Source: http://www.readingonline.org/research/eyemove.html , some sentences highlighted by P. Rösler (23.03.2006)
Influential Studies in Eye-Movement Research
Eric J. Paulson
Kenneth S. Goodman
Authors' Note: The studies discussed here do not constitute the entire corpus of useful and informative research in the area of eye movement but were chosen because they are representative of the valid, reliable, high-quality work that exists and because each has contributed significantly to the body of knowledge about perceptual process in reading. These studies form the base for research that continues to yield insight into vision and perception in the reading process.
This review is divided into five parts:
Perhaps the first concrete insight into the reading process made
possible by eye-movement research was provided in 1891 by Landolt,
one of Javal's colleagues at the
Dodge (1900) constructed an experiment to explore the type of information the eye picks up while it moves. Two pieces of cardboard were positioned one behind the other, and a slit measuring 4 mm wide was cut into the center of the piece in front. Subjects, who sat before the cardboard pieces, were to fixate first on a point to the left of the slit and then to make a single eye-movement to a point to the right of the slit. The slit itself was not visible from either fixation point. Six different colors were placed on the rear screen five times each to determine whether the subjects saw the color through the slit as they executed the required eye movement. Dodge reported that "when the eye movement was unbroken, the observer was unable to tell what had been exposed or even that anything at all had broken the black of the perimeter" (p. 461). His results indicated that since no useful information is received during the movement of the eyes, research should concentrate on the pauses the eye makes.
Edmund Burke Huey and His Contemporaries
Huey provided what is possibly the first physical record of readers' eye movements. In a procedure that sounds more than a bit uncomfortable for the subjects, a plaster of paris cup with a hole in the center was placed on the cornea of one eye, much as a contact lens. The cup was attached to an aluminum pointer which responded to the slightest eye movement. As a subject read, the pointer traced the movement of the eye on a piece of paper. In addition to demonstrating that the eye regresses a small percentage of the time, Huey's study showed that the first fixation in a line is frequently not at the first word but at the second or even third; likewise, the final fixation is usually not at the last word. Huey's data also demonstrated that readers fixated on anywhere from 20 to 70 percent of the words in a line. This remarkable research, undertaken more than 90 years ago, provided evidence that reading is not a passive process of word-by-word identification, but that readers make choices about where and when to fixate while reading.
Buswell and Judd (Buswell, 1922, 1937; Judd & Buswell, 1922) photographed readers' eye movements in what was for their time a relatively nonintrusive manner. The procedure consisted of photographing a beam of light reflected first to a subject's cornea from silvered glass mirrors, and then from the cornea through a camera lens to moving kinetoscope film. The changing positions of the beam of light were recorded on the film, which provided an "accurate record showing the position and duration of each fixation of the eye while the subject reads" (Buswell, 1922, p. 12).
Judd and Buswell also deserve acknowledgment for the sheer amount of data they collected, analyzed, and disseminated. In addition to providing copious illustrations of what the data look like (there are 90 plates in Judd & Buswell, 1922, alone), they support their conclusions well. Their findings include evidence that not only do different readers read differently, but individual readers read differently in different circumstances. Ever mindful of pedagogical implications, they asserted that readers "need to be made aware of the fact that reading habits should be flexible and properly adapted to the purpose and the type of material which is read" (Buswell, 1937, p. 143).
Judd and Buswell's contributions to our understanding of the reading process are also significant. Their work led them to conclude that reading is not simply a matter of bottom-up word identification but a perceptual process that involves interpretations on the reader's part. On the subject of word identification, even today a subject of great dispute, they argued for the primacy of context in determining the meaning of a word:
[I]n real life the word will always turn up as a part of a sentence and...will have a peculiar shade of meaning through its contrast with other words or through its special relation in the total idea conveyed by the sentence. The notion that a word and its meaning are two fixed pieces of experience that can be tied together is a purely mechanical theory and not adequate... (Judd & Buswell, 1922, p. 4).
Tinker's landmark 1936 study investigated the reliability and validity of eye-movement research as it applies to reading. One of his primary concerns was whether the artificial situation that necessarily accompanied eye-movement studies conducted in the laboratory caused subjects to alter significantly their reading strategies and processes. He had 57 college students read one version of a reading test at a table away from the eye-movement apparatus and then read another version of the test while under typical eye-movement recording conditions. The results were encouraging for eye-movement researchers: “Although some subjects did better and some poorer before the camera, the group as a whole gave an entirely typical performance in the photographic situation” (Tinker, 1936, p. 742). Tinker's conclusion that eye-movement research can reveal authentic reading behavior has allowed workers in this area to extend their findings to situations outside the laboratory.
Despite the exciting work of these early investigators, the studies undertaken at the beginning of the 20th century were followed by a long hiatus, blamed by some on the influence of the prevailing behaviorist doctrine of the time (Rayner & Pollatsek, 1989). By the late 1960s, however, eye-movement recording apparatuses, while operating on the same basic principles as earlier equipment, became much more sophisticated. Microanalyses of eye behavior now became possible. Accordingly, more recent eye-movement research is characterized not by broad generalizations, but by smaller scale contributions to our overall knowledge about the role of the eye in reading.
The Physiology of Eye Movements
A central question since the inception of eye-movement research concerns the amount of information the eye can process with each fixation. This issue became even more important when Dodge's conclusion that we see nothing when the eye is actually in motion began to be replicated empirically (see, for example, Wolverton & Zola, 1983). This suggests, of course, that the only text information available is presented during fixations.
From physiological studies we know several basic facts about how the eye processes information and about the physical constraints that limit how this information is presented to the brain. During a fixation, the eye has access to three regions for viewing information: the foveal, parafoveal, and peripheral. The foveal region is the area that we think of as being in focus and includes 2 degrees of visual angle around the point of fixation, where 1 degree is equal to three or four letters (thus, six to eight letters are in focus). The parafoveal region extends to about 15 to 20 letters, and the peripheral region includes everything in the visual field beyond the parafoveal region. The fovea is concerned with processing detail, with anything beyond producing a marked drop in acuity; words presented to locations removed from the fovea are more difficult to identify (Rayner & Sereno, 1994).
Eye Movements and Perception
Many studies have suggested that the shorter the word is, the more likely it is to be skipped. In order to ascertain whether it was length or a syntactic feature that was responsible for the likelihood of a word being skipped, O'Regan (1979) recorded the eye movements of subjects, each of whom read a pair of sentences that began in the same way but ended differently (for example, “The dog that growled the most was friendly” vs. “The dog that growled ate many biscuits”; “He claimed the ladies the maid knew lived in New York” vs. “He claimed the ladies met many times to discuss”). In these sentences, subjects skipped the substantially more often than they skipped three-letter verbs; for example, in the first pair of sentences given, the was more likely to be skipped in the first sentence than ate in the second. O'Regan summarizes as follows:
The conclusions to be drawn from this experiment are, first, that local eye movement parameters (saccade size, regression probability, number of fixations, and perhaps fixation duration), are controlled sufficiently rapidly to be influenced from moment to moment by information concerning the lexical category of a word in peripheral vision.... [I]t is clear that some systematic influence of sentence structure exists (p. 59).
It seems that when readers use their implicit knowledge of the structure of language along with their constant predictions about upcoming text, they sample from the syntactic cueing system (Goodman, 1996). This enables them to read more efficiently—to skip words that have been confirmed parafoveally.
In the early 1980s, Just and Carpenter recorded the eye-movements of 14 college students who read 15 short excerpts from Time and Newsweek magazines. The subjects were asked only to read normally and to recall what they could of each paragraph after it was finished. The researchers found that readers fixated an average of 67.8 percent of the words, with content words being fixated 83 percent of the time and function words 38 percent of the time (Carpenter & Just, 1983; Just & Carpenter, 1980). This work provided further evidence that not only is every word in a text not fixated, but the syntactic and semantic components of each word play a role in determining whether fixation occurs. In addition, data from these studies concerning the differing lengths of fixations provided evidence for two major assumptions: first, with the “immediacy” assumption, Just and Carpenter asserted that the reader tries to interpret each word in a text as it is encountered rather than holding it in abeyance and assigning meaning later; second, their “eye-mind” assumption holds that the reader's eyes remain fixated on a word as long as the word is being processed.
In 1981, McClelland and O'Regan explored whether the usefulness of information available in the parafovea was dependent on readers' expectations about what the next word in a text would be. In one experiment whose paradigm simulated eye movements, they examined target word-naming reaction time when preview information was used. A sentence from which the last word was missing was displayed on a computer screen. The sentences were of two types: those for which a group of judges (who did not take part in the experiment) accurately predicted the missing word, and those that the judges felt did not allow prediction of the word. When subjects reached the last word, they pressed a button that initiated first a 100 millisecond (ms) display of a preview item, then a 100 ms blank space, and finally the target word. Subjects were to name the target word as soon as they read it. Preview items were of three types: a series of Xs, an item similar to the target word except that the second and penultimate letters were replaced by other letters of the same shape, or the same word as the target word.
Results of this study demonstrate that the speed and ease with which readers can name a target word from a parafoveal preview depend on readers' expectations. As McClelland and O'Regan state, "a priori expectations and context greatly increase the benefit subjects gain from a preview of a word in parafoveal vision" (1981, p. 634). Further, they assert that "our experiments have clarified one point: The ability to derive benefit from the preview we receive of upcoming words in parafoveal vision depends on a prepared mind" (p. 643). That is, readers are able to make use of text information that they have not fixated on but which they have predicted. Fuzzy input is enough to confirm a prediction.
To support the idea that unfixated words are indeed perceived and processed, Fisher and Shebilske (1985) performed an experiment that made use of a yoked control group. Half of the 60 university undergraduate students who participated had their eye movements monitored while reading sentences (as well as short essays) such as “Pets have funny names such as my favorite dog, Jingles.” If those subjects failed to fixate, for example, the words funny and dog, the remaining subjects (the yoked controls) would be shown the sentence as “Pets have _____ names such as my favorite _____, Jingles.” The researchers then examined the percentage of skipped versus unskipped words that subjects could report. They reasoned that if words that are not fixated are not perceived, the first group of subjects would recall as many of the skipped words as would the second group. In fact, this was not the case—the ratios of reporting nonfixated words to fixated words in the first subject group were 1.0 for sentences and .97 for essays, while the ratios for the yoked control group were .40 and .45, respectively. That the yoked controls were able to “recall” a word at all is a function of their ability to infer the words from the context. The indication is that even though a word is not directly fixated, it is still perceived. Fisher and Shebilske concluded that their results “support the generality of the hypothesis that expectations based on contextual constraints can interact with parafoveal information to determine the guidance of fixations” (p. 154). In other words, predictions from context are used by the brain to direct the eye to fixate or not to do so.
Balota, Pollatsek, and Rayner (1985) used a boundary technique to explore the influence of context and parafoveal information that was either visually similar or dissimilar to a target word that the reader would subsequently fixate. For example, in the sentence “Since the wedding was today, the baker rushed the wedding cake to the reception,” cake had as a parafoveal preview either cake, cahc, pies, picz, or bomb; the preview was changed to cake during the reader's saccade immediately preceding. The researchers found that readers were significantly more likely to skip the visually identical or visually similar parafoveal previews (cake, cahc) than the semantically related, visually dissimilar, or anomalous parafoveal previews (pies, picz, bomb). They conclude that “the data imply that when the word is skipped, only the beginning two or three letters of the parafoveal word were actually identified. Thus, on these occasions, a strong context helps readers to fill in information that is not totally available in their parafovea” (p. 374).
Readers are able to sample phonological and orthographic information in the parafoveal field of vision and use it, together with their expectations for the upcoming material to be read, either to skip text or to fixate on it for a shorter than average period. It is this sampling—as opposed to a thorough processing of each and every letter—that makes possible the efficient use of the least amount of information necessary to make sense of the text and move on.
Different parts of words carry different types of semantic information, and some of these parts are more useful than others for the process of word recognition. Underwood, Clews, and Everatt (1990) examined the process whereby a fixation location is determined by the information distribution of the word in the parafoveal preview. For example, underneath is the only word of its length that ends in neath, which makes the end of the word a “zone of high information.” The end of engagement, however, is “redundant,” because ment can attach to a great number of words. Underwood, Clews, and Everatt selected target words with zones of high information at the beginning or end and embedded them in short stories. The location of readers' fixations on the target words were recorded. The researchers proceeded with the expectation that if readers consistently fixated on the zones of high information then they must be processing morphological information parafoveally. This is indeed what happened, as the researchers explain:
The target words used in our sentences varied in their distribution of information. Being given the first few letters of some words would not be sufficient to identify them, and, likewise, the final few letters of some words did not provide a unique suggestion as to the identity of the word. The distribution of the information had its effect upon the location of the first fixation upon the target.... A redundant beginning induced a first fixation further from the word's beginning. This variation in the initial landing position is evidence of parafoveal processing of the distribution of information in the word, because until that fixation had been made only parafoveal processing could deliver the information necessary to guide the eyes to one location or another (p. 58).
This conclusion indicates that readers are able to sample semantic information from the parafoveal field, which enables them to use as textual cues the most informative part of a word—a good example of one of the numerous ways readers make efficient use of text.
To demonstrate the importance of phonological information, Pollatsek, Lesch, Morris, and Rayner (1992) undertook a project designed to determine whether homophones provide a better parafoveal preview than do visually similar words. Of interest was fixation duration on the target word in a sentence. Each target word had one of four corresponding preview words, either a homophone, a visually similar word, a completely different word, or an identical word to the target. The text was displayed on a computer screen. When subjects began reading the sentence, one of the preview words would be in the target word's position until their eyes crossed the boundary point (the parafoveal preview), at which time the preview word would be replaced with the target word. In other words, the preview word would be in the sentence until the readers began the saccade that would take them to that word, at which time the computer would change the preview word to the target word; because the eye picks up no information during a saccade, the reader would not be aware that there had been a change. The dependent variable was processing time of the target word once it was reached, with a shorter duration of fixation being attributed to the usefulness of the parafoveal preview. The results indicate that readers fixated for less time after a homophone preview than after a visually similar preview. The authors' central finding was that “when a parafoveal preview word was a homophone of a target word, there was a greater preview benefit than when the preview was a non-homophonic control word that was as visually similar to the target word as was the homophone” (p. 158).
To further confirm the effects of contextual constraint, or predictability, of text, Rayner and Well (1996) asked subjects to read sentences that contained a target word classified as high, medium, or low constraint. Their method of determining the predictability of the target words was similar to that already described of Fisher and Shebilske (1985): judges who did not participate in the study were given a cloze sentence or paragraph and asked to fill in the blank. Target words that were produced by the judges over 60 percent of the time were considered highly constrained, or predictable, and target words that were produced less than 10 percent of the time were considered unconstrained. The eye movements of eighteen adult participants were recorded as they read a range of sentences, from highly constrained to relatively unconstrained, with word length and frequency controlled. The study's results indicate that the low-constraint words yielded longer fixation times than the medium- and high-constraint words, and that readers were more likely not to fixate on high-constraint target words than on medium- or low-constraint target words.
This work effectively confirms findings of other eye-movement studies that show that “highly constrained target words are skipped (i.e., not directly fixated) more frequently than unconstrained words...(and) when target words are fixated, fixation time is shorter on constrained than unconstrained words” (Rayner & Well, 1996, p. 504). The researchers conclude that "predictability of a word (or the amount of contextual constraint for that word)...will affect both fixation time and word skipping" (p. 507).
The Past, Present, and Future of Eye-Movement Studies
The long, barren gaps in a century of research are not a unique phenomenon in studies in education. In part, they reflect the changes in research paradigms, research methodologies, and views popular during each period. Most of the research of the 1980s and '90s, for example, is rooted in a view of reading as rapid, automatic, word recognition, and eye movements are studied within a set of assumptions arising from that view of the reading process.
In the decades from the 1930s to the late '60s, reading research was dominated by views that put little focus on how the eyes functioned in reading. The feeling was that we already knew what we needed to know. Still, there is a remarkable continuity over the long history of eye-movement research. There is a tangible reality to eye movement that transcends researchers' assumptions. Interpretations of findings are certainly affected by the vantage point of the researcher, but the eyes provide data that ultimately must be accommodated.
In some graduate programs, students are advised not to read or cite research that is more than five years old. That leaves these young researchers ignorant of a knowledge base that has provided a foundation for many investigations and findings. It is also unfortunate when workers in the field come to know a body of research only from reports of it in the current literature. Relying on third-party interpretations alone can lead to widespread misunderstanding or misrepresentation of the original researchers' actual data and findings.
Summaries of the findings of eye-movement research are now being used to argue for a word-recognition, as opposed to a meaning-construction, view of reading (e.g., Adams & Bruck, 1995). To put this current view into context, we need instead to look directly at the research of the past and present and think about what new research directions could help us more completely to understand how the eyes work in reading.
Paulson can be contacted at the Department of Language,
Fisher, D.F., & Shebilske, W.L. (1985). There is
more that meets the eye than the eye mind assumption. In R. Groner, G.W. McConkie, & C. Menz (Eds.), Eye movements and human information
McClelland, J.L., & O'Regan, J.K. (1981). Expectations increase the benefit derived from parafoveal visual information in reading words aloud. Journal of Experimental Psychology: Human Perception and Performance, 7(3), 634-644.
J.K. (1979). Moment to moment control of eye saccades as a
function of textual parameters in reading. In P.A. Kolers,
M.E. Wrolstad, & H. Bouma
(Eds.), Processing of visible language (vol. 1).
Pollatsek, A., Lesch, M., Morris, R.K., & Rayner, K. (1992). Phonological codes are used in integrating information across saccades in word identification and reading. Journal of Experimental Psychology: Human Perception and Performance, 18(1), 148-162.
Underwood, G., Clews, S., & Everatt, J. (1990). How do readers know where to look next? Local information distributions influence eye fixations. Quarterly Journal of Experimental Psychology, 42A(1), 39-65.
Wolverton, G.S., & Zola, D. (1983). The temporal characteristics of visual information extraction during reading. In K. Rayner (Ed.), Eye movements in reading: Perceptual and language processes. New York: Academic.
Reading Online, www.readingonline.org
Posted January 1999
© 1999-2000 International Reading Association, Inc. ISSN 1096-1232