2. Background isformation lear that, for reading to proceed at all efficiently, we muse be able to recognize and understand the meaning of most (if wotall) of the individual words that we encounter. In ater sections of this book we explore the processes of recognizing words and understanding sentences in normal reading, he biclground infornpation we present in this Hirst section of this book will help you understand the complex information-processing activities chat occu when you read and understand text.1 INTRODUCTION AND PRELIMINARY INFORMATION Reading isa complex skill dhot is prerty muuch taken for granted by those who can do it. About 35 years ago (when cognitive psychclogists first became interested in studying reading) one of che authors, then a graduare student, got into an elevator in the Engineering Department at a famous university in the northeastern part of the United States with a copy of Sinith’ book Understanding Reading (1971) under his arm. A bright young freshman engineering studens, upon seeing che book, was quick to remark:"“Oh, radiyg: Hlearaed how to do that 15 years ago." ‘That remark is Pretey consistent with most people’ atticudes about reading. Those who can do it eake it for granted. Yer itis an extremely complicated process chat és sometimes difficult to Jearn (particularly in comparison to the ease with which children learn to speak). And illicerste adults fine attempss © learn to read agonizingly frastruing. Anyone resding this book is likely co be fumiliar with 30,000 or more words and can generally recognize most of them within a fraction ofa second, A skilled reader can do this despite the fet that letters that make up the words are often printed in different type fonts. Ia he case of hand written letters a reader can still read and comprehend despite rather dramatic differences in style and legibility. In being sble ¢o read and identify words in spite of all this variability, a skilled reader is able to perform a feat that is well beyond the capability of the most powerful computer programs available today. Bue this is not all. Skilled readers ean identify words that have different meanings in Aifferemt contexts, Consider the use of the word luxer in the following ewo sentences: Jobn knew the boxer was angry when he started barking at him. ad Jol knew the boxer was angry when he started yelling at bim. 2) ‘These two sentences are identical except for a single word which disambiguates the appropriate meaning of the word boxer, The most common meaning far boxer is a dog. Since dogs bark and people don'r, boxer in sentence LL clearly reférs to a dog, Likewise, im sentence 1.2 the face thae the boxer iy yelling leads us to believe that the sentence is referring to a person, If you are very observant, you may have noticed dhat there are actually two ambiguities in sentences 1.1 and 1.2. Not only is the word boxer ambignous, the pronoun he is ako smbiguous. Most of the time, in sentences like 1.1 and 1.2, we associate the pronoun with the most recent antecedent, Hut if the sentence read4) Background information “The boxer scared John when by started yelling at him, a3) ‘we would sos likely associate the pronoun with John. Yet notice even here that itis not completely clear who fies. If we replace the word scared with the word attacked we woukl probably understand the sentence quite differently, ‘The paint of this cicasion is chac we can easily understand the meaning of these different cemvence, despite the fret that individual words bave more thas. ene meaning and pronouns €29 selia i moe cham one referent, Coupled with this fact is che ebservation that we can easily under -san puts, ious, aan anetaphors, Por example: Joho thang the billbowr was wart on the landscape. a4 Here none of ws woul believe that dhe titertl meanin; ceaily understand the senienice to mean thy as we can easily compreh: # oF the word wnt was intended, We quite at the billboard was ugly and spoiled the scene, And, just or in seritence 1.4,s0 the idiomatic xl the snetap! tbe snot nace of Johu hit the aail on the head with his answer, as presents a difficulty only for non-native veaders of En, the sentence and find it nonsensical. Thus, skilled readers are very good at combining the nearing of individual words to derive the meaning of sentences and paragraphs ond hore sages aud books, Readers can draw inferences by relying on whae they alkeady know te el ps ee ae and from reading words chey can form images of scenes and appreciae pany Ip understan ‘We have been arguing that the feats of = skilled reader ene teu oe ty fil computers cannot do whata skilled reader ean dss, Be tresive. Very power hi machines (or more specifi e ens ‘entioned that a skilled reader handles almost effortesly Hon : jo skilled ccompli complex task? And how is the skill acquired? These are the comma questions eh Accomplish me vost part ore wil focus on the sll vader in sttempring to expla he we eek Fo primary rationale is chat we raust understand she sil ele can end eating Our lestand how i¢ is I before we sequined, anc ous primary orientation in dis book is a cagnitiy govah processing point of view (i.e, understanding the component mechansaty ‘inde, informacion: the eeoninder ofthis chapter ie attempt to place the rst ofthe book fons Foun). do this by ft dssusing how vse have hisrially viewed ee eevee We vil an overview of the hunan information-processing system, diseusing Me hen we will present mechanisms may be involved in reading, First, however, we bri at types of processing icy di a reading—that of a cognitive psychologist. Neuss out perspective of etish who attempt a literal intecpretation of What is cognitive psychology? Gognitive psychology is che branch of experimental psychology that sendies how the mind works. Cognitive prychologits are interested in a wide range of mental aetvities fom attention ee memory to languoge tearing concepts to decision making. Within the cognithe pening wolbox shere are & number of cook that are used to ty and investigate the topes eee ee Foremost among these tools ate what ate now generally referred to as belavorlexperinte tis rerun is usually used in contrast to breiu-imaging stds. In behavioral experiments people (referred to as participants) are asked to perform some kind of task. ‘The amount of time that it takes them to complete the task (or respond co a stitmulus), and the response they actually make {and itsIntroduction § accuracy), are typically measnzed. We will discuss various types of response tinte tasks relevant for reading in Chapter 3, but a¢ this point we only want to alert you to the general technique, In brain- inning studies, which have become very popular in che tase 20 yeors or so, participartts are asked to perform some type of task while correlates of neural activity in their brains are measured, The third main tool used by cognitive psychologists is that of producing computer simnulations of the issue of interest, As we will sea, a number of models or computer stinukations thet are relevant to anderstanding reading have appeared within che last few years. We will extensively discuss behavioral techniques and the results of computational modeling throughout the book, but will briefly introduce brain-imaging techniques here (for 2 yood overview, see Dehaene, 2009}. Currently, ewe general types of measurement of braia activity ars employed. ‘Two of these techniques, positron emission tomegeaphy (PET) and fienctional magnetic resonance imaging (MIRD, sre good at localizing where in the brain certain activities occur, but their temporal resolution iy not very precise. They both measure che hemodynamic response, the increase in blood flow to active neurons in the brain, This response occurs ovet an approximate span of 6 seconds, varying, across individuals. As we will see later, many of the important processes involved in reading occur within 250 milliseconds (or one quarter of a second), well beyond the temporal resolution of AMTU and other blood flow-based brain-imaging cecluigues. Seill, such studies have contributed valuable information about the location of the neural circuits involved in teading and ate beginning co examine the neural bases of reading disorders, such as dyslexia (Gee Frost et al., 2009, for a review). Other neurophysiologicsl measures have a millisecond temporal resolution fine enough for studying reading processes. Flectroencephialogrphy (EEG) measures changes in electrical poten tials at che scaip that result when large networks of neurons prepate to fie in response to a stinvelus, These event-related potentials (ERPs) reflect brain activity from the start of a task (e-g., word recognition), providing a contingous, online measure of the time course of cognitive processes 25 they unfold. Arguably the most complex brain-imaging tectmology is magnetcencephalogrphy (MEO), which measures the magnetic consequences of electrics! activity in the brain to offer not only fine-grained cemporal resolution, but also high spatial resolntion, which enables the localization of neural activity ¢o porticular areas in the brain, MEG studies have contributed novel Gindixgs abou the location and time course of brain activity during word reading (see Halderman, Ashby, & Perfetti, in press, for a review). However, the highly sophisticated dita acquisition and modeling techniques that MEG requires necessatily limit its use. As informative as these methods ae, niost of what is known about the complex processes invalved in reading has been discovered in carefully conceolled behavioral experiments. Thus, mnost of the research we discuss in this book is behavioral research. Historical overview of reading research The roots of cognitive psychology can be traced to the esablishmenc of Wunders laboratory in Leipzig in 1879. Workers in Wundt’s hnboratory were keenly interested in questions related to meniory, perception, and accion. Shortly thereafter there was considerable interest in the process of reading, which reached its apex with the publication of Huey’s (1908) The Psychology aid Pedagogy of Reading. A perusal of the chapters in the first part of his book (the part dealing with the psychology of reading) will reveal that che chapters bear a remarkable similarity 1@ the copies covered in the present volume and most other modern books dealing with the psychology of reading, Huey and his contemporaries were inrerested in eye movements in reading, che natere of the perceptual span (how much information can be perceived during a fixation of the eye), word ion processes, inner speech, reading comprehension, and reading rate. Hucy’s snarvelously6 Background information couene ann! concise description of his findings and those of his contemporaries prior to 1908 is still a joy to rex. Many of dhe basic facts we know about eye movemenss during reading were discov- ered by Huey and contemporaries using cumbersome and seemingly archaic cechuigues in comparison the sophisticated devices currently available to record eye movements during reading, Yet dhe discoveries have stood the test of time and have held up when replicated using tone accurate revoning systems, A contemporary of Hey, the French oculs, Emile Joval,firse feted that daring reidings our eyes do not move smoothly across the page as our phenoreeno- Jopicsl impressions would imply. Rather our eyes make a series ‘of jumps (or saranfes in Prench) slong, the fine, Betwweest dhe mumps the eyes remain relatively sill, for about a quarter ofa second, ins yehal rs velon ted ce ais u fications, fu onler to study how an cc information ean be percei d in a single eye fixation, the tachisto- Scope wars devired. ‘The (scope (si soften called) ina device chat allowsan experimenter to contol hhow sich) information is presented eo a pucieipant, a8 well as the domtion of the exposure. relic, as i¢ has been supplanted by high-speed computers with millisecond contol However by varying the amount of information avaikible in the tscope and by presenting it for a brief duniion (to preclude any cye movemend, early researchers hoped to infer the size of the perepinat span or the area of effective vision during a fixation. Hueys book also describes chsic experinnents by Cattell (1886) and by Erckniann and Dodge (1898) on word recognition and two full chapters in the book are devoted to the role offnner speech Hoey’ ted observations en inner speech and word recognition provesses kaye lhegely sto of time. ‘The cscope is wow largely a historie in reading. ed the test ork related tothe eoguitive provesses involved in veacing continued fee a few years after ehe publication of Huey’s book. Howevet, serious work by psychologists on the reading provess pretty much came to a bale a few years after 1913. In that year the behaviorist revolution in experimental brschology began. According to behaviorist doccine,the only thing worthy of study by experi- mental psychologists were activities that could be seen, observed, sed measured, Since ¢ ni : Drocesies involve in sles reading cannot be cbsrved and decry wweasured, inesnet nnn ne rae. be se, soe well-known imestgation oF ey movemane en reading by Ian and Vinker were carried out beowsen 1920 and 196), bu for ene men pie deh ‘ing by ae ely peripheral components of rainy, Tht ty e7 mrowmane oe be sean a I line een anid dineedly measured, and were by Buswell aad ‘Tinker, bus ateempts t ofthe mind were real non-existent. BNO lee the activity oF de ee vo detiviey In essence, work on the cognitive processes asociated wi 1920s and did not begin again until the 1960s. Smal) ‘wonda, cece {0a standstill in the republished tx 1968 i secured sa relevant! We hadht learned s whee oe HUEY’ book: yas proce volved in tsiding in he 60 years beeen dhe nia pap cata about the cognitive : : cation of the work 1 second sppeerance of the book. In zeldion oo the work on eye ovements daring ate the resarchers such s Boswell ad Tinks, some Work on reading did coma NE ean auetion- But mos ofthis work was conducted in Schools of Eatacey Where pe BE ineceva a is geusrally on more applied aspects of reading. ‘Ibus there was re the: prj method! to teach reading, and mary of che standardized reading developed during that period, However, work an the mental proce almost non-existent. Tosiy sony prpchologios oe interested in rnin. Why did dhs Ohne ae ptr eon appean co ave Been dhe eof bshavorn so ecoun forage in any reasonable way. ‘The prontise of behaviorism: was always thac if. Psychologists coun) ng stand the laws of learning and behavior in simple tasks (ike knee jerks and eye blinks), those would be generalizable to more complex tasks like language processing, In 1997, 9) see Wher lot more PPOPriage tod ere reading was St still im cxisteneg S25 aSS0ciated with Place? ay.Introduction 7 decided it was high ime that the belaviorists produced on this promise, and he published Vial Behavior, which was an account of language from a behaviorist viewpoint. The linguist Noam Chomsky (1959) wrote a scathing review not only of the book, but of behaviorism in general. In essence, Chomsky argued chat belaviorise principles could not account for language learning or language processes in general, Around chat sime cime, he also publisled Syruacic Structures (1957), which was a rndical depareure from traditions linguistic theory. In that work he suggested dhac dhe study of language ard the mind are intimately related, and he presented an elegane theory of sranimat, Meny psychologists, disillusioned with behaviorism, became very interested in the rela~ tionship between Chomsky’ theories of language and work on cognitive processes was underway after a hiatus of over 40 years, ‘A number of ether factors contribuced to the re-emergence of the study of cognitive processes around 1960, However, they are beyond the scope of our current discussion. Out of the burgeoning interest ia language processes in genera, interest in the reading process began cce agin around 197), Since the mid-[960s numerous scholarly journals desling with cognitive processes wud hunnan experimental psychology have been founded andl nearly every issue of these journals contains at least one article relaced to reading. Ju addition, a number of textbooks dealing wich reacding appenred starting sround 1985, Clearly there is now considerable interest among cognitive psychologists in smeying reading. Cognitive psychologists studying reading approach the issue from slightly different perspectives Some have a background rooted in perception research and see the study of word recogninon, for example, as a mieans to study percepesal processes or pattern recognision using well-defined stimuli, Others approach the study of wading with a background in memory processes and verbal learning theory, or a background in Tinguistics. They tend co approach the study of reading by examining comprehension processes. Still others are interested in reading int and of self because they believe, as Huey pointed out over 100 years ago, that to understand what the mie does daring reading would be “the acme of a psychologist’ achievements, since it would be to describe very many of he most intricate workings of the human mind, as well 2s to unravel the tangled story of the most seinarkable specific performance thae civilization bas learned” (Huey. 1908, p. 6). Ic is onr contention hat this diveesity of interests and backgrounds is healthy and can easily be accommodated within the information-processing approach because it views reading av 9 highly complex process relying on a nnnber of sub-processes. Thus it is unlikely that there will be a single insight that will answer all of che questions aboue a complex skill like reading, Rather, breakthroughs will come front researchers working on different sub-components of the reading process, This is not fo say that all of the information obtained by cognitive psychologists will not ced to be put together, for it will. Rather, it provides a clear justification for examining different component processes of complex skill, Critics of the infrmation-processing approeh offen argue chat attempts to isolate component processes of resing result in tasks very unlike real reading, For example, to study word recognition processes cognitive psychologists often present a word for a very brief duration (say 50 muilliseconds, one-twentieth ofa second). A participant in such an experiment may be asked to pronounce the word or make some type of decision abour it (Is ita word? Does it belong to a certain category of dhings? fs it larger than a breadbex?). In uraking decisions about the word, participants push one burton for a yes response and another for a io response, Aduitedly these tasks are untike reading, Yer, to respond appropriately, participants may well be using the same processing miechanisins that they use during reading. Pechaps an analogy will help, Suppose were interested! ine studying walking, If we study the motor responses that people make when they take two steps, crities may say “But chac’s not walking. When you walk you gu a loug way.” True, but are the motor responses8 Background information differen when you take ewo steps? Not completely! What cognitive psychologists strive to do is set uup experiments in which the same processing mechanisms are used in the task derived as in eading—sonietines more sucessfully than others, In the chapters in this book, as with our prior book, we will place the greatest weight on hose kinds of experiments that most closely match the risk of eading., Te ingen tu: pone out dat the primary methodology of the cognitive psychologist is empir- ration, “Theories are models of proceses such as reading are also critically important lwvauie ley Help to formule de kinds of research questions to be asked. Bur the ulsimate testis invent io which comuasting theoretical positions ave tested against each other. vinetines tuvcare’ ic vanlertaken purely tor iniosmation-gathering purposes (50 as 0 form the model or theory). The point is that what we will eke as evidence i not siraply somebedy or other’ idea or intuigion, but des obtained gooundeerk fer well-controlled experiments. With these points iv mind we now turn te 4 description @f the human information-processing system. Overview of the human information-processing system In this section we well present an overview of the traditional view of the buman information processing system, We provide a stage analysis, but we caution you that notall cognitive psycholo- gists would agree with the notion of distinct stages. Indeed, the best way to describe the human information-processing system has always been somewhat controversial (for some alternative perspectives see Anderson, 1976; Broadbent, 1984; Craik & Lockhart, 1972; MeClellind & Ramethart, 1986; Rumelhare & McClelland, 1986). ‘The type of mode] thar we will present consists of three distinct stages of processing (Sensory store, short-term memory, and long-form miemory) in a more of less passive system. We focus on phenomena and analyses that have stood the test of time, and overlook inany current debates in cognitive psychology, since our primary intention is to give the flavor of the wnformation-processing approach and to introduce termi nology thar virtually ail cognitive psychologists vse (modern textbooks, such as Fysenck & Keane, 2010, provide current overviews). In addition, some of the controversy over details is peripheral to understanding reading, Figure 1.1 shows an example of a typical stage model of the human information-processing systems. It consists of three stages, and each stage has distinct finsctions and characteristics. However, prior fo discussing such a system, we must discuss something about the initial sensory registration of printed words by the eyes and subsequent pattern recognition processes. Sensory | visual onic mamary store | Dutton een emery * snorcteim | sometines called overs | working memory Components induc: tenga | die memory serail memory FIGURE 1.1 A simple overview of the bumau information-provessing systemIntroduction The retina and visual acuity Vision depenck on a pair of speciatized organs (the eyes) whose neural receptors can be thought of 3s being a part of the brain that has extended outside of the cortex. Patterns of light falling on the sensory: nearons in the setina resulr in the sensation of sexing. When you look at a page of text (lke the one you are currencly reading), you are not able to see all of the words on the page equally well. Thi is because of acuity lbnitaigns. In ters of acuity. a horizoutal line of text fling on the retina can be divided into three regions foveal, parafoveal, and peripheral. The foveal area subtends about 2 degrees of visual angle around your fixation point (the specific location that your eyes are centered on} the parafoveal area sbtends about 10 degrees of visual angle around fixasion (5 degrees to the left and right beyond che fixation pomne); the peripheral area includes everything on the line of text beyond the parafoveal region. Acuity is greatest in the center of vision (the foves) and drops off markedly in the sparafavea and even more so in the periphery: This is because of the anatomical structure of the retina ‘The retina is composed of two types of receptors called rods and cones, The fovea consists almose entirely of cones. With increasing discnce from the fovea, che density of cones decreases and the density of rods increases, ‘Thus the peripheral region of the tetina is composed entizely of rods. The patafovea contains a mixture of rods and cones. These ewo eypes of receptors serve drama: cally differen: fsctions. The cones ate specialized for processing derail and for acuity. In addition to permitting fine discrimination of detail, cones also setve in the discrimination of wavelengths or ue, The rods, on the other hand, are specialized for detecting movement and permit discrinina- tion of brightoess or shades of gray. The rods are particularly important for night vision; when you enter a dark room, at first you feel as if you cannot see anything, However, after a short while (unless tie 200m is eotally dark) your rods adapt and you ean see. The most important point to be gleaned fron: the description of rods and cones is that the acuity necessary for discriminating fine detail (as is necessary in reading) is available only in che center of vision. A sitnple experiment can demonstrave this quite clearly. Ifa word or letwe string briefly appeared on a video monitor of a computer, and you were asked to say what word (or lecters) appeared there, your accuracy in doing so would decrease as the stimulus was presented farcher fom your point of fixation. In che experiment the stimulus is presented briefly enough {about 150 milliseconds or less) chat it is virmally impossible for you to move your eyes to look dicectly at it, Figure 1.2 shows hew performance in such x task would depend on how close c© 100% Relative density eny Chance 10 3 © 3 19 Fovea Degrees from fovea FIGURE 1.2 [Relative density of cones (solid Tine) ane rodk (dashed line) across the visual field. Dotted Fine shows the aceutacy of identifying a target word exposed briefly to the right or left of fixation10 Background information fixation the stimulus was presented. We have ako plotted on the figure the relative distribution of rody and cones in the retina, Note that the accuracy fametion in our expetiment is very similar to. the dist bution of cones in the ret “The purpose of this demonstration is to convince you that, race the five details of letters and words as we read, we must move our eyes so jy 10 phe dhe fovea over that part of the text which we want te read. in onder diner Pattern recognition processes Alice ws recognition fh nue aur eyes 240 place the favea on the word or words that we want to rexd, pattern 13. Actually the pattern recognition process for a word may have begun on the Iiior fixation when te wonl wis in parafoveal vision, as we shall see in Chapter 4, What we are concensd with in thissection is how the brain goes about recognizing the letters and words which, aust be procesed for usto real. To take a simple example, how do we recognize the printed leer “A'? Two major theories of pattern recognition have been proposed (for a recent discussion see Grunger, Rey, é& Dufan, 2008), ‘The Gest, tewiplate maiching theory, snygyests that we have stored in. our brains 9 representation of every pattern that we can recognize. ‘Thus we recognize the fetter A by comparing the pattern of excitation ftom the cclls in the retina to 2 template stored in memory. Iethete is a match between representation and the stimulus, che letter A is perceived. While templace matching theory works quite well in computer pattern recognition devices that read letters and digits in highly constrained contexts, such as the digits that specify the code number on your checking account checks, itis well known (Crowder, 1982; Neisser, 1967) that such a system would fil to recognize instances of the letter A that were slightly deviant wish respect to shape, size, or orientation, as Figure 1.3 shows. The major problem for the theory in its most rigid focns is that it suggests we have a template for every vatiacion of every pattern we are INPUT = TEMPLATE @ tb) fc) (a) FIGURE 1.3 illusttation of the difficulios with a simple template-matching system for character recognition. (After Neisser, 1267.)Introduction 11 able to recognize. The simple parvern for the letter A, for example, can appear (as we mentioned earlier) in a number of different cype fonts and handwriting variations. Yet we are able to recognize it quite easily. It seems unwieldy to think chat we have so many patterns stored in our head. ‘One way to make the template matching theory more workable is to assume that, before comparisons of new input to a stored comparison tale place, the input is “cleaned up” or normal ized, The normalization process would separate essential information in che stimulus from mose non-essential information. For example, variations in size could be taken care of before compari- sons £0 a template occurred by transforming all images wo a standard size. Accounting for varistions in orientation prior co the matching process alse seems quite reasonable. It is somewhat hucder to undezstand how the normalization process would fil in missing gaps and eliminate fuzzy squiggles @s in handwriting) chat are irrelevant to recognizing the pattern as the letter A. While the normalization process gives plansibility co the cemplave matching cheory, the second cheory, feature detection theory, is more parsimonious in accounting for how the pattern recognition process deals with such variation, and is generally considered to be a more viable account. ‘The starting point for feature detection theory is the idea that there are many common elements for leters (consisting of horizontal, vertical, oblique,and curved lines) and thar we analyze these conspo- nent elements in recognizing a pattem, The letters C and G and the letters Q snd (, for example, have a great deal of similaeity and feateral overlap. The distinguishing feature berween the C and G js the horizontal line ehat is present in the G, but not in the C, The distinguishing feature berween the © and Q is the oblique line present in che Q, but absent in the ©. According eo feature detection theory, wher a letter is analyzed the Fist step is to prepare a list afts eatuzes and this is is compared with the list stored in memory. This process then is avalytcal in that we put together the different elements of she pactern until recognition occurs. Template matching theory, on che other band, is more of a fiolistic process. As we implied above, the feature detection theory is considerably less cumbersome since che analytic processes would rely on a small nurnber of features that are commen toall rype fonts (the distinguishing feature for C versus G remains invariant over different type forts and kandwriting styles) rather chan having a diffrent template for ench style of print. What type oF evidence is there for feature detection theory? Three types are consistent with the cheozy: (1) physiological data fiom animals, 2) stabilized image research, and (3) visual search data. The best-kuowi physiological evidence in favor of feature detection theory comes from: work by Hubel and Wicset (1962) on the visual system of the cat. Via electrical recordings, Hubel and ‘Wiesel were able to examine the rate of firing of individuat cells in the visual system a8 a fianceion of what che cat was looking at The most important finding feom their werk for our purposes i that they demonstrated that cortical cells in che visual system fired differentially depending on the stinnutus, Cells that were specifically respousive co different visual shapes —tine, edge, und slit detec tors, as well as more complex detectors —were all discovered. It is easy to generalize from these results (although it should be done cautiously) and suggest chat there are ikewise feature dececcors in hunions specialized for firing when the various horizontal, vertical, oblique, and curved lines making up the leteers of our alphabet fallin front of our eyes. Ina stabilized image experiment an image can be kept at the same phe on the retitma (via some tather sophisticated technology), so that to whatever extent the eyes move, the stimulus moves a corre sponding amount ia the same direction. Even when we are asked to hold our eyes very stil, there is 2 slight movement or ueuior of the eyes (called nystagaas). The stabilized image echnique, however, keeps the stinulas on the same renal location. Apparently, nystagmus is impoctane for perception because uncler these conditions perception gradually blanks out so that the observer no longer sees the stimulus. IElight remains in the same place on the retina, the cells n that location become fatigued and stop firing, and perception of the stimulus ceases, Movement of a stimulus or movement of the eyes will place a stimulus at a new location on the retina, resulting in new neural firing, What is12 Background information incevesting is that when the image is stabilized, perception does not blank out instancaneously, Rather, itis gradual. Coherest parts, including lines sharing the same orientation, disappear at the same timie, wor in which the stinvolus fades is inconsistent with a cemplate theory.annl pravisies evidh portance of feaaues at various levels of absiaetion, The th vidence for feature detection theory is che work on visual search originated 5 shows simple stimuli in the task, Pasticipanes are asked to find the er Z, Actin ont dic the letter Z is much easier to find (parsicipanes are over twice as as depicted in F ice ford FIGURE 1.4 Perceptual fragmentation in “seopped images” The figures on the lefea the ozhers are cypical products of frogmentation, (From Pritchard 1961.) ¢ stimulus patterns; STANDARD CONDITION NCNCONFUSABLE CONDITION CONFUSABLE CONDTTION TARGET = Z TARGET = Z TARGET = 2 RerwKe casocs ‘onan PTHSHG UBSQ0Q, WYLKWY sTvcaH gong XWLLY nUTRYD scponc yazw2t KIREGD czoqus nasavian cazran buscee YNLXLE POLKRF poquce xs ETIEWR ecogau Locus FIGURE 5,5 Neisser’s visual search task. A display is exposed and the subject muse scan vertically, from wp co bottom, until finding a tanger site, Time co reach the target is recorded as a fimction af the posidion: ofthe eargetIntroduction 13 fast) in the middle column than in either of the other colunins, The reason, of course, is that the: distractor lesters in che niiddle colurnn ace sll lewers that are quite dissimilar to che target and do aot share many features with it, Ternplace matching theory would suggest that there should be no Uiffereuce between the columns since the number of matches that muse be made would be equiv- alent in eacle, However feature devection theory can account for the zesult since virtually che samme set of feature detector cells would be firing when the distractors are confusable whether or not the target letter was present. On the other hand, when the target is embedded in dissiunilur letters, the cells thae Fire for the letter Z will only fire when the target is present. While there are criticisms of feature detection as a theory of object perception in general, it provides 2 reasonably sutisinctory model of how ledters and words are processed, One issu that we have not diseussed, which will become relevant in & later chapter, is whether pattern. recognition ptocesses ovcur in a serial or panilfel manner. Ifinformarion is analyzed serially, the pattern recogni- tion processes occur one at a time, Thus, in Neisser’ visual search task, feature analysis would occur for ong letter and when it wos complete would begin on the next and so on. If information is analyzed in parallel, the sensory information from various spatial locations is analyzed sinnultane- ously. Thus, within acuity limits, all of the letters im each row in Figure 1.5 would be analyzed at ‘once. The experiments on woul identification thac we will discuss in Chapter 3 alse argue for paralle! processing of letters within words, However, we will argue that serial processing is ako important and oceues frequently én reading. Indeed, we will argue that words are identified serially as we read, In pactioular, we don't process the meatng of word a+ (the word to the right of fixa- tion) uncil we have processed the meaning of word «(the currently fixated word, In this section we have argued that the pattern recognition processes that operate on print can best be conceptualized in terms of feature-analytic processes, We return naw co @ more detailed exantination of the informarion-processing system depicted in Figure 1.1. The sensory store ‘The lowest level in the inforniation-processing system is generally referred to as the sensory store. For audicory information, thy store is referted to as echoic memory {for a discussion see Cowan, 1984). For visual information, the store is referred to as Hanis memory, leonie memory is considered to be a temporary and transient memory store in whicls mach of the information physically avail- able in che stimulss is still availible after the display has gone off, In one of the most widely cited experiments in cognitive psychology, Sperling {1960} demon~ strated the existence of 9 visual sensory store. He presented participants with diree rows of four letters each for a brief duration (say 50 mnllisecond or one-twenrieth of a second). [a a contsol condition participants were instructed eo report as many oF the lercers a¢ possible, and the average number of letters reported was 4.5, [0 the experimental conditions Sperling ased a partial report technique in which he cued participants as ¢ which sow they should report. ‘Thus if, after the display went olf, parcivipauts heard a high-pitch tone they were to report the top row, « mmedinm tone the middle row, and a low pitch the bettoin reve The tone was presented at various delays after che disappearance of the letters fom: the screen. Sometimes the tone occurred sinultaneously with the offset of the leteers and sometimes it occurred up eo a second after the Jetters disoppeared. Sperling’ reasoning was that if.we have an iconic memory {although he did not call ie such}, we should be able to use the cue to"read cut” tine lettexs from the cued row. IC the tone actually occurred. priog to the omsct of the lemers, then we would be able to nse chat information €o focus on the ielevant sow Sperling’ question was whedher or not we can nse the cue to examine inforination after it kas physically disappened fiom before our eyes. IEso, then we must have an iconic memory in Which all or most of the information present in the stinsulns ie sill available. In fact, what Spedding, | |14. Background information found was that parccipants could report on average 3.3 letters ftom the row that was signaled. Since the participants hod no way of knowing beforehand whick row was going to be signaled, it amust be Une case that parricipimes could get ronghly 3.3 letcers from each of the rows, which means that chey nut bave had approximately 10 lurers available in their memorg system. Keep in mnind that when participants in the control condition were asked to report as anany letters fiona thy aliply 2s they coukd, they averaged 4.5 letters, How do we account for the discrepancy l-iyseeu rhe cise partial report triak and the whole report condition? If we assume that iconic snenseny lesa hues caparity (pay Te feters fiom a 3 % 4 array) for a very short duration (pechaps a quunter of a sreenl) anil the the rive at which people can “read ou” levers is slow (ay 3 to 3 Levers ima quarter of veromd, then th «liphy, perl par c divarepaney can be explained. Ifthe letters are taken front the whole we i limstet by the “read ove" process. On the other itand, if te tone cues the pants that only one row is relevant, hen the participant has dauie “read oue” most of dat ro ince Sperling’ demonstration then: have been literally hundreds of experiments investigating the characteristics of iconic memory. This research has revealed that the primary characteristics of iconic memory ane that (1) it has a hye capacity, 2} it has a duration of roughly a quarter of a second, (3) itis pre-cateyorical (it stores physical form, not meaningtul objects), (4) itis interfered ‘with by new information, and (5) the “rand out” rate is relatively stow. ‘The stutus and function of iconic memory have been highly debatable for soine cine (see Coltheary, 1980; Haber, 1983; Tuevey, 1977). Some workers in the field inave acgued that it is an spiphenomenon of the bighly sterile and controlled experimental hboratotysafter all, when does a stimetus appear before our eyes for a fraction of a second only to disappear completely? Such individuals tend to argue that iconic memory, like our appendix, has ne functional uiility. Others ‘would argue chat all biclogical mechanisms are adaptive and just becouse we do not know for certain. what function the icon serves, does not mean chat ie does not have a role in processing. With respect to reading, itis not ac all clear what function iconic memory might serve. Clearly, Participants in iconic memory experiments are not reading in the sense that we would normally think of reading. In reading, che stimulus does not disappear from in fiont of our eyes after only a brief exposure (unless perhaps we try reading in a lightning stozan or we are participants in “dlisap- pearing text” experiments, which we will describe in Chapters 4 and 5), At one time it was thought that something like an iconic memory might play « role in integrating visual information sero the eye movements we make during reading. We will discuss that idea in Chapter 4, bur eo anticipate chat discussion slightly che available evidence argues against stich a conchision, Indeed, the fact chat we make eye movements so frequently is » problem for the utility of iconic memory in reading. Recall chat the duration of iconic memory is roughly a quarter of a second, which i about the rate at which we make eye movements in xeading, Given that infornuation in iconie memory is disrupted by new information, and that eye movements occur at the rate they do, plus the fact that the information we want to read is available to us continuously (we can always look back with our eyes), it does not seer that iconic memory plays any rote in reading, Ac this point you may be asking yourself If iconic memory plays no role in reading, why was it discussed in such detail? Indeed, why was ie discussed a all? There are two reasons why we troubled to present thy dewils of iconic memory, First, ou primary purpose in this section is to present an overview of the human information-processing system. Eatlier informstion-processing models of the seading process (Gough, 1972; Mackwouth, 1972; Massaro, 1975; Mitchell, 1982) utilized the concept of iconic memory as the initia! stage of registration of visual information during reading. Recill that we argued tha at chis point we are not so concerned about tise extent to which the differene stages ate accurate and useful in understanding reading; they are presented more to give youa flavor for the approach, This brings us to our second reason far discussing, iconic memory Jn such detail the nocion of bugis (or temporary storage units) in information processing turns outIntroduction 15 co be highly useful. Actually, she icon is ttele more than a buffer in which inforsuation is held for soine lacer processing setivit, The short life of the icon. in fact suggests that the visual features of the princ are of litle use ouce the eyes are no Ionger looking at them, As we shall see later, che notion of a buffer has been very aseful in various types of research related to the reading process. Hopefily, by discussing iconic memory in such detail, you will have a sense for how such a concept may be usefull in designing experiments and theorizing about reading. To summarize, iconic nemory is the initial stage in an information-processing model. It is highly transient, bue with a lange capacity. We have ao argued that its usefulness for understanding, reading is limiced since the stinwahts i always available to uy in reading, However, the concept of a bulfer-like store bas been very usefil in experiments related to the ceading process Short-term memory According to the standard view of the information-processing syscemt, the transient mature of iconic memory requires us to get information vegisteted by the sense organs into a more permanent repre sentation. The structure that contains this representation is short-term meniory (or STM). Considerable information is lost before it can be cramsfetved to STM because che “reed out” rate fiom iconic intemory is quite slow. However,a certain amoune of information is transferred. to STM. But STM has problems of its own. First, and most importantly, ic has a limited capacity. The capacity of STM is ahout seven plus or minus two items (Miller, 1956). Notice diac we said items, not letses, words, oF digits. Indeed, we can Jearn to short-circuit to some extent the lintted capncity of shert-verm memory by various types of chunking statygies. we say the mumber 967835241 to you, and ask you to recall it in the same order chat it was presenced, chen if you treat each individual digit as an item you will have a difficult tiue recalling it. Quite simply, for mest people, STM will be overlonded. However, if you treat che number as three 3-digit mumbers (967-835-241), then you will most likely be able to recall the number with 100% acemacy: Another way that we deal wich the capacity limitation of STM is vin a process called retearal, When you look up a telephone number in the phone book, you often find yourself repeating, (reltearsing) it over and aver to yourself (often silently, but sometimes aloud) so that you won't forget it, Such a strategy is another way to hold information im STM. Notice alse dhat we said you offen repeat che number ever to yourself silently. For a long time it was considered that STM was exclusively an acoustic store, That is, even information coming in the visual modality was assumed 10 be receded into acoustic or auditory information. ‘The reason for this was that the kinds of errors that participants make in recalling information in STM tended to be related seoustically, not visually, to the information actually presented. We now know that there are visual and seniandic codes in STM, Still, for linguistic stimuli, shortcrerm mentory is primazily acoustic, 8 evidenced by the fuct that we try and remember phone numbers from the telephone beck by rehearing them subvocally. This aspect of short-term memory tuens ont to be particularly impportane for understanding che role of subvocal or inner speech in reading. “The fact chat we engage in various strategies (some of thent unconscious) co short circuit the limited capacity of STM has Jedd some workers to talk about a workhag memory (Baddeley & Hitch, 1974) rather chan « passive store, That is, short-term memory can be considered a flexible work= space whose limited capacity can be allocated to either storage oF processing, Information in STM ‘ean remain there as Jong as it is being worked on. Working memory, in the sense of a flexible workspace for processing, is alse heavily iavolved in reading, Words are integrated in this raemory, and as we shall see later, comprehension processes are initiated lrere, “Ty summarize, short-term memory has a limiced capacity. However, we can hold items shece for long periods of time via reliearsal. We alse develop elliciene strategies for dealing with che limited capacity, STM is alo primarily acoustic in nature, Whereas iconic memory was argued to have16 Background information Timitedt uselilness in anderstancing reading processes, the characteristics of short-term semory, amay be insportint in understanding inner speech im reading and comprehension processes. Long-term memory “Vhe rte at which we ean set up progranns to trausfce information from STM to long-term) memory (or 11M, is scutively slow in achition to the rate at which new information enters STM, Because of this ‘onnislersble infia pation is lost, However, its yenerly believed that once information exiters “TM ie saored shere pernunenily: Paventy mder a local aneschetie whose brain is electrically stimulated Can remember things bey long, siuce dioughe dhey had fergouen,and even relive memories of events tlt occurred a long while in the pas (Penfield & Roberts, 1959). Information in TM is not orga ized i a hapliveard fashion. Indecdl, LEM is highly orgmnized and much of the macerial that we ‘cannot retrieve has heen mists, wot lost. Thus the priate retrieval key to najor problem with UTM is getting the appro- urmation stored there. This is not surprising given the vast amount of new information we process ind store i LTM each day. In addition there is evidence that the new information We lei interferes with our ability to retrieve previously stored information. Likewise, information already stored in LTM can interfere with retrieving newly leaned information, Many cognitive psychologists find it useful to distinguish two types of long-term inemory: episodic mesuory and semantic nremary (Tulving, 1972), Episodic memory is the rrenery for sequences of events in your life. Semantic inemory, which is more importance for understanding reading, seen in the right half of the visual field arrive inidally in the left bemisphere. Material presented 8 the center of vision is simultaneously available to both hemispheres of the brain, With respect 0 the left and right cars the same pattern holds as for the presentation of information to the left oF right of fixation in the visual field; information presented co the right ear goes primarily to ¢he lef hemisphere and information presented to the left ear goes primarily to the right hemisphere. From experiments in which stinmli ate briefly presented in the left or tight visual field (or leét or sigh ear), we know that words are processed more efficiently when prevented in the right visual fiel (inplying lefe hemisphere processing), whereas faces and pictures are processed more efficient when presented in the left visual field, From such experiments itis often argued that he left hemi sphere operates in a serial and analytic fashion, whereas the right hemisphere operates in a parallel and holistic fashion, although there is no compelling reason for this conclusion. ‘The second way in which we know about the functions of the two hemispheres is froro research on brain-damaged patients who have one of the two hemispheres missing (fom birth or duc « brain injury) or from “split-brain” patients whose corpus callosum has been severed as a treacrnet for epilepsy. With respect to language the basic evidence is that, if certain regions of the left henié sphere are damaged, in most people iangoage functions are impaired, while right hemispheniIntroduction 19 damage does not produce [anguage impainiuent. For the “splic-brain” patients the evidence is chat linguistic information that is put into the left hemisphere is processed normally, whereas there is Fle comprehension of lingnistic information put inzo the righ hemisphere, Recently & grest deal about the reading process has been learned by examination of patients with brain danage (and Known lesion sites). In Chapter 11 we will review the evidence obtained from such patients. Finally, experiineats using brain-imaging eechniques (such as PET, fMRI, and MEG) have yielded evidence regarding areas of dhe brain involved in Ianguage processing, We will omic the deals of the physiology, such asthe locazion of language-specificstes;the interested vender should consult Frost et al., 2009; Bagh et al., 19%, What is reading? In this chapter so far we have presented preliminary information necessary to understand how cognitive psychologists think about reading, This brings us to a critical question. What do we mean by reading? Ie is cbvious that, to many people, xeading isan all-encompassing activity chet can take on diffexent forms. For example, when you look at a map 10 tell you how to get from one place to another, are you reading? When you proofread a paper for typouzaphical mistakes, are you reading? When you look at a computer program to find your programming error, are you reading? When you scan a newspaper for the [aest results of the stock marker, are you reading? We will tke a conservative view here that none of chese activities is what we have in rind s reading. ft is also obviously the cae that when you read a novel ona 5-hour aeplane trip you maya times be reading slightly differeatly chan when you sead this book: 4 hours into your trip you find chat you are only: hal€-way dhrough the book so you start skipping over sections of the book that seem redundant, Jooking only for relevant and new information so that you can finish it before reaching your desti- ration, You would have a rather diffcule time understanding a textbook if you read in such a fashion, yet we ean generally read most novels a chat way and sill end up understanding the story 1g we will discuss skimming and the adjustments the reader makes under such conditions. Howevet, apart foomn that chaprer, we will focus on the rather carefit type of killed rending that ocenrs wien you real te comprehend, 2s when you read a Cextbook or news- paper suiele or 1 narrative in which you do nct skim pacts of the text, Our focus will aso be largely tonilene reading, While iis clearly the cate that beginning readers spenel more time reading aload cham sitently, for the most part apoxt people spend far more time reading silently than reading aloud, Te would be easy this poiat so get into lengthy argument cbout what js and js tot ceasing, We do not with to do so. Hopeflly, its cleat what ve have ir mind by reading, If forced :o provide a definition of reading, 36 would probably sty something likes wading i the abity to extent oval ‘formation fons the page aed comprehend the meaning ofthe ext. By focusing on the careful reading of a newspaper article, for example. we do not wish to imply that the other activities that we mentioned ate not anteresting, Our bias is that activites such as proofieading ond skirnuning probsbly involve different from normal silent reading. Occasionally we will examine suivh tcsks, wover, our central concera is how people tead during normal silent reading, “This brings us to a second critical question. What i she best way te seudly reading? The answer ds on which aspect of the reading process you are interested in at coynitive psychologists incerested in word recognition gener~ ‘on a video monitor and ask them to make some kind of We shall discuss such tasks in detail in Chapter 3. Other Fh in reading have devised clever techniques to age, H researchers ace interested in how rmuich of kd want to examine how well readers can answer In ovr chapter on speed readin strategies and processes that are to the question is that it depen studying. We mentioned earlier th: ally present isolated words ¢0 people Judgment about of response to that word. rescarchers interested in the role of inner speed detec mine its role int understanding written fange dhe text the reader comprehended, then they wou!20 Background information questions about the concent of the text, Techniques used to study inner speech and comprehen- sion will be discussed in Chapters 7 through 9 I'dhe goal is to stay the coypitive processes chat occur duriny normal silent reading of text on a pinnent-to-moment basis, then any technique thar has readers do something different, such as reanninye words in isoladion or reading text out loud, may significantly distort the component process in silent reading one wishes to study, such as worl identification of the role of acoustic codes in readin, White 3c phursble that che componenies of reading do not change radically from task to there is no giasaneee of it. Thus the relevance of any technique is sn oper question if we do hot know how the proceses work during silent reading of text. This brings 1, to oor fivored technique for studying cognitive processes during accual silent lng—eye movement recording. Reconliag of eye mevemenis has a long history in expeti- nenral psychology. shonid be clear front an earlier section of cis chapter [nthe last 30 years eye movement data have beer widely used co infer moment-to-tmoment cognitive processes during, reading (Rayner, 19783, 1998, 20009}, Te is nave fairly clear that where readers look, and how long they look there, provides valuable information about the mental procestes associated with under standing a given word or set of words (Rayner, 1978a, 1998, 2008). Bye movement recording eam be acconiplished in a variery of ways, but often involves shinting a beam of invisible (intiared) Tight ‘onto the eye that is reflected back from the cornea of reting to.a sensing device, or using an infrared camera co forin an image of the pupil whose location is determined by a computer. With this anethodology teaders are free to look at any part of the text for as long.as they wish. As mentioned above, the technique also has a great deal of ecological validity in that participants in eye movement experiments are actually engaged in the task we wish to study, namely reading. This is not to say that eye nrovement recording is fice fiom criticism. In order to wnconfound smoveanents of the eyes from moventents of the bead, it has offen been necessary to stabilize the head. “This is offen done via the use ofa bitebar (which consist of dental compound that is very soft when 4 participant initially bites into it, but chen quickly hardens to provide a dental impression chat keeps the head «cll In other cases forehead and chin rests are used and participants generally read fora a display placed directly in front of thom, Some crities have suggested thac the rigid constraints on head inovenrent, plus the fact chac in reading outside of the eye movement faborstory we often look down atthe cext (rather chan straight alzead), wil lead to different reading strategies. It has even been suggested that the mere face that our eye movements are being recorded will make us conscious of them and lead us to do something different when we read under such circumstances. Our impression is that these concerais ate all l-founded. Indeed, Tinker (1939) demonsteated quite somie time ago that the reading rate of participants when asked co sit in a soft easy chair and read a book chapter did nor differ oo the reading rate obtained in the eye movement laboratory. Furthermore, recent advances it eye move= ment echnology have made it possible to record eye movernents without stabilizing the head in any wy although generilly greater precision it avnilable when the head is stabilized) ‘All of the authors of this book have been participants in experiments using eye movemene recordings. Our firm impressions are that reading in the eye movement laboratory is virtually the same as reading outside of it, and ir is definitely our sense and intuition thot this latter technique provides a much better approximation of reading itself than any other cechnique. But we de not want to argue that eye movement recording is the only way to study skilled reading, Many of che techniques that will be mentioned chroughout this book provide vsefial information and the bese type of evidence would be converging data ia which information obtained from a number of che techniques converge on the tame answer to a giver question. Our intention is to use converging evidence fiom a number of sources to understand reading, but the data that will receive the greatest emphasis fiom us will be those that are obtained while the participant is silently reading connected text.rather chan simply being engaged in one of the clever tasks cognitive psychologists have devisedIntroduction 21 Models of reading While there ate anany facts about reading chae have been learned by cognitive psychologists (and ‘our emphasis in this book will be to rely on the faces that have beea learned), many people find cognitive psychology somewhat frustrating because there is often conflicting evidence on a single issue. There are many reasons why this may be die case, including the fact dhac our experiments se sometimes not very good. But another reason is thar cognitive psychologists often have different models or theories of how some mental process works, What are model and theories? Let's borrow from Carr (1982) in defining these ewo concepts. A theory isa set of principles (assump- sions or rules or laws) tat together constitute a verbal or mathematical description of an inter- esting phenomenon, and an explanation of how or why the phenomenon happens. A theory defines the important choracteristies of a phenomenon that are thea included in a model of che ior working parts of a real-life process phenomenon, A model represents a description of the m uch as ceasling). The description captures the most important characteristics of each pares opeta- tion, although it might leave out large amounts of detail, Currently there are 2 number of models of the reading process that (in our opinion) vary in the extent to which they vapture importance aspects of the skill (Rayner & Reichle, 2010; Reichle, 2012). We shall not attempr to deseribe the various models of reading here (see Chapter 14), Rather, let us simply charseterize chem as being primarily (1) botem-up, (2) top-doum, or (3) dsrtereciive inedels, facidertally, shete thaee eypes of models are charncteristie not only of the reading process, but also of descriptions of most of the tasks and phenomena that cognitive psychologists typically investigate. Some books on ceading (Just & Carpenter, 1987; Smith, 1971) begin by presenting their audience wich a model of weaving and then interpreting relevant evidence within the framework of that model, Other bouks (Crowder, 1982; Crowder & Wagner, 1992; Downing & Leong, 1982; Gibson & Levin, 1975) snanage to avoid presenting a model of reading altogether and prevent only the facts as interpreted by the authors (in some cases the rationale is that a single model cannot capture che complexicies of reading or the varieties of types of reading). Srll other books (Mitchell, 1989; Perferti, 1985; Rayner & Pollatsek, 1989; Taylor & Taylor, 1983) present evidence first and then, on the basis of the evidence, deseribe a model of the reading process. In general our strategy will be to present you with the facts os we see them. Our bias is that there isa danger in presenting. the model Gs and chen fiteing the facts to the anedel because such a stategy often makes it sound 4s if we know more than we really do, We also suspect that researchers often become committed toa particular model of che reading process and then the model itself becomes more important than the data that are collecced. However, as will be clear in Chapter 6, we will rely very heavily on a mode! of eye movement consrol in reading that we developed, and will se it as a convenient way of presenting some critical isstes to you. We feel eae most iiodels are File more than general frameworks for understanding research on reading that provide some biases on which aspects of reading are really important. Our discussion of models below will indicate many of our biases and provide che “bare bones” ofa general framie- otk. This fianework will seguire more detail as we progeess though the boak, In the final chapter, Chapter 14, we will briefly «ry 1 suuninarize the Gamework that has evolved throughout the book, ip bottom-up models (Gough, 1972), most information flows ia a passive manner through the humian information-processing syste. The major idea is that this flow of information is very fast and that knowledge we have stored in memory has little impact on how the processing takes place. In contrast, proponents of top-down models (Goodman, 7970; Smith, 1971) fect that the passive flow of information threngh the prosessing system is relatively dow because there are numerous bottlenecks (places where the architecture of the system forces us to slow down). Accordingly, to short-circuit these bottlenecks these models sues chat we rely heavily ou information stored in22 Background information memory (general information thac we have abour the world) to help speed up our processing. The primey way in which readers short-circuit che information-processing bottlenecks is to formulate hypod auc what they will next reid. This view of reading, often referred to as the hypothesis testi: modi of vending, was once very popular. However,a great deal of evidence aow sugents that the visual processing of text is actually very fixt and thy extent to which readers engage in hypodh- y, oF gutessing behaviors seenis to phy a niinimal wie in che process of zeading. We will return 1 this sme ae Narious points throughout che chapeers that follow, For now, Jet ws simply ieve 1s botwom-aup view of reading more accurately characterizes ntuch of the vwiilable evitene. Having sai this, we hasten co point out that we do nict think top-down processes play ao role in roading, “they clearly do, et point owe thae we by sour model of the reading process can best be described as a bortosnsup mode! in which the reader gets some help fron top-down processes We have toll you brielly what bottom-up and top-down models are, but we have not yet mentioned imeractive models, Invernctive models (lust & Carpenter, 1980; McClelland, 1986; Rammiclhart, 1977) allow for all sorts of communications between top-down snd bottom-up processes. These types af models have become very popular in the last 30 years. Proponents of these models claim thae such niodels are very good in accounting for the data on reading processes. Critics of ehese models argue that, while chey may be able ro account for lots of data, they are very uuconstrained and heave do sot predict in advance very well what the outcome of any particular experiment might be. In contrast, dhe major virtue of most bottom-up models is that chey are very good at making clear predictions about performance measures, The view of reading chat we will be presenting to you will largely be a bostomeup view, but with some influences from top-down proceses, Notice Ghat we have used che worl process 2 numiber of tines in this discussion. Ekewhere in this book we will make a distinction between the proves of reading and the product of reading. The product of reading is the information that gow stored away in memory:it is what gets comprehended during reading, The major emphasis ia book is on the process, eather dhin che product, of reading {although the latter will be discussed in Chapters # and 9) becie, fromm onr point of view, the most iniportant ding tw undersiandl about reading is che process. This iv a bias thet not everyone would age with. For example, educators vould undoubtedly argue that knowing the best way so teach children to read is more important than understanding the process of skilled reading, While we appreciate their opinion, our sense is dat if we can understand che process of skilled reading, we may well be able to provide usefi) informacion to educators about the end preduct of what they are trying vo teach (Rayner, Poorman, Perfetti, Pesetstey, & Seidenberg, 2101, 2002), In essence, we believe thar understanding the end produce (cilled reading) should provide firm: conclusions about how to instruct novices ro become skilled in the task. Hopefully oer discussion in Chapters 10 and 11 will highlight some of the ways that we believe research has made clear how children should be instructed to learn to read. Some cognitive psychologists who study the product of reading would ako want to angus with tas concerning our bias tovnrd understanding the process of reading, To their way of thinking, what people remember from what they read may be more important than tow they go about the process of reading, However, our response to such a poine is that understanding, che process by which some mental structure is created almost logically eneails understanding thot structare. In contrast, understanding what gets stored in inetnory may not reveat muich about the processes that created the structure, ‘Thus umdertanding what is in memory as a result of reading discourse may not be unique to reading—escentially the same structuces uray be cteated when people listen to discourse. We are not saying that understanding the product of reading and how that product gets remembered is not important, Its just cha reading is a remarkable skill chat must be understood quite apart fiont issues like general comprehension skills and intelligence.2 WRITING SYSTEMS Perhaps the place to start our detailed discussion of reading is ac the beginning of the reading process: the printed (or written) page. A carefull snalysis of the information contained in the squigakss on the page will help in understanding the task that confionts the reader. In the course of chis analysis we will inteoduce several linguistic concepss relating co both spoken and written language dat are necessary for a uveaningtul discussion of reading, A general discussion of writing systems will ako help to puc te task of reading English in a broader context Before plunging inta a discussion of weiting systems we might hazard a definition of what writing is. At first blush the exercise seems silly, since we all know what writing means. One defi nition is dat writing “isthe ‘fixing’ of spoken Inagaage in a permanent or semi-permanent form” Diringes, 1962, p13). sens that such a definition is too broad,since we wouldn't want to count a tape or phonograph recording of speech as writing, even thoagh it anay serve roughly the same finetion as a written transcript. Somehow we feel chat writing implies that the record is to be perceived by chy eyes. What abou Braille? Most people would call cat a writing system, so that the eyes aren't necessary. Will any code do? For example, is listening to Morse code wading? We doubs that most people would accept that, So, inherent in writing is some sort of spatially arranged message Chat is usually perceived by the eyes, but could be perceived by a tactile sense, (While reading, systems for blind people, such as Braille, are certainly writing systems, they fall largely beyond the seape of this book since the perceprual systemi for encoding the message is so different than in ordinary reading.) ‘The major problem in defining writing comes in trying to determine whether we agree with the requirement that writing “fixes” speech, and iso, whac we mean by ie. Some people might find that requirement too restrictive, and deems as writing any visual message that conveys meaning, That definition seems to0 loose, since most people would probably not call a painting such as the Mona Lisa” writing, buc would reserve che term weiting for graphic displays that are understood 38 a code for a spoken message. There is disagreement, however, about how literal a transcription of the spoken word the geaphie display has to be, in order to count as writing. Consider the following communication: there is a pictuce of Cluristopher Columbns followed by a picnare of thiee ships followed by 80 suns followed by a pictore of an island, ete. Mast people in our culture would probably be able to deduce thar these pictures stood for a narrative about Colutsbus for some explores) sailing across the ocean, making many cays to do so, discovering Jana, exe. 1s such a zepresensation writing? This appears to be a borderline case: some people would1B Background information i views, langel¥ reading; however, there is substantial disagreement about what the role is, In some views, largel hased on de small regulscity effect which appears only for low-frequency words, itis very ae (eis, Seidenburg ot al., 1984u). However, there are data that lead to a different conelsion. ia «ven bigh-fixquency words may be biased by sound codes (Pollatsck etal, 2005) and warts even pictdlo wend) can be mischasified a their homophones (Van Orden, 1987). A majot problen iv dee idinys ov the sole of sound in werd encoxting is that irreyolarites in most langage a twins i ic ibavot leur whether ie wvasonable ro expect a large regularity effect These i679 simple resolition to tle problem, ‘The position we have taken is that the data are Sinan «cooper camputions model whersin entries in the lexicon are excited by both the ie svn toute sal the inditeet valesto-sound route, with the recognized word being, the entry thi has acerued the mot combined excitition. hat is, we see the sound system as heavily invol ee lexical access, At present, however, rexsonable peaple can hold almost diametrically opposite von an the subject The conmos yrourd forall postions is chat diet visual access is hmportant that sound encoding plays soine part. 1c ‘There is also. evidence Uh moxphentically complex words ate probably locked up in two a possibly with the help of morphemiic rules (eg., Toft, 1988; Taft & Forster, 1976), Thus word see “ppeats C9 vole three systems: the disect visual route speling-to-zoumnd route, and a wiorp " of the decomposition route—that isa dtect route ancl owo more comiructive Proce Since mos of Mt evidence duat wort processing is automatic comes fou the study of relatively short, frequen! fi.e,, those for whieh the. direct is nots? Toute coukl predominate}, itis possible that word processing. i ™ automatic for word whose access reli ies more heavily on the more constructive routes. sndowt [We should emphasize cae this chapter has forthe most part deale wish + ately marrow wind of word perception; we have discussed {G) skilled readers (b) of English (.) reading isolated pr . nts WE ons However, our bef dicusion of cros-cultal sands ted us to conehete char che points Stewed wate genenlly true (or other writingsystems. We have focused on English, since it has since ther? ¢ handwriting is ofen quite messy, sentential context may en print, Second, since lettors are may be ‘more important in deciphering it eh not transparent visual units in print, more constructive processes needed in addition to automatic letter detection- “ Let us close with some comments on the relevance of the study of sill readers to the proce ofleneniog to reid. Fits che better we understand the word identification process in skilled reade she better we understand what the goa! of reading instruction should be. However, event a per ‘nderstandng ofthe skilled weadee nay Notes 1 In all of te above experiments dhe locatio 2 In addition to the models discuss! in & Perception hive been proposed, ‘Thes (Davis, 2010), the SERIOL model (Whin 1996) 2 of che initial consonane yas preserved. fais chapter, other models dealing with various aspects oF we lude the Bayesian Feader (Norris, 2606), the SOLAR ‘teats: ‘ney, 2001), the Multiple Read Our model {Grainger & J andthe Overlap model (Gomes, Ratelifl,& Peres, 20038).PART II Skilled Reading of Text im eading text we do much more than identify words, However, identification of words is clearly Ah important fas step in comprehending text—as the eyes move across he printed page, presum= “SY Words ore frst identified aud then ghied together ino larger structures such = phases, Pantences, and parspraphs, From these Inger steuctures we are able to comprehend che text we ace ling. We are able co infer the gist ofthe text or certain relationships. and store is information whg2M913. The cena tase in eeading wescaci isto understand how all his accomplished a the reader. we are really to understand the process of reading, we would like to froth details OF this cognitive activity fron, moment to moment. For example, iF sentence such as “The mn Bit he dois read, we would He to kaow when and how exch word wa ented, when x the seadet identified the man as the actor and the dog :s the recipient of ehe action, and when ‘nil how the reader realized that the sentence was grammatically correct but mildly absurd, A encral genet ofthis book is that dhe record of how the eyes move during lene reading of text Sy fr che best way to study the process of reading, Other methods ave useful (such as the single ‘pathos described in Pare but wll ds the process of eating so wach that one {RS Whether the conclusions dawn fn dhew would geneiize e9 normal stent reading one “W movements can be measured relatively unobtrusively when someone is slewty reading dog RY Allow us to study real eading, Ir addition to being unobirasive, the eye move cond tee, He us quite a bit oF insight inzo the cognitive provesss of veading (we ope the res copa ipters will document). However, understanding th sonhip of eye Processes in reading requires mastering some technic 7 , j ae Plan of the nexe three clupiess js as follows. In Chapters 4 and 5 we dss bev wa "ation is extracted fiom the printed page. Chapter 4 presents some basic facts abo ye tigg, DEP and discusses what information is extracted from the page on a single shace oe fixa- mae en 5 Continues the diceson by examining the flew of information when the ees Waltes the page during the silled act of silent reading. More specifically, we try to det nin informe RE SoeRttve events contzol eye movements, and attempe co relate the ssquon oft on ation from the printed page to che movements of the es. Sines these 0 men " Pree tition of visual information, they naturally focus on the identification of words, Chapter Re us 2a implemented computational model of the control of eye movements in. reading, ‘This del allows us to ineegeave many of the facts that were presented in easter chapters and to how they interact in determining when and where the eyes move in normal reading, Word never plore Be SSCSTHE WORK OF THE EYES When wre read we have the impression that our eyes {and mind) sweep continuously across the text Sept for a few places in which we encounter difficulty, ond at those poinss we patie to consider SihGt we have just read or regress (go back) co reread earlier material, However, that impression is f.lision, Fist, che progeess of the eyes actoss the page is not continuous. The eyes come to rest 1 Petiods that are usually between 150 and $00 ms; these periods when the eve is close to immo~ ate cslled fixations. Between the fixations are periods during which the eyes are moving Lidl: hese eye movements are called saccades after che French word for jump, Saccades aze listic movements {Le., once they start, they cannot be altered). When we read our eyes generally 20 forward about sever 10 nine character spaces with each saccade. The dasation of a saccade UA Peading vaties with the distance moved, with 4 typical succade cakicng cbout 20-25 ms.Since, for {Practica purposes, no visual information i extracted fiom the printed page during saccados ett of saccadic suppression (Matin, 1974), all visual information is extracted daring fixations. + Pattern OF information extraction during reading is thus a bit lke seeing a slide show-You see “lds for about a quacter ofa second, there ita brief off sime, and rhen a new slide ofa different ono the Page appeass for abouc a quatter ofa second, This parteen of Gxations and secs sang IME 19 reading. ‘The perception of any static display Gea plete or ascane ee the Wilh Mttough the pattern and timing of fixations differs fom that in ending (Rayne 1 moyen Cave. 8 Well, 2007), An exception is when the eyes tack a mowing rs ftos ura and gas tt-U this ease the eyes move relatively smoothly (and much slower than during saccade Seful visual informsation is extracted during the eye movement for 800d way in which our subjective impression 1 an illusion is thar the eyes do not move arid thtougi the eext as rclentiessly 48 we think. While mose saccades in resding ine forrt oF ee 10-15% move backwards and ate termed regressive saccades (or regeessions for short). Thin} ae HONS this way: singe we make about four co five snecades per second, we make a regression wove every 2 seconds. This makes it likely that we are generally unaware of most regressions Mile we aze probably aware of some regrewins that refect major coniion,requiting ust go reece erable distonce inthe txt to straighten things ovr we are likly 70 be waar ofmost ions which are quite short, only going back a few characters (to the preceding word or two), Snother type of eye mnoveiment in reading is the return sweep When the eyes move om near Moat oF one tine vo near the beginning of the nex:. Although vette, sweeps are right-to-left “Sments they are not counted as regressions, since they ate moving the reader forward through BR ee92. Skilled reading of text, the text, Return sweeps ave actually quive complicated as they often stare Give ¢© seven character spaces fiona the exxl ofa fing and they generally end on abous the third co seventh charveter space of the next line. Rewrn sweeps often Zall short of theie goal and chere is often an additional shore Linh ter lef saccaile after dhe large retara sweep. However, the leftmost fixation is still somerimes 1 the second word othe line, ‘Thus most of the time about 8096 of che line falls between the er we will explains why readers may often fail te fixate the beginning sux cud won of lines) “Phe ssnall saccades following return sseceps are probably corrections for iu aiming the eyes; ix is difficule wo execuse @ long saccade perfectly, with the eyes usually shootings the target position, Sitice thy detaifs of such motor execution are peripheral to our concerns here, mort of the interest in eye movement records is on what the eyes do on the middle four-fifths of the line, OF eomse, if one wanes to ger global measures of reading such as the overall ing speed, return sweeps anust be counted as well Another poine about the general properties of eye movernents during reading is that the wo eyes are not always perfecdy aligned on the same position ina word, For a long rime i¢ was assumed thar the two eyes typically land on che same letter in a word, or that they were perfectly aligned However, while on over 50% of the fixations the two eyes are aligned on the same letter, ehey axe on different letters quite often and sometimes the two eyes are even crossed (Liversedge, Rayner, White, Findlay, & McSorley, 2006; Liversedge, White, Findlay, & Rayner, 2006). While this is au iniporeant fice about the characteristics of eye movements during reading, it is also the case chat how long che eyes teniain in place isn’t dramatically affected by whether or not the two eyes are on the same letter Yukasz, Liversedge, White, & Rayner, 2006; for a complete review see Kirkby, Webster, Blythe, & Liversedge, 2008). In chis chapter, and chroughout the rest of this book, our focus will be on eye movements during silent reading, However, we need co poine our that there are both similarities and differences, beeween the eye movements readers make in silent and oral receding, Much of what we know about eye miovements during oral reading stems from seusinal work reported by Buswell (1922), but thers have been soine recent investigations of eye movements during oral reading (Inhoff, Solomon, Radach, & Seymour, 2011; Lanbrock & Kliegl, 2011) using much beter and more ‘accurate eye-tracking syseems than Buewell had available. Nevertheless, most of Buswell’s findings have held up rather well. What are the differences between silent and oral reading? The average fixation duration in oral reading is abour 80 rns longer than in sileat reading, the average saccade length is shorter overall, and there ave mote regressions. All of these findings are undoubsediy related to the fact that readers don’t want their eyes to get too far ahead of their voice. Thus there se places in the eye niovernent record where she reader is obviously keeping the eyes in a holding Pauterd so that this doesn’t happen, ‘The eye-voice span, the distance the eyes are abead of the voice, is generally the focus of research on eye movements during oral reading. The main finding ts that the eyes are typically about two words ahead of the voice, and if the eyes get too far ahead renders (unconsciously) slow down the eyes’ movements so as to not get too far ahead. extreme fixation on at. tT Te suunuiarize, the eyes move forward (about seven to nine character spaces on average) 3 ling, bur not relentlessly x0. They pause for periods of approximately 200-250 ms, and move backward about 10-15% of the time. In this chepter we will discuss in considerable detail mach of ‘the cognitive processing, during allthis activity and its relation to the ongoing pattern of eye move ents, ‘This topic is interesting in itself, as itis at the core of understanding, visual cognition in reading and visual cognition more generally. In addition, understanding the details of the work of the eyes in reading is an invaluable tool for understanding che process of reading, We claim, in fact, that eye movements are by far the best tool to naderstand the process of normal silent resding (which undoubtedly accounts for well over 90% of the reading adults do}. fea‘The work of the eyes 93 by Chapter ( we discussed alternative methods for studying, ending of text (as opposed to indi- vidual words). The current chapter deals with how visnal information is extracted fiom exe. Thus swe will focus on what useful information readers extract during fixations and on how the eyes ave guided dhrough text. Necessary to understanding both topics is some basic information about eye movements in reading, ‘These data will be far more meaningfol, however, if we make them concrete, by examining an example of an eye movement record. Before we do, it is important co note chat readers’ eye movements are very much influenced by the lexical properties of the fixated words. So, for example, how long reders look ata wort is stongly influenced by factors like woud frequency. We will document these findings in more detail Jater, but for now it i ieaporeane to keep in mind that frctors like how frequent the word is in the English language has a large impact oa how long the eyes remain ou any given word. Basic characteristics of eye movements Figure 4.1 shows part of a page of text with a record of a reader’ eye movements superimposed oon the text. The average saccade length is about 8.5 characters, but the range is [ character to 18 characters, The aversge fixation duration is 218 ms, but the range is 66 to 416 ms, Notice that, for the mest part, words are fixated only once, However, eiviigh is fixated vice and pain and feast are not fixated at all. Since a fixation lands on or near almose ail, wouds, it appears tha: a nnajor purpose of eye movements is to bring all words close to the fovea, the region in che center of vision thot is best for processing fine detail (See Chaptee 1). However, what is causing the variability? Why are some words not fixated while others are fixated twice? Is this just miscalculation of the eye movements as in reearn sweeps, or dogs it reflect something deeper? Similarly, why do fixation durations differ? Does a long fixation dime on a word indicate cha the readcr is taking more time processing the fixated word, of are these vatiations in fixation time random? Assuming that fixation times are not random (which indeed they ase not), what fixation time de we use to index the procesing time for a word? If there is a single fixation on a word, there is litle choice: we simply measure the fixation duration on the word (this is typically referred to as single fixation duration), However, consider the ease of draiustortr in Figure 4.1. There are a number of posible candidates to measure processing rime here, The first is the duration of the first fixation (first fixation duration) which is 277 ms. ‘The second is gaze duragion, which is the comal fixation time on the word before the eye moves off {or 277 ms + 120 ms = 397 ms). (This measure assumes chac the second fixation was needed to finish processing the fixated word,} Another possibility és the total viewing time, which: includes later fixations on the word that are the result of regressive saceades. In the ease of brcnstona the sotil viewing time would be 576 ms. (This measure axswmes that the regression wos made in order to cuntinue processing the word in some way, although sometimes it means that the reader needed to verify that che original encocling of de word was correct.) Two other frequently ased measures are single fixation duration (when only a single fixation is made on a word, as in the case of sweet and remand) and go-past time (onretimes called regression path duration: the amoune of time that it takes the reader to move forward in the cext, inchiding regressions back to prior text a6 in fixations 41-44 in Figure 4.1). Variation of reading measures The recom in Figure 4.1 is typical of adult readers, Figure 4.2 shows the distributions of fixation times and saccade lengths from 9 large corpus of data fom adult readers. As can be seen in9A Skilled reading of text Roadside joggers endure sweat, pain and angry drivers in the name of 12 3 2ae 224 213 me 2s 76 Nee fimoan Ahealthy body may seem raward cnough for most people. However, oD wR Mw Bw 7 ew 8 301177 19558 NN, 7 2g vee 199 for ali those who question the payoff, some recent research on physical a mo 2 Bm a 218 27a 108 280 225108 208 activity and creativity has provided some surprisingly good news, Regular 20 a0 2 8 a % Fe 21 8 = (OT 100 wo 00 Ft a8 ats bouts of aerobic oxercise may also help spark a brainstorm of creative ee ee 7 48 2 | 250 271 won 250 S205 227 zea 10 ate { 60 i v9 thinking. At least, this is the conclusion that was reached In a study that nr) 62 83 Sao? a) 200 as p28 416 200 177, 413 708 220 8 28 FIGURE 4.1 An excerpt Gom a pusage of text with fixation sequence and fixation durations indicated Figure 4.2, both the average saccade size and average fixation duration of our ticle segment (and the vaviabitity as well) are reasonably in agreement with the larger aggreyation of data Text differences ‘The averages and distributions in Figure 4.2 should not be regorded as numbers engraved in stone: reading measures such as reading rave, migan fixation duration, mean saccade tength, and percent of regressive fixations vary front text w text, Table 4.4 shows some of the variabilicy for adults reading text ont various topics, with more difficule text requiring longer fixations, regressions, and hence a slower reading rate. aller saccades, moreThe work ofthe eyes 95 4 12 10 Percentage 9 100-200-300 400 500600 Fixation duration (ms) 2 0 Percentage o 5 we 1s 20 25 Saccade Fength FIGUREA.2 Frequency distribution of fication duration (uppet graph) and saccade lengil (lower geoph) for eight college-age readers. Leturn sweeps of the eye have heen excluded front the distribution. Short fixations following the retmm sweep, which ace followed by corrective snccades, have also been excluded Typographic differences Is the pattern of eye movements dependent on typographic Feattces, such as letter size, type of font, length offline? Tinker (1963, 1965) studied this question in some detail for English (see Morrison & Inhoff, (981, for a review of this work). His data are complex, but the following brief sununary96 Skilled reading of text TABLE 4.1 Viriabilisy in adutes resting differen types of text Tops ction dination’ Saveade fag Regressions (24) EVPM Vii tie cans 3 305 Hewyriper ariel: 6 wa Hiseny 4 313 Pye luslopy " 308 Panelist lireraune 19 10 x5 Hcountnies a " 26H Mathematics 73 18 2s Physics 60 7 Be Biology oe {8 233 At 78 W 288 Mews fixaion churction, ican scesle feng, pram jons dae were reprewionn, ane woud per minete readin, time (WPM) for 1H good college-age readers reatling different eypee often “Joa. ‘tn charscter spaces (4 character spices = 1 of visual " Pezcentage of total feation that wens so pressions oF fi captuses the essence. First, the type of font made a minor difftvence, although all of the fonts that Tinker studied were (subjectively) relatively easy co read, There are some fonts that appear to be pathologically difficule (such as che elaborate script used in German known as“fractur”}, and these slow the reading process appreciably. Indeed, a recent experiment in which a stondird font (Times New Homan) was compured with a more difficult fone (ld English) revealed dhat the more difficule-to-encodg font led to more and longer Fi shorter saceades, and more regressions Rayner, Reichte, Soud, Williams, & Pollatsck, 2006), Slaecery and Rayner (2010) also found that clearer text Gin wich dhe letters were in sharper focus) led ¢o faster reading than when the cest was not as clear. Second, it is difficult to make inferences about low the size of the charicters influenced yeading, speed from Tinker’ data siuce the size of a character and the number of charscters per line were confounded in mos of the studies: there were more characters per line when the print was smaller (Morrison & (nhoif, 1981), However, Tinker varied line length (keeping the size of the chatacters constant) in une study, and the differences he observed for differing size of characters appear to be explained by line-length effects. He found that there was an optioul Ime size of approximately 52, characters. This optimality is parsimoniously explained by a trade-off between two opposing factors, Fhe tine is too long, return sweeps become incressingly difficule to execute and readers inay wind pon the wrong line. On the other hand, as we will shordly see, renders can extract information from more than one word on a line during 2 fxation. Iflines are too short, readers will not be able 10 rake fall advantage of the fact that they can extract information from more char one word per fixation, ‘The optimal line length thus appears to be the best compromise between these opposing desiga considerations, We should remark, however, chat all these effects are relatively minor, so that the furdamental conclusion to be drawn fiom the work on typography is that reading appears to proceed at about the sime rate if the type fone, size, and length of line employed are atall reasonable, Viewing distance effects In reacling, the average saccade is about seven to nine leteer spices long, or about 2 degrees of visual er, the value of seven co nine letter spaces appears ta be angle at a normal reading distance, Howe;The work of the eyes 97 more fundamental in that the average saccade size is seven to uine letter spaces regardless of the retinal size of the text (as long as che letters are not too big or tao small) “inus, for exssmple, eegard Jess of whether a given textis 36 cm or 72 cm from the eyes,the average saccade length is still about eight letters even though eight leteers subtends twice the visual angle at 36 enn as it does at 72 can (Morrison & Rayner, 1981; O'Regan, 1983}. This face suggests that the visibility of the text is relatively invariant to absolute size over an extended range of distances. AS a ret, data on siccade lengch are cypically expressed in letter spaces, which appears co be the natural mecrie in reading, rather tan degrees of visual angle, A recent clever manipulation (discussed in tnore detail later) les shown that when letters outside the center of vision are larger on each fixation (thereby compen siting in some sense for che poorer acuity with letters outside the center of vision), how far the eyes move is still driven by number of characters (Miallet, O'Donnell, & Sereno, 2009). The fact that the distance of the cext (and henes tne absolute size of the letters) makes Hirde difference in saccade size is probably duc to a usde-off between two actors: (a} when the vext is nearer, the letters are biggee and easier to see; however, (b) when the text is nearer, a given letter will be (archer fiom the center of fixation, hence harder so see (see Chapter 1). Of cours, there are limits; the text will be impossible to read ifa mile away or against your fice, Orthographic differences A question related to typographic differences is whether the writing system influences the process of reading. The information we presented in this chapter so for concerning eye movements is based on data collected from readers of English. Do the characteristics of eye movements change wlien people read text which wses other writing systems? The answer to this question is clearly “yes” as experiments dat have examined the patterns of eye movements of Chinese aud Japanese readers have demonstrited. However, a major problent with comparing saccade sizes in English with either of these languages is what to use as che unit of measurement, The previous section implied that the letter (or letter space) is the fundamental unit ‘of measurement for English. However, there are no letters per se in cither Chinese ot Japanese: the characters stanel for syllables and /er morphemes (see Chapter 2). fone measures by characters (Le., a letter is a character), then eye movements of Chinese and Japanese readers ten to be much smaller chan eye movements of readers of English. Chinese readers move their eyes about two characters on average (Shen, 1927; Stern, 1978; Wang, 1935). (Remember that a character is a morpheme rather than a word, so that chis i less than two words) Readers of Japanese text, which is made up of morphemie charsetcrs (Kanji) and syllabic characsers (Kana}, move their eyes about theee and half characters (Ikeda & Saida, 1978). This again is less than three and a bal words, since it offen takes several characters to make 1 wonl. Since the average saccade in English is about seven to nine characters (about a word aad a hal} it appears tht the average saccade Jengeh is ifanything a bit lessin English than in Chinese and Japanese ifone equates for uuraber of words or morpliemes. Readers of Hebrew sso have smaller saccades (about five and a half characters on average) that readers of English (Poliatsek, Bolozky, Well & Rayner, 1981). Hebrew differs structurally and orthographically fom English in some impoctane ways. First, as discussed in Chapter 2, not all vowels ae represented orthographically in Hebrew: In addition, many fimetion words in Hebrew are clitic, meaning they are attached like prefixes or suffixes to content wouls. The net effect of these differences is that Hebrew sentences normally contain fewer words and fewer letters than their Enyiish counterparts. hi shore, alchough Hebrew is basically an alphabetic system, che inti jon is also more densely packed than itr English. “The average saccade lengths of Chinese, Japanese, and Hebrew readers suggest that the informs sional density of the text determines how far the eyes move in each saccade, This finding seems.98 skilled reading of text consistent with the fact dc, for readers of English, as the text becomes more difficult (and hence the informational densicy i greater), saceade length decreases, However, it is an open question the diiferences in isformationa! density acros: languages are best thought of in terms of she ceusity of the meaning or the arnount of visual information per character Gneayured pechaps hy Ui number of sikes or lines in the characrer). For Hebrew the characters seem of approxi- nately equal complesity to Koglis, xo the differences between Hebrew and English are more likely to be expliined by differences in the amount of meaning per character. However, the Chinese | systenss are so different from English that itis hand ro say which eype of infor ypaniese wr tnational density is operating to produce the differences in reading, We suspect that both the visual and semantic factors are contributing, Fixation durations ior seaders of Japanese, Chinese, and Hebrew are foirly similar t those of readers of English, Despite the fict dhat reading in these languages is slower when measured super- ficially, reading rates, when measured in tormns of amount of meaning extracted per unit time, seem to be equivalent. In fnct, whew the reading rate in Hebrew was based on the number of words ja the English translations of the Hebrew sencences, the average reading rate for the Israeli and English speaking participants was nvarly identical (Pollatsek et al., 1981). One final dimension of orthogeaphies isthe direction in which the characters proceed. As we pointed out in Chapter 2, there were no clear conchisions that could be drawn about the effect of the direction of print on the eye movements or the efficiency of reading. In general, the results were consistent with the hypothesis thie differences in the direction of print do not matter and that all differences observed in reading speed were due to the more familiae orthography being read more easily (Shen, 1927; Sun et al, 1985). A similar conclusion follows from laboratory experiments which manipulated the dixection of print. Tinker (1955), for example, found that readers of English initially read vertically arranged English 506 slower than horizontally arranged text, However, with 4 weeks of practice their reading speed was only 22% slower chan for che horizontal text. Ins number of experiments, Kolets (1972) likewise showed chat, with practice, teaders of English ean read text arcanged in a right-to-left fashion faisly well. Children Jearning to read can also read from right to left as easily as they raid left to right (Clay, 1979), While relatively short amounts of practice in the laboratory did not abolish differences in the arrangement of text (Kolers, 1972), they also suggest thae differences between arrangements of print, if they exist, are Tikely to be quite small. There is sorne physiological reason to believe that a horizontal arrange- ment in any Lnguage may be better: visual acuiey falls off faster in the vertical direction than the horizontal direction, However, given that no direction of text is preferred over any other sugzests: that this physiological fact may have a relatively small effect on reading speed. A few comments about saccades and fixations At the beginning of this chapter we claimed that reading was a slide show in which the eyes remained in place for a certain peried of time (the fixation) and then moved quickly with no visual information extracted during the move (the saccade). While these claims are essentially true, they are slight oversimplifications. We will briefly discuss the complexities,so that we can set the record straight, However, for the rennainder of the chapter and book these complexities are so insignificant that we can safely deal with the stide show metaphor. Saccades First let us consider the assertion we made that no visual information is extracted during a saccade, ‘You can demonstrate for yourself that little is perceived during saccades by looking in a mirror andThe work of the eyes 99 trying 10 watch your eyes move. You will norsee them doso. This reduced percepribibiy of simi ation during saccades was discovered over 100 years ago (Dodge, 190; Holt, 1903), Why don’t we see anything during the saccade? Fist, the eyes are moving so fast during a saceade that dhe image “painted” on the eyes by a fixed scinmalus would be largely a smear and thus highly unintelligible. Flowever, we are ot aware of any smear. Thus there must be some ntecka- nisms suppressing the largely useless information thar is“painted” on the retina ducing the saccade, One putative mechanism is central anesthesia. That is, itis possible that when the brain knows that the eyes are making (or about to make) a saccade,it sends outa signal to Ue visual system to ignore (or attenvate} all input from the eyes until the saceade is over, There is in fact evidence (Main, 1974} chat the thresholds for stimuli shown dusing or even a bit before a saccade begins and after it ends are raised (with the effect much more pronounced for stimuli presented during a saccade}, This increased threshold before and after the saccade is not of much importance for reading since the lesters seen in text are fir above Unresholé. Thus it is not clear whether these relatively sinall chreshold effects would mean char the ability to exerace information ftom the text would be alered significantly. (That is, it might be like the difference between reading with a 6l)-watt bulb and yeading with a 150-wate bulb.) However, the threshold effects are more ikely ¢o be significant with the moving eye, where the contrast between the light and dark patts of the satear would be far less. For many years centeal anesthesiz was accepted as che main mechanism by whieh information during saccades was suppressed, However, anore recent experiments indicate cbat a different mech nisin explains at least part of the suppression. It can be demonstrated char under certain (unnat- ural circumstances, visual input during the saccade van be perceived (Liteal & Smith, 1968} when the room is totally dark: prior co and afier the saccade, and a pattern is presented only during the saceade, 1 sinesred image of the pattern is perceived (Campbell & Waste, 1978), Since the blur is thus seen if no visual stimulation precedes or follows is, the implication is that the information available prior to and after the saccade during nozinal vision masks the perception of aay informa dion acquired during the saccade, This phenomenon hs been related to laboratory phenomena of anasking such as those used in subliminal priming experimetits (See Chapcer 3). In sum, while we can’: say for sure that absolutely 00 visual informacion is extracted during saccades in reading, the evidence indicates chat if visual information gets in daring a ssccade, itis of little pructical innpor- tnee. Indeed, Wolvertan and Zole (1983) replaced the cext with a mask during each saccade as people read and ic was nor perceived nor cid it affece reading in any way, Definition of fixations ‘Our chim chat the eyes are inunobile during a fixation isa bit of an oversinmplification, As indi cated in Chapter I, very sinall rapid movements, called izemaz or nystagmus, go on constantly 00 help the nerve cells in the retina to keep firing. However, these are 30 sunall as to be of litde prac- tical importance in studying normal reading. There are also somewhat larger movements called microsaccades and deifis. While the ressons for these movements ate not completely clear, it appears that the eyes occasionally drift (ie., make « sinall and cather slow movement} due to less than perfect contol of the oculomotor system by the nervous syscem, When this happens, there is often a small (one charicter or less) mnicrosiceade (ie.,a much more rapid small moversent) to bring the eyes back to where they wore. Some of the drift is ako acenally vergence between the wo eyes, serving to reduce fixation disparity between the eyes, Many experioionters assume that these snuall movements ie “noise” and adopt scoring proce- duces im which shese sinall movennents are ignored, For example, some scoring procedures take succesive fixations that are separated by a character or less and himp them together as a single100. Skilled reading of text fixition, Some msicrosaccades nay be ander cognitive control (like ochez saccades) and thus sont: experimenters believe that microsaccades should be created no differently from ocher saccades, Another alternative is a more sophisticated pooling procedure in which fixations are pooled ifthe intervening ssecadle is a character or less and at least one of the fixations is short (LOG voi or es Mont eye movenserit data in reading bave been adjusted using some sort of procedure that pools wires at least some stuall drifts and microsaccades. In some cases the eye 1; system» is not sensitive enough to detect these small movements, so such amoventents are aitlonnitically ignored. Others with urore sensitive equipment decide on some sort of criterion for pooling. Since deifis and miccosaccades are relatively uninteresting aspects of the eye movement record for reading, aud since there is enough complexity in the data without ‘worrying about dhesn, aur subsequent discussion will ignore diets for the most part. sone Fiscal ul snoveritent reconli Summary ‘We inave summarized the basic facts about eye movements in reading, ‘The eyss move about four of five times per second and jump an average of about seven to nine letter spaces each time they move for readers of English. However, the eyes inove back about 10-15% of the time and there is large variability is both the extent of the forward motion and the amount of time they stay in a Sixation. Since virtually all che ioformation is extracted during the fixations, che inerest in fixations is on how their duration reflects the processing of information during the fixation. Since saccades cexist to move the eyes to another fixation, the interest in saccades is the exten to which the direc tion and size of the saccade reflect what is being processed. The perceptual span Since the eyes move four to five times a second during reading, it scems reasonable to assume thor they move to new loeations on the page because the amount of information that can be extracted from a given Gxation is linited, However, some advocates of techniques for increasing reading speed claimn that many of our eye movements are not necessary andl that large amounts formation can be extracted from a single glance (see Chapter 13). Thus, if we are to under stand which view is true—whether eye movements ate a centre! functional part of the reading process or just bad habic picked up from old-fashioned readmg methods—we have to discover how much information from the prineed page is obtained fom an individual fixation during silent reading of text. As we will see, dhe constart movement of the eyes is not a bad habit: the region from which we can vblain usefel information during each eye fixation is relatively small. One reason why some people uy believe a lage amount of information can be extracted from a single fixation is that it often seems 10 us thac we can see muamy words on the page ar che same time, However, chis is an illusion. Many of the words are seen on a fixation only in the sense that the reader knows that some word-like object is in a given location. The brain takes the detail extracted fiom each location and incegrates then: somehow mto a perception that the detail from a wide area is seen on each fixation. We discuss this integration process in the next section. In this section we will briefly describe various atcerapts to determine the size of the effective visual field (or perceptual span) on a fixation in reading. We will fst review tachistoscopie tech- niques, chen tzc) juues bascd on eye movements, aud conclude the section by discussing che technologically more sophisticated gnze-contingent moving-window technique:The workef the eyes 101 Fixed-eye techniques “The tachistoscope (¢-scope), which we introduced ia Chapter L, was designed in part to determine how muuch nsefil information could be acquized during an eye fixation in reading, Psychologists hoped to mesure the perceptual span by asking paticipants co report all they could see when a sentence was exposed briefly, say 100-200 ms, Since such an exposure duration is brief cnomgh to preclude the posibility of an eye moveanent during the presentation, che technique measures how much information can be reposted fiom a single fixation, ‘Thus, co some extent, the technique simulates a single fixation in wading, An expetiment by Marcel (1974} serves co illustrate the logic of the method and ies attendant probleins. Marcel had participants read a short fiagment of a passage in a tscope, When chey reached the final word of the fragment, they read it aloud. The pronunciation of this word caused the text ¢o disappear, and HK ms later some more worls were presented for 200 ms, just (0 the right of where the pronotmced word had been. The participant: task was to report as many words from the second set 23 possible. ‘This second set of words was not actually text, but a sequence of words thet varied in how closely they approximated normal English, When the sequence of words was essentially random, pasticipants were able to zeport jast over ewo words (or roughly 13 char- acter spaces}, while when the sequences were close to normal Euglish, dhey reported three or four words (18 to 26 character spaces). Since the stinmli in this last condition are most like normal text, perhaps three to four words provides a good estimate of the perceptual span in reading. There are three potential problems with this type of research, First, the delay between the offset of the passage fragment and the onset of the target words is quite different from anything encoun- tered during normal reading. The delay is usually about 600 ms (sbout 500 ms to begin pronun- ciation of the last word of the first fragment plus the 100 ms experimental delay), Second, eye positions were not monitored so that the experimenter did not know where the participants’ eyes really were, The thicd and geeatest problem, however, is that the expetimenter has no control over the extent to which the participant is consciously guessing, In Marcel’ experiment, for example, since the participants attempted Co report what had been seen, there was little conerol aver the speed of the response. Thus better performance on the sequences that closely appzeximated English may have been becouse te participant could guess which words were likely to follow from the constraints of the text (possibly aided by partial information obtained from the stimulus). fa contrast, the use of random sequences gets around the guessing problem but may disrupt the normat reading situation, Another tzchistoseopie technique that has been drequently used €o make inferences about the perceptual span in reading (Feinberg, 1949) involves asking a participant to fixate some point and then identify stirwuli (words or ketters) presented at various distances from fixation so briefly ¢hae no eye movement can occur, On the basis of the tesults from such experiments (see Figure 4.3), estimates of the perceptual span have generally been in the range of two or three words, or about 16-20 characeers (Feinberg, 1949; Woodworth, 1938}. A strength of the latter method is that, by the use of isolsted words in che visual field, one can limic the guessing problem and get a better estimate of whether the word can be identified on the basis of the available visual information. The method has its problems, however, As we discussed in Chapter 1, Sperling (1960) deronserated that we are able to see much more than we ean retain and later report, Thus, what participants report from a brief word of letter presentation can not be taken asa complete specification of whist they actually saw: Even if the verbal report coincided with ‘what che participant sav, there is no particular reason to believe thet the estimate of the perceptual span obtained from either ype of tachistoscopic presentation diteussed here actually ecincides with chat ofa fixation in reading. A second problem is thac she responses are noc timed. ‘Thus one102. Skilled reading of text el Teck FIKATION POINT AREA OF FIGURE 4.3 Example of how the perceprial span is extimazed from tachiscoscopie acuity daea, (From Taylor, 1965.3 discovers whether the word cant be identified on the basis of the available visual information but not whether ic can be identified 2s qnickly as it needs to be in normal reading. Even if all the guessing problems could be removed, there might be 1 real difference between the perceptual span in silent reading and in t-scope presentation of words or sentences, The perceptual span in reading could be Jagger either becouse the contextual consteaine in text allows a teader to identify words with less visual information chan in t-scope presentations, or because the requirement to hold the eyes still interferes with normal perception. On the other hand, dhe percepteal syan in reading coukl be sinaller because dhe rapid sequence of fixations and dhe complexity of the surrounding stimulus pattern may lead co “tunnel vision” (Mackworth, 1905). Primitive window techniques A somewhat different technique involving experimental contol of what is see on a given fixation iy to present text but to limit the amount that is visible to a reader at 2 given: moment.The work of the eyes 103 Poulton (1962) had participants read aloud from text over which a mask coneiining a “window” was passed. Only the text in che window could be seen, Thus the text was immobile and the window passed over it, allowing only a certain antount to he seen sit once. The speed and size of the window varied systematically on different trials and readers’ eye movements were recorded. Newnnn (1966) and Bourn and de Voogd (1974) reversed the procedure by having the partici~ ponts hold fixation and che text moved on a screen fiom right to left. The size of the window was manipulated by varying the number of letters on the screen at any moment, These experiments typically found chat smaller windows create greater disruptions in reading, than Langer windows. These techniques, however.are suspect since they diseapr normal reading: the reader's natural eye movements were inhibited (in the Intter case, fixation had co be maiutained; in the former, the reader had to follow the moving window); in neither situation could the reader re-examine text (via regressions); in addition, these particular experiments suffered becauve the participants wore required to read the text orally. Estimates based on reading where natural eye movements are allowed The teclniques mentioned so far seem to be unsatisfactory, They involve casks that disyupt the normal reading situation. In addition, they provide rather discrepant estimates of the size of the perceptual span with the estimated size ranging ftom one or two t four words. It would cleacly be better ifone could estimate the perceptual span directly fromm pormul silenc reading. One siunple technique for estimating the perceptual span fiom natural reading is measucing die average number of words per fixation. That is, one simply records eye movements during reading. and divides the nuraber of words read by the auniber of fixations used to read those words (Teylor, 1965). Using such a technique Taylor estimaved the perceptual span for skilled readers 10 be 1.11 words, While this suethod is simple and unobtrusive, itis unformnately based on the asumption W the percepeual spans an successive @xations do mot overlap. In other words, it assumes tae a given word or letter is never processed on mote than one fixation, As we shall see this assumption 3s fe. Gaze-contingent display change techniques ‘The gazu-contingent moving-window technigue introduced by MeConkie and Tayner (1975) uses the idea of the moving-window techniques discussed before~to manipulate what is seen on a given fixation —but does so in the task of worm silent readinng, where the participaues can move cheir eves wherever and whenever they wish. This is accomplished by presenting che text on a video monitor and making display changes in this kext as the participant is moving his or her eyes. ‘This research relics on sophisticated eye-tracking equipment interfaced with. a computer, which is also interfaced with the video monitor. The position of the reader’ eye is sampled every milli- second by the computer and changes in the text are made contingent upon the location of the eye. Because this type of research has been very influential for our understanding of skilled reading and tray provided clear answers concerning the size of the perceptual span, we will describe it in some devil (for further review see Rayner, 19783, 1998, 20009), In the prototypical moving-window experiment a version of nnutilated text (in which every letter from the original text is reploced by anether etter) is initially displayed on the video monitor. However, when the reader fixates on the text the display is itimeditely modified by the replice- meat of letters within a certain region around the fixation point with the corresponding letters fiom the original text. This creates an experimentally defined window region of nocumal text for the reader co see on that fixation, When die reader makes eye movement the text in theTOA Skilled reading of text window area returns to this wareadable form and 2 new window of normal cext is ercated a¢ the location ofthe new fixation. This, wherever the reader looks, there i 3 window of nocitial eae wad in background of mutilated text. Table 4.2 shows a line of cext and four successive fixation’ for one wacier neler proving-window conditions.' (Because of the sophistication of the tavent, the display « Header dees net 3 equip’ “ges eas be mate in about S-UW nis; cis change is rapid enough that the the changes that are taking place) wuuption in this wsearch is that when che window becomes smaller than the reactors perceptual spun. reading will be disrupted, By varying the size and location of the window The basic Teqion, the experimenter es determine what area of the text the reader is actually exteactinis Useful inforonation Gout on a fixation, Iby varying the type of information in the background area the experiner m1 yan or destroy various types of informacion that may be potentially uveftd) during rensiing and thus be more analytical about the type of information that a reader i$ extracting fom a region of the visual fickl In the original moving-window experiment McConkie and Rayner (1975) had participants tead text when the window was 13, 17, 2t, 25, 31, 37, 45, or 100 character spaces in width: A window size of 17 meant shat the reader hud normal text for she Ieter directly fixated and for Sletser spaces on either side as shown in Table 4.3, (With a window size of 100, the entize line was TABLE 4.2 An example ofa moving window Frsatins Ramp sumer 1 oorthelogy means perseticeE seNEENOSE EERK RHR SUCKER. ee ae 2 KempocORNXk sions personality diaxaunc: KexK OOH sOBETEEE. GE aa x 3 Gereemeces saci sonoma Kiagnesis from Name HxKKKEK. KAX x X 4 Jeecaoanoor romeo seocomeEnE JonmioGEGE xKom hand writing. joe oc eS ‘The ovtensk represcits the location of foarion yn four suecersive fixations TABLE 43 An example of line of text and the various text patteros derived fam it Text Graphology means personality diagnosis from hand writing. this is a xB SoucoonERE: soca eeonality diagnosis zKxx xxxK xmucKE. MEK Re x x XRAANXEHLENEXNAAKAIAKONAL LEY Adagnogs SKXKRARXK KARE K AIH IOCLE RRR, Ws Caojkaiazp wsore Jsmconality diagnosis tnaw keri mnlfira. Ykle le ve Chojkaiaqpausoreajsnconality diagnosisatnawakoriamalfirqaaavklealeao bs Ubtxwyayvo titel xiblonality diagnosis abyt widn hbemedv. Avel ol f be fib ixwysyvoatifdiaxiblonality diagnos isaaby tawfdnakbemedvaaaawelawlat —=aerororors SO On cach Line a window of tire 17 is shown, asuming the eezder is Fxating the letter XS ~ Loters repliseed with Xi —apaces preserted X= Lowers reploed with Xs spaces filed WS = Letiots replaced. with silk leew —=paces praerved. VE = Letters teplaced wih similar lerens—spaces filled. DS = Lotte seplaced with dissnniarlesess—spaces preserved. DE= Leers replaced with dinsinifr letesrs—spaces billed. digi,The work of the eyes 105 slosost always present.) The participanes ako read with che six different types of eext murilations shown in Table 4.3. The texts were 500-word passages and the participants were told rat they would answer questions afterwards that would test their comprehension of the passages. MeConkie and Rayner found that reducing the size of the window had a substantial effect om reading speed as reflected in verious measures of eye movenient behavior, increasing reading time by as much 6.600%, bur had no effect on readers ability to answer questions about the text, Rayner and Bertera {1979} subsequently used window sizes runging froma | letter space to 29 letter spaces. Wish windows as small as 7 letter spaces, resders can usually see litde more than one word ata time. This reduces their reading speed to about 60% of theie normal reading speed. However, ehey can still read with pormal comprehension. Rayner and Bertera found no effect on comprehension unless the window was reduced to only one character (in which case readers are literally reading, letter-by-tetter) “The first question chat McConkie and [ayner asked was how large the windov had to be made in order for readers to be able to read normally (i.e. at both normal speed and comprehension). The answer was 31 lecter spaces or 15 letter spaces to each side of fixation. When the window was reduced to less than shae, reading rate was reduced. In other words, the experiment indicates that the perceptual span extends co something like (5 letter spaces from the fixation point, This finding has been subsequently replicated by a eumber of studies with the extent of the span te dhe right of fixation being 14-15 letcer spaces (DewBuurman, Boerseina, & Gerrissen, 1981; Rayner & Bertera, 1979, Rayner, Castellano, & Yang, 209%; Rayner, lnboff, Morcison, Slowiaczek, & Bertera, 1981). Incerestingly, Miellet et al. (2009) also ceplicated che basic findings concerning the perceptual span when they used a navel gaze-contingent paradigm in which the letters outside of che center of vision were larger on each fixation (see Figuee 4.4). ‘Their resules basically demonstrate thae acuity gone never get rid of the wowwmoaorthe image £] uld neve vevget he mage from hl S m er get Bd of gv ton his mind’ s eye LG Of the sn moines ove. Ee 1mage from Ps ein FIGURE 4.4 Graphical depiction of the parafoveal magnification (2M) paradigm, from Miellet et al., 269, The location of each fixation is indicated with an arrow and che corresponding display for hat fixation is represented. Consecutive lines sepresent the chronological onder of fixations106 Skilled reading of text linunations, while important, are Vhs Wier bu ‘ ot the ynain factor influencing the size of dhe readers appear to extract some sort af useful information out te caf fixation (eee below for why only information to the right of fixation ig mainly relevsnt to uve information beyond thet. Bor what kind of information i ie? Des they extract reaping, of words out dhat fi, oF only sone information about che compomtene leteets or do they merely exteiet sone information shout where wards begin and end that mighe be asehl i kuvowing where to phtve the next fixanion? One way to atuick the question of how fir fiom fixation diffe exoucted is by experimentally manipula text, MeConkie and By Comparing ewo kinds oF altered displays outside dhe window: In one, all letters in words were replaced by Xs but che spaces betwen werds were preserved, while in the other, the spaces were replaced by Xs as well. By comparing performance in these two background conditions ovte ea tell how for from fixation the presence of spaces makes a difference. When the window sige Wis 25 chaeaeters or fewer (12 or fewer leer spaces to the right of fixation), reading was faster when spaces were present among the Xs in the background than when they were not.On the ether hands when die window size was 31 or greater (15 or more letter spaces to the right), there was no diflet~ ence between the background conditions, ‘has it appeats that readers use the information of where spaces are out te about 15 charicter positions from fixation, movements into that region. MeConkie and Rayner also attempted to determine how far from fixation information about the shapes of letters and words is extracted. They compared backgrounds ini which the leteere wert visually similar to the letters in the text (having the sane parcern of ascenders and descenders) with backgrounds in which che letters were visually dissimilar (see Table here, if there is percepiwal spar. about 14-15 leer spaces #2 rent kinds of information can be probably to help guide their eye 4.3). The same logic applies difference berween these evo background conditions at a certain window size, then some information bout the shapes of letters and/or words is being extracted beyond the end of the window. Their data indicated that letter shape information was fot extracted as far ont 35 word boundary information, since there were differences between these two background condi- tions only for windows up to 21 Jetter spaces (10 to the right). eis worth noting that the "“swindow Of consciousness” for letter information is significantly smaller than that, extending little beyond the fixated word, If the fixated word is preserved, and dhe background vaguely resembles normal text (eg.spaces are left berween the words aud all letters in de background are replaced by similit letters), the reader is rarely aware of seving anything other than normal text (even those who know thar it isn’c uormaly, However, they are often aware that they are reading slowly and that something is holditys chem back. Further studies have greatly increased our understanding of the percepeual span (ee Raynes 1998, 2009). We should point out that single sentences weee employed in many of these exper tnemts, since it is technically difficult to make display changes rapidly and not have a lot of flicker in the text display. Portunavely these sentence-reading experiments have closely replicated dhose using passages of text, so that we can be reasonably confident dat the data frum the sencence- reading experiments is a youd approximation so what would be obtained under more naceral reading conditions. Auother question is whether the perceptual span is symmetric, In the original McConkie and Rayner experiments, the distance that normtal text was extended was the same on both sides of fixation so that it sos not possible to east whether readers extract more information from one side of fixation than the other, To test the synmmetty of the perceptual spon, MeConkie ond Rayner (19760) independently varied the lef and right boundaries of the window of normal text and found that when the window extended 4 lester spaces co the lefi of fixation and 14 to the right,‘The work of the eyes 107 resting was virtually ay fast as when the window extended 14 letter spaces in each direction. In contrast, when the window extended 14 letter spaces to the leit of fixation and 4 to the right, renting was mackedly impaired. Thus, for readers of English, the perceptual span is asymmetric, with informacion froma the right of fixation being used much further out. Rayner, Well, and Polasek (1980) and Rayner, Well, Pollatsek, and Bectera (1982) Arcther extended che work on the size of the perceptual span, Their major finding was that the left and right boundaries of the pecceptual span are somewhae differently constituted. They compared conditions in which the wiudow was experimentally defined by the number of visible letters with conditions in which the window was experimentally defined by che number of visible words. ‘They found that the major determiner of the left boundary is the beginning of the currendy fixated word. That is, when the left boundary of the window was manipulated, the speed of reatling could be predicted by knowing whether the currently fixated word was visible: beyond ensvring that the beginning of the fixated word was visible, che number of lecters to the left of fixation had virtually no effect.On the other hand, the right boundary of the peceptual span does not appear to depend on word boundaries. When che window to the right of fixation was varied, the major dererminant of reading speed was the number of letters visible: given that a certain number of letters were visible, it made ite difference whether whole words were preserved or whether a word was partially visible (even che fixated word). For example, the ceading rate was the same when the right boundary of the window was three lesters to the right of the fixated letter as when the right boundary was defined to be che end of the fixated word, in spite of the Fact that, in the former case, the fixated word was not entirely visible abouta third of the time. The fact that reading speed did not appear to depend on whether the right boundary of the window maintains the integrity of words (see Table 4.4) suggests that readers acquice information about parts of words from parafoveal vision. Rayner et al. (1982) reporced more deuailed evidence about the partial word information that readers obtain fron parafoveal vision. They asked pacticipants 0 read when (1) only the fixated word was visible and all other letters the tight of fixation were replaced by another letter, (2) the fixated word and the word to the right of fixation were visible and all other leteers were replaced by another letter, or (3) the fixated word wos visible and partial information about the word £0 the right of fixation was visible In the thitd condition one, two, or three letters of the word to the right of fixation were visible (ee Table 4.4). When the first three letters in the word to the right TABLE 4.4 Exanuples of conditions in the [Rayner et al (12) seucy and wading sates associnved with them in words pet minute Winslow size Sexton Revi rate tw (3.7) An experiment 2x exmtoune ax nan ax 212 wan 2W (9.6) An experiment was sonouDEE xx me mE 309 WoR 30 (15-0) An experiment was conducted xx xxx aux 339 wpm ot An experimmnn sone saeseconer ax oo emt 207 KOR on An experiment wax .2o0G00G XC 2006 25N 308 WpRL ton An experiment was condisoum xx aor sux 260 wom Jn the W conditions word imegrity s prserved, while in the L conditions the right bbounshey is determinad by the number of lewers visible, The vulues in patenthesst ake the average number of lcers visible in the W eonaitéons, ta all cases, dhe fixated let he second ein experennt106 Skilled reading of text of fixation were vuble and che ather letters were replaced by visually similar letters, reading race was ot mul different from when the entire word to the right of fixation was visible. This result inliesies ia word informacion 1s utilized during reading and that an individual word may cone fixation. These experinents also indicated that letter information Seas: ning letter spaces from fixation (see uso Underwood & McConkie, 1985). The nsovings-window webnique demonstrates that information beyond 14-15 letter spaces to the Fight of figution is of heale use inv normal reading. One posible reason for this is that the reader is busy exnugh proceving the information shat i closer co fixation, so chat he or she has bite use for any ore thlornation, One variation of the moving-window technique, the moving-misk technique (Rayner & Bertera, 1979), demonstrates that parafoveal information is of litle value even when you. need to have it because faveal information is not available. The moving-unask eechriique is the inverse oF the moving-winlow technique: the normal text is displayed outside of the center of vision and a visual niask moves ia synchrony with the eyes, making it impossible for the render to obtaiu uscfl information foveally (Rayner & Bestera, 1979; Rayner ct al., 1981). Thus, foveal vision is completely masked (see ‘Table 4.5), and an artificial scotoma of the retina is created, (There are people who have such scotomas due cicher to retinal damage oF to braint damage and they have gzeat difficulty reading.) Rayner and Bertera (1979) found chat when foveal vision (i.¢., che central seven letters around the fixation poing) was masked, rencing was sll possible from parafoveal vision but ata rate of only 12 words per minnze (ve also Fine & Rubin, 1999; Rayner et al, 1981). When foveal vision and part of parafoveal vision (i.e., the central 11+17 letter spaces around the fixation point) were masked, reading was akviost impossible. Readers in the experiments knew that there were words (or ac least knew there wore strings of leuers) outside of the center of vision, bue could not tell what chey were, They were more likely co be able to identify short function words like the, ad,and 4, particularly when they were at che beginning or ead of the line, The errors chat readers made when foveal and parafovesl vision were masked indicated that they were obtaining information about the beginning ietters (and sometimes ending letters} of words ia parafoveal vision, as well as letter shapes and word lengdh information, and eryiag to consinuct coherent sentences out of the information available, For exansple, the senuence The pretty brareler attracted suich attention was. cead as The priest broghr much anmustion and The banuer waped above the sane monnnent as Tie banker watched te snoa mostiai. There was aiso uo indication that the gist of she sentesces was compre ended if the words were not identified. Let us briefly recap what we know about the span of pezception during reading so fr, We row that its hindted,and the limitation en the right side appears to be chiefly due to Timtations in perception: even when foveal information is eliminated, renders still extract litthe useful info tion about letters and words beyond abont 14-15 letter spaces. On the left side, information is extracted front a smulller arey, genceally including at mose she word currently fixated on {although see Binder, Pollatsek, & Rayner, 1999, for a discussion of situations in which readers may attend to words to the left of fixation). Furthermo: window Ine pues prone th a research using the gaze-contingent movi TABLE 4,5 An example of a moving mask of seven letters An @xKXXMIRK was conducted in the lab. An axpordXXXXX¥X conducted in the lab. AN expesimenXxXxxwXndueted in the lab. ‘The astersk marks dhe location of fisation on ehree succersive Gxations.‘The work of the eyes. 109 paradigns (Pollatsek, Raney, LaGase, & Rayner, 1993) demonsteated that readers do nor obtain useful information below the custently fixaced line. Thus readers appear to limit theie attention to words on che line currently being read. Pollatsek et sl, also found that if the task was changed to a visual search task (wheze participants had to find specific target words in. paragcaphs), hen participants ean obtain useful information below the curently fixated line. Reading sill aso influences the size of the perceptaal span, since beginning, readers (Huikid, Bertram, Mydnii, & Niemi, 200% Rayner, 1986) and dyslexic readers (Rayner, Murphy, Henderson, & Pollatsek, 1989) have smaller spans than more skilled readers. Presumably, difficulty encoding the fixated word leads to sumaller spans for both beginning and dyslexic readers, Older renders read more slowly than younger college-age readers (Lauibrock, Kliegl, & Engbers, 2006; Rayner, Reichle ee al., 2006) and cheir perceptual span seems ¢o be slightly sinaller and less asymmetrical chan younger readers (Rayner, Castellano, & Yang, 2009a}, The perceptual span in other writing systems Before going on to explore what information is extracted trom the right of fixation, Yet es briefly discuss what is knows about the perceptual spar in other orthographies. Within the alphabetic writing system, moving-window experiments have been done in Dutch (DeaBuurmou et al, 1981), Finaish (Hiiki6 et al.,2009),and French (O'Regan, 1980), wish identical results to those in English. Moving-window experiments have also been conducted wich Chinese, Japanese, and Hebrew readers. Net only does the weiting system affect eye movement characteristics, it also infinences the size of the percepeual span. As noted in Chapter 2, the weiting system that is the most different from English is Chinese, Researci using the moving-window cechiique lus denionsteated tha for readers of Chinese (which is now typically read from loft to vight in mainland China), the percep teal span excends from one character to the left of fixation to two to thos chotaetecs to the right (Chen & Tang, 1998; Inbo& Liv, 1998). Obviously eis is much smaller chan the petcepewal span in English if one uses ce character as the unit. However, as we noted eavFier, Chinese characters are not the same as English lestets,and when one considers thar mast Chinese words are made up of two characters (although there are also many one-charicter words, and some three- co four- character words), then the arnount of information being proceswed on cach fixation, when measured in terms of words, is probably not that ditferenc from English. Ikeda and Saida (1978) used the movinig-window technique to stady Japanese readers (Remember, Japanese is a hybrid language consisting of morphemic characters, Kanji, and syllabic characters, Kana.) They found dhat the perceptual span extended sbout six eharacters to the right of fixation. Thus, for die Japanese writing system, the percepaual spun is considerably smaller hau. for English if ene equates a Japanese character wich a letter. The percepnmal span is even shorter when ideograpbic charicters are used. Osaka (1987) used the moving-window teehnige and found that the perceptual span for Japanese readers is smaller when the text consists primarily of logographic Kanji characters than when it consists primarily of syllabic Kana characters, However, Japanese text is considerably more dense than English, leading to the observation that more infor~ mation is processed per fixation. Thus, while it is bad to compare actoss languages (since the perceptual span in English seems to be defined mainly in terms of ewes, it appears that die perceptual span is roughly ewo co cites words (dhe fixated word and generally two words to che right of fixation) in the writing systems thar have heen examined. With Hebrew text, Pollatsek et al. (1981) found chat, for native Israeli ceaders reading Hebrew, their perceptual span was asynimetric to the left of Sxation, whereas when these sume readers read English, their perceptual span was asymmetric to the right of fixation. Thus the asymmetry of the110. Siilled reading af text window is not“hard-wired.” Not only does it vary fromm language vo language, but bilingual readers van after dhe area from which they excract information when they switch fom language to Je The major difference between Hebrew and English, of course, is chac Hebrew is read tion right to Teli, ‘That means thas the dominant patcern of eye movernents is the opposite in the langatages. An important implication is that readers concentrate their attention on the material lat 3s in the direction where they are about to move their eyes, The perceptual span in Braille As long as we are talking about other writing systems, ic might be of some interest to digress for a moment to discays what is known about how tactual information is “read” by the blind. The most common systent for alphabetic languages is known as Brille (after its inventor's name). In Braille 33> 2 mutrix of raised dots reprevents a letter; dots thus can potentially appear in any one of six locations, and che peters of presene and absent dots defines the letter. The arrangement of the letters js from left to right as on a pristed page with spaces between the words. For many Brsille veaders the size of the perceptual spat is one leer (Bertelson, Mousty & D'Alimonte, 1985b). They read with one finger, one Jeter at a time. Braille readers also typically don’t skip words, and maintain physical contact with the page even on return sweeps (although they move more quickly ‘on the sweeps than when they read a line of text). Some Braille readers ue the right index finger to read and die Jelt index finger mainly asa place marker to help them find ehe appropriate line on the return sweep (Bertelson et al., 1985b; Mousey & Bertelson, 1985). This enables eliemn to increase thelr reading speed by zhnost 30% over when they read with only one finger. Other Braille readers (the most skilled), however, appear to use both index fingers to extract information. Some will keep their two index fingers adjacent to cack other (on adjacent letters) while they read the entire text. However, a more typical pattern is to move the two in synchrony on adjacent letters in the middle of a line, but then continue moving the right index finger to the end of the line white moving the lefé one co the beginning of the next fine and seart reading the next line at the same time that they are reading the end of the previous one (Bertelson ct al., 1985b). The right index finger usually rejoins the Jeff on the next Tine after a word or two is read by tue left. The pescepuust span of these most skilled Braille readers thus appears to be two letters, ac least some of the timng, since they can read mote dan 30%6 faster with ewo fingers than with one, However, the details of what is happening are somewbar unclear, Since using the lefe index finger as a place marker provides appreciable benefit in itself, itis hand to know exactly how nich benefit is actually a resule of extracting information from both fingers simultaneously Using this nwo-band method, however, che best Brille readers can read about 100-140 words per minute (Bertehon et al., 1985b). What is a reader doing on a fixation? ‘Weare closing in a bit on what information the reader is extracting on a fixation. The information to the left of the fixated word in English (or to the right in Hebrew) seems to be largely ierelevane because the reader is not attending to it. The moving-window and mask experiments suggest that information further than about 14-15 eharacter spaces to the cight of fixation is no: used because of acuity and capacity limitations in processing text. However, we still need tv discuss how the formation from the fixation poine tw the right-hand boundiry of the perceptual span is used Rayner et als (1982) experiment cited earlier makes it clear thar more than the fixated word is procested, When the window only included the fixated word, participants read at only about 200 words per minute in contrast to about 330 words per minute when there wat no window TheThe work of the eyes 111 simplest conceptual model to handle that fact would be that readers make sure co encode the fixated word on eich fixation but shat on some fixations they may also encode another word or two. However, other data from Rayner et al. (1982) indicated that reality is more complex. Since readers were not particularly bethered by incomplete wosds—in Gt, che major variable affecting reading speed was the nomiber of lewexs available to the tight of fixation—readers nse be doing, something more complex than extracting words as visual units Gee alse Chapters 3 and 5). One possibility is that words are encoded only a limited distmce fiom fixation, but thie lester information is extiacted farther out, This conclusion emerged from a study by Rayner (1975b) which used another wype of gaze-contingent teclinique (dhe boundary technique), We will discuss dhis sandy in some dewail because the technique it introduced has been used rather extensively since the original scudy co examine how far away Gomi fixation different types of infor mation are obtained and how information is combined actoss saccades (for reviews see Rayner, 1998, 2100). In this technique che experimenter attempts to determine what kinds of information ate acquited fom 2 particular word focation in a paragraph (called the critical word location or CWL) when readers fixate different distances fiom it, This is accomplished by changing the contents of the CWL when a saccade crosses an invisible boundary location. The logic of the method is chat ifa certain aspect of che stimnulus in the CWL has been encoded in the parafovea and then changed when the word is fixated, some disruption of normal reading would be expected. {In particular, we might anticipate a longer-than-normal fixation afer the change had been made, since the reader would have #0 resolve the conélict in the information obtained from the two fixa- tions. The advantage of the boundary technique over the moving-window technique is that more precise connol over parafoveal information is possible since one word is selected for manipulation. Tn addition, since the region ofabnorinal text is suuall, nonual reading is even more closely approx imaced, in Rayner’ experiment the stiulas in che critical word location when it was fixated (the target word) was atways a word that fit into the text. However, the stinulus in the CWE before the boundary had been crossed was sometimes 2 word and sometimes a nonword (ee Figure 4.5 0 get an idea of the possibilities). Rayner was able to observe a large number of instances on which che reader’ eyes fixated different distances to the left of the CWL on the fixation prior to the stimulus change and then directly on the CWL after the change. It was assuined that ifthe reader's fixation was sufficiently far to the left of the CWL no infurnation would be acquired fiom chat region. [f this were the ‘ase, the reader would then fail co notice any of the different types of display changes. (We use the L The old captain put the chovt on the... 1 8B M. The old captain put the chart on the... 8 2 Key: Location of the boundary which triggers @ change in the display. ‘1—Location of the last fixation prior to crossing the boundary. 2ALocation of the first fixation after crossing the boundary. Alternatives in target location for base word chart: cchart—identical word (W'-Ident) chest-—avord with similar shape and letters QW-SL) ebovt—nonword with similar shape (N-S) chovt—nonword with similar shape ane letters (N-SL) chyft_nonword with similar letters (NAL) FIGURE 45 Boundary study: Au example of the type of disphy change thet occurred in Rayner’s boundary experinsent112. Skilled reading of text words notice and. detect to mean that there is some effect on the eye behavior of the tender, rather chau that he readey is coasciously awate of these changes; in face, in such experiments readers are Lypically not aware of de changes) Qn the odher hand, ifthe fixation were closer to the CWL the eae sight eltain sume information, perhaps word-shope or letter information or even meaning, vl aP cic ainuulus change eased a change in that type of infonnation 2 longer fixation would veal However, aC the stimalas change were of a type which did not cause a change in the type of acquired, no change would be detected, and no disruption of reading ini neativan ne rea wont cu Since some of the initially displayed stimuli in the CWLys were nonwords, this raises an inter cooing quewion: [low near to the CWL did the nader’ eyes have to be before the nomworduess of we COWL atfected reading? One way to examine this was to examine the fixa- c chinge, grouped! according to how far they were from the CWL, and calonlate the average fixation duration at cach distince, Rayner found dae the existence of @ nonword in the CWL did not affect the fixation duration unless die CWL was no more chan three leer positions to the right of the fixation point. Ifthe CWL began four of more letter positions to the right of the fixation point, the worduess of its cemporary occupant had no effect on dhe length of this fixation. The durations of fixations on the CWL immediately after the display change were also exam- ined, chsified according (1) the type of display change that had occurred and (2} the Jocarion of the previous fixation. These data are shown in Figuie 4.6. Reading was unaffected by any stimulus change if che fixation prior « crossing the boundary was mere than 12 letter spaces to ete left of the CWL. If the previous fixation was 7 to 12 fetter spaces to the left of the CWL, readers did pick up infornmation about the shape of the word or its component letters and information about the identity of che extreme lerters of the stimulus in the CWL, since if eititer of these chinged when the boundary was crossed, the fixation on the earge: word was increased. In conenasc, if the initially displayed stimulus bad the same word shape and the extreme Iecters were the same 38 the target word, very lire disruption wos noted. Finally, she fixation on die target word was affected by che wordness of the preview when the preview was as much as six characters away from Ure Heuer sein, in nv prior v0 the displ 500 Pe E 300 = we . § — q 350 5 ' B 250 § 300 3 i Z © aso 5 200 x0 1816 15-13 12-10 9-7 6-4 ST 0-2 WSA3 W210 97 GH 21 one pr aton Ragone ol he CL rag Peenacce won (om ot dle pation tte (ne. of character positions) meet FIGURE 46 Dora from Rayner’s study. 'mel on the loft shows fixation tinte of last fixation prior co erossiag tle boundaryspanel on the right shows grze duration on target Word after crossing the boundary In our eximple, the base word is chart and previews of chart, chest, ebay, chant, aud chyft represent the Weldent, W-SL, N-S. N-SL, and N-L, conditions, respectivelyBe 124. Skilled reading of text they don't have a valid preview of the word. Two additional points are imporeant to sone ‘willl respect (© preview benefit, First, Henderson and Ferreita (1990) used the boundary tw nie anl inanipolsted foveal processing difficulty white also vacying the avaihbiley of parafoveal nation. ‘Shey foul thac difficule foveal procevsing led to the reader obzaining 10 een Jneview fivny the word to the right of fixation, These results suggest eaat the ainount o infer (ow procesved ot a fixation is somewhat variable and can be influenced by factors an Jevnah of the curently fisted word and the difficulty associated with processing the fixal co (ve ako Kennison & Cllion, 1995: Rayner, 1986; White, Rayner, & Liversedge, 20051). Seon readers typically obtain preview benctit from the word thee they are going «© mere ot (McDonald, 2006), Thus, while they generally obtain preview benefic from word n+1, th a rash 8 obtsin preview benetic fron word 142 (Angele & Rayner, 201; Angele, Slattery, Wand aie Royner, 20018; Klicgl, Risse, & Laubrock, 2007; Rayner, Juhise, & Brown, 207 Iter sandy with Chinese indicated that Chinese readers ste more likely ta obrain preview ben nes tha word n+? than ore English eeaders (Yang, Wang, Xu, & Rayner, 2000), Furthermore itappe ube naders of English (see below), Chinese readers obtain semantic preview benefit fom we not yet fixated (Yan, Richtes, Shu, 8 Klisgl, 2009; Yang, Wang, Tong, & Rayner, 2011)- effzets probably have to do with the fact that word nt ar fixation point in Chinese dhan English, : 0 \d word n-t2 are typically loser & Word skipping , ord ‘Another index of how fo co the tight of fixation words can be identified is word skipping. W Jength has the largest influence on word skipping: short words ate reliably skipped mor), longer words (Brysbacce, Deieghe, 8 Vira, 2003; Rayner, Slattery, Drieghe, & Liversedge, ‘Thus three-letter words are skipped about 67% of the tine, are fixated most of the tinte (Rayner & McConkie, 1976), Si left of the fixated word is largely ignored by the reader, ilentified before its skipped ot the seader simply has having seen it. Since word dipping is a ubiquitous pat ‘ms oF the word to the right of fixation is reasonably common if guessing does not account fo ofthe skipping. At times words can be skipped from reasonably long distances. In a later bow en experiment by Balora, Pollatele and Rayner (1985; see also Dri she, : ov! Pollatsek, Rayner, & Balota, 1986), it was found that che CWL was occasionally skipped (ald ” tes: thaa 1% ofthe time) when the priot fasion sas greater an nine character spaces fora beginning ofthe CWL. Thus it appeats chat the Treaning ofa word in the parafavea can sometin” be extracted faint for fiom fixation, although this is noe usually the ease even with highly P' able words. In the [alota ot al. (5985) experiment the ta setttsnce comtext. Skipping occurred much less ince, as i ifs word s skipped it ether must have made a guess as to what the word is speation ut of the eye movement record, ident! rio eget word was highly predictable from the p ‘ Gequently when a word other than that pre ference suggests that variables such as the predictably fixation words can be encoded and meaning extract ity in more detail in the nexe chapter. err In the previous two sections, much ofour discusion wee deen by soune well-knowa e#P27 ‘ments that were first reported nezcly 40 years ago. We focused on them because in my hae they forged new ground in our understanding of skilled reading, Fucthecmore, the results will discuss effects of word predictabilThe work of the eyes 113 fixation. Thus this measure appears to be a more sensitive measure of whether lexical infocmation was extracted than the durarion of the prier fixation. Rayner’s results were originally interpreted as evidence that word shape information is obtained fiom parafoveal word that the reader cannot identify. However, mbsequent research (to be discussed soon) has demonstrated that when word shape effects emerge, it is really because words thar begin with the same Jesters and share the same overall shape fs in Rayner’ study) have many letter features in common. Rayner’s results also suggest that the meanings of words to the right of fixa~ tion are not extracted very far from the point of fixation, since the reader appears to be unaware that a nonword was present ifit started further than theee to six letter spaces front it. This conclu- sion is reinforced by a study (MeConkie & Hogaboam, 1985) in which participants were reading silently with their eye movements monitored, At certain places in the text the serzen went blank and participants were asked to report the hast word that they had read. There is a guessing problem here, since the participant nay be able w figure oc a word not actually seen on the basis of prior context. Nonetheless, the results are consistent with the boundary studies reviewed in this section, McConkie and Hogaboam found (ee Figure 4.7) that the word readers reported most frequently. vwas the word that they had lase xired on. although the word co the right of fixation was some= times reported. However, words to the left of che fixated word or ewo or more to the right of the fixated word were rarely reported. More on preview benefit “The experiment by Rayner was the fire to examine the nature of preview benefit in reading: when readers have a valid preview of a word prior to fixating it, they spene Jess time on it chan when 187 No Mask: Mask: x During Fixation During Fixation & ; i rk iv e xx xX KX XK XK K x KX XK XX RR K KX KX 34 orl2345 S43 270123 45 107 Ho Mak vik x During Saccade During saccade XX ry x x x § x x x x f Bad y ERR “ KX KX X KM XK KX KX 5 4-3+-2-10 123 4 5 54-32-1012 3 4 5 FIGURE 4.7 Frequency distribution of the location of the last read word with respect to the location of the hast fixation on which text was prevent. 0 represents the lasc word fixared and 1 represents the word to ies right, In the Mask conditions (right pinels),a mask cane on when the text went off, while in the No Mask condition (left panel). the texe just went blank. Distance is measured in word units, without regard fir word JengthThe work of theeyes 115 ay soa tee test of time? the basic MeConkie and Rayner studies with the moving, window % bet replated numerous tines ad for several alphabetic writing systems sk omni, he percept sun hited, extending fom the beginning of che curently Fate ond to about 18 cars se the right of fixaon, The area withit wick word ifn es plas even me ee 4 Teeaders can sometimes identify the word to che Tekh fared word (and sommes my identify evo words to he right of fixation, particu when the fixated word and the next WO ate J. Tn fact, a8 we mensioned cackies rae short words ae 1s often do not fixate either the frst ot ast word ofa line in text Apparently, the last word of © is often fully processed in the parafovea, It is som od een ‘ewhat harder to understand why the first is oye line is sometimes not fizaced. One possible explanation is dat the fies fixation om a line reemeached bya fivard) rear see fa reader's perceptual span mirrors the direction of eye rane (5 with the verack renders discussed eater, it could be dhat covert San U shifee thove neat che frst fixation so that che span anclades the word r° the left of the fixated word on aceasions, tow teaders can identify words that they do not fixate, for most content word often no word yond the fixated woud is filly identified (cf, Klicgl, 2007: Kliegl, Nuthmana, & Enghert, 2006), meas ite seen thar preserving some Fetters it al word aids reading ie appears that tea formation about a word can be encoded or id veal to aid identification of ino the subsequent fiation. We now cm of whac we know about how lation is integrated across fixations, a parafove wn one Fixation ant to a discussion Int " fegration of information across eye movements the last section suggested tha xe Ire rs sneneding xabion, Apo ndican tee words sebtoeesed on mose than one Bxatin i the face that ce perceptual span ie abou double the fines ofa saccade for oilers of English. This comparison # HOt completely fir, however, an pe Peteeptal span is nor an aves ging the maxima distance that information teeny eed. Nonetheless, the discrepancy between the perceptual span and the size of the age saccade chy foces the conclusion thatthe eyes are maviT FO 0. area of text that has been Pressed to some extent. | | eegateton of infermation 268 saccades is by wo means 9 conscious proces Sos We A TY notaware of our eye smoversents. Each eye movement changes the pattem of light on che sand ge ofthe wonk we ae ooking 2% We never have "ror 30 followed by a break fn impute s giseussed excice explains why you Gerailed understanding of why the S nel converging pieces of data fiom. 1 some words are processed lly on one fixation and chen finishes tne and ete pore secon OF ne to the of having stinmuls input for 2 quareer of 8 secon | ont see Sak ta “The research on noc surresion . i cl erween the slides. However, we Tabs between Grations esate Gopechow the besin is able €O smooth out the discrete POs from each eye fiation and creat a feeling of 8 sominuous coherent perceptual world, in ‘ inforrantion geoue «words obtained on exo succes fan, the Fst when de words Br the osafovea and che second wien ii i rere aad ite intepeion process i wef Yeading, de porafovesi preview of the word shold Tacibate lave foveal processing of the word, © shall das discuss integration of information 30% ations largely in terms of such Giitation 96) that parafoveal previews BLE Jter identifica thas by fg esa own since early as Dodge ( afore Pr on arom acca + However that facilitation, in itself, is not necessarily evidence for gra saccades, id to document inte~ Since ¢ [eee woed may have bee fly ieetied se proves, What ete to domes 'on across saccades is eo make the paralo foveal target stimuli similar but not idens weal peeview and * ‘Mical, and to determine whether there still Gacthieation from the preview: Bi cre IEEE116 Skiled reading of text An experimental teeluique requiring participants co name isolated words (originateca by Hayter, IWAh) has produced a [ot of information about integration across fixations (Balow, 8 Payers, 1005; MeClelland & O'Regan, 1981; Rayner, MeConkie, & Ehrlich, 1978; Ras-per, Me Cambie, Re Zooks, 18). Iisa i tion of the boundary technique. Participants are asced con a conval fixation point, and vées a letzer string appears in parafoveal vision, they are (eit, During the saccade the initially displayed stimoius is rephceet ine as ist as posible, ‘The parafoveal stimulus is thus visible for ajye toh to make wy eye mowesne sr asord which they are te Jy 201 mn tions when the paraleweal Lecter stein appears aml the eye mover eye Sham extended sone, participants are almost me Ser a appro begins. Im spite of the fice that it ic visible for ware of the identity of dhe parsfavea! word ond ore rurely evon aware that there has beers a vy change! Thos they hove no trouble deciding which word to name. ° Figure 4.8 shows the basie pattern of retulee from che experiments. Ifthe stimuli presented &_) 9 w+ are identical, chore is facltation in naming the targer word (compare. 4 fixation » and fix ¢@ when a row of aterisks or unrelated letcers ate initially presensed parefoweally). More impor ea aye is the fact chae facilitation occurs even if the parafoveat preview only has some lesters in comaics,, With the target word. Av one would prediet fromm the perceptual span experinents, the amour exp siulus occurs. That i, there is mor, ficilitation depends on how far into che parafovea the ficilitation when the initial scimmulus is 1 degree fiom fixation chan when ie ts a€ 3 degrees ane. hardly any facilitation at 5 degrees (.c., 15 letter spaces) from fixation. Thus the results indicstet that participants can use parcial parafoveal information abouta word co aid their recognition of daa ¢ word when itis Iator fixated, but only when the preview is less chart 13 letter spaces from fixatieyny some, ‘The fact that chest in the parafovea facilitates the later identification of chart implies that toformation excracted from chest was useful in the lacer identification of chart. Indeed, itis proeey- 500, 450 Naming time (ns) 400, Visual anghe FIGURE 48 Mean naming tine 95 4 Finetion of intially displayed alternative and visual angle, it our example, tive base word is art and previews of chart, des, chou, chovt, and cyfi represent the Weldent, W-SL, N-5, N-SL, and N-L conditions, respectively. The asterisk preview: was 2 row of asterisksThe work of the eyes 117 lear thae chest and chart are orchographicaily simiiar, and this nuere fact does provide some facilita- tion (Williams, Peres, Pollatsek, & Rayner, 2006). However, the deeper issue is: Why does the fact that ewo words are orthographically similar yield preview benefit? In the rescareh literature five ned in detail: (1) some of the visual fates of chest porential sources of facilitation have been exam are stored and aid later identification of chars (2) some aspect of the meaning of diest has been encoded by che parafoveal preview of chart which facilitates later identification (although that seems improbable in this particulac example); (3) some sort of sound codes (e.g., phoncmes or syllables) activated by cfest (pethaps the initial /cl/) aid later identification of the word dius (4) some of che letters are identified and these abstzact lester identities (not the Jeter forms) are whatare facilitating; and (6) the lexical entry of chart is partially activased by the parafoveal preview of chest which aids in the later identification of cfurt. The distinction we wish co draw berween 1 and 4 is that the jaformation about letters is in a visual form in 1, but 2 more abstract form in. 4. The distinction we wish wo draw between 2 and 8 is that in 2, some aspect of the word's meaning a8 well a5 its identity ix activaced in the parafovea, Of conrse, these hypotheses are not mutwally exclusive, as facilitation could have arore than one source. As we will see the evidence points at present to a relatively simple answer: there is no evidence for either of the first ewo mechanisms being operstive,so that, by a proces: of elimination, 3, 4, and 5 appear to be the sources of integra- tion. We will discuss the evidence against the first two in turn, 4s visual information integrated across eye movements? ‘The idea that detailed visual information is integrated across eye movements may seem the ‘most plansible, since it vorresponds t0 ove intuitions that we see a singke seamless word witen viswal informacion fom two fixations is brought rogether into a single representation of the visual world (McConkie & Rayner, 1976b). Thac i, readers may obtain gross feaeural information from parafoveal vision during a fixation and store it in a temporary visual buffer, an integrative visual butler. The visnal inforniation stored in che baffer would then be used as » base to which new information is added when che region (previously in parafoveal vision) is fixated. The aligga~ ment of tie information from the two fications would presumably be based on (a) knowledge about how far the eyes moved and (b) the commonality of the peteerns from the two fixations, Of coune, all this computation would generally be unconscious, since we are usually not aware of moving our eyes. The integrative visual buffer in reading can be thonght of as being lke iconic memory (see Chapter’ 1), excepe chat information is preserved in the visual buffer across eye movenntents. While this view of information integration is pethaps the most intuitively appealing of the alternatives, the evidence against it is quite strong. Ficst, Rayner et al. (1978) showed that proper alignment was not necessary in order te obtain the results shown in Figure 4.8. Recall thae align= ment or justification of two suecessive images in the buffer should be based on keeping erack of how far the eyes moved, However, Rayner et al. found that the same partern of results was obteined in an experiment in which she scinwutus pattern rather than the pacticipants' eyes moved, That is, appeared in parafoveal vision and, after 2 period of time approxi an initially presenzed scinnuhy mating the sum of che saccadic latency and saccade duration (about 200 ms), the target word tobe ye che named appeared foveally and the parafoveal stimulus simultaneously disappeared. Natice that sequence of events on the retina is the same as when the eyes move to a parafoveal stimulus: an initial stimulus impinges on che parafoveal rerina followed by a stimulus in foveal vision, fn one condition, however, an eye movement intervenes between two retinal events while in the other condition the eye movement is sinnulated by moving the stinalus rather than the eyes, Ifkeeping: track of how far the eyes move is iraportant for the integration process, performance should be‘VIB. Skilled reading of text much worse in the no eye movement condition than in the standard eye movement condition. However, ayer et al found ne unajor differences between these two conditions. More dainagingg to the integrutive visual buffer notion were ewe experiments which dizectly tested whether visaal features could be integrated, The first demonstented that changes in the visual foren of the information hud no effect if ee mscaning was net altered (Iayner, MeConkie, & Zola, Hany. Rayer ot ail, found choc changing, the case of a word between the preview and target (eg. CASH chanyped to sass} had no elfect on how long it took ¢© pame the word, even though ere vere cll clear fvviliiningg elects (rom parafoveal previews. The second ested ime-gration of visual infinite ina sbifloven! way. O"Regan ancl Levy-Schoen (1983) prevented half of the features of | excl leer off a wre) oy one fixition ane! dhe other half on the subsequent fixation, (Both stimuli were in lhe i Frticipamts in this condition were rarely ever able to identify the target worl, fie the two halves were presented in the sime spatial locaton one after the other in quick succession, participants readily identified the marge word. Thus the visual infor mation that can be integrated within a fixation cannot be integrated when a siceade incervenes, At this point you may well be saying to yourself that all of the experhnents we have described in this section do nor reilly involve participants in the task of reading, Perhaps, as we have pointed ‘on before, dhe tasks used in these experiments encourage a strategy that is different from what normally happens when we read, However, it turns out to be the case that a number of expexi- mens in which participants are aetwally reading (see below) yield results consistent wich the couclusions we have reached from the experinents described up to this poi ‘The question of whether integration is dependent on keeping track of how far the eyes move bas been cested in the reading situation as well, In shese experiments (McConkie, Zola, & Wolverton, 1980; O'Regan, 1981) participants were reading text, and at selected points the entire line of text was shifted to the left or right during the saccade. In che gormal state of perception, the distance that the image has moved fiom fixation to fixation is determined by the distonce that the eyes moved, If the alignment of the visual information obtained on ewo successive Aixations is dependent on this caleulation oF how far the cycs have moved, chen considerable disuption should be produced when the text is shifted. Even if itis shifted only a few characters there should be manive disruption, since he Ietter information ix the two images will conflict in all locations. Since small corrective saccades sometintes cocurced after che shift the shift was possibly somerintes registered in the brain (if noc in consciousness). However, chese eye movements could have been because the eyes landed on a position other than intended rather than the shift actually being registered in the brain, Nonetheless, shitting the text qwo or three character poxitions resulted in sto conscious awareness of the shift and produced negligible effects on reading speed and compre- hrevsion Similarly, the issue of whether integration occurs by integrating the visual forms on two susces- sive Fxations was tested by MeConkie and Zola (1979). They had participants read passiges printed in AITeRaAtlaG cAsE,and changed the case of every letter during certain saccades, 90 that succes sive visual images would not be similar. Thus cel? oa fixation u would appear as CaSe on fixation 1 and cdsfon w!2, These changes were not noticed by readers and they had viruaally no effect ‘on comprehension or on reading speed (for a snore recent demonstration sev Slatery, Angele, & Raynes, 2011). In addition, the basic finding that parcial infornsasion facilieates naming of the Gxoted word (og, chest ficilicates naming of chart) parallels the finding described in the previous section {Rayner ot al, 1982) chat reading, was faster when the first two to three letters of the word to the right of fixation were visible than when they were altered. Io sunuuary, the basic findings that emerged ffom the parafoveal naming experiments have been cortoborated in experiments involving silent reading, The two experimental situations «lius provide convergent validity for the conclusions, combining the ecological validity of the reading‘The work of the eyes. 119 situation with the more tightly controlled naming experiments iat which the response is eranspar— antly tied to word identification. Partial encoding of meaning “There ate two different ways in which one might think the reader extracts partial meaning from a word. The frst is that the whole word is processed, Lut only dimly: That is, the aetivacion from the physical stimulation does net lead to identification of the word, but may lead to a vague idea of the meaning of the word. Perhaps a semantic feature is activated. The second way is that a specific mcan- ingfal segment of the word,a morpheme,is identified, We consider cach of chese possibilities in cam, Semantic preprocessing ‘As.we look around the world we feel we have a vigue idea of what things seen only in the para foves and periphery are, For example if we are not directly looking at a dog we may be aware that it is on animal, and have a vague idea of its size, but may noc be able to make a precise identification of je. Moreover, there is evidence in picture perception that there are possibly unconscious influ ences of such partial meaning om processing. For example, in picture pesception, the eyes quickly nove se regions judged co be informative (Antes, 1974; Mackeworth & Morandi, 1967) or seman~ sically anomalous (Becker, Pashier, & Lubin, 2007; Loftus & Mackworch, 1978; Rayner, Castethano, & Yang, 20092; cf., Henderson, Weeks, & Hollingworth, (999). These phenomena suggest that something similar may be going on in reading, However, itis important to point one that there are rather substantial differences in the stintalus pattern berween text and a picnure (Loftus, 1983; MeConkie & Rayner, 1976b). With text the pactern is rather homogerious, made up of leters and spaces, and it is likely that lateral masking of swords and Letters (by adjacent words and letters) is mach greater in text, A single distinctive and informative featare of an object i a piewure may convey meaning in a way that no single visual Reawne of 2 word does. It may well be these distinctive features that allow for rough semantic classifications of objects and guide che movement of the eye in piczure perception. Another reason that semantic preprocessing seeurs like an attractive explanation for parafoveat preview effects is because of the unconscious priming experiments described in Chapter 3 (@-g Allport, 1977; Balota, 1983; Marcel, 1983). lin these experiments words are briefly presented followed by masks in che foven, If conditions are set up right, she patticipanc will be unable to identify the word, but the speed in identifying 4 semantically relased word thac follows will be increased. Marcel (1978) suggested on the basis of the foveal priming studies that meaning is sismat- rancously available fron a number of places on 2 page. For exarnple, Marce] noted chat if you eurn the page of a book and are reading the top ue, something at the boeron of the page may catch your eye. He fiueber argued that this is ealy possible if its meaning has been analyzed indepen- dently of where atcention is. A key assumption in his inference is that a brief foveal presentation of a word is analogous to » word in parafoveal vision during reading. The analogy may be misleading, Although a briefiy presented foveal word and a parafoves! word are both visually degraded, they are degraded in differeat ways. Brief foveal words are degraded by their duration and by backward masking: parnfoveal words in eext are degraded by acuity and lateral masking, in reading normal rext these scuity and lateral masking considerations make it difficuls to identify words at increasing distances from the fixation point. The phenom- enon of foveal masking is still poorly understood: however, it appears that there is some sense in which the stimulus is fully identified, but something about the mask dissociates it from awareness and direct avcess. On the other band, it seems implausible that partial semantic aecess cam occur120 Skilled reacting of text fiom wygue information about a word such as global shape, enge,or knowing a letter or wo. One posible exphnaion far the phenomenon that Marcel described—something a¢ the bottorn of the Irigy catching your eye when you turn che page—-is chat When you begin to move your eyes to Ining, den to the top oF the page you may make a shore fixation near the bottom of the page. “That this phenomenon nay be explained by something similar to che foveal masking experiments cr thio by sernutie preprocessing in she parafovea or periphery, Ju reading, there is wo clear evidence supporting semantic preprocessing. One attempt to demon~ Meme seonuitic prepiocessing, wes a variant of the semantic priming technique described in the previous chapteh A semantically ambiguous word such as ban: is presented in the fovea and one of two words that coukl disambiguate che word, sie or money. is presented in the parafovea, Both words ate presented briefly and the participant is ested on which meaniag he or she associates with che foveal wore, dum. 1° participants perform at abuve chance levels ia choosing the ineaning suggested by the parafoveal word, then ic iumplies that the meaning of the parafoveal word has been processed. Infact, participants perform above ehance. However, we already know chat parafoveal words can be identified from outr previous discussion of skipping. The key question is whether partial meaning can be processed. The way this has been examined (Bradshaw, L974; Inhoff, 1982; Inhoff & Rayner, 1980; Underwood, 1980), 1981), is to test both for which sense of bank the participant selects and. to test whether the parafoveal word has been identified. If participants can select the appropriate meaning at above chance levels, even when the parafoveal word has not beea consciously identi~ fied, one would have evidence thar senuantic preprocessing has taken place, Unforcunately the results from these experiments are not completely consistent. Some huve found above chance performance and others have uct. However, even in those that obtained above chance peefor- mace, it was net much above chance. [n addition, the experiments that obtained above chance performance are difficult to evaluate as certain factors (such as eye location, guessing, and readout feom iconie memory} were nor contolled. Iu sur, there is Title clear evidence for semantic Preprocessing from these experiments (see also Rayner & Morris, $992). Rayner, Balota, and Pollatsek (5986) provided a more direct test of semantic preprocessing in reading using the boundary technique described earlier, The scimulus that appeared ia the target location before the base word (song) was fixated wos either a visually similar nonwword (sep), seman Cically sssociated word (éwie), or an orthographically and semantically different control word (doo). While the visually similar preview facilitated processing of che bese word relative to the control condition {fixation time on the base word was reduced), there was no difference bervieen the condi- tions eniploying semantically related snd uncelated parafoveal previews. That i, there was no evidence for semantic priming. in these conditions. In contrast, the pairs of related works presented sequentially in the fovea produced the usual semantic priming effect when the second word was named, More recently, other areempes to determine if there is semantic preview benefit using che boundary para digi (Alteriba, Kanzbe, Pollatsek, & Rayner, 2001; Hydini & Huikid, 2005) have hisewise failed 60 fiod evidence in favor of such benefit in reading, Interestingly, while there is little evidence for semantic preview benefit in English, there is some evidence for semantic preview benefic in Chinese (Yan, Richter, Shu, & Kliegl, 2009:¥ang, Wang, Tong, & Rayner, 2011) and German (Hohenstein, Laubrock, & Klieg), 2010). Whether the latter studies ane due to charscteristics of the writing system rtemains 10 be determined: Chinese maps more directly to meaning than does Bnglish and in German the target words (all nouns) have a cnpital leeter at che beginning of che word. Identification of morphemes ‘We appear to be down to four posibilities for integration of infoemation across saccades in readings the entire Lexical entry is activated, sound codes are activated, meaningful subunit (a morphewte)The work of theeyes 121 is activated, or merely some of the leters are activated, Bofore discussing the involvement of morphemes we need to review more details of the parafover! preview experiments, Rayner, McConkie, & Zola (1980) demonstrated that significant facilitation was produced when the first vo of three Jetters were constant actoss the two fixationt (e.g, chest-ciarg). No facilitation was obtained when only che first letter was constant across fiasions nor wes there facilization when all leeters were the same except the fest letter (¢.g., board-hoar). Thus it appears that encoding the beginning letters of the word is crucial to obtaining parafoveal facilitation. Interestingly, Inhoft (1987) found chis was true when readers had to read from right ro left so that the parofoveal preview yas ¢o the left of fixation and thas the beginning letters were farthest from fixation, He also found that when practiced participants eead teat ftom right to left,a preview of che Legitming three letters of a six-leter word provided facilitation in ceading, O£ course, when reading from right to lef the beginning levers are furcher away from fixation so it is not just chat the beginning letters of the word to the right of fixarion are close (0 the current fotiom point; there is something important about those lerters. The pattern from moving-window expechnents in which only the first part of woud n+ was exposed also indicates that the information fiom the first two or three Fetters of a word provides much of the parafoveal benefit, pacticulsly if the remainder of dhe word consists of letters that are visually similar to the real letters of the word; if the remaining letters are not visually similar, readers do not read as well as when the entire word 4 is present (Inhot?, 19899; Lima &¢ Inhoff, 1985; Rayner et al., 1982). While the first ew letters of a parsfoveal word provide significans preview benefit, studies using transposed letters (e.g, swisching the letters st to 15} have indicated that veacers do get some useful information fron the letters at the ends of words, particularly shorter words (Johnson, Petea, & Rayner, 2007). Since information from the first two to three letters of a word appears to provide much of the preview benefit, che logical plice to look for extraction of a morpheme fiom a parafoveal word is at the beginning, Moreover, i ako suggests that it would help to Jook for relatively short morphemes. Lima (1987) hypothesized thae the begining letters may Excilitate, at least in part, because they aid in idencitying the initial morpheme of a word. She tested her hypothesis using prefixed words, since most of the prefixes in common use have from one to three letters and because prefixes form snaall set of highly familiar word-initial letter patterns. In particular she wanted ¢o determine whether there was any evidence thar"'prefix seripping” (see Chapter 3) could begin before a word is fixated. ‘Words with prefixes (auch as mistmst) were compared with pseudloprefixed words (ch as pisress)- The stitnali were matched on number of syllables, word lengrh, and ward frequency ania sentence frame was prepared into which cithey of the words would fit (The teenager’ abrupt/abstnd anstter in Lima’s experiments the boundary technique was used. Prior to the display chemge, the CWL contained the letters common to the evo words plus random letters or Xt (abusb of atecee) oF simply a string of rondo lewets or Xs (Lamsbf or zasaixa). Whicn the teader’s snecade crosied she boundary, the word abipt or absuef (depending on che condition) vas displayed at che CWL. Lima found that participants looked at the rorger word for less time when the initial letters of the target word were present than when they were noc. Site ako found, as mentioned in Chapter 3, that prefixed words were fixated for a shorter amount of time than psoudoprefised words, However, the benefit of dhe parafovea! preview was the same for prefixed and pseudepre~ fixed words, There are ewo possibilities for this equality. If one assumes chac prefix stripping 1s che first step of the only route to identification of both kinds of words (which then has ¢o be followed by a second access ity the case of pseudoprefixed words), chen prefix stripping in the parafovea is tenable: the parafoveal preview would start off che identical first stage of word identification (prefix stripping) in thie two cases, However, we argued in Chapter 3 that it is mote plausible to assume that access of pseudoprefixed words can go on ditectly as well as having co go through the false start of prefix steipping. If this is the cave one would expect greater parafoveal benefic for prefixed words,122. Skilled reading of text since acceys of them would be aided by identifying the initial morpheme as well as the first three fetiers, Since the parafoveal benefit did not differ berween prefixed and pseudoprefixed words, we have some evidence thac inorphemes are not extracted in the porafovea. Furthermore, in 3 replica~ “s study, Kainbe (2004) likewise obtained no evidence for morpholog- wal preview lemelitin kuglish, tncerestingly.in Hebrew (in which it can be argued that morphemes val roke it word processing). experiments using che boundary technique have 1° of inprphological preview benefit (Deutsch, Frost, Peleg, Pollatsek, & Rayner, 20005; Denisa h, Frost, Polltsek, & IRaytier, 2001, 2005), An experiment employing, compound words provided additional evidence against morphemic units an paralinveal infiorniation extraction in English, Inkof (1989b) employed six-lester compound swords such as cmdayy, As with Linn’ experiment, be employed preview conditions in which the whole word comfy, die first morpheme conse, oF no letter information axxxes:, was present in the pansfoven, InhofF employed ewe conerols: pseudocompound words such as carpet, where the Grse three letters were also 9 word but nora morphemic subunit, and monosyllabic words such as priest. He found die sume preview benefit in all dheey cases, indicating that neither the first morphente nor the first syllable was a significant anit in integration s¢ross saccades. Inhoff’s results appear to concradict those of Lima and Pellaseh (1983). Lima and Pollatsek found that a preview of the first morpheme speeded lexical decision more thin a preview of che beginning letters when they did not for a nierpleme, However, die morpheme preview in the Lima and Pollatsel experiment ‘was foveal and thus the integration was not across ewo fixations. tion and extension of Lin phy a more ce hts Sound codes Let us next consider the possibility that she reader is extracting some sound-based code from the parafoveal stimulus such as the initial phonemes or the first syllable of the word. This possibility seems particulirly sppealing since many of dhe studies thar we have described with incividval words required participants to name the word that is present on the second! fixation. Perhaps infor mation acquired from the parafoveal word permits the participant to begin to form the speech snusculiture properly for saying the word, This would reduce the time needed ¢o initiate am utter ance when the earget word occurs in the fovea following the eye movement. There is now good evidence that sound codes are used to integrate information across eye movements and are purr of the benefit readers obtain from preview information, For example, Pollatsck, Lesch, Mortis, sad Rayner (1992) found that sound codes are used in integraring information acess saccades, A homophone of a target word (beech as a preview for beach} presented asa preview in che parafovea faeilitaved processing of the target word seen. fon the next Gxation more than a preview with a non-homophone matched in orthographic similarity (fexh) to the target word. Because the orthographic similarity of che preview so the target also plays a pare in che fheilitative effecr of the preview, however, codes other than phonological codes (such as abstract letter codes) are preserved across saccades, Further evidence for the rule of phonological coding in integrating information across saccades will be presented inv Chapter 3. Letters vs. words ‘The evidence availible chus suggests that parafoveal previews help in ewo ways (Blanchard, Pollatsek, & Rayner, 1989}. First, che word in the parafovea may be fully identified (and perhaps skipped). Second, it nay only be partially activated, with this partial activarion speeding Tater identification of a word. We have reviewed rather convincing evidence that visual codes do not play any signif