Ординатура / Офтальмология / Английские материалы / Eye Movements A Window on Mind and Brain_Van Gompel_2007

although here and there problems of divided processing and limited resources may also play a role.

Principal viewpoints on the role of attention in human information processing appear to oscillate between two extremes: Posner & Petersen (1990) postulate the existence of an integrated attention system that is anatomically separate from the various data processing systems performing operations on specific inputs. This attention system is composed of a network of anatomical areas that carry out different functions; attention is thus a property neither of a single center nor of the brain as whole. Posner (1990) organizes the attention system into three subsystems responsible for: (a) orienting to sensory events; (b) detecting signals for focal (conscious) processing; and (c) maintaining a vigilant or alert state.

An alternative view is advocated, for example, by Allport (1992), who rejects the idea of one integrated functional system. Instead, he emphasizes the existence of a multitude of qualitatively different mechanisms that are involved in implementing visual-attentional selectivity. Allport discusses two questions that were at the core of dispute for more than 25 years: the question of early vs late selection and the question of which processes require attention (e.g. automatic vs controlled). He arrives at the conclusion that the quest for unified answers to these questions is hopeless, as “ there is a rich diversity of neuropsychological control mechanisms of many different kinds (and no doubt many yet to be discovered) from whose cooperative and competitive interactions emerge the behavioral manifestations of attention” (p. 203). This general statement also applies to the domain of reading, where, as we will see below, different mechanisms of selection and preferred processing co-exist and serve different specific purposes.

In our discussion we will focus on processes and mechanisms tapped by (a) Posner’s functions and (b) visual selection of information and processing of selected information at the expense of other potential visual input. Even within this limited domain, it is interesting to observe subtle but theoretically important differences between theorists’ positions on what the term “attention” shall refer to. We can illustrate this using the example of attention as a directable resource vs attention as a controller, a mechanism or an assembly of mechanisms for visual selection (Johnston & Dark, 1986).

According to a popular, almost canonical, definition, selective attention is a limited resource, something that can be “directed” to or “focused” on a certain location or object. Posner & Peterson (1990) not only specify operations of disengagement, movement and re-engagement of attention but also make suggestions about the neural circuitry underlying visual attention shifts. “The parietal lobe first disengages attention from its present focus, then the midbrain area acts to move the index of attention to the area of the target and the pulvinar is involved in reading out data from the indexed locations” (pp. 28–29, our italics). A second important component of this view is that attentional processing is confined to a relatively narrow area, often described as having a focus and a sharp margin (James, 1890). Posner (1980, p. 172) introduced the metaphor of “a spotlight that enhances the efficiency of the detection of events within its beam”.

From a slightly different angle, attention can be conceptualized as a controller that directs or allocates a limited resource. For example, Yantis (1998) begins his review on the control of attention by emphasizing that the visual system accomplishes object

recognition by visually selecting a relevant or salient part of the image and then concentrates processing there before moving on to another part. He continues by stating that “ The mechanism that accomplishes selection is called visual attention ” (p. 223). This view is also quite popular in the attention literature and it is not unlikely that one is confronted with multiple references to the concept of visual attention within the same publication: First, as a mechanism that does selection and then as a entity that is moved as a result of this selection, or, similarly, as a beam of enhanced processing due to resource allocation.

Given these (and other) different ways to specify the term “attention”, we will adopt a minimalist definition, using “attention” as a synonym for visual spatial selection and preferred processing of visual input, leaving details unspecified until specifications are required and become useful for our discussion. At some points in this chapter it will become difficult to draw a clear line between visual processing terminology (e.g. letter identification, lexical processing) and attention terminology (attention allocation, shift of attention). This is a consequence of the custom of using the concept of attention simply to describe perceptual and cognitive processing. Also, in many current discussions, the necessary distinction between attention as a description of what has been observed vs an agent that causes what can be observed is hard to establish, a difficulty that our chapter shares with much of the literature in the field (Johnston & Dark, 1986; Neumann, 1987).

Below we will raise a number of questions around the issues of attention and spatial selection in reading. We hope that it will become clear that, although these questions may appear straightforward, attempts to give clear answers can turn out to be quite complicated. With these issues in mind we review some basic attention research as well as relevant work within the domain of reading itself.

2.2. What is the purpose of spatial selection in reading?

It is basic knowledge in the field that the efficiency of processing during reading across the functional visual field is not simply a function of visual factors like acuity and lateral masking (see Radach, Kennedy & Rayner, 2004, for a recent compilation of empirical work with a focus on parafoveal processing in reading). As the classic studies of Rayner, McConkie and their colleagues have shown, the “span of letter identification” during a given fixation extents about 8–9 letters to the right and about 4 letters to the left. The fact that this asymmetry reverses when Hebrew readers read from right to left is seen as a consequence of the dynamic allocation of visual processing resources or “attention” (see Rayner & Sereno, 1994 and Rayner, 1998, for comprehensive reviews). An intriguing question is whether the types of spatial selection relevant for linguistic processing and oculomotor control are the same or different. If they turn out to be the same, this would raise the further question of how attention is divided between the two tasks (Hoffman, 1998). If they are different, then we need to ask further questions about the nature of this difference and about what specific mechanisms accomplish selection to serve both linguistic and visuomotor processing.

A look into the relevant literature suggests that there is solid empirical evidence in support of different control processes for the two types of selections. In fact the fundamental architecture of the human visual system suggests the existence of this division. Ungerleider & Mishkin (1982) proposed that the processing of an object’s qualities and its spatial location rest on different neurobiological grounds. They distinguished a “ventral stream” of projections from the striate cortex to the inferior temporal cortex from a “dorsal stream” of projections terminating in the posterior parietal cortex. This distinction is widely known as the “what” vs “where” – systems of visual processing. An alternative perspective on modularity in the cortical visual system was put forward subsequently by Goodale & Milner (1992), placing less emphasis on informational input and taking more account of output requirements, the purposes of visual processing. They refer to the ventral and dorsal processing streams as two visuomotor systems, named the “what” vs “how” system. In this dichotomy, the dorsal system is concerned with processing for the purpose of motor control, including, of course, the preparation and execution of eye movements. Goodale & Milner (1992) note that “attention needs to be switched to particular locations and objects whenever they are the targets either for intended action or for identification ” (p. 23) and emphasize that “ spatial attention is physiologically non-unitary and may be as much associated with the ventral system as with the dorsal” (p. 24).

On an information processing level, the neurobiological segregation of a ventral from a dorsal system has its analogy in the distinction between “selection for object recognition” (e.g. La Berge & Brown, 1989) or “selection for perception” (Deubel, Schneider, & Paprotta, 1998; Kowler, Anderson, Dosher, & Blaser, 1995) vs “selection for action” (Allport, 1987; Neumann, 1987). The latter mechanism can be seen as solving the problem of parameter specification: Space-based motor actions like saccadic eye movements require a mechanism that supplies the motor system with the spatial parameters of the intended target object. Therefore, selection-for-space-based-motor-action is considered a central function of visual attention (Schneider, 1995). Following this view, we can assume as a working hypothesis that, in normal reading too, two qualitatively different forms of selection are operational. Whether the term “attention” is used to characterize either form or whether it is being reserved for describing selection for the purpose of recognition (i.e. linguistic processing) is merely a question of terminological preference.

2.3. What are the units spatial selection is operating on?

McConkie & Zola (1987) characterize the process of word-selection as being based on a two-dimensional representation of the stimulus array provided by early visual processing. This stimulus structure can be conceived as a four-level object hierarchy (letters, words, lines, pages) consisting of objects that are relatively homogenous in size and shape. Within this rapid process of parsing of the available visual information, low spatial frequency word-objects, bounded by empty spaces, are assigned the role of potential saccade targets. This proposal can be seen as the first specification of “preattentive” processing in the domain of research on continuous reading.

be processed concurrently. That is starting from the beginning of a fixation, or does “perceptual attention” need to be shifted away from its initial focus and re-allocated at a new location? This controversial point, which, as pointed out in the initial part of this chapter, is central to current models of eye-movement control in reading, will be considered in the following section.

2.4. What is the time course of spatial selection?

Estimations of how long it takes attention to “move” from one stimulus location to another are often based on the slopes of search functions in serial search tasks, used to compute the amount of time spent per item in the visual display.4 The time per unit suggested by this procedure is usually in the order of about 25–75 ms (Egeth & Yantis, 1997), with 50 ms as the value initially reported in the classic study by Treisman & Gelade (1980). Other techniques, such as the attentional blink paradigm, have yielded estimates that are in the order of 150 to well over 300 ms and are thus similar to typical fixation durations (Horowitz, Holcombe, Wolfe, Arsenio, & Dimase, 2004; Theeuwes, Godijn & Pratt, 2004; Ward, 2000). In a recent review, Wolfe (2003) discussed the apparent contradiction between these vastly different estimates using the metaphor of a car wash: Cars enter the car wash in a sequence, but at any point in time several cars are being washed in parallel. Using this metaphor, the 50-ms-per-item slope corresponds to the rate at which “cars” pass though the system. However, it may still take 300 ms for every car to get washed, leading to a 300 ms “attentional dwell time”. Logan (2005) recently used a clever experimental technique to separate cue-encoding time from the time it takes to switch attention. He estimated a time interval of 67–74 ms for cue encoding and 76–101 ms for attention switching.

Eriksen (1990) concluded from a large body of research using attentional cueing paradigms that an enhancement of processing at a cued location (taken to indicate a shift of attention) starts to develop after 50 ms and continues to grow up to an asymptote at about 200 ms after the cue. As Kinchla (1992) put it: “In terms of the spotlight metaphor, it is as if the spotlight went off at one point and then gradually came on again at the target location” (p. 726). Most interesting in this context is the conclusion that the shift of attention itself takes time in addition to the time for the processing event that triggers that shift. Taken together, there appears to be consensus in the field that, if an attention shift is assumed to take place, there must be a specific time interval assigned to its preparation and execution.

A related question that is central to our line of discussion concerns the point in time during an eye fixation in reading when letters and words are being “attended” within and outside the focus of foveal vision. For one type of visual selection in reading, selection

4 Of course, this approach rests on the assumption that visual search is indeed strictly serial. Alternatives involving mechanisms of parallel processing have been proposed, however, for example, by Duncan & Humphries (1989), Wolfe (1994) and others. Findlay & Gilchrist (1998) have provided empirical evidence that makes it unlikely that search displays are scanned item-by-item by an attentional beam.

for the purpose of visuomotor processing (a special case of “selection for action”), the answer to this question appears straightforward. Pollatsek & Rayner (1982) reported an experiment where spaces between words were filled at various times contingent upon the onset of a reading fixation. Filling the first space to the right of the current fixation position had no effect when it occurred later than 50 ms after the beginning of the fixation, suggesting that basal information on the visual configuration is acquired very early. This is in harmony with the idea advocated above that early vision provides an array of potential saccade targets (initially perhaps in the form of low spatial frequency word-objects) among which the processing system chooses on the basis of visuomotor constraints and/or linguistic processing.

More interesting is the demonstration by Rayner, Inhoff, Morrison, Sloviaczek, & Bertera, (1981) that reading can proceed quite smoothly when the entire text is masked after 50 ms. Recently, Rayner, Liversedge, White, & Vergilino-Perez (2003) showed that there are robust word frequency effects when fixated words disappear after 60 ms. These findings can be taken to suggest that “readers acquire the visual information necessary for reading during the first 50–70 ms of a fixation ” (Rayner, 1998, p. 378). This appears to be in contradiction with results by Blanchard, McConkie, Zola, & Wolverton (1984). They changed the word displayed during a fixation in sentence reading at various intervals (50, 80 and 120 ms from fixation onset) and received responses by readers who reported to have seen only the first word, both or only the last word. These results are compatible with a model that distinguishes the registration of information by the visual system from its utilization for linguistic processing (McConkie, Underwood, Zola, & Wolverton, 1985). The idea is that, whenever language processing requires new input during a fixation, the visual information that is currently present will be used. Within this model one could accommodate Rayner et al.’s findings by assuming that visual information is registered throughout the fixation (including the first 50–70 ms) and can then be stored for cognitive utilization in terms of cognitive processing. The data presented by Blanchard et al. pose a problem for models assuming a sequential word-to-word movement of attention, because late during a fixation there should be no report of the replacement word, as attention is supposed to have moved away already.

2.5. What is the nature of the relation between attention and eye movements?

It has often been claimed that attentional orienting precedes eye movements to a specific location. However, observations that allow such a conclusion to be drawn are not necessarily informative about the question whether this relation is one of causality or rather one of coincidence, as covert and overt orienting will usually happen to have the same goal (Hoffmann, 1998). The question whether attention and eye movements can operate concurrently on different regions or objects has been studied in dual task paradigms. Participants are asked to move their eye to a specific location and, at the same time, to detect or identify a visual target located at varying distances relative to the saccade target. Performance in the identification task is taken as an indirect measure of attention allocation. If it is possible to spatially dissociate attention and saccades, subjects should

be able to attend to one location (i.e. show a high identification performance) while making a saccade to another.

Studies using this methodology found ample evidence in favor of close links between attention and saccades. For example, Hoffman & Subramaniam (1995) instructed subjects to make a saccade to one of four fixation boxes located left, right, above or below a central fixation point as soon as they heard a tone. The direction of the saccade target was held constant throughout the experiment. Before each trial, an arrow was presented that pointed to one of the four possible locations at which a target letter could be presented for a very short time interval. The arrow was used to direct attention to a specific location, and, as indicated by a detection-only control condition, this manipulation did indeed enhance performance in cue-target match trials. The key result of this study is that targets were detected better in the dual task situation when they occurred in a location to which a saccade was being prepared even in comparison to locations the central arrow was pointing to. This can be taken as evidence that visuospatial attention is required to execute a saccade, rendering the endogenous “attentional” cue ineffective (Hoffman, 1998; Hoffman & Subramaniam, 1995).

Deubel & Schneider (1996) reported experiments using stimulus materials that were similar to a reading situation in that they presented a horizontal string of letters separated by blanks. A central cue designated a specific member of the string as the saccade target. Before the onset of a saccade toward this goal, a discrimination stimulus was presented briefly within the string. Their results, again, indicated a high degree of selectivity: discrimination performance was nearly perfect when the saccade was directed to the critical item, but close to chance level when the target was located only one item to the left or right of the discrimination stimulus. Importantly, conditions designed to provoke a decoupling of “recognition” attention from saccade programming failed to alter the principal results.

Taken together, these studies indicate that object-specific coupling of saccade programming and object discrimination is strong and mandatory, at least in paradigms using dual tasks with explicit oculomotor and recognition instructions.5 This has led to the currently dominant view that a pre-saccadic shift of attention is functionally equivalent to the selection of the peripheral item as the target of a subsequent eye movement, thus constituting the initial step in the programming of a saccade (see Deubel, O’Regan, & Radach, 2000, for a detailed discussion). However, it remains to be seen whether this conclusion extends to other experimental conditions, and, most importantly, whether it is an adequate description of what happens in normal reading.

5 One study claiming that voluntary saccades can be programmed independently of attention shifts is by Stelmach et al. (1997) who used a temporal order judgment as the indicator for attention allocation. However, in agreement with most other studies, they found that under conditions of exogenous cuing, attention clearly “moves” much faster than the eyes. Their finding that under endogenous cueing more than 300 ms is needed to redirect attention to the parafovea is very interesting but perhaps irrelevant to a possible role of attention for reading saccades, as these need to be programmed within a much shorter time interval.

Elaborating on the research described above, Godijn & Theeuwes (2003) recently studied the association of attention with eye movements in a paradigm involving the execution of eye movements to a sequence of targets. They used a dual task paradigm to independently manipulate saccadic target selection and the allocation of attention to a particular location in space. In one experiment, the primary task required the execution of a sequence of saccades to two simultaneously cued object locations of a circular display with eight locations, and the secondary task required the identification of a letter that was presented for one of four time intervals from 47 to 129 ms following the saccade cue. Letters at the locations of both the first and the second saccade target were recognized more accurately than letters presented at a non-cued location, and this occurred even at the shortest dual-saccade cue-letter presentation interval. Five experiments using different task variations and controls provide compelling evidence for the view that attention was allocated (in parallel) to more than one distinct object location, and that parallel attention allocation preceded saccade targeting.

3. Consequences for visual processing and oculomotor control in reading

A number of important conclusions can be drawn from our discussion above for theories and models of oculomotor control in reading. First, it has become obvious that we need to distinguish visual selection for the purpose of letter and word recognition from visual selection for the purpose of movement preparation. As illustrated in Chapter 12, current SAS models assume that “attention” (for the purpose of lexical processing) moves sequentially in a word-by-word fashion. This attention shift is triggered by the completion of lexical processing on the current word. However, a preliminary stage of processing, referred to as L1, is assumed to trigger the programming of an eye movement to the neighboring word. As the authors emphasize, in their theory, “attention is decoupled from eye movement control”. As noted above, the idea that saccades are programmed toward locations that are not (yet) “attended” for the purpose of recognition puts SAS models at odds with the mainstream of research on relations between attention and eye-movement control (Deubel, O’Regan, & Radach, 2000).

But what exactly does “decoupling” mean? It may mean that in fact there is a dissociation between attention for letter recognition and attention for oculomotor control, in which case one specific form of attention shift would be associated with either stream of processing. It may also mean that the selection for movement is left to pre-attentive processing, an idea that brings the SAS theory a step closer to many proposals made within the family of PG models.

The problem of how “covert” (perceptual) and “overt” (visuomotor) attentional processing is being coordinated in reading has received little emphasis in prior theoretical discussions. However, the example of interword regressions shows that this question is far from trivial. Here, SAS theory suggests that many regressions occur when attention is still at word N while the eyes have already advanced to word N + 1 (Reichle, Rayner, & Pollatsek, 2003). If word N turns out be more difficult to process than expected on the