-generated movements across an appropriately textured surface (stiff brush bristles), (ii) self-generated movements with no relative movement of tactile texture, (iii) passive movement of his hand across the surface, (iv) movement of the surface across his hand held steady by the same means used to move it in the second condition. This provides a new way to test the Von Holst and Mittelstaedt account of perception of self-movement (comparison of reafference with motor copy), and to evaluate the relative contributions of corollary discharge, movement reafference and tactile flow-field.
B2.5 Balance: IW used to fall over if he did not visually monitor himself or focus on a vertical surface. He no longer has this problem. Use of Wing's balance board in different attentional conditions may identify the mechanism or strategy he has evolved in his self-rehabilitation, and thus help in the physiotherapy of similar patients.
NATURALISTIC STUDIES OF ACTION SLIPS (Sellen)
Introduction
Slips can be modelled as a dissociation between a conscious, intentional system and unconscious, automated action. The characterisation of slips as "liberated automatisms" has benefitted from collections of naturalistic errors from which taxonomies of error have been derived (e.g., Norman, 1981; Reason & Mycielska, 1982). Sellen's theoretical work on action slips has also focused on the analysis and categorisation of naturalistic error corpora, collected mainly through diary studies and other methods of self-report (summarised, in part, in Sellen & Norman, 1992). The first area which Sellen's work has addressed is that of detection of everyday action slips. The second area is more applied and specialised in nature, focusing on slips committed by anaesthetists.
A. Detection of Action Slips
The evaluation of action and the detection of slips have been largely ignored in the psychological literature. In an initial exploration of the possible underlying cognitive mechanisms (2.60), a corpus of over 500 everyday slips and mistakes was collected by asking subjects to keep a diary, focusing on the details of error detection and identification. The errors constituted a diverse set, from slips at the sensorimotor level, to memory lapses, to errors in judgement. The tasks involved were "everyday" tasks and were thus also diverse, spanning a wide range of human daily activities.
On the basis of this collection, a theoretical taxonomy of detection modes was proposed, consisting of three broad categories: Action-Based detection, Outcome-Based detection, and detection through Limiting Functions. Action-based detection describes errors detected by comparing incipient action, executed action, or memory of action to an internal reference. Errors detected on the basis of outcome describes detection which results from observation of the consequences of action as they are manifest in the external world. Detection by a Limiting Function describes errors detected because constraints in the external world prevent further action. Each category or "mode" of detection has implications for the kind of feedback used in realising an error, and in the reference against which the feedback is compared. The taxonomy provides a general, descriptive framework for understanding these various mechanisms, and the cognitive levels at which they operate.
B. Errors in Anaesthesia
The psychology of human error was applied to the problem of errors in complex tasks, such as the job of an anaesthetist (2.59, 2.62). In collaboration with anaesthetists at the Royal Adelaide Hospital in South Australia, work began by helping in the re-design of a form for anaesthetists to provide anonymous report of "incidents" which occurred on the job. An incident was defined as any accident or error, whether or not it could have been prevented, and whether or not it had detrimental effects. This study, called the Australian Incident Monitoring Study (AIMS), has so far resulted in a corpus of over 3,000 incident reports.
Initial analysis of these reports identified a range of factors which contributed to the incidents, from workplace policies and procedures, to equipment and product design. The nature of the errors that actively contributed to the incidents were also classified according to whether there were problems in knowledge and judgement, misapplication of rules, or slips due to habit or inattention.
The next phase of the research involved the initial classification of the incidents into subsets of interest. One such subset was defined as those incidents in which drugs were inadvertently swapped. Analysis of the causal and contributing factors underlying "wrong-drug" errors identified distraction and haste as the two most common factors (2.16). Closer analysis of the errors suggested that most of the incidents were in fact due to simple slips, as opposed to errors resulting from bad judgement or inexperience, and an initial approach was given for the prevention and management of such slips.
FUTURE PLANS
Work on analysis of the AIMS data has only begun. Dr. John Russell, an anaesthetist at the Royal Adelaide Hospital, will be visiting the APU in the summer of 1994 in order to carry out further analysis and classification of the AIMS data. We will be collaborating with Prof. John Senders, an expert on human error, and Dr. David Duthie, an anaesthetist at Papworth Hospital. Roy Patterson from the APU will also be helping to analyse those incidents in which auditory warnings were reported to be a contributing factor. The hope is that we will be able, as a team, to outline a variety of practical ways in which errors can be prevented or managed, from the redesign of anaesthetic equipment, to suggestions for the implementation of hospital procedures and policies. In the course of this analysis, we also hope to learn more about the nature of human error in complex tasks. Errors in anaesthesia have much in common with errors in other kinds of situations, such as nuclear power plants and aeroplane cockpits. One focus of the research, about which little is currently known, will be "collaborative errors", or errors in which more than one person is involved.
In addition, we are planning a study at Papworth Hospital with eight trainee cardio-thoracic anaesthetists. The trainees will be given palm-top computing devices into which they will be asked to enter a log of anaesthetic incidents that they observe or participate in, over the course of two weeks. This should provide a more complete record of the errors that occur in the anaesthetic domain, and a more accurate estimate of their relative frequency than AIMS can deliver. We hope that the study will also begin to raise awareness of potential errors in this field, and will lead to the establishment in Britain of a reporting system similar to AIMS.
COGNITIVE CONTROL AND INTEGRATED MENTAL ACTIVITY IN COMPLEX TASKS (Barnard, Blandford, May)
Introduction
If we are to develop forms of cognitive theory that are readily applicable in the context of complex tasks, these theories must have a number of capabilities. First, they must, of course, adequately capture and predict regularities in cognitive behaviour. Second, theories that provide an integrated viewpoint would be of significant benefit. They avoid the problem of having to deploy different theories and techniques to analyse the many issues that arise in a typical applied context. Third, there must be a practical means of communicating and delivering the theory into that applied context. Fourth, the specific part played by cognitive theory must be clarified -- particularly in interdisciplinary contexts where theories and concepts of fundamentally different types all have "something to say" about how issues should be resolved.
For this area, the development of applicable cognitive theory is based upon the Interacting Cognitive Subsystems framework (ICS). In the previous reporting period (1986-89), ICS was applied to human-computer interaction in the context of simple command and menu based systems. The research emphasis was on the acquisition by novice users of the conceptual structure and lexical content of the dialogue tasks. Over the course of the last five years, the emphasis has been upon extending the scope and applicability of the ICS approach. One extension has been to provide a more systematic treatment of learning that goes beyond the novice user. Another key extension has been an examination of visually rich settings, including multimedia and multimodal forms of interaction. Both have contributed substantially to the development of integrated theory.
The basic extensions to the approach have largely been accomplished with intermittent internal research assistance (A. Green, Tweedie & Scott) or by graduate students (Duff & Lee); the consideration of theory application in an interdisciplinary context has had more substantial post-doctoral support through two successive rounds of Esprit Basic Research funding covering the period 1989-1995. This project, AMODEUS , has supported the work of A. Green, Blandford & May for this area and that of Whittington & Blandford (see R. Young's section of this programme). The project has been scientifically and financially co-ordinated by Barnard. It has supported around £2M of international research effort covering collaborations between two dozen scientists working in seven countries. Its core aims have been to develop modelling techniques; to examine how their outputs might best be integrated; and to consider the transferability of the techniques to real design contexts.
A. Cognitive Control in the Longer Term Development of Knowledge and Performance
Some empirical work on longer term learning was conducted prior to the current reporting period. That work has now been both summarised (2.120) and extended in a number of new directions. Much laboratory research on human-computer interaction has been "over-focused" upon the performance of novices. The work reported below goes some way towards the provision of a more balanced focus of effort. The unifying theme has been theoretical and empirical work examining the use of different forms of mental representation in complex tasks as skill develops from novice, through intermediate, to expert status.
Duff initially examined how novices and expert users dealt with the problem of using an unfamiliar digital watch, observing that the two groups resolved the uncertainties of what action to perform in rather different ways. He went on to manipulate the form of knowledge that might help to reduce uncertainty in action specification in a range of device control tasks (2.161). These included exploratory learning, learning based upon sequential procedures, and learning based upon the provision of propositional knowledge about the way the device worked. These all produced different patterns of performance and different forms of transfer when new tasks were introduced (2.19, 2.94; Duff, 1989). Duff was able to make use of ICS, and its associated analysis of uncertainty within propositional representations, to make the counter-intuitive prediction that the provision of more knowledge does not necessarily lead to a greater or more persevering reduction in uncertainty - a prediction which was supported by his data (2.161). As a part of the AMODEUS project, A. Green also made direct use of ICS and its derived analysis of the resolution of item and order uncertainties. Green took a series of specific examples from classic work on novice and expert problem solving performance, in the domains of physics and mathematics, and related them to a series of examples from the HCI literature, including the experiments of Duff. In an encouraging demonstration of the potential generalisability of theoretical concepts derived from HCI research, Green showed that both forms of "paradigm" were open to a common theoretical understanding (2.31).
Lee (2.171) developed several paradigms for studying the development of knowledge and performance in interactive tasks. In a broadly similar approach to that taken by Duff, Lee compared the methods used by novice, intermediate and expert users of a commercial word-processing system. Each developed a characteristic, and often stable, pattern of method utilisation. The most efficient functions were frequently under-utilised. In an ingenious experimental manipulation, Lee showed that providing knowledge of how functions worked was not in itself sufficient to change the patterns of usage. However, providing users with knowledge of how to reformulate the problem to make its representation compatible with their mental representation of the computer function did cause users to shift to a more efficient pattern of function utilisation (2.44). Once again the specific form of representation appears to have a crucial effect on what is learned and the dynamics of its processing. In a series of studies with strongly visual (menu) interactions, Lee showed that users acquire specific forms of knowledge that reflect the uncertainties they have had to resolve in the development of task representations (2.45). Following leads in the literature on the causes of error, Lee was also able to show (2.43, 2.171) that certain forms of "action slip" have a high initial incidence which reduces as users acquire more experience. However, with even greater experience the incidence of the same type of errors actually increases again. Such a change may well reflect a qualitative shift in the pattern of cognitive control from one with a focus on the acquisition of the interface task itself, to a pattern of cognitive control more typical of a focus on the underlying domain of work (in this case, stock market trading).
B. The Representation of Visually Rich Environments and Dynamic Deployment of Attention in Complex Tasks
Most theoretical contributions in the area of human-computer interaction stress the importance of some form of task representation. Although there have been modest extensions to task based approaches to help deal with visually rich interactional settings, this tends to be accomplished by adding features to the task representation. Interacting Cognitive Subsystems assumes distinct types of mental representation. It is assumed that task representations are generally held in a semantic form, but that these are processed in conjunction with two visually based codes: a raw visual code and a higher level structural description of items in an object-based code. Since the processing of visual scenes and the deployment of attention is fundamentally dynamic, an analysis was developed based upon an analogy with thematic processing of language. Halliday's systemic grammar was taken as a point of departure. Entities in a visual scene are assumed to have a psychological subject-and-predicate structure associated with them. The mental representations constructed and manipulated then reflect the form and content of the information processed from the scene on the basis of thematic transitions among superordinate, basic and constituent units of object representations (2.51).
This approach enabled the modelling of the deployment of visual attention to be brought together at an object-based level with both bottom-up influences of the visual scene and top-down influences of the task. Initial experiments (2.32) suggested that search time in an array was related to: (i) the complexity of the predicate structures; (ii) the number of items in the scene at a given level of structural decomposition that shared the same psychological subject (similarity); and (iii) the extent to which the task and setting forced thematic changes in dynamic processing that required the encoding of particular elements of the structure. People do not need, for example, to encode the visual details of items that are always presented in the same physical location. Quite detailed predictions based upon this model have been supported by the results of further experiments in icon search tasks (2.52). The form of analysis is by no means restricted to simple search tasks. It has now been applied to several concrete design issues in complex displays in many different applications (e.g. see 2.50, 2.79, 2.149, 2.154, 2.155). The potential for inter-relating task-based properties to the use that visual information is put also reinforces the scientific message from the work of Wright (detailed under the Language and Communication Programme).
The work on structural descriptions of visual form was initiated by May & Barnard during the first AMODEUS project. During the second project, it has been extended to multimedia and multimodal settings for interactive performance. As a part of the requirements for European reporting, detailed papers have already been produced on the application of the analysis to film theory and dynamic video-clips as well as the cross-modal blending of data streams involving vision, dynamic gestures and voice (e.g., see 2.151, 2.152 which include 2.84 and 2.173). The results of initial experiments are also encouraging. In a setting which experimentally simulated key aspects of some computer-supported work environments, video sequences were shown of an actor making deictic reference (hand gestures) to a task-based array of items. Consistent with earlier generalisations, performance in this complex setting depends upon the extent to which the gestures eliminate uncertainty of item identity and location (2.69).
C. Approximate Modelling of Cognitive Activity and its Application
A large proportion of the cumulative empirical work carried out by Barnard and his colleagues has been oriented towards the longer term theoretical objective of developing modelling tools. The core idea is that production system technology, embodying psychological principles, could be used to generate approximate models that describe mental activity in complex tasks. These theoretical descriptions could then underpin the processes of both scientific prediction and design analysis. At the end of the previous reporting period the general concepts had been outlined and a number of local examples implemented in a running expert system (reviewed again in 2.81 and 2.82). ICS formed the theoretical basis of approximations that described basic configurations of mental processes; the procedural knowledge embodied in those processes; the memory records used; and a characterisation of dynamic cognitive control. Descriptions in this attribute space were generated by explicit rules and were themselves operated on by explicit rules to generate predictions or design advice (see Barnard et al. 1988).
For the first AMODEUS project, a target was set to generate a modeller capable of dealing with a broad range of settings. The technical work of specifying the detailed theoretical constructs needed for this implementation was extensive (2.84). It involved not only the basic theory, but also how best to scope the approximations over conceptual phases of mental activity (goal formation, action specification and action execution) and longer term learning (novice, intermediate & expert). Naturally, the specification drew upon much of the work reported above and in previous APU progress reports on this topic. By the end of the project a running modeller was produced embodying several hundred rules. The software and its associated documentation and evidence of performance (2.153) was delivered to the CEC as a scientific deliverable (associated with 2.154). The final system was capable of addressing seven areas including experimental topics covered in the last progress report (lexical command and task structures); experimental work covered in the previous two sections (2.19, 2.32); and a range of practical scenarios drawn from HCI design problems (e.g. 2.51). Key developments in the modeller, not previously reported in the literature, and the underpinnings of its generalisability have now made their way through to journal publication (2.50). Further work on re-implementing the modeller is currently underway (see 2.152).
D. The Application of Modelling in Design and Interdisciplinary Contexts
A central objective of the second AMODEUS project has been to investigate how theoretical techniques might best be transferred to, and deployed in, the context of design, software development and usability evaluation. A number of different strands of research have again been pursued.
Much emphasis has been placed on using modelling techniques in the context of concrete design scenarios. This itself involves ensuring that theoretical analyses have relevance and appropriate scope for application -- itself a matter of some methodological debate to which APU has recently contributed (2.66, 2.122 see also R Young's section of this programme). May & Barnard (2.79) have drawn together a number of scenario analyses to illustrate how ICS-based concepts of approximate modelling can provide conceptual support for the resolution of practical designs and their evaluation. As a part of the AMODEUS project, the problems of transferring both ideas and the specific tool have been the subject of systematic research. For obvious reasons of maintaining independence, these studies of transfer and tool use have been conducted by our collaborators rather than ourselves. Journal publications by others are about to appear detailing both the general nature of the transfer problems associated with ICS-based modelling as well as issues associated with the specific use of the expert system. So, for example, Buckingham-Shum & Hammond (in press) have recently reported a study in which they had human factors students attempt to apply the expert system tool, developed in the first AMODEUS project, to a specific design problem within its scope. They found that the students could make use of the tool without necessarily having a deep understanding of the underlying ICS theory. It was nonetheless clear that effective deployment did rely upon the tool user knowing all about the allied structural description techniques for tasks and displays.
As with most of the other areas of APU's work dealing with attention and cognitive control, the theoretical ideas and tools outlined here must be put to use in an interdisciplinary context. Interdisciplinary work is challenging because different disciplines make use of different terminologies, have different bases for their theoretical ideas and assign very different priorities to the resolution of specific points in a given problem space. In the specific context of human-computer interaction research, there are cognitive resource-based models, knowledge-based cognitive models, embedded user models, abstract task models, domain models, design space representations, formal models of the computer system, structured human factors methodologies, heuristic evaluation and so on. Quite clearly, they cannot all be used to analyse all issues. The problem of context sensitive selection of tools and techniques itself needs to be addressed, as does the allied problem of ensuring that the owners and advocates of individual approaches actually understand one another.
A new approach to these issues is being developed by Barnard, Blandford and Harrison. The approach makes use of formal logic to specify abstract properties of interactions among agents operating in some specified domain of application. The agents may be of any type - human or computer. As with conversational analysis, the focus of the modelling is upon the conjoint sequential behaviour of the relevant agents. Interactional events are defined, as are the constraints upon their sequential organisation. Such constraints may deal with properties like detours in trajectories, the disengagement of individual agents from the interaction, or issues of interactional focus (2.70, 2.85, 2.142, 2.143). This provides a means of defining design requirements - not individually for the system component or the user component but upon the whole "interactional trajectory". The requirements are initially specified at a high level of abstraction, and as they are successively "refined", the properties of the refinement highlight specific needs for detailed models and evaluation techniques.
The cognitive implications of these interactional analyses can be very direct. So for example, section A.1 above referred to an action "slip". An example is that of a computer user attempting to perform an action in an unselected window. Interactional analysis of this setting indicated that the transitions between users' cognitive states and the corresponding system states (how the display indicates that a window is active) were not reflected in the interactional trajectory. A principle was therefore proposed to mark interactional transitions by separate "interactional" events. This was then refined to the levels of user and system properties. The idea was to make the window border have a "dynamic or fizzing" quality at readily definable transitions - when users or system appeared to disengage from continuous interaction or change focus within it. Experiments incorporating this way of governing dialogue transitions produced dramatic reductions in the rate of action slips (2.43, 2.124). Furthermore, it was not simply a matter of the fizzy border having more attention grabbing properties throughout. People do adapt to the continuous presence of such properties. Empirical tests have confirmed that, to have its full effect, the property must be active only during an interactional transition (see Ch. 4 of 2.171).
An unanticipated research thread has emerged through this interdisciplinary collaboration. The above analysis works via a top-down specification of the requirements on the interaction. The initial assumption was that when refinement reached a point where the properties of interest dealt solely with a cognitive or system agent, then further analysis would rightly lie in the province of a specific form of cognitive or system model. However, using our more informal definitions and specifications (2.84), the formal modellers of computer systems, on their own initiative, have recently sought to develop specifications of the ICS model of the human information processing system. This has been done by modelling key ICS axioms using a deontic extension of modal action logic. In this way properties of a cognitive model can be specified in the same terms as the model of the computer system. In principle, this should support bottom-up inferences concerning the properties of interactions (2.162, constituent part of the D7 scientific deliverable associated with 2.152). In the longer term this development has far wider potential. It offers the exciting prospect of being able to provide precise means of establishing the consequences of cognitive theory without using simulation.
E. Getting the Message Across
As with previous periods, considerable effort has gone into disseminating and summarising our research for different purposes. One is to present longer chapters for the research community to draw out issues that cannot easily be expanded upon in individual research papers (2.81, 2.118, 2.120). Another is to provide tutorial material for non-specialists in research. One overview was initially directed at human factors students (2.82), was then incorporated into a book chapter designed for electrical and software engineers (2.80) and reprinted a third time in a book oriented to the large US market for computer science students and HCI professionals. In interdisciplinary and multinational research projects the provision of short summaries assumes particular significance (2.179). A series of summaries of AMODEUS cover well in excess of 200 technical reports, software artifacts and publications (2.123, 2.150, 2.151, 2.152, 2.154, 2.155). A book has been edited with the specific aim of bringing together basic and applied material for an interdisciplinary audience in the European computing and telecommunications industries (2.3).
Two new developments are noteworthy. First, Barnard, May and Tweedie have been developing interactive technologies for presenting information. They have explored combinations of textual, animation and video technologies for communicating the products of research. Animation techniques have been developed for explaining dynamic information flow in the ICS model (2.177). Coupled with interactive presentation technologies, the resulting software has been used to support public talks. The basic technologies and techniques have been described in 2.179. The same techniques were used by May (2.172) to demonstrate dynamic perceptual phenomena and theoretical explanations. This interactive document is available in electronic form over the internet - which reflects the second major development in information dissemination (anonymous File Transfer Protocol - or FTP).
The AMODEUS project now makes the bulk of its non-copyright research material available electronically through anonymous FTP. In the first 18 months of its operation, some 100 documents have been made available in this way and 2,835 copies of them have been taken electronically by 139 sites from 25 different countries. Whilst these new developments will not draw effort away from the production of traditional refereed contributions to the archival literature, it will certainly enhance the scope and rapidity of information exchange and impact. Unlike academic journals, the network can do much to make research material easily accessible by commercial sites.
FUTURE PROPOSALS
Introduction
Over the course of the last three years, Barnard has begun to alter the balance of his research interests (see the work reported and proposed under the Cognition and Emotion programme). Since the mid seventies, the specific topic of human-computer interaction (HCI) has played a pivotal role in his research. It has provided tightly constrained topics for empirical and theoretical enquiry. It has also, of course, provided a more or less continuous supply of additional resources in the form of five research grants. Barnard's approach has involved a blend of theoretical, AI, and experimental work in the context of leading edge technological applications. This particular blend can now only really be sustained within large collaborative interdisciplinary teams. As a part of the re-balancing of effort, Barnard will discontinue managerial involvement in large scale externally funded projects in HCI. The dependence of the research programme upon direct links to leading edge technology and the application of psychological theory in this context will be reduced accordingly.
Three proposals are itemised below. These draw upon the same types of techniques as previous work and it is intended to continue work on the same general areas of the programme on attention and cognitive control. The proposals are no longer organised around external projects in HCI but upon the development and application of the Interacting Cognitive Subsystems framework (ICS). The first proposal nevertheless covers the consolidation of prior work on HCI. The second covers two related approaches to the specification of cognitive theory. One involves the embodiment of theory in AI systems for reasoning about cognitive activity and the other will explore the use of formal methods to specify abstract properties of ICS. The final proposal encompasses three topics for empirical investigations. One topic concerns the relationship between central functions and the dynamic processing of complex visual scenes, including the allocation of attention within such scenes and the use of memory resources to connect scenes. A second topic concentrates upon cross modal blending of visual and auditory information. The third topic is focused directly on unpacking cognitive control aspects of central executive functions - realised within the ICS approach as reciprocal processing of schematic and propositional meaning. These three topics share two underlying themes. They are all directed at examining exchanges among different forms of mental representation within the ICS framework and hence the integration and control of cognitive activity. Within that overall pattern of activity, all three topics are concerned, in one way or another, with the definition, and selection, of coherent streams of mental data.
A. Consolidation of HCI as a Case Study in Applied Cognitive Psychology (Barnard, Scott, May)
Much of Barnard's work on human computer-interaction has been carried out in the context of externally funded projects. The AMODEUS project is scheduled to submit its final report in the summer of 1995. Accordingly, substantial effort will have to be devoted to reporting current work throughout the first half of the next reporting period.
Barnard has for sometime been working on a larger synthesis treating most of his previous work in HCI as case study in the development and application of applied cognitive theory. The outline synthesis was presented as an invited address to the BPS Cognitive Section Meeting in 1993. A research monograph has been in preparation for sometime, and a large proportion of the material is already summarised in the form of interactive text and demonstrations (2.177). The end of the AMODEUS project should see the finalisation of a few remaining empirical, conceptual and demonstrational links. The existing monograph material will then be revised and completed.
Work in progress should also enable us to complete some tutorial aids for students of human-computer-interaction. The existing multimedia software has already been used on the MSc course in HCI at University College, London. Over the course of the next eighteen months, the software will be revised to allow students to use it individually without a tutor. It will also be supplemented with specific instructional material on the analysis of mental representations, and further worked examples of analyses carried out in AMODEUS on concrete design problems. Some of this work may be carried forward through a long-standing collaboration with Jørgensen in the psychology department at Copenhagen University. It is important to note that what is being proposed is not a major software project. We currently have the multimedia tools and expertise in their use (Jon May), to produce revisions and updates very quickly.
B. The Specification of Cognitive Theory (Barnard, May)
An enduring set of issues for psychology revolves around the problem of specifying theory. Much interesting and important theory still relies upon verbal presentation and relatively informal means of arriving at a prediction. Other theory can be very explicit - calling upon mathematical modelling techniques, or symbolic, connectionist or hybrid forms of simulation. Many such paradigms are both widely applicable and very powerful. However, tractable simulation usually involves a restriction in scope and requires additional assumptions about knowledge representations to be bound into the simulation. Typically these additions are not in themselves part of the formal theory. As Fox et al. (1992) have pointed out, this makes it hard to separate out the contribution of the core theory to the behaviour of the simulation. They have stimulated new debate by advocating a role for abstract specification of the core theory and its functional separation from additional assumptions. As we move toward more complex, multiprocessor, architectures based upon distributed underlying processing capabilities, these problems become even more severe. The two proposals listed below reflect similar concerns to those of Fox et al., but are pursued by different means. The objective in both cases is to provide means of making theory explicit in an abstract but unambiguous form, while enabling it to be scoped appropriately to generate predictions in specific circumstances.
B1. Rule-Based Approximate Modelling: The kind of rule-based modelling (section C of past work above), is based upon a theoretically driven approximation of the functioning of the ICS multiprocessor architecture. Although not a simulation, the required theory is abstracted into explicit rules for reasoning about cognitive activity. It is also appropriately scoped in that the rules are also explicitly laid out (a) for establishing task and other external conditions and (b) for the conditions in the approximations for which particular predictions about behaviour apply. The UK product in which our past work was implemented has recently been discontinued. In this context and that of the alteration of Barnard's research interests, it is proposed to start a complete re-implementation of the theoretical modelling rules in a new environment, but this time, scoped for the prediction of the results of laboratory experiments. Initially, it is proposed to concentrate on those issues in the deployment of visual attention and multimodal blending that are described below. This particular proposal requires the full time support of a post-doc and is heavily dependent upon having access to skills in both AI and ICS. Jon May has both set of skills. Accordingly funds will be sought to support his further involvement. Failing that, and with great reluctance, this particular proposal will be dropped.
B2. Formal Methods and Human Cognition: In section D above, it was noted that a group of computer scientists have become interested in the formal specification of the conjoint behaviour of users and computer systems. They use the same form of logic to represent abstract properties of both a computer interface and the human interacting with it, and then seek to model the behaviour of a complete "system" composed of these constituent agents. Armed with an appropriate schema for the discharge of a proof, this could form a basis for formal theory-based reasoning about cognition. In this context, it is interesting to note that the approach has two really interesting features. First, it is an appropriate level of abstraction for representing "core theoretical claims", yet is technically capable of refinement into increasingly detailed claims. Second, the approach has a built in means of scoping the theoretical claims about cognition. In order to model the interaction, it is necessary to model the salient features of both parties. In this case the system model and the interaction must represent the relevant aspects of the environment in which cognition is occurring and the tasks it supports. In technical sentiment, this could not be further removed from Suchman's (1987) approach to situated cognition. Nonetheless, broadly similar issues and problems may be addressed precisely.
At present, it is unclear whether formal proofs can meaningfully be discharged in this context. By the end of the AMODEUS project, sufficient specification work will have been done for ICS and system models to exist for a number of very specific cases in multimedia communication and air traffic control. We propose to use this material as a basis for exploring the issues raised in greater detail.
C. Empirical Proposals (Barnard, Scott)
C1. Central Functions and the Dynamic Processing of Complex Visual Scenes: Our general approach (work completed, section B) assumes that the dynamic deployment of visual attention can usefully be analysed by considering how the most central processes within the ICS framework control thematic transitions in structural descriptions of visual scenes and the objects they contain. The deployment of attention is related to the number of objects in a scene sharing a common psychological subject description and the depth to which predicate evaluation must proceed to ensure target discrimination (2.52). Depth is not, however, a unitary dimension. The same number of items, with the same degree of depth, can be embedded within markedly different constituent organisations (e.g. object grouping). We propose to extend our recent experiments on icon search by testing the hypothesis that intermediate structures must be processed in attentional transitions which cross levels (up or down) in the structural description of visual scenes. This will be accomplished by systematically manipulating (a) the way in which constituents at various levels are grouped together and (b) the relationship of the target's description to those of the surrounding constituents.
Classical searches for targets in arrays represent only one form of attentional transition. In computer displays, theatre and film, the change of scene or objects within a scene can be either continuous (panning; scrolling of information in a computer window) or discrete (cuts; opening new window environments). When discrete changes are made, only some seem to support thematic continuity in scene processing. Using alternative "film cuts" in short computer controlled animations (c.f. Frith & Robson, 1975; 2.173), we propose to investigate how thematic continuity is maintained by systematic substitutions of individual subject and predicate components presented within the current visual field.
C2. Cross Modal Integration of Information: Work on thematic continuity in the processing visual scenes provides one way of examining how central mechanisms interact with incoming streams of information. Information from different modalities (or sources) is systematically correlated in well known ways. Lip movements are correlated with voice sounds and such correlations can lead to well known blending effects with cue conflict (e.g. McGurk & MacDonald, 1976) or specific attentional effects in infants (Bahrick, 1988). Similarly, expressed meaning may be systematically related to concurrent facial expression, tone of voice or physical gestures. Within the ICS framework the possibilities for information flow among subsystems are clearly defined and constrained such that different forms of information blend at different levels within the subsystem organisation. Tone of voice and facial expression are integrated at the level of schematic rather than propositional meaning (Teasdale & Barnard, 1993) whereas information derived from lip movement and speech blend at the level of structural descriptions of speech form. Concurrent verbal and deictic reference will be resolved at the level of propositional meaning and so on.
It is hypothesised that the blending of information derived from different modalities or sources is determined by the properties, and the arrival timing, of the current psychological subject at any level of encoding within the ICS architecture. If these move outside limits determined by the particular mental code involved, then thematic coherence cannot be established and two data streams will be established. According to the model, only one stream at a time can then be selected for further processing. A well known instance is when a soundtrack goes slightly out of synchrony with a film. Like the concept of perceptual centres, cross-modal synchrony must rest on internal attributes of the data streams, rather than everything starting and stopping at the same time. If perceptual centres can be found in any data stream and if cross modal information is to be attributed to the same source or event, these different perceptual centres must be aligned, or synchronous. Initially, we propose to investigate cross modal blending by studying the internal properties of verbal and visual data streams, manipulating relevant properties of form and timing within computer controlled displays. Video-recordings of real faces and more coarse grained animations of faces should be subject to different influences. The high quality images of faces should provide conditions for phonological cue conflict. Coarse representations in animated facial characters should be associated with higher level attributional blending at a propositional level of representation. The relevant psychological subject and predicate structures at these levels should differ and thus the acceptability of thematic transitions should be influenced by different parameters.
C3. Unpacking Cognitive Control Aspects of Central Executive Functions (collaboration proposed with Fowler at the Department of Psychology Fulbourn Hospital): The previous two experimental topics both focused upon rather different relationships between central processing activity and incoming data streams. As a part of a recent theoretical exercise, Barnard (1994) has developed a set of inter-related proposals concerning the analysis of central executive functioning within Interacting Cognitive Subsystems. Central executive functions are implemented in terms of processing exchanges between propositional and schematic levels of meaning. One part of the analysis addresses the issue of what might happen if the temporal synchrony of the exchanges between propositional and schematic levels were to break down. It is hypothesised that a breakdown in the synchrony of exchanges between propositional and schematic levels would lead directly to a separation of information that should be correlated within the same data steam. In an analogous way to the separation of lip movements and soundtrack, the individual would be "aware" of two distinct data streams. Only one would be predicated to the sense of self - and the other could potentially come to form the basis of some delusional system about alternative sources of agency to which "voices in the head" may become attributed.
These particular proposals about asynchrony within central executive functioning bear some similarities to Frith's (1992) information processing approach to understanding schizophrenia. Fowler & Barnard intend to explore the asynchrony hypothesis by monitoring a population of schizophrenics over time. Although delusional systems may become stable and assume a chronic schematic pattern, the asynchrony should be most marked when a patient is "in episode" compared to periods when "not in episode". It is proposed to monitor their performance on specific tests requiring schematic evaluation - such as Shallice's cognitive estimates test, and tasks involving the generation of random numbers within defined instructional constraint. Pilot work on this latter tasks indicates that slips comparable to failure on cognitive estimates occur at a significant rate - even in the normal population. It is also proposed to use the case grammar techniques, outlined in the cognition and emotion programme, to study schematic properties of self and agency within this particular population and to assess whether or not such properties shift systematically as patients move in and out of episode.
COMPLEX COGNITION IN LEARNING AND USING INTERACTIVE DEVICES (Blandford, Howes, Whittington, Young)
Introduction
This area of research is aimed at understanding, through modelling, cognition in tasks more complex than those typically studied by laboratory investigations in Cognitive Psychology. The central underlying theme is that detailed computational theories and models have much to offer in helping to illuminate human cognition, but that, in order to be helpful, those theories and models have to be developed in ways sensitive to both the psychological and the computational constraints.
Most of our research has exploited the existence of the cognitive architecture called Soar. Soar is a problem-solving architecture constructed as a nested set of problem spaces. Problem solving consists of the repeated application of an operator to a state in a problem space to yield a new state. When an impasse occurs which blocks this process from continuing, Soar automatically sets up a subgoal with its own problem space, the purpose of which is to resolve the impasse and allow processing to resume in the original space. Closely associated with this goal-subgoal structure is a mechanism for learning by chunking, which stores the results of processing in a subgoal as new rules, so that in future occurrences of a similar situation, the rules will apply and the impasse will be avoided. Soar has its origins among attempts in Cognitive Science to build what are now being referred to as "Integrated Intelligent Architectures", information-processing organisations which combine two or more of the basic components of AI, such as planning, knowledge representation, vision, etc. In his 1990 book on Unified Theories of Cognition, Newell lays out the Soar account of cognition, and makes the case for theories which integrate diverse findings from different areas of psychology into a coherent whole, and for Soar as a candidate for such a theory. Soar is useful in our investigation of cognition because, among other reasons:
• Soar's emphasis on unification provides a good basis on which to build models that integrate different aspects of cognition in the performance of a complex task, as is explained below.
• The close intertwining of learning and problem solving in Soar makes it feasible to include learning within the phenomena covered by models of cognitive skills, in a way that until recently was not possible.
• Soar's being a strongly constrained architecture provides some much-needed help in alleviating the problem of "theoretical degrees of freedom". Instead of, as in other approaches, having the theoretician simply posit a model for a task, with Soar it becomes more a matter of discovering how Soar would do the task. This is what Newell calls "listening to the architecture".
Soar investigators constitute a distributed research community located at a dozen or so sites in the USA and Europe. The existence and responsiveness of that community (e.g. to technical queries or scientific debate conducted over email) has been of great help to our research. In 1992/93 Young and Howes spent an academic year at Carnegie Mellon University, Pittsburgh, Pennsylvania, working at one of the principal Soar sites. We have contributed to making the case for Soar as a theory of cognition (2.119), and to debate about its theoretical status. In collaboration with Frank Ritter of Nottingham University, Young has been developing a tutorial introduction to cognitive modelling in Soar, which has now been presented several times in both half-day and full-day versions.
A. Towards an Integrated Model of Learning and Performance (Young, Howes)
Much of our research has been done as part of two consecutive projects funded by the Joint Councils Initiative in Cognitive Science and Human-Computer Interaction. The aim of the projects has been to develop models of users of interactive devices within an appropriate cognitive architecture, in a way that both advances the art of user modelling and also contributes to a fundamental understanding of the cognition involved in performing realistically complex tasks. The second project is still continuing, while the first was assessed after the final report as being very good (with one of the three external referees awarding it an outstanding).
Our work on the first project began from the observation that even experts' skill at using an interactive device, such as an Apple Macintosh personal computer, is display-based and situated: It is evoked in the context of use, and utilises information from the display, but is not necessarily accessible to the user away from the device. Drawing on a prior analysis (Howes & Payne, 1990a) which extends the formalism of "task-action grammars" to handle aspects of display-based skill, we examined what would be required for a Soar model to acquire the task-action mapping rules for a display-based task. The result was the Task-Action Learner model (TAL). "Task-action mapping" refers to the sequence of physical actions (moving the mouse, clicking a button, …) required to carry out a unit task on the device. The model simulates a person learning to use an interface by trying to perform tasks, guided where necessary by advice from an external tutor, and reveals how the mappings from simple tasks to actions are dependent upon the detailed dynamics of the device's display (2.40, 2.75; Howes, 1992). It exhibits at least four empirically observed phenomena of HCI:
• It learns consistent interfaces more easily than inconsistent ones. For example, it needs to consult the tutor less often in cases where the methods for using the device are consistent with the model's semantic categories.
• It speeds up with practice, in part by learning the location of particular items on the display.
• It is sensitive to the actions afforded by the device display, so that highly interactive devices are easier to learn.
• It remains dependent upon the display to cue its memory for methods, as do even very skilled users.
The TAL model exhibits at least two kinds of interactivity: (1) The repertoire of low-level actions it considers at each step is determined by the affordances offered by the device display. (2) It relies upon information presented by the device to flesh out the details of an action, which can therefore be represented in a general semantic form to be instantiated appropriately in a range of different task situations. This work on learning task-actions mappings extends our earlier research on Programmable User Models in HCI by having the analyst instruct the model rather than program it, and by assessing the learnability of an interface rather than its plain usability.
Because it depends on advice, TAL performs guided learning, while people often engage in genuinely exploratory learning. To understand this ability, we again drew on some of our earlier work (2.54, 2.55; Howes & Payne, 1990b; 1990c) and extended it to a new model (2.144). It is becoming clear that recognition knowledge (as against, e.g., recall knowledge) can by itself make a major contribution to guiding the user during exploration of an unfamiliar interface. Such knowledge is easier to acquire, and is "cheap" in the sense that it is needed anyway to support the learning of the more complex knowledge required for control or recall. The model of exploratory learning uses heuristics based on recognition knowledge together with control knowledge which is learned as it becomes available. The model (2.39) exhibits behaviour known as final-first learning, with these characteristics:
• It learns by exploration through interacting with the device.
• It learns the later parts of the sequence first.
• It speeds up with practice.
• It acquires display-based (or "situated") knowledge.
The latter two characteristics are shared with TAL, though the speed-ups occur for different reasons. TAL becomes faster with practice because it avoids the task decomposition and automatises the control decisions that it initially makes deliberately. The exploratory model performs the task faster because it learns the menu structure.
Stepping back from the individual models, we have also made a preliminary attempt to articulate the space of models used in modelling cognition within Human-Computer Interaction. The work is a precursor to the development of an integrated model (see below, under Future Work) (2.114, 2.115). The intention is to provide a theoretically motivated and conceptually consistent overview of the literature on cognitive modelling.
B. Approximate Modelling of Cognition and its Application (Young, Blandford, Whittington)
Some of our work has been done as part of the AMODEUS project, a large, multidisciplinary, European collaborative project which has been co-ordinated by P J Barnard and is described more fully in his section of this Progress Report.
B1. Programmable User Models: Within the context of the Amodeus project, our approach builds on the earlier idea of "Programmable User Models" (PUMs) developed under an Alvey project. This approach postulates that designers can gain insight into the problems of usability by attempting to instruct an architecture with human-like constraints how to perform tasks with a device corresponding to the design they have developed or plan to implement. Initially we focussed on the planning component of the PUM, developing a scheme for making means-ends analysis available ubiquitously in Soar (2.178). When applied to a simple HCI situation, this yielded predictions of a class of conceptual errors (2.68), and also generated a small space of possible behaviours. Some of these behaviours do the task "correctly", i.e. as intended by the interface designer, some achieve the task but do so inefficiently, and some lead to irrecoverable error.
We analysed the knowledge needed by users to perform a given task with a proposed device — such as opening a document within an application running on an Apple Macintosh personal computer (2.67) — and how that knowledge can be expressed in an Instruction Language and compiled into rules for the Soar architecture (2.5). Starting from what the designer intends as the correct sequence to accomplish the task, we analyse what knowledge the user needs in order to know to do each step at the right time, and to know how to do the step. This knowledge is then expressed in the Instruction Language, compiled into rules, and run as a cognitive model. As the model runs, it acquires new rules which proceduralise the performance of the task. That procedure, reconstructed through this cycle of analysis and synthesis, may or may not correspond to what the designer originally specified, and in either case can be revealing of issues concerning the usability of the proposed design.
B2. Analysis of Concrete Scenarios: As noted in Barnard's contribution to this programme of the Progress Report, we have continued to develop a method for assessing cognitive (and other) modelling techniques by applying them to concrete behavioural and design scenarios. In addition to the methodological aspects (2.66, 2.122), we have shown how our approach explains a persistent error committed by people who use multi-window computer screens, in which they type input intended for one window whilst another window is still active (2.64).
B3. Design Rationale: In collaboration with partners at Rank Xerox Cambridge EuroPARC, Young has worked on the development of an analysis and associated notation for representing design rationale. Normally when some object or device is designed, all that is left at the end is a blueprint or the artifact itself. All the reasoning, argumentation, exploration, and justification that went into its design gets lost. The lack of that information makes subsequent processes of maintenance, modification, re-design, and explanation much harder and less efficient, and makes it difficult to re-use parts of the design for other purposes. Our technique, known as design space analysis, attempts to redress those defects, and aims to represent why a design is the way it is (2.46, 2.48). The notation centres on the use of Design Questions, Options, and Criteria, and provides an explicit representation for part of the design space surrounding the artifact. Considerations of usability, and more general psychological aspects, enter mainly as criteria, or as justifications supporting the assessment of design options against criteria. Our approach has been influential within the general growth of interest in design rationale. It is used for various purposes within Amodeus (2.146), for example to analyse the role of analogy and metaphor within interface design (2.47).
C. Devices and Agents
Some work on people's understanding of complex devices draws its inspiration from Young's research in the early 1980s on users' mental models of pocket calculators. (See also 2.56) Elizabeth Churchill, the holder of an MRC Collaborative Research Studentship with Rank Xerox EuroPARC, showed both by modelling (2.126) and empirically (2.160) how different instructions to calculator users, even where these lead to the same overt performance on the training material, yield different internal representations of the skill which generate observable differences on diverse tasks. She also showed that in using such devices, people form only shallow, local plans which specify in detail only the immediate next steps, with the direction of future steps sketched only loosely and left to be determined, opportunistically and interactively, by the behaviour of the device. Working with another (non-APU) graduate student, John Bradshaw, we showed how adding an explicit depiction of purpose to the kinds of representations used in Artificial Intelligence and Knowledge Engineering for dealing with complex devices, enables us to facilitate reasoning about the device for the purposes of design, or diagnosing faults (Bradshaw & Young, in press). Typical applications of Bradshaw's work are to the understanding of central heating systems (2.6) or the triply-redundant hydraulic system of the BAe 146 aircraft (2.60).
Since joining the APU, Blandford has maintained an involvement with research on agent architectures for interaction and collaboration, drawing on her PhD thesis work on the design and implementation of an Intelligent Educational System to support the teaching of decision making in design (2.4). The focus of this work is on the architecture of the agent, and the knowledge it needs to enable it to engage in a mixed-initiative dialogue with a user and solve problems collaboratively (through a process of negotiation) (Blandford, in press, a). As well as being equipped to deal with knowledge in a non-prescriptive way, the agent takes account of motivational factors in deciding what to say to the user. The agent architecture is based on an 'action cycle' which governs changes to the internal state of the agent, causing it to acquire goals and become committed to actions on the basis of its beliefs about the current context (Blandford, 1993a). The relevant aspects of context include mutual commitments (what the system and user have agreed to do together), open topics of conversation and the agent's current beliefs (including those about the user's beliefs). The agent is largely opportunistic, and does not engage in advance planning. In deciding what to say or do to satisfy its current goals, it selects the action which it believes will best satisfy its values while also contributing to satisfying its (achievable) goals. In the current implementation, actions include deciding what to say or do, saying something, doing something (generally updating information in a spreadsheet which supports the problem solving) and receiving and understanding input from the user. Values include, for example, keeping the interaction varied and purposeful, encouraging the user to reflect on their beliefs, and doing the minimum (Blandford, in press, b); this agent (implicitly) derives pleasure from having the user learn well and expending the minimum of effort! Results of an informal evaluation of the current implementation are encouraging (Blandford, 1993b). Future work will include developing fuller knowledge structures to support richer interactions and testing the same basic architecture in different domains.
FUTURE PROPOSALS
Our future work falls into two parts, both of which are concerned with various kinds of integration.
A. Integrating Different Aspects of Cognition
Current theories in cognitive psychology offer only fragmented accounts of human learning and performance in a real-life domain such as Human-Computer Interaction (HCI). Models derived directly from the results of experimental psychology focus primarily on just a single aspect of cognition (e.g. perception, motor control, planning, or problem solving). But because many everyday activities — such as those involved in using a technological device, for example withdrawing money from a bank cash machine, or discovering how to use an unfamiliar software application on a familiar computer like an Apple Macintosh — draw on many aspects of cognition at the same time, such fragmentary models are either too narrow and fail to predict important phenomena, or else are too loosely scoped and make incorrect predictions through being applied outside their proper domain. One of the advantages of carrying out research on Cognitive Science in an applied area such as HCI is the opportunity it provides to redress this shortcoming, by turning the problem around, focussing on the application, and drawing in the cognitive capabilities needed for handling the domain. The challenge is to offer an integrated account of learning and performance in such a task. There are major theoretical and technical difficulties to be overcome in providing such an account, but the way has been pointed (and inspiration provided) by the initial proposals for "integrated cognitive architectures" such as Soar, and Anderson's ACT-R. Our own research with Soar over the past few years — for example on planning in the context of HCI, and on exploratory learning — has made enough progress to confirm its potential, but most of the work remains to be done.
A1. Modelling Cognitive Capabilities: Our approach, being pursued in the second round of funding from the Joint Councils Initiative, and to be continued beyond, has two parts. First, we focus on the provision of various generalised cognitive capabilities. For several of them, we already have some understanding of how the capability can be modelled. Examples are:
• Mini-planning, by which we mean a limited capability to perform simple problem-solving and planning. As already mentioned, the planner makes general use of means-ends analysis.
• Interactivity, here meaning the user's exploitation of, and reliance upon, information and affordances offered by the device, and therefore shared by the user and the system. We have already explained how the TAL model exhibits at least two kinds of interactivity.
• Acquiring task-action mappings. As already described, TAL models the learning of task-action mappings, and their generalisation over semantic categories of the task.
• Rapid initial learning refers to the dramatic speed-up with practice observed over the first few trials on a task. The speed-up has been demonstrated most clearly in the case of TAL, but it occurs equally with means-ends analysis, and indeed is ubiquitous in Soar-based models.
• Exploratory learning has been described at length above.For the kind of integrated model we are aiming at, further capabilities are needed. Examples are:
• Goal representation. The present models have no representation of what they are trying to accomplish when performing a task. They have an encoding of the task, but no notion of what that means in the domain. Giving a model a representation of what it is trying to accomplish will overcome two of the limitations of TAL. First, the model will be able to recognise when it has reached a goal or subgoal. (At present, TAL has to be told explicitly when it has finished each level of a task.) Second, the model will be in a position to assess the correctness and quality of the advice or guidance it receives.
• Variety of knowledge sources. An integrated model needs to be able to utilise a range of different kinds of knowledge: lexical knowledge, semantic knowledge, display knowledge, locational knowledge, knowledge of methods, domain knowledge, and so on.
• Visually-driven processing. Much of the use of interactive devices is driven strongly by the information presented on the display. Existing Soar models do not reflect this dependence at all well. TAL, for instance, employs a notion of visual frames, but although helpful, this is inadequate for capturing typical cases of visual layout. We are encouraged by how far our colleagues Barnard and May have been able to go with the (non-Soar) modelling of visually-guided behaviour by employing a plain hierarchical representation of visual structure combined with simple rules for processing that structure (e.g. 2.51). We will, of course, not attempt to model the low- and medium-levels of visual perception that parse the raw image into a structured visual description. Our interest is not in that process of parsing, but in the interpretation and use of the resulting parse. The requirement is for a model to be able to combine knowledge of the task with a description of the visual layout in order to guide its further problem-solving and behaviour.
A2. Combining Capabilities: Second, for the integration itself, the main challenge is posed by the observation that models implemented independently, even in Soar, inevitably make certain assumptions or pre-empt certain decisions in ways that make it difficult to combine one capability with another. So the methodology of this part of the research takes the form of an iterative cycle in which we repeatedly:
(a) Analyse existing approaches (our own and others') for providing the relevant cognitive capabilities, in order to uncover the core of each approach.
(b) Discover a means to realise each capability in a way that respects and exploits the Soar cognitive architecture, whilst remaining compatible with the other capabilities as well as with a broad range of empirical properties of cognition.
Several of our existing models provide illustrations of early steps towards integration, especially in that they combine learning and performance. Howes' model of exploratory learning, cited earlier (2.39), is one example, where the final performance, of being able to perform a set of tasks fluently, depends upon learning the hierarchical menu structure. Another is an initial investigation into modelling the use of deliberate analogy in performing and learning a new task. With colleagues at the University of Colorado at Boulder, we constructed models in Soar and in Anderson's ACT-R for a simple analogy task, that of inferring how to start a new program on an Apple Macintosh given that the user already knows how to start some other program (2.57). (The comparison between the two architectures was instructive.) The more interesting properties of the final performance Soar model — such as that there is positive transfer to subsequent instances of analogy, even though the mapping rule learned in each case is specific to the given task — depend crucially upon the process by which it was learned.
Within a couple of years, we expect to be able to model the performance of a user who draws on a variety of sources of knowledge and deploys a range of tactics in order to learn to cope with an unfamiliar software application. The model would explore and learn relevant parts of the command structure of the application, use simple lexical knowledge in relation to the semantics of the task it is trying to perform, apply knowledge it has of other related applications in order to make "informed guesses", and when necessary simply try things out to see what happens. Starting from this initial stage of uncertainty and trial-and-error, the model will with experience become "expert" at a range of tasks performed with the new application, and will be in a position to use its newly acquired knowledge of the application to support the learning of yet others.
A3. Accounting for Detailed Data: Having such a model will put us in a position to relate the models to empirical data in a different way than we have done so far. At present, the models are constrained by data in the form of empirically supported regularities, such as those mentioned earlier: that people learn consistent interfaces faster than inconsistent ones, that they speed up with practice, that even experts remain dependent upon the display, and so on. The integrated model will make it possible to perform comparisons with more detailed data, such as studies of individual learning trajectories and experiments chosen to probe particular aspects of the account.
A4. Implications for Basic Cognition: Because of its integrational nature, the modelling based on the Soar cognitive architecture often has implications for basic 'laboratory' phenomena of lasting concern in cognitive psychology, many of which seem to cry out for detailed modelling: resource-limited cognition (which is dealt with at length elsewhere in this Report, e.g. by Barnard); implicit and explicit learning; dual-task performance (again, dealt with in detail elsewhere, e.g. by Duncan); and so on. Soar offers a distinctive theoretical perspective on such phenomena, and when successful, that kind of modelling yields both an integrational account of the phenomenon in question and also serves to modify and further specify Soar as a psychological architecture. For example, to the currently confused distinction between 'implicit' and 'explicit' cognition (which is rapidly threatening to become one of those diffuse ideas in psychology which mean something different to each investigator), Soar brings to bear two relevant contrasts: (1) All permanent knowledge is encoded as content-sensitive pattern rules in LTM, and is not directly accessible to awareness, whereas dynamic information about the situation being dealt with is encoded as data in working memory (WM) and is available to awareness (though it may or may not be verbalisable, depending upon the nature of its content); (2) Rules which work directly to modify the current state correspond to automatic processing, and are not under the voluntary control of the person, whereas rules which modify the state via the current operator correspond to deliberate processing, and are under the person's control. The cognitive activity in the performance of any actual task results from the interaction among these different sources and ways of deploying knowledge. Consider what happens in a task where the correct action depends upon making distinctions not initially known to the subject — as is the case in experiments requiring the control of complex systems, the learning of artificial grammars, and so on. One possibility is for the subject more-or-less passively to let the experience of different cases accumulate. The kind of performance that will result will depend greatly on the precise circumstances of the task, but in any case the resulting ability will take the form of non-articulated skill, distributed across the rules in LTM — one clear sense of "implicit" knowledge. Alternatively, the subject may be able to pursue a more active problem-solving approach, forming and testing hypotheses, and asking herself questions about the relevant features on which the discrimination depends. As a consequence, she may be able to formulate an explicit regularity ("after two UPs I have to go DOWN") which is held in WM and comes to play a role in determining her behavior — a clear case of "explicit" knowledge. The possibilities, and the interplay between the different representations of information and with the learning that results from them, all depend on the precise details of the experimental task and the subject's strategies in a way that requires detailed cognitive modelling to tease apart.
In the longer term, the integrational approach can sometimes suggest new accounts for old phenomena. One such possibility concerns the nature of limited working memory (WM). Although the limitations of WM are usually interpreted as reflecting a limited resource, it is possible to find a functional explanation for many of the phenomena, and one can argue on general grounds that where both accounts are available, the functional account is preferable (2.121). Functional explanations include, for example, arguments that any system with the known capabilities of humans will exhibit the phenomenon in question, or that the phenomenon arises from a design trade-off such that humans are better off with the "limitation" than without. Another functional explanation is to show that only a limited amount of WM is needed for a task. In Howes' model of exploratory learning (2.39), for example, because the learning is guided by knowledge acquired and stored in long term memory, only a bounded amount of working memory is required independent of the size and shape of, e.g., the menu structure to be learned. Most existing search techniques, in contrast, require a goal stack that grows with the size of the space to be searched. Over the next few years, it will be interesting to see how far this general line of argument can be carried.
B. Integrating across Disciplines
Another issue is integrational, though this time in a different sense: across disciplines, rather than just across psychological areas. A major opportunity and challenge arises from the collaborative projects in HCI we have been engaged in during the past few years. As a result of the Amodeus projects, we are just beginning to understand the relationships between different kinds of modelling derived from different disciplines: formal system analysis done by computer scientists, cognitive analysis by psychologists, design space analysis, and so on. The question then arises of how to combine them, in order to paint a more complete picture of what happens in HCI, whether for practical reasons (giving guidance to interface designers) or theoretical (understanding the nature of human-computer interaction). The work poses difficult conceptual puzzles, such as how to reconcile the view of the user implicitly assumed in most HCI work in computer science with the view proposed by more psychological work on user modelling. There are also considerable technical obstacles, such as trying to understand and interrelate the different languages used by the different disciplines (mathematics, logic, cognitive task analysis, design rationale, …). The potential reward, though, is a deep, cross-disciplinary understanding of cognition in tasks requiring interaction with a complex device, that respects both the psychology of the person and the nature of the device.
B1. Multi-Disciplinary Modelling: One of several ideas within Amodeus about how such an integration might go is sketched in a recent paper which discusses the multi-disciplinary analysis of an "Undo" function for a collaborative computer application (2.65). The paper summarises six different modelling analyses of the same issue from different perspectives, two being psychological, two having to do with system modelling, one focussing on design, and one directly on the interaction between user and machine. It then proceeds to build an account of how the different threads can be woven together to tell a story that is richer and provides more insight into the user (and system) issues than can be provided by any one perspective alone. Current and future research is aimed at extending this work, to discover if there are other and perhaps better ways to fit the pieces together.
B2. Formal Modelling: At the present moment, a somewhat more formal method, using descriptions based on algebra and modal logic, looks promising for integrating at least the psychological and system models. The idea is that the behaviour of relevant aspects of the computer system be described by means of logical expressions that specify the transitions between system states. On the user side, the effects of possible user actions would be similarly described. In many cases the user's knowledge of the system state is indirect, in other words the user knows that the computer holds certain information, but does not herself know what that information is. For example, in the case of an address book held on the computer, the user knows that the computer has names and addresses, but does not know the actual addresses. Because of this indirectness, the user's knowledge has to be represented by a modal rather than a classical logic. The appropriate kind seems to be an epistemic logic, because the information to be represented is analogous to that involving beliefs about other people's knowledge (as when representing situations such as that Fred knows that George knows Susan's birthday, but does not know it himself). It then becomes feasible to deduce, for example, that the user can know that if she follows a certain strategy she will succeed in getting certain information (such as Murgatroyd's address) from the computer. Such deductions are more powerful than the simulations (of user and of device) that we are performing at present. Indeed, simulation interestingly enough turns out to be a restricted case of deduction. It emerges when the deduction takes a particular form, namely an alternation between inferences based just on the system representation and inferences based just on the user representation. This work is closely related to the approach proposed by Barnard in his Section B, Formal methods and human cognition.
C. Devices and Agents
No future work on these topics is planned.
STUDY AND MODELLING OF TRANSPORT SYSTEM USERS (Brown, Chapman, Groeger)
Introduction
The importance of psychological research on driving: The driving task is of immense social importance (see 2.11, 2.87). Recent figures indicate that the vast majority of the adult British population hold driving licences (e.g. 90% of all males and 50% of all females between 30 and 60 years of age). The female driving population is also growing rapidly. Meeting this continuing demand for independent mobility poses both economic and theoretical challenges. Maintaining and extending the road network is a substantial fiscal pressure on any economy. Further growth in the motoring population almost inevitably leads to a growth in congestion, a reduction in the cost effectiveness of road transport and a consequent increased demand for further extensions to the road network (2.7, 2.87).
Casualties also have economic consequences. In 1989, over 5,200 people died on British roads. It is estimated that about 70,000 suffered serious injuries, about 400,000 suffered slight injury accidents, and in the region of 2,000,000 damage-only accidents occurred that year. The personal cost of road traffic accidents, in terms of the psychological stress and trauma caused to the survivors, remains unquantified, but must be substantial. The economic costs are rather more clear: for 1989 the cost of these accidents was estimated to be 7.4 billion pounds. Although the fatality figures to be released in 1994 are likely to reduce to somewhere in the region of 4,400, the comparatively stable incidence of injury, increased actual cost of accidents and larger numbers of younger and older (i.e. higher risk) drivers in the population all suggest that road traffic accidents will continue to constitute a substantial social and economic problem, whether or not Government targeted reductions in casualty rates are achieved by the year 2000 (see 2.86).
Another aspect of the driving task which is worthy of note is the role of developments in vehicle and road technologies. Recent pan-European research initiatives such as DRIVE and PROMETHEUS have attempted to bring together industrial research and development teams, academic researchers and traffic engineers in order to develop intelligent vehicle and road applications for the transport system of the last years of this century and beyond. Such applications are intended to improve safety, efficiency and congestion by reducing the scope for driver error (2.53). The importance of behavioural considerations in the development and evaluation of such applications cannot be overstated, nor can the potential for wealth creation and employment opportunities afforded by the equipping of hundreds of millions of vehicles with new technologies (2.53, 2.128, 2.175).
Contributing to these societal goals of safety, efficiency and wealth creation, whilst also building and extending the theoretical basis of psychological research on driver behaviour, have been the aims of the Unit's long-term programme of research in this area. The work had its origins in studies conducted by Bartlett and his colleagues during World War Two. The continuous programme of research that followed this initial work may be divided into three broad areas of enquiry: the acquisition of skills, the effects on performance of task-induced and environmental stress, and problems arising at the interface between transport technology and its operators. Throughout this programme, the emphasis has been on attentional, perceptual and cognitive skills, rather than on lower-order vehicle control skills, although the latter have necessarily been included as important dependent variables in many studies, particularly those concerned with the development of methodology. In the current period, the work included a range of attempts to investigate different aspects of drivers' learning, drivers' appreciation of risk and decision making, the role of memory in drivers' assessments of danger and error, and drivers' self-assessments of skill. These theoretically driven research interests have benefited from substantial external funding, to the tune of around half a million pounds, in the period of this report.
A. Investigations of Drivers' Learning
Despite the fact that in the region of 95% of those learning to drive take at least some formal training before their driving test, and the fact that drivers with different levels of traffic experience are differentially involved in accidents, there have been few systematic studies of drivers' learning (for discussion see 2.100, 2.131, 2.134). Two approaches have been adopted in our research, one which seeks to examine the whole range of what it is drivers are required to learn, and one which seeks to examine the effects of instruction under controlled conditions (see 2.101, 2.170 for discussion of how the two interact).
A1. Learning During Training: The series of investigations involved the video recording of each and every driving lesson taken by 20 teenagers learning to drive for the first time. The content of these video tapes was analysed, and each manoeuvre, statement by the instructor and error by the pupil was recorded. Unusually in a British context, these drivers did not drive between lessons, and thus the Driving Instruction Database provides a unique record of the early learning experiences of these novice drivers. The collection, coding and initial analysis of these data was carried out as part of our efforts within the CEC funded Generic Intelligent Driver Support project by Groeger and Grande (2.109, 2.139).
The analysis of the database has been continued by Groeger and Clegg, with funding from the Department of Transport. A variety of results have emerged from these analyses, a few of which are particularly worthy of note. Drivers undertake in the region of 2,000 manoeuvres when learning to drive. Almost all of these are commented upon by the driving instructor at the outset, while at the end of the course the rate of instruction per manoeuvre is much reduced, but only for certain manoeuvres. From other analyses it would seem as if the withdrawal of instruction (or "instructor independence") is well described by a power law for each pupil studied. This raises an important issue, as yet unresolved, about the nature of practice in complex tasks such as driving, and where it is most appropriately placed on the traditional "massed-distributed" continuum (see 2.100, 2.134, 2.135). These analyses also make clear that certain manoeuvres are performed relatively infrequently and rather erratically during the course of lessons, with the result that they appear to be learned less rapidly. This is especially the case for manoeuvres which require decisions to be made about the possible actions of other road users, and those which require some assessment of danger. The comparative infrequency of such manoeuvres, and their more or less random dispersion across lessons, raises the possibility that higher-order decision making skills cannot be adequately acquired by drivers during traditional courses of driving lessons. This prediction carries with it the implication that vehicle control skills, which are frequently performed in a wide variety of contexts, will be quickly acquired and relatively enduring.
A2. Learning under Controlled Conditions: Research carried out as part of our efforts within the CEC funded ARIADNE project by Groeger and Kuiken (University of Groningen), using the fixed-based interactive driving simulator at Groningen, has investigated drivers' learning under more controlled circumstances. A series of studies revealed that in the absence of augmented feedback drivers performed consistently, but inaccurately, in the simulator (2.167). Where instruction was made contingent on the occurrence of errors of various types, the rate of those errors referred to in the instruction decreased. Improvements were not observed where the support given was not contingent on performance (2.145). More interesting from a theoretical point of view was the fact that, while drivers were aware that their performance had improved in the supported scenarios, they were unaware that their performance in other scenarios had also altered. By careful manipulation of elements of the scenarios used, it was clear that such transfer occurred only for those scenarios which were functionally similar (i.e. support on curves altered performance on curves and roundabouts but not at intersections; whereas subjects reported that their performance changed on curves, but not on roundabouts or crossroads). As another example of "implicit" learning, behaviour changes at green traffic signals carried over to junctions where the driver had right of way, but not to junctions where the driver did not have right of way (see 2.111, 2.141, 2.168).
The results from these two lines of work have clear practical applications, with respect to the redesign of driver training curricula and to the development of effective in-car performance support devices (see 2.102, 2.106, 2.138). They also serve to echo, if not extend, current theoretical debates in human learning (2.105).
B. Drivers' Decision Making and Appreciation of Risk
B1. Decision Making on the Approach to Traffic Signals: Our investigations of drivers' decision making on the approach to signal-controlled junctions was initially motivated by practical rather than theoretical concerns. As part of their efforts to understand why 40% of all accidents take place at urban intersections, the Department of Transport funded a joint research project between the APU and a traffic engineering group at University College London. Laboratory-based empirical work by Groeger, Grande and Brown made extensive use of a "time-to-coincidence" task. This revealed that drivers consistently underestimated how long it would take to reach a set of traffic lights towards which they were travelling (2.137, 2.140). Estimated and actual arrival times were related by a power law. Work built upon this paradigm revealed that drivers were confident when taking "stop" or "go" decisions when confronted with amber onset, even though their estimates of when the junction were reached would have been highly inaccurate (2.149, 2.166). Furthermore, the studies strongly suggested that these two decisions were typically made at different points on the approach, rather than at the same stage. Recent reanalysis of these data suggest that "go" decisions are made earlier on the approach to junctions than "stop" decisions. That is, in a manner which is highly adaptive, drivers decide to continue with an approach to a junction when the lights have changed if they believe they are closer to the intersection than they actually are. They decide to stop when they are more capable of accurately estimating when the junction will be reached.
B2. Appreciation of Risk: Research by Groeger and Chapman, supported by General Accident and the Economic and Social Research Council, revealed that only some drivers know some of the types of traffic situations in which traffic accidents are more likely to occur (2.36, 2.99, 2.107). That is, earlier findings which suggest a systematic relationship between subjective and objective risk considerably oversimplify the relationship. This project also developed a laboratory based "judgements of traffic scenes" task which successfully discriminates between drivers of the same age with different amounts of traffic experience, and also those with equivalent amounts of driving experience who are different ages (2.108). Thus, for perhaps the first time in the driving domain, separate effects of age and driving experience were demonstrated in a laboratory task. This has had very substantial effects on the way we conceptualise drivers' understanding of danger and the risk-related decisions they take. Differences on the "judgements of traffic scenes" task appear to be corroborated by the way these individuals actually drive. Work is currently underway to adapt the task for possible use in detecting newly qualified drivers who may be particularly at risk (research funded by Department of Transport as part of our investigation of Early Learning of Driving Skills).
B3. Decision Making and Traffic Violations: A further strand of our research on drivers' decision making, and ways of influencing it, comprises our efforts within the CEC-funded DETER project. A novel decision-making task (Animated Interactive Driving Environment - AIDE) has been developed within this project and has been used to demonstrate that provision of normative information is effective in reducing some, but not all, types of traffic violation. Rather more powerful effects were found of simulated in-car feedback which indicated that a violation had been detected. These effects have recently been corroborated by studies of driving under real traffic conditions, while the normative feedback effects are the subject of a large-scale field trial currently underway in Norway (see 2.174).
C. The Role of Memory in Drivers' Assessments of Danger and Error
C1. Drivers' Memory for Risk: Although driving experience has long been considered an important variable in attempting to predict or explain behaviour, very little attempt has been made to study drivers' memory for information related to the task. Chapman and Groeger have, as part of the research project supported by General Accident and the Economic and Social Research Council, shown that feelings of risk dramatically affect drivers' memory for traffic situations (2.90, 2.159). A number of results have emerged from this programme of research which has involved both laboratory and on-road research and made use of free recall, cued recall and recognition paradigms. Overall, subjects show a bias towards recalling dangerous situations. When this is controlled for, it has been shown that where risk is to be anticipated, drivers are better able to recall and recognise risk-related information (2.159). The results of these studies further suggest that more risk-related details of risk-related objects are spontaneously recalled. These results have been interpreted in terms of "attention focusing" where, as is documented in the eyewitness testimony literature, subjects attend particularly to threat-related information during encoding. By comparison, aspects of the driving scene which are not related to dangers are recalled and recognised rather poorly when a perceived threat is present.
C2. Memory for Errors Committed and Criticism: In addition to continuing our interests in issues directly related to safety, attention has also devoted to the errors drivers commit and what they tell us about the skills involved in driving (2.13, 2.14, 2.33). Recently, this research has been linked to another main interest, i.e. the relationship between emotional arousal and recall of driving related information. Thus, for example, Groeger and Bekerian have shown that drivers are aware of having been criticised during a test-drive, but are unable to remember where or when that criticism was made.
The strength of this work on drivers' memory for aspects of the driving task lies in its use of standard recognition and recall paradigms, to some theoretical effect, to investigate memory in contexts where emotional arousal is naturally integral to the task. The findings are also important from a practical point of view. Drivers' poor recall of critical feedback strengthens our contention that normal driver-training practices need to be revised to take account of everyday learning and memory processes and phenomena. Similarly, it would seem that reports of accidents which rely on participants' and non-participants' recall of events are likely to be systematically biased towards certain types of information. Since most actuarial data on traffic accidents are based on just such information, the information on which theories of accident causation are based may be partially misleading.
D. Drivers' Self-Assessments of Skill
As part of the CEC funded Generic Intelligent Driver Support systems project, Groeger and Grande had the opportunity to develop methodologies for assessing the accuracy of, and influences upon, drivers' assessments of their own skills (2.165). The basic result, that drivers consider themselves better than novice drivers, was unsurprising in light of the frequently reported finding that drivers overestimate their ability. However, large sample sizes, the provision of feedback and evaluation of performance by a qualified driving instructor, and the use of repeated self assessments across different driving tasks revealed important and novel findings. Although drivers considered themselves better than novices, this opinion was not shared by an independent assessor. Feedback provided by the independent assessor reduced, but did not eliminate, this bias towards a positive view of one's own ability. Interestingly, in spite of this feedback's effect on self-assessments, there is remarkable stability in how drivers evaluate themselves over time (2.110). The consistency of this relationship is enhanced by greater amounts of traffic experience, which suggests that self-assessments in highly practised tasks may become rather resistant to feedback over time. Overall, this consistent view of one's driving is less positive depending on the numbers of reported accidents per mile driven in the recent past. There is also a strong suggestion that affective reactions to critical feedback depend, at least in part, on the consistency with which one's performance is viewed over time.
It appears that there is a consistent way of viewing one's performance which seems specific to a particular task, which strengthens with accident-free experience of that task, and which helps to predict which affective responses are likely to different types of feedback. This suggests that what we are observing is the operation of a "task-related self". This is important in theoretical terms, as is the possibility that self-assessment techniques may prove useful indirect tests of what has been learned in some, but not in all situations (see C1 above). That this task-related self is quite stable over time may help to explain why, as shown by investigations of drivers' recall of criticism (see C2 above), the effects of isolated, situation-specific feedback on one's driving may not be very enduring.
Conclusions
Over the past 40 years or so, the Unit has made a unique contribution to this field of research, and its publications have had a notable influence on the way in which behavioural problems in transport systems have been approached, particularly in the UK but also internationally. This long-term influence has been reinforced by Brown's continuing advisory contributions to national and international bodies concerned with transportation issues. His appointments as Extra-mural Professor of Traffic Science at the University of Groningen from 1988-91 and as an Honorary Research Fellow of the UK Transport Research Laboratory for 5 years from 1992 created more specific opportunities to inject the Unit's views on and findings from this field of research into larger and internationally respected laboratories elsewhere. In recent years, Groeger's research contributions to European programmes of work on advanced transport technology have considerably strengthened these broad influences of the Unit's approach to transport studies. For example, research on the relative risks of taxi and minicab travel in London (2.163, 2.164) will lead to legislative changes during the next session of Parliament.
The Unit's substantial contributions to this field of research were recognised by the award of an OBE to Brown in June 1991 for his "services to transport safety research"; by the 'Distinguished Foreign Colleague' Award presented to him by the Human Factors Society in 1990 for his "outstanding contributions to human factors"; and by the A.R. Lauer Traffic Safety Award presented to him in 1993 by the Human Factors and Ergonomics Society "in recognition of his efforts to stimulate psychologists' interest in driver behaviour and to convince traffic authorities of the importance of a driver-centred view of traffic problems". The point was made that "his work proved seminal in establishing that workload is an important concern in the driving domain". With Brown's retirement in April 1993 and Groeger's departure to an appointment at the University of Leeds in April 1994, the Unit's (and Council's) contributions to this field of research have effectively come to an end. In many respects this is regrettable, since there is now no institution in the UK with a comparable long-term commitment to researching attentional, perceptual and cognitive aspects of transport system design and usage in a way that bridges the gap between theory and application. The result is likely to be a concentration on short-term, highly specific, applied transport problems, the solution of which may go unreported in the mainstream psychological literature and hence fail to contribute substantially to the development of psychological theory.
PUBLICATIONS
Authored Books
2.1. HOUGHTON, G. (in press). The Production of Spoken Discourse: A Computational Model. New Jersey: Ablex.
Edited Books
2.2. Bermùdez, J., MARCEL, A.J. & Eilan, N. (Eds.) (in press). The Body and the Self. Cambridge, Mass.: MIT Press, Bradford Books.
2.3. Byerley, P.F., BARNARD, P.J. & MAY, J. (Eds.) (1993). Computers, Communication and Usability: Design Issues, Research and Methods for Integrated Services. Amsterdam: Elsevier Science Publishers, B.V.
Refereed Articles
2.4. BLANDFORD, A.E., Cross, N.G. & Scanlon, E. (1994). Computers and the development of design decision making skills. Computer and Education, 22, 45-56.
2.5. BLANDFORD, A. & YOUNG, R.M. (1993). In J.L. Alty, D. Diaper & S. Guest (Eds.), People and Computers VIII: Proceedings of the HCI '93 Conference (pp. 111-122). Cambridge: Cambridge University Press.
2.6. Bradshaw, J.A. & YOUNG, R.M. (1991). Integrating knowledge of purpose and knowledge of structure for design evaluation. IEEE Expert, 6, 33-40.
2.7. Brookhuis, K. & BROWN, I.D. (1992). Ergonomics and road safety. Impact of Science on Society, 165, 35-40.
2.8. BROWN, I.D. (1990). A down-to-earth view of proposals for a grand unification of traffic sciences (GUTS). Journal of the International Association of Traffic and Safety Sciences, 14, 109-111.
2.9. BROWN, I.D. (1990). Drivers' margins of safety considered as a focus for research on error. Ergonomics (Special Issue), 33, 1307-1314.
2.10. BROWN, I.D. (1993). Driver fatigue and road safety. Alcohol, Drugs and Driving, 9, (Nos. 3-4), 239-251.
2.11. BROWN, I.D. (in press). Conflicts between mobility, safety, and environmental preservation, expressed as a hierarchy of social dilemmas. Journal of the International Association of Traffic and Safety Sciences (IATSS Research).
2.12. BROWN, I.D. (in press). Driver fatigue. Human Factors.
2.13. BROWN, I. & GROEGER, J. (1990). A way with errors (Editorial). Ergonomics, (Special Issue), 33, 1183-1184.
2.14. BROWN, I.D. & GROEGER, J.A. (Eds.). (1990). Errors in the operation of transport systems. Ergonomics (Special Issue), 33. London: Taylor & Francis.
2.15. Chelazzi, L., Miller, E.K., DUNCAN, J. & Desimone, R. (1993). A neural base for visual search in inferior temporal cortex. Nature, 363, 345-347.
2.16. Currie, M., Mackay, P., Morgan, C., Runciman, W., Russell, W., SELLEN, A., Webb, R. & Williamson, J. (1993). The "wrong drug" problem in anaesthesia: An analysis of 2000 incident reports. Anaesthesia & Intensive Care, 21, 596-601.
2.17. Desimone, R. & DUNCAN, J. (in press). Neural mechanisms of selective visual attention. Annual Review of Neuroscience.
2.18. Driver, J., Baylis, G.C., GOODRICH, S.J. & Rafal, R.D. (in press). Axis-based neglect of visual shapes. Neuropsychologia .
2.19. DUFF, S.C. & BARNARD, P.J. (1990). Influencing behaviour via device representation: Decreasing performance by increasing instruction. In D. Diaper, D. Gilmore, G. Cockton & B. Shackel (Eds.), Human-Computer Interaction - INTERACT '90 (pp. 61-66). Amsterdam: Elsevier Science Publishers, B.V.
2.20. DUNCAN, J. (1990). Goal weighting and the choice of behaviour in a complex world. Ergonomics, 33, 1265-1279.
2.21. DUNCAN, J. (1993). Coordination of what and where in visual attention. Perception, 22, 1261-1270.
2.22. DUNCAN, J. (1993). Similarity between concurrent visual discriminations: Dimensions and objects. Perception and Psychophysics, 54, 425-430.
2.23. DUNCAN, J. (1994). Selective attention in the primate visual system. Canadian Psychology, 35, 104-105.
2.24. DUNCAN, J. (in press). Cooperating brain systems in selective perception and action. In T. Inui & J.L. McClelland (Eds.), Attention and Performance XVI. Cambridge, MA: MIT Press.
2.25. DUNCAN, J. (in press). Target and nontarget grouping in visual search. Perception and Psychophysics.
2.26. DUNCAN, J. & Humphreys, G. (1992). Beyond the search surface: Visual search and attentional engagement. Journal of Experimental Psychology: Human Perception and Performance, 18, 578-588.
2.27. DUNCAN, J., WARD, R. & Shapiro, K. (1994). Direct measurement of attentional dwell time in human vision. Nature, 369, 313-315.
2.28. DUNCAN, J., Williams, P. & BROWN, I. (1991). Components of driving skill: Experience does not mean expertise. Ergonomics, 34, 919-937.
2.29. DUNCAN, J., Williams, P., NIMMO-SMITH, I. & BROWN, I. (1993). The control of skilled behavior: Learning, intelligence, and distraction. In D.E. Meyer & S. Kornblum (Eds.), Attention & Performance XIV: Synergies in Experimental Psychology, Artificial Intelligence, and Cognitive Neuroscience (pp. 323-341). Cambridge, MA: The MIT Press.
2.30. GOODRICH, S., Henderson, L., Allchin, N. & Jeevaratnam, A. (1990). On the peculiarity of simple reaction time. Quarterly Journal of Experimental Psychology. 42A, 763-775.
2.31. GREEN, A.J.K. (1994). Interacting Cognitive Subsystems: A framework for considering the relationships between performance and knowledge representations. Interacting with Computers, 6, 61-85.
2.32. GREEN, A.J.K. & BARNARD, P.J. (1990). Iconic interfacing: The role of icon distinctiveness and fixed or variable screen location. In D. Diaper, D. Gilmore, G. Cockton & B. Shackel (Eds.), Human-Computer Interaction - INTERACT '90 (pp. 457-462). Amsterdam: Elsevier Science Publishers, B.V.
2.33. GROEGER, J. (1990). Drivers' errors in and out of context. Ergonomics, (Special Issue), 33, 1423-1429.
2.34. GROEGER, J.A. (1990). Saying something different: Levels in the monitoring and repair of speech. Belfast Working Papers in Language and Linguistics, 10, 66-84.
2.35. GROEGER, J. (1994). The working memory man: An interview with Professor Alan D Baddeley, FRS. The Psychologist, 7, 58-59.
2.36. GROEGER, J.A. & CHAPMAN, P. (1990). Errors and bias in assessments of danger and frequency of traffic situations. Ergonomics (Special Issue), 33, 1349-1363.
2.37. Henderson, L. & GOODRICH, S.J. (1993). Simple reaction time and predictive tracking in Parkinson's disease: Do they converge on a single, fixed impairment of preparation? Journal of Motor Behaviour, 25, 89-96.
2.38. Henderson, L., Kennard, C., Crawford, T., Day, S., Everett, B., GOODRICH, S., Jones, F. & Park, D.M. (1991). Scales for rating motor impairment in Parkinson's disease: Studies of reliability and convergent validity. Journal of Neurology, Neurosurgery and Psychiatry, 54, 18-24.
2.39. HOWES, A. (1994). A model of the acquisition of menu knowledge by exploration. In B. Adelson, S. Dumais & J. Olson (Eds.), Human Factors in Computing Systems ("Celebrating Interdependence", Proceedings of CHI '94 Conference, Boston, MA, 1994) (pp. 445-451). New York: ACM.
2.40. HOWES, A. & YOUNG, R.M. (1991). Predicting the learnability of task-action mappings. In S.P. Robertson, G.M. Olson & J.S. Olson (Eds.), Reaching through Technology - CHI '91 Conference Proceedings: Human Factors in Computing Systems (pp. 113-118). New Orleans, Louisiana: Addison-Wesley.
2.41. LAVIE, N. & Tsal, Y. (1994). Perceptual load as a major determinant of the locus of selection in visual attention. Perception & Psychophysics, 55, 1-15.
2.42. LAVIE, N. (in press). Perceptual load as a necessary condition for selective attention. Journal of Experimental Psychology: Human Perception & Performance.
2.43. LEE, W-O. (1992). The effects of skill development and feedback on action slips. In A. Monk, D. Diaper & M.D. Harrison (Eds.), People and Computers VII (pp. 73-86). Cambridge: Cambridge University Press.
2.44. LEE, W-O. (1993). Adapting to interface resources and circumventing interface problems: Knowledge development in a menu search task. In J.L. Alty, D. Diaper & S. Guest (Eds.), People and Computers VIII: Proceedings of the HCI'93 Conference (pp. 61-77) Cambridge University Press.
2.45. LEE, W-O. & BARNARD, P.J. (1993). Precipitating change in system usage by function revelation and problem reformulation. In J.L. Alty, D. Diaper & S. Guest (Eds.), People and Computers VIII: Proceedings of the HCI'93 Conference (pp. 35-47). Cambridge University Press.
2.46. MacLean, A., Bellotti, V. & YOUNG, R. (1990). What rationale is there in design? In D. Diaper, D. Gilmore, G. Cockton & B. Shackel (Eds.), Human-Computer Interaction - INTERACT '90 (pp. 207-212). Amsterdam: Elsevier Science Publishers, B.V.
2.47. MacLean, A., Bellotti, V., YOUNG, R., & Moran, T. (1991). Reaching through analogy: A design rationale perspective on roles of analogy. In S. Robertson, G. Olson & J. Olson (Eds.), CHI '91 Conference Proceedings (pp. 167-172). New York: ACM.
2.48. MacLean, A., YOUNG, R., Bellotti, V.M.E. & Moran, T.P. (1991). Questions, options, and criteria: Elements of design space analysis. Human-Computer Interaction, 6, 201-250.
2.49. MARCEL, A.J. (1990). What does it mean to ask whether cognitive skills are prerequisite for learning to read and write? A response to Cossu and Marshall. Cognitive Neuropsychology, 7, 41-48.
2.50. MAY, J., BARNARD, P.J. & BLANDFORD, A. (1993). Using structural descriptions of interfaces to automate the modelling of user cognition. User Modelling and User Adapted Interaction, 3, 27-64.
2.51. MAY, J., BARNARD, P.J., Boecker, M. & GREEN, A.J. (1990). Characterising structural and dynamic aspects of the interpretation of visual interface objects. In ESPRIT '90 Conference Proceedings (pp. 819-834), (Brussels, November 1990), Dordrecht: Kluwer Academic Publishers.
2.52. MAY, J., TWEEDIE, L. & BARNARD, P.J. (1993). Modelling user performance in visually based interactions. In J.L. Alty, D. Diaper & S.P. Guest (Eds.), People and Computers VIII: Proceedings of the HCI '93 Conference (pp. 95-110). Cambridge: Cambridge University Press.
2.53. Michon, J.A., Kuiken, M.J. & GROEGER, J.A. (1992). Smartening road traffic: The role of GIDS. IATSS Research, 16, 106-113.
2.54. Payne, S.J. & HOWES, A. (1992). A task-action trace for exploratory learners. Behaviour and Information Technology, 11, 63-70.
2.55. Payne, S.J., HOWES, A. & Hill, E. (in press). Conceptual instructions derived from an analysis of device models. International Journal of Human-Computer Interaction.
2.56. Payne, S.J., Squibb, H. & HOWES, A. (1991). The nature of device models: The yoked state space hypothesis and some experiments with text editors. Human Computer Interaction, 5, 415-444.
2.57. Rieman, J., Lewis, C., YOUNG, R.M. & Polson, P.G. (1994). "Why is a raven like a writing desk?" Lessons in interface consistency and analogical reasoning from two cognitive architectures. In B. Adelson, S. Dumais & J. Olson (Eds.), Human Factors in Computing Systems ("Celebrating Interdependence", Proceedings of CHI '94 Conference, Boston, MA, 1994) (pp. 438-444). New York: ACM.
2.58. ROBERTSON, I., Tegnèr, R., GOODRICH, S. & Wilson, C. (in press). Walking trajectory and hand movements in unilateral left neglect: A vestibular hypothesis. Neuropsychologia.
2.59. Runciman, W.B., SELLEN, A.J., Webb, R.K., Williamson, J.A., Currie, M., Morgan, C. & Russell, W.J. (in press). Errors, incidents and accidents in anaesthesia. Anaesthesia and Intensive Care.
2.60. SELLEN, A.J. (in press). Detection of everyday errors. Applied Psychology: An International Review (Special Issue on Error Detection edited by D. Zapf & J. Reason).
2.61. WARD, R., GOODRICH, S.J. & Driver, J. (in press). Grouping reduces visual extinction: Neuropsychological evidence for weight-linkage in visual selection. Visual Cognition.
2.62. Williamson, J., Webb, R., SELLEN, A., Runciman, W. & Van Der Walt, J. (1993). Human failure: An analysis of 2000 incident reports. Anaesthesia & Intensive Care, 21, 678-683.
2.63. WING, A.M., GOODRICH, S., VIRJI-BABUL, N., Jenner, J.R. & CLAPP, S. (1993). Balance evaluation in hemiparetic stroke patients using lateral forces applied to the hip. Archives of Physical Medicine and Rehabilitation, 74, 292-299.
2.64. YOUNG, R. M. (1994) The unselected window scenario: Analysis based on the Soar cognitive architecture. In G. D. Abowd (Ed), Proceedings of the CHI'94 Research Symposium.
2.65. YOUNG, R. M. & Abowd, G. D. (1994) Multi-perspective modelling of interface design issues: Undo in a collaborative editor. In People and Computers IX: Proceedings of the Conference on Human-Computer Interaction. Cambridge University Press.
2.66. YOUNG, R.M. & BARNARD, P.J. (1991). Signature tasks and paradigm tasks: New wrinkles on the scenarios methodology. In D. Diaper & N. Hammond (Eds.), People and Computers VI: Proceedings of the HCI '91 Conference (pp. 91-101). Cambridge: Cambridge University Press.
2.67. YOUNG, R.M., HOWES, A. & WHITTINGTON, J. (1990). A knowledge analysis of interactivity. In D. Diaper, D. Gilmore, G. Cockton & B. Shackel (Eds.), Human-Computer Interaction - INTERACT '90 (pp. 115-120). Amsterdam: Elsevier Science Publishers, B.V.
2.68. YOUNG, R.M. & WHITTINGTON, J. (1990). Using a knowledge analysis to predict conceptual errors in text-editor usage. In J.C. Chew & J. Whiteside (Eds.), Empowering People. Human Factors in Computing Systems (pp.91-97), (CHI '90 Conference Proceedings, Special Issue of the SIGCHI Bulletin). Seattle, Washington: ACM Press.
Submitted
2.69. BARNARD, P.J., MAY, J & Salber, D. Deixis and points of view in media spaces: An empirical gesture. (manuscript submitted to Behaviour and Information Technology).
2.70. BLANDFORD, A.E., Harrison, M.D. & BARNARD, P.J. Understanding the properties of interactions. (Manuscript submitted to International Journal of Human Computer Studies)
2.71. BOURKE, P. A. Measuring attentional demand. (Manuscript submitted to Quarterly Journal of Experimental Psychology).
2.72. BOURKE, P. A., DUNCAN, J., & NIMMO-SMITH, I. A general factor involved in dual task performance decrement. (Manuscript submitted to Quarterly Journal of Experimental Psychology).
2.73. DUNCAN, J., BURGESS, P., & EMSLIE, H. Fluid intelligence after frontal lobe lesions. (Manuscript submitted to Neuropsychologia).
2.74. DUNCAN, J., EMSLIE, H., Williams, P. & Johnson, R. Intelligence and the frontal lobe: The organization of goal-directed behavior. (Manuscript submitted to Cognitive Psychology).
2.75. HOWES, A. & YOUNG, R. M. A learning and performance model of display-based task-action mapping. (Manuscript submitted to Cognitive Science)
2.76. Humphreys, G.W., Romani, C., Olson, S., Riddoch, M.J. & DUNCAN, J. Non-spatial extinction following lesions of the parietal lobe in humans. (Manuscript submitted to Nature).
2.77. MARCEL, A.J. (a) The influence of vision on tactile sensation. (Manuscript submitted to Perception)
2.78. MARCEL, A.J. (b) Blindsight: A problem of visual consciousness or visual function? (Manuscript submitted to Brain)
2.79. MAY, J. & BARNARD P.J. The case for supportive evaluation during design. (Manuscript submitted to Interacting with Computers).
Invited Chapters and Commentaries
2.80. BARNARD, P.J. (1991). Applied cognitive psychology: Research for human-computer interaction. In A.C. Downton (Ed.), Engineering the Human-Computer Interface (pp. 28-61). McGraw-Hill.
2.81. BARNARD, P.J. (1991). Bridging between basic theories and the artifacts of human-computer interaction. In J.M. Carroll (Ed.), Designing Interaction: Psychology at the Human-Computer Interface (pp. 103-127). Cambridge: Cambridge University Press.
2.82. BARNARD, P. J. (1991). The contributions of applied cognitive psychology to the study of human-computer interaction. In B. Shackel & S. Richardson (Eds.), Human Factors for Informatics Usability (pp. 151-182). Cambridge: Cambridge University Press.
2.83. BARNARD, P. (in press). The contributions of applied cognitive psychology to the study of human-computer interaction. In R. Baecker, J. Grudin, W. Buxton & S. Greenberg, S. (Eds.), Readings in Human-Computer Interaction, Second Edition, Los Altos, CA: Morgan Kaufmann. (Reproduction of APU 2205)
2.84. BARNARD, P.J. & MAY, J. (1993). Cognitive modelling for user requirements. In P.F. Byerley, P.J. Barnard & J. May (Eds.), Computers, Communication and Usability: Design Issues, Research and Methods for Integrated Services (pp. 101-145). Amsterdam: Elsevier Science Publishers, B.V.
2.85. BLANDFORD, A., Harrison, M.D. & BARNARD, P.J. (1993). Integrating user requirements and system specification. In P.F. Byerley, P.J. Barnard & J. May (Eds.), Computers, Communication and Usability: Design Issues, Research and Methods for Integrated Services (pp. 165-196). Amsterdam: Elsevier Science Publishers, B.V.
2.86. BROWN, I.D. (1990). Accident reporting and analysis. In J.R. Wilson & E.N. Corlett (Eds.), Evaluation of Human Work: A Practical Ergonomics Methodology (pp. 755-778). London: Taylor and Francis Ltd.
2.87. BROWN, I.D. (1990). On the social dilemma of motorway safety. In T. Benjamin (Ed.), Driving Behaviour in a Social Context (pp.663-670). Caen: Paradigme.
2.88. BROWN, I.D. (1991). Prospects for improving road safety during the 1990's. In Ergonomics, Safety and Health: Perspectives for the Nineties. Leuven, Belgium: Leuven University Press.
2.89. BROWN, I.D. (1993). Methodological issues in driver fatigue research. In Driver Impairment, Fatigue, and Driving Simulation. Applecross, Western Australia: Promaco Conventions Pty Ltd.
2.90. CHAPMAN, P.R. & GROEGER, J.A. (1992). Subjective risk and subsequent memory. In G.B. Grayson (Ed.), Behavioural Research in Road Safety: II (pp. 44-52). Crowthorne, UK: Transport Research Laboratory.
2.91. Clegg, G., Warr, P., GREEN, T., Monk, A., Kemp, N., Allison, G. & Lansdale, M. (Eds.). (1988). People and Computers: How to Evaluate your Company's New Technology. Chichester: Ellis Horwood Ltd.
2.92. Desimone, R., Chelazzi, L., Miller, E.K. & DUNCAN, J.D. (in press). Neural mechanisms for memory-guided visual search. In C. Nothdurft (Ed.), Structural and Functional Organization of the Neocortex. Springer-Verlag.
2.93. Desimone, R., Chelazzi, L., Miller, E.K. & DUNCAN, J. (in press). Neuronal mechanisms of visual attention. In T. Papathomas (Ed.), Linking Psychophysics, Neurophilosophy, and Computational Vision. Cambridge, MA.: MIT Press.
2.94. DUFF S.C. (1992). Mental models as multi-record representations. In Y. Rogers, A. Rutherford & P. Bibby (Eds.), Models in the Mind: Theory, Perspective and Applications (pp. 172-186). London: Academic Press.
2.95. DUNCAN, J. (1993). Selection of visual information in the control of behaviour. In A.D. Baddeley & L. Weiskrantz (Eds.), Attention: Selection, Awareness and Control: A Tribute to Donald Broadbent (pp. 53-71). Oxford: Clarendon Press.
2.96. DUNCAN, J. (in press). Attention, intelligence and the frontal lobes. In M.S. Gazzaniga (Ed.), The Cognitive Neurosciences. MIT Press.
2.97. Eilan, N., MARCEL, A.J. & Bermùdez, J. (in press). Self-consciousness and the body: Interdisciplinary issues. In J. Bermùdez, A.J. Marcel & N. Eilan (Eds.), The Body and the Self. Cambridge, Mass.: MIT Press, Bradford Books.
2.98. Gothelp, H., Farber, B., GROEGER, J.A. & Labiale, G. (1993). Driving: Task and environment. In J.A. Michon (Ed.), Generic Intelligent Driver Support (pp. 19-32). London: Taylor & Francis.
2.99. GROEGER, J.A. (1990). Concepts of danger: The unknown risks we run. In H. Bohm (Ed.), Psychological Statistics and Models of Accidents in Traffic Systems (pp.59-72). Bremen: Commission of the European Communities.
2.100. GROEGER, J.A. (1991). Acquiring and retaining driving skills. In M.J. Kuiken & J.A. Groeger (Eds.), Report on Feedback Requirements and Performance Differences of Drivers (pp. 3-14). Deliverable ADA2, DRIVE Project 1041: Generic Intelligent Driver Support Systems, Traffic Research Center, University of Groningen, The Netherlands.
2.101. GROEGER, J.A. (1991). Meeting drivers' needs for adaptive support: Personalised Support and Learning Module (PSALM). In M.J. Kuiken & J.A. Groeger (Eds.), Report on Feedback Requirements and Performance Differences of Drivers (pp. 89-94). Deliverable ADA2, DRIVE Project 1041: Generic Intelligent Driver Support Systems, Traffic Research Centre, University of Groningen, The Netherlands.
2.102. GROEGER, J.A. (1993). Degrees of freedom and the limits of learning: Support needs of inexperienced motorists. In A.M. Parkes & S. Franzen (Eds.), Driving Future Vehicles (pp. 77-88). London: Taylor & Francis.
2.103. GROEGER, J.A. (1993). Driving research at crossroads. In J.A. Santos (Ed.), Human Factors in Road Traffic (pp. 47-68). Lisbon: Escher.
2.104. GROEGER, J.A. (in press). Activities of lone watchkeepers on an ocean-going voyage. In H. Bohm (Ed.), Psychological Care for Operators Working in High Tech Transport Systems in Isolated Conditions. Bremen: Commission of the European Communities.
2.105. GROEGER, J.A. (in press). Degrees of freedom and the limits of learning: Support needs of inexperienced motorists. In A.W. Parkes & S. Fransen (Eds.), Driving Future Vehicles. London: Taylor & Francis.
2.106. GROEGER, J.A., Alm, H., Haller, R. & Michon, J.A. (1993). Acceptance of intelligent in-car devices. In J.A. Michon (Ed.), Generic Intelligent Driver Support (pp. 217-228). London: Taylor & Francis.
2.107. GROEGER, J.A. & CHAPMAN, P.R. (1991). The unknown risks we run: Feelings of danger and estimates of accident frequency when driving. In G.B. Grayson & J.F. Lester (Eds.), Behavioural Research in Road Safety (pp. 131-138). Crowthorne: Transport and Road Research Laboratory.
2.108. GROEGER, J.A. & CHAPMAN, P.R. (1992). Developing an understanding of danger: Contributions of experience and age. In G.B. Grayson (Ed.), Behavioural Research in Road Safety II (pp. 37-43). Crowthorne, UK: Transport Research Laboratory.
2.109. GROEGER, J.A. & GRANDE, G.E. (1991). Support received during drivers' training. In M.J. Kuiken & J.A. Groeger (Eds.), Report on Feedback Requirements and Performance Differences of Drivers (pp. 15-42). Deliverable ADA2, DRIVE Project 1041: Generic Intelligent Driver Support Systems. Groningen, The Netherlands: Traffic Research Center, University of Groningen.
2.110. GROEGER, J.A. & GRANDE, G.E. (in press). Psychological and performance correlates of self-assessed skill. In G.B. Grayson (Ed.), Behavioural Research in Road Safety III. Crowthorne, UK: Transport Research Laboratory.
2.111. GROEGER, J.A. & Kuiken, M.J. (1993). GIDS functions: Adaptive support. In J.A. Michon (Ed.), Generic Intelligent Driver Support (pp. 129-135). London: Taylor & Francis.
2.112. HOUGHTON, G. (1990). The problem of serial order: A neural network model of sequence learning and recall. In R. Dale, C. Mellish & M. Zock (Eds.), Current Research in Natural Language Generation (pp. 287-319). London: Academic Press.
2.113. HOUGHTON, G. & Tipper, S.P. (in press). A model of inhibitory mechanisms in selective attention. In D. Dagenbach & T. Carr (Eds.), Inhibitory Mechanisms in Attention, Memory and Language. London: Academic Press.
2.114. HOWES, A. (in press). An introduction to cognitive modelling in human-computer interaction. In A. Monk & N. Gilbert (Eds.), Perspectives on Human-Computer Interaction. Academic Press.
2.115. HOWES, A. (in press). Cognitive modelling: Experiences in human-computer interaction. In T.L. Nyerges (Ed.), Cognitive Aspects of Human-Computer Interaction for Geographic Information Systems. The Netherlands: Kluwer Academic Publishers.
2.116. MARCEL, A.J. (1992). The personal level in cognitive rehabilitation. In N. von Steinbüchel, D.Y. von Cramon & E. Pöppel (Eds.), Neuropsychological Rehabilitation (pp. 155-168). Berlin: Springer-Verlag.
2.117. MARCEL, A.J. (in press). Bodily experience and self: Explanatory priority. In J. Bermùdez, A.J. Marcel & N. Eilan (Eds.), The Body and the Self. Cambridge, Mass.: MIT Press, Bradford Books.
2.118. MAY, J., Byerley, P.F., Denley, I., Hill, B., Adamson, S. Patterson, P. & Hedman, L.R. (1993). The enabling states method. In P.F. Byerley, P.J. Barnard & J. May (Eds.), Computers, Communication and Usability: Design Issues, Research and Methods for Integrated Services (pp. 247-290). Amsterdam: Elsevier Science Publishers, B.V.
2.119. Newell, A., YOUNG, R.M. & Polk, T. (1993). The approach through symbols. In D.E. Broadbent (Ed.), The Simulation of Human Intelligence (pp. 33-70). Oxford: Blackwell.
2.120. Wilson, M., BARNARD, P. & MacLean, A. (1990). An investigation of the learning of a computer system. In P. Falzon (Ed.), Cognitive Ergonomics, Understanding Learning and Designing Human-Computer Interaction (pp. 151-172). London: Academic Press Ltd.
2.121. YOUNG, R.M. (in press). Functionality matters: Capacity constraints and Soar. In D. Steier & T. Mitchell (Eds.), Mind Matters: Contributions to Cognitive and Computer Science in Honor of Allen Newell. Hillsdale, N.J.: Lawrence Erlbaum Associates.
2.122. YOUNG, R.M. & BARNARD, P.J. (1992). Multiple uses of scenarios: A reply to Campbell. SIGCHI Bulletin, 24, p.10.
Conference Proceedings
2.123. BARNARD, P.J. (1993). Modelling users, systems and design spaces (ESPRIT Basic Research Action 3066). In M.J. Smith & G. Salvendy (Eds.), Human-Computer Interaction: Applications and Case Studies. Advances in Human Factors/Ergonomics Vol. 19A: Proceedings of HCI International 1993 (pp. 331-336). Amsterdam: Elsevier Science Publishers B.V.
2.124. BARNARD, P.J. & Harrison, M.D. (1992). Towards a framework for modelling human-computer interactions. In J. Gornostaev (Ed.), Proceedings of the International Conference on Human Computer Interaction, EWHCI '92 (pp. 189-196). Moscow: ICSTI.
2.125. Bradshaw, J.A. & YOUNG, R.M. (1991). Evaluating the behaviour of the BAs 146 hydraulic system using the Doris system. In B. Neumann (Ed.), Tenth European Conference on Artificial Intelligence (pp. 739-743). John Wiley & Sons Ltd.
2.126. CHURCHILL, E.F. & YOUNG, R.M. (1991). Modelling representations of device knowledge in Soar. In L. Steel & B. Smith (Eds.), AISB91: Proceedings of the Eighth Conference of the Society for the Study of Artificial Intelligence and Simulation of Behaviour (pp. 248-255). London: Springer-Verlag.
2.127. BROWN, I.D. (1991). Highway hypnosis: Implications for road traffic researchers and practitioners. In A.G. Gale, I.D. Brown, C.M. Haslegrave, S.P. Taylor & I. Moorhead (Eds.), Vision in Vehicles III (pp. 459-466). Amsterdam: Elsevier Science Publishers B.V. (North-Holland).
2.128. BROWN, I.D. (1991). Introduction to DRIVE Project V1041: Generic Intelligent Driver Support (GIDS). In Y. Queinnec & F. Daniellou (Eds.), Designing for Everyone: Proceedings of the 11th Congress of the International Ergonomics Association (pp. 1533-1535), (Paris, 1991). London: Taylor & Francis.
2.129. BROWN, I.D. (1991). Overview of the Conference and the Road to VIV IV. In A.G. Gale et al. (Eds.), Vision in Vehicles III (pp. 481-486). Amsterdam: Elsevier Science Publishes B.V. (North-Holland.
2.130. BROWN, I.D. (in press). Reducing accident risk for the heavy goods vehicle driver. In Proceedings of the 12th World Congress of the International Association of Accident and Traffic Medicine (Helsinki, Finland, 23-25 June 1992).
2.131. BROWN, I.D. (in press). The case for 'long-term care' in driver training. In Proceedings of AGAM Nederland Conference on 'New Perspectives on Integrated Traffic Training in Europe' (Best, The Netherlands, 6-8 May, 1992).
2.132. Gale, A.G., BROWN, I.D., Haslegrave, C.M., Kruysse, H.W. & Taylor, S.P. (Eds.). (1993). Vision in Vehicles IV. Amsterdam: North Holland.
2.133. Gale, A.G., BROWN, I.D., Haslegrave, C.M., Moorhead, I. & Taylor, S.P. (Eds.). (1991). Vision in Vehicles III. Amsterdam: North-Holland.
2.134. GROEGER, J.A. (1991). Learning from learning: Principles for supporting drivers. In Proceedings of the 24th ISATA International Symposium on Automotive Technology and Automation (pp. 703-709). Croydon: Automotive Automation.
2.135. GROEGER, J.A. (1991). Supporting training drivers and the prospects for later learning. In Advanced Telematics in Road Transport: Proceedings of DRIVE Conference, Vol. 1 (pp. 314-330), (February 4-6 1991), Amsterdam: Elsevier.
2.136. GROEGER, J.A. (in press). Introducing ADA: Adaptation and Instructional Support for Drivers. In R. Haller (Ed.), Proceedings of Seminar on MMI Basic Research (Karlsruhe, Germany, April 17-18, 1991). Brussels: Prometheus Office.
2.137. GROEGER, J.A. & Cavallo, V. (1991). Judgements of time-to-collision and time-to-coincidence. In A.G. Gale et al. (Eds.), Vision in Vehicles III (pp.27-34). Elsevier Science Publishers B.V. (North-Holland).
2.138. GROEGER, J.A. & CLEGG, B.A. (1993). What is learned by drivers during training, and how can we improve it? In W. Barta (Ed.), Proceedings of "Novice Driver Education" (Edmonton, April 22-23, 1993). Edmonton: The University of Alberta and Alberta Motor Association.
2.139. GROEGER, J.A. & GRANDE, G.E. (1991). Too little too soon: Limitations of training and the need for continuing driver support. In Y. Queinnec & F. Daniellou (Eds.), Designing for Everyone: Proceedings of the 11th Congress of the International Ergonomics Association (pp. 1492-1494), Paris, 1991. London: Taylor & Francis.
2.140. GROEGER, J.A., GRANDE, G. & BROWN, I.D. (1991). Accuracy and safety: Effects of different training procedures on a time-to-coincidence task. In A.G. Gale et al. (Eds.), Vision in Vehicles III (pp.35-43). Elsevier Science Publishers B.V. (North-Holland).
2.141. GROEGER, J.A. & Kuiken, M.J. (1993). Performance histories and adaptive instructional support. In Advanced Transport Telematics (Proceedings of the Technical Days Volume II) (pp. 340-343) (Brussels, March 8-10, 1993). Commission of the European Communities.
2.142. Harrison, M.D. & BARNARD, P.J. (1993). On defining requirements for interactions. In A. Finkelstein (Ed.), Proceedings of the IEEE International Workshop on Requirements Engineering (pp. 50-54). New York: IEEE.
2.143. Harrison, M.D., BLANDFORD, A.E. & BARNARD, P.J. (1994). The requirements engineering of user freedom. In F. Paterno (Ed.), The Design, Specification and Verification of Interactive Systems: Proceedings of the Eurographics Workshop (pp. 181-194), (Carrara, Italy, 8-10 July). Pisa: CNUCE-CNR.
2.144. HOWES, A. (1993). Recognition-based problem solving. In Proceedings of the 15th Annual Conference of the Cognitive Science Society (pp. 551-556). (Boulder, Colorado, 1993). Hillsdale, N.J.: Lawrence Erlbaum Associates.
2.145. Kuiken, M.J. & GROEGER, J.A. (in press). Reducing drivers' speeds on bends: Differential effects of feedback. In A.G. Gale, I.D. Brown, C.M. Haslegrave, I. Moorhead & S. Taylor (Eds.), Vision in Vehicles III. Amsterdam: Elsevier Science Publications.
2.146. MacLean, A., YOUNG, R., Bellotti, V. & Moran, T. (1992). Design space analysis: Bridging from theory to practice via design rationale. In Proceedings of Esprit Conference 1991 (pp. 720-730) (Brussels, Nov 25-29, 1991), CEC.
2.147. MARCEL, A.J. (1993). Slippage in the unity of consciousness. In Ciba Foundation Symposium No. 174 - Experimental and Theoretical Studies of Consciousness. Chichester: John Wiley & Sons.
2.148. MARCEL, A.J. (1994). What is relevant to the unity of consciousness? In C. Peacocke (Ed.), Philosophy of Mind: Proceedings of the British Academy, Vol. 83.
Technical Reports and Theses
2.149. Allsop, R.E., BROWN, I.D., GROEGER, J.A., & Robertson, S.A. (1991). Approaches to modelling driver behaviour at actual and simulated traffic signals. Contractor Report 264, Transport and Road Research Laboratory, Department of Transport. Crowthorne, Berks: Transport and Road Research Laboratory.
2.150. BARNARD, P.J. (Ed.) (1990). Assimilating Models of Designers, Users and Systems (AMODEUS), Periodic Progress Report No. 1, and Associated Deliverables. Brussels: CEC, September.
2.151. BARNARD, P.J. (Ed.) (1993). AMODEUS 2: ESPRIT Basic Research Action 7040, Periodic Progress Report 1, and Associated Scientific Deliverables. Brussels: CEC.
2.152. BARNARD, P.J. (Ed.) (1994). AMODEUS 2: ESPRIT Basic Research Action 7040, Periodic Progress Report 2, and Associated Scientific Deliverables. Brussels: CEC, July.
2.153. BARNARD, P.J., BLANDFORD, A.E. & MAY. J. (1992). Demonstration of expert system capability. Documentation to support and accompany D19, Constituent part of scientific deliverables associated with APU 2962, pp 136.
2.154. BARNARD, P.J., Coutaz, J., Hammond, N., Harrison, M., Jørgensen, A., MacLean, A. & YOUNG, R. (1992). AMODEUS ESPRIT Basic Research Action 3066 Final Report, D23. Brussels: CEC.
2.155. BARNARD, P.J. & Harrison, M.D. (Eds.) (1991). Assimilating Models of Designers, Users and Systems (AMODEUS), Periodic Progress Report No. 2, and Associated Scientific Deliverables. Brussels: CEC, September.
2.156. BARNARD, P.J. & MAY, J. (1993). Real time blending of data streams: A key problem for the cognitive modelling of multimodal systems. AMODEUS working Paper UM/WP 10; May 1993, pp 25. Constituent part of scientific deliverable D2 associated with APU 2963.
2.157. BOURKE, P.A. (1993). A general factor in dual task performance decrement. Unpublished PhD Thesis, University of Cambridge.
2.158. BROWN, I.D. (1992). Fatigue and driving. Unpublished review, commissioned by the Department of Transport via the Medical Commission on Accident Prevention and submitted by the author in January, 1992.
2.159. CHAPMAN, P.R. (1992). Subjective risk and memory for driving situations. Unpublished PhD Thesis, University College London.
2.160. CHURCHILL, E.F. (1993). Models of models: Cognitive, computational and empirical investigations of learning a device. Unpublished PhD Thesis, University of Cambridge.
2.161. DUFF, S.C. (1990). Task and device representations in the use of interactive systems. Unpublished PhD Thesis, University of Cambridge.
2.162. Duke, D., Duce, D., BARNARD, P.J, Harrison, M.D. & MAY, J. (1994). On the integration of cognitive and system models, AMODEUS, ID/WP26, pp. 21. Constituent part of the D7 scientific deliverable associated with APU 3122.
2.163. GROEGER, J.A. (1991). Expectations and experience of taxi and minicab use. Technical Report to Department of Transport on CON 3030. London: The Suzy Lamplugh Trust..
2.164. GROEGER, J.A. (1991). Relative risks of taxi and minicab travel. Consultant's Report to Department of Transport on CON 3030. London: The Suzy Lamplugh Trust.
2.165. GROEGER, J.A. & GRANDE, G.E. (1992). Meeting the support requirements of drivers with different levels of traffic experience: An evaluation. Deliverable ADA3, DRIVE Project 1041: Generic Intelligent Driver Support Systems. The Netherlands: University of Groningen, Traffic Research Centre.
2.166. GROEGER, J.A., GRANDE, G.E. & BROWN, I.D. (in press). Decisions at simulated traffic signals. Report to the Transport and Road Research Laboratory, (CON/98 34/35).
2.167. GROEGER, J.A. & Kuiken, M.J. (1992). Instructional support: Scenario and support definition. Deliverable 211, DRIVEII Project V2004: Application of a Real-time Intelligent Aid for Driving and Navigation Enhancement (ARIADNE). Rover Group Ltd., U.K.
2.168. GROEGER, J.A. & Kuiken, M.J. (1993). Effects of instructional sets. Deliverable 212, DRIVE II Project V2004: Application of a Real-time Intelligent Aid for Driving and Navigation Enhancement (ARIADNE). Rover Group Ltd., U.K.
2.169. GROEGER, J.A., Kuiken, M., GRANDE, G.E., Miltenburg, P., BROWN, I.D. & Rothengatter, J.A. (1990). Preliminary design specifications for appropriate feedback provision to drivers with differing levels of traffic experience. Deliverable ADA1, DRIVE Project 1041: Generic Intelligent Driver Support Systems. Traffic Research Centre, University of Groningen.
2.170. Kuiken, M.J. & GROEGER, J.A. (1991). Report on Feedback Requirements and Performance Differences of Drivers. Deliverable ADA2, DRIVE Project 1041: Generic Intelligent Driver Support Systems. Traffic Research Center, University of Groningen, The Netherlands.
2.171. LEE, W-O. (1993). Incremental change in the development of expertise in using interactive systems. Unpublished PhD Thesis, University of Cambridge.
2.172. MAY, J. (1993). The part-whole problem in perception. Electronic document, AMODEUS Pres 4, October; pub/amodeus/usemod/pres4.hqx @ ftp.mrc-apu.cam.ac.uk
2.173. MAY J. & BARNARD, P.J. (1993). Modelling the user's interpretation of dynamic displays. AMODEUS: Working paper UM/WP9; May 1993, pp.28. Constituent part of scientific deliverable D2 associated with APU 2963.
2.174. Muskaug, R. & GROEGER, J.A. (1992). On-site Tutoring and Enforcement Systems. Deliverable 120, DRIVE Project V2009: Detection, Enforcement and Tutoring for Error Reduction (DETER). The Netherlands: University of Groningen, Traffic Research Centre.
2.175. Piersma, E.H., Kuiken, M.J., van Winsum, W., GROEGER, J.A., Stove, A.G. & Verwey, W.B. (1992). Specification of the Requirements for a Second Generation GIDS System Prototype. Deliverable 01, DRIVE II Project 2004: Application of a Real-time Intelligent Aid for Driving and Navigation Enhancement (ARIADNE). Rover Group, U.K.
2.176. PRICE, M.C. (1991). Processing and awareness of masked stimuli. Unpublished PhD Thesis, University of Cambridge.
2.177. TWEEDIE, L., BARNARD, P.J. & MAY J. (1993). AnimICS v 5.0. Electronic document, pub/amodeus/usemod/AnimICS_5.hqx @ ftp.mrc-apu.cam.ac.uk
2.178. YOUNG, R. M. & WHITTINGTON, J. E. (1990) Interim report on means-ends analysis in Soar. ESPRIT Basic Research Action 3066, Amodeus Project Document RP5/WP3 (also in Deliverable D5).
Dissemination
2.179. BARNARD, P.J. (1990). Research on Human-Computer Interaction at the MRC Applied Psychology Unit. In J.C. Chew & J. Whiteside (Eds.), CHI '90 Conference Proceedings: Human Factors in Computing Systems (pp. 379-380), (Special issue of the SIGCHI Bulletin). Seattle, Washington: ACM Press.
2.180. TWEEDIE, L. & BARNARD, P.J. (1992). The Interactive Talk: A new tool for presenting complex theory. Psychology Software News, 3, 43-45.
REFERENCES TO OTHER WORK
Allport, D.A. (1980). Attention and performance. In G. Claxton (Ed.), Cognitive Psychology: New Directions (pp. 112-153). London: Routledge and Kegan Paul.
BADDELEY, A.D., Bressi, S., Della Sala, S., Logie, R. & Spinnler, H. (1991). The decline of working memory in Alzheimer's Disease: A longitudinal study. Brain, 114, 2521-2542.
Bahrick, L.E. (1988). Intermodal learning in infancy: Learning on the basis of two kinds of invariant relations in audible and visible events. Child Development, 59, 197-209.
BARNARD, P.J. (1993). Applying cognitive theory: Human-computer interaction as a case study. Invited Lecture presented at the BPS Cognitive Section Annual Conference (Cambridge, England. September).
BARNARD, P.J. (1994). Unpacking the central executive functions. Paper presented at the International Conference on Working Memory (Cambridge, England, July 20-22nd).
BARNARD, P., Wilson, M. & MacLean, A. (1988). Approximate modelling of cognitive activity with an expert system: A theory-based strategy for developing an interactive design tool. The Computer Journal, 31, 445-456.
BLANDFORD, A.E. (1993). An agent-theoretic approach to computer participation in dialogue. International Journal of Man-Machine Studies, 39, 965-998
BLANDFORD, A.E. (1993). Applying the WOM to WOMBAT: evaluation of a tool to support learning about design evaluation. Design Studies, 14 228-246
BLANDFORD, A.E. (in press, a). Teaching through collaborative problem solving. Journal of Artificial Intelligence in Education.
BLANDFORD, A.E. (in press, b). Deciding what to say: an agent-theoretic approach to tutorial dialogue. In R-J. Beun, M. Baker & M. Reiner(Eds.), Dialogue and Instruction. Springer-Verlag.
Bradshaw, J. & Young, R.M. (in press). Shared causal knowledge as a basis for communication between expert and knowledge acquisition system. In J. Boose, B. Gaines & M. Linster (Eds.), Proceedings of the 2nd European Workshop on Knowledge Acquisition. Bonn.
Buckingham-Shum, S. & Hammond, N. (in press). Delivering HCI modelling to designers: A framework, and case study of cognitive modelling. Interacting With Computers.
Driver, J. & Spence, C. (1994). Spatial synergies between auditory and visual attention. In C. Umilta & M. Moscovitch (Eds.), Attention and Performance XV.
DUFF, S.C. (1989). Reduction of action uncertainty in process control systems: The role of device knowledge. In E.D. Megaw (Ed.), Contemporary Ergonomics 1989 - Proceedings of the Ergonomics Society's 1989 Annual Conference (pp. 213-219). Reading, England. London: Taylor & Francis.
DUNCAN, J. (1981). Directing attention in the visual field. Perception & Psychophysics, 33, 20-28.
DUNCAN, J. (1984). Selective attention and the organization of visual information. Journal of Experimental Psychology: General, 113, 501-517.
DUNCAN, J. & Humphreys, G.W. (1989). Visual search and stimulus similarity. Psychological Review, 96, 433-458.
Fox, J., Cooper, R., Farringdon, J. & SHALLICE, T. (1992). Building computational models of cognition. In B. Silverman (Ed.), Proceedings of Workshop on Expert Judgement, Human Error and Intelligent Systems (10th European Conference on Artificial Intelligence, Vienna).
Frith, C. (1992). The Cognitive Neuropsychology of Schizophrenia. Hove: Lawrence Erlbaum Associates.
Frith, C.D. & Done, J. (1986). Routes to action in reaction time tasks. Psychological Research, 48, 169-177.
Frith, U. & Robson, J.E. Perceiving the language of films. Perception, 4, 97-103.
Funahashi, S., Bruce, C.J. & Goldman-Rakic, P.S. (1989). Mnemonic coding of visual space in the monkey's dorsolateral prefrontal cortex. Journal of Neurophysiology, 61, 331-349.
HOWES, A. (1992). Learning task-action mappings by exploration. Unpublished PhD Thesis, University of Lancaster.
HOWES, A. & Payne, S.J. (1990a). Display-based competence: Toward user models for menu-driven interfaces. International Journal of Man-Machine Studies, 33, 637-655.
HOWES, A. & Payne, S.J. (1990b). Semantic analysis during exploratory learning. In J. C. Chew & J. Whiteside (Eds), CHI'90: Proceedings of the Conference on Human Factors in Computing Systems (pp. 399-406). New York: ACM Press.
HOWES, A. & Payne, S. J. (1990c). Supporting exploratory learning. In D. Diaper, D. Gilmore, G. Cockton & B. Shackel (Eds), Human-Computer Interaction - INTERACT'90 (pp. 881-885). Amsterdam: Elsevier Science Publishers.
Kinsbourne, M. (in press). Awareness of one's own body: A neuropsychological hypothesis. In J. Bermùdez, A.J. Marcel & N. Eilan (Eds.), The Body and the Self. Cambridge, Mass.: MIT Press, Bradford Books.
LAVIE, N. (1992). Perceptual load and physical distinctiveness as determinants of the locus of attentional selection. Unpublished dissertation (Hebrew).
McGurk, H. & Macdonald, J. (1976). Hearing lips and seeing voices. Nature, 264, 746-748.
Norman, D. A. (1981). Categorization of action slips. Psychological Review, 88, 1-15.
Reason, J. T. & Mycielska, K. (1982). Absent Minded? The Psychology of Mental Lapses and Everyday Errors. Englewood Cliffs, N.J.: Prentice-Hall.
Rizzolatti, G. & Gallese, V. (1988). Mechanisms and theories of spatial neglect. In F. Boller & J. Grafman (Eds.), Handbook of Neuropsychology (Vol. 1).
Robertson, L.C., & Lamb, M.R. (1991). Neuropsychological contributions to theories of part/whole organization. Cognitive Psychology, 23, 299-330.
Rubens, A.B. (1985). Caloric stimulation and unilateral visual neglect. Neurology, 35, 1019-1024.
Suchman, L. (1987). Plans and Situated Actions. Cambridge: Cambridge University Press.
SELLEN, A.J. & Norman, D.A. (1992). The psychology of slips. In B. J. Baars (Ed.), Experimental Slips and Human Error: Exploring the Architecture of Volition (pp. 317-339). New York: Plenum Press.
Shepard, R.N. & Cooper, L.A. (1982). Mental Images and their Transformations. Cambridge, Mass.: MIT Press, Bradford Books.
TEASDALE, J.D. & BARNARD, P.J. (1993). Affect, Cognition and Change. Hove: Lawrence Erlbaum Associates.
Tipper, S.P. (1985). The negative priming effect: Inhibitory priming by ignored objects. Quarterly Journal of Experimental Psychology, 37A, 571-590.
Treisman, A. (1988). Features and objects: The Fourteenth Bartlett Memorial Lecture. Quarterly Journal of Experimental Psychology, 40A, 201-237.
Tsal, Y. & LAVIE, N. (1988). Attending to colour and shape: The special role of location in selective visual processing. Perception & Psychophysics, 44, 15-21.
Tsal, Y. & LAVIE, N. (1993). Location dominance in attending to color and shape. Journal of Experimental Psychology: Human Perception & Performance, 19, 131-139.
Tsal, Y., Meiran, N. & LAVIE, N. (1994). The role of attention in Illusory Conjunctions. Perception & Psychophysics, 55, 350-358.
Collaborations
Barnard
UK based
Eldridge - Rank Xerox Cambridge
Fowler - Clinical Psychology, Cambridge
Palmer - Clinical Psychology, Norwich
Harrison - Computer Science, York
Murray - Psychiatry, Cambridge
Outside UK
Jørgensen - Psychology, Copenhagen
Byerley, Böcker - SEL Pforzheim
Foa - Psychology, Pennsylvania
Linehan - Psychology, Seattle
Barnard, R Young (Esprit AMODEUS)
UK based
Harrison, Duke - Computer Science, York
Hammond, Buckingham-Shum - Psychology, York
MacLean, Bellotti - Rank Xerox, Cambridge
Duce - Rutherford Appleton Labs
Outside UK
Jørgensen, Aboulafia - Psychology, Copenhagen
Bernsen, Ramsay, May - Cognitive Science, Roskilde, Denmark
Nielsen - Business School, Copenhagen
Verjans - Expert Systems, Leuven
Löwgen, Sjøberg - Computer Science, Linkoping, Sweden
Coutaz, Nigay, Salber - LGI, Grenoble
Faconti, Paterno - CNR-CNUCE, Pisa
Darzentas, Spirou - Mathematics, University of the Aegean, Greece
Brown
UK based
Maycock - Transport Research Laboratory, Crowthorne
Outside UK
Michon - Traffic Science, Groningen
Duncan
UK based
Burgess - Psychology, London
Freer - MRIS, Cambridge
Hall - Medicinal Chemistry, Cambridge
Humphreys - Psychology, Birmingham
Johnson - Rehabilitation, Cambridge
Pickard - Neurosurgery, Cambridge
Robbins - Experimental Psychology, Cambridge
Outside UK
Bundesen - Psychology, Copenhagen
Desimone - NIMH, Bethesda
Orban - Neuroscience, Leuven
Seitz - Heinrich-Heine University, Dusseldorf
Shapiro - Psychology, Calgary
Goodrich
UK based
Harrison - Psychology, Charing Cross & Westminster Medical School, London
Henderson - Psychology, Hertfordshire
Driver - Psychology, Cambridge
Ashburn - Rehabilitation, Southampton
Groeger
UK based
Stove - Philips Research Labs
McMurran - Rover Cars
Parks - HUSAT, Loughborough
Allsop - UCL, London
Outside UK
Rothengatter - TRC, Groningen
Michon - NISCALE, Leiden
Verwey - TNO Institute for Perception, Netherlands
Bosser - Munchen
Cavallo - France
Esteve - CNRS-LAAS, France
Vallet - INRETS, France
de Santos - Braga, Portugal
Fuller - Dublin
Lynch - CARA Software, Dublin
Muskaug - Institute of Transport Economics, Norway
Andrew - Volvo, Sweden
Alm - VTI, Sweden
Houghton
UK based
Tipper - Psychology, Bangor
Lavie
UK based
Driver - Psychology, Cambridge
Outside UK
Tsal - Psychology, Tel-Aviv
Ivry - Psychology, Berkeley
Driver - Psychology, Cambridge
Marcel
UK based
Evans - Anaesthetics, John Radcliffe Hospital, Oxford
Garvie - Speech Therapy, Cambridge
Cole - Clinical Neurophysiology, Southampton
Eilan - King's College, Cambridge
Bermùdez, King's College, Cambridge
Campbell - New College, Oxford
Outside UK
Tegnèr - Neurology, Stockholm
Bisiach - Psychology, Padova
Berti - Psychology, Bologna
Ladavàs - Psychology, Bologna
Pizzamiglio - Psychology, Rome
Paillard - CNRS, Marseille
Sellen
UK based
Duthie - Papworth Hospital, Cambridge
Outside UK
Runciman, Webb, Williamson, Russell - Anaesthetics, Royal Adelaide Hospital
R Young
Abowd - Computer Science, Carnegie Mellon University
Lewis, Polson, Rieman - Cognitive Science, University of Colorado
Ritter - Psychology, Nottingham