Fixation locations in reading:
Considering influences on saccade lengths and landing points.
Saccades and Fixations:
Saccadic movements occur to transfer a reader’s eyes from one location in text or a visual scene to the next location. This serves to bring a new stimulus onto the foveal region, which is the central 2o of vision. When the saccade lands the eye remains relatively still for a short time, which is called a fixation. Fixations are generally regarded as when information is accessed from the visual scene. This is because there is a suppression of visual information access during saccadic movement (Matin, 1974, as cited in Rayner, 1979). However, during a fixation, information may be accessed from the parafoveal regions as well as from the fovea. Saccades travel at up to 500o per second but their speed cannot be altered voluntarily (Rayner, 1998). It also seems that, once started, the saccadic target landing point cannot be altered. The purpose of this review is to discuss what influences the choice of this landing point.
The two most important questions regarding saccades involve considering where the saccade moves to after a fixation and when the eyes make this movement; how long they remain fixated at each point. I intend to focus on the first of these two issues, discussing what is common about saccadic landing points and what features affect where fixations occur. I intend only to consider saccadic movements when reading text within the scope of this review, although saccades are also involved in eye movements when studying pictures or any other complex stimuli. Other types of reading such as music reading have also been previously studied (Rayner, 1998), as well as cases in which saccades operate differently, such as when typing. Although most research covered concentrates on saccadic information regarding reading English text, I do intend to refer to research involving other languages.
Saccade length appears to be affected by many different types of information available from the text. These aspects include basic characteristics such as text quality, font and spacing; word length and the positioning of spaces; and conceptual difficulty of the text, possibly incorporating word frequency and higher-level factors of comprehension. I intend to review all of these aspects, with specific reference to whether lexical pre-processing influences saccade landing locations. This relates to the various suggested models of eye guidance which fall into 3 main categories, which I shall outline briefly.
Models relating to saccadic movements:
The most simplistic model types are called minimal-control or constant-pattern models. These maintain that no characteristics of the text direct the eyes, and only the overall difficulty of a piece of text can alter saccade length. The eyes are moved in a regular pattern, with only a small variation between fixations (for example, Bouma and DeVoogd, 1974, as cited in Rayner, 1979). The second type of model is known as a stimulus-control, high or low-level model, which suggests that textual characteristics influence eye movement. In low level models only visual characteristics are available for input, with saccadic landing position being influenced by parafoveal information relating mainly to word length. High level models assume that semantic or syntactic information can guide saccades and fixations to facilitate reading of the text (Rayner and McConkie, 1976). Similar but separate models, known as process-monitoring models, hold that a mechanism exists which adjusts saccade lengths so that information is acquired at a rate corresponding to the rate at which the text is being processed. Eyes are moved to a new location when there is sufficient capacity to take in new information or alternatively the eyes are moved to a location reflecting how far ahead words have already been processed by parafoveal identification (McConkie, 1979, as cited in Rayner, 1979).
However, different models may apply to fixation durations than for saccade length. These mixed models have been suggested whereby durations are influenced by a process monitoring system but where saccades are directed is only determined by low level visual factors, not using processing information. It does seem that there are separate processes that decide where and when to move the eyes, with fixation duration being unrelated, apart from by overall text difficulty, to fixation location (Rayner and McConkie, 1976). Although I only consider these models briefly, most of the evidence I review for factors influencing saccadic amplitudes and landing points is in some way related to high or low level stimulus control models.
Characteristics of fixation locations:
The mean saccadic length is 7-9 characters (Rayner, 1998), however, there is great variability between saccades for a single individual for different words, as well as variation between people. These saccades tend to land somewhere between the middle and the beginning of words, with a shift towards the beginning the longer the word is (Rayner, 1979). This was generally known as the ‘preferred viewing location’ (Rayner, 1979), which was later distinguished from the ‘optimal viewing position’ (O’Regan and Lévy-Schoen, 1987, as cited in Rayner, 1998), defined as the location within a word which allows shortest word identification time. This is not necessarily the same place as the preferred viewing location.
So far, this information mainly pertains only to an incoming saccade initially entering a word, however, many words receive more than one fixation, which tends to be further towards the end of a word (Rayner, 1979), but saccades are also sometimes directed to the left of a previous fixation. These kinds of backwards movements are known as regressions; with a distinction being made between those that traverse between words and those that remain within the same word. Within word saccades tend to be due to a corrective programming when a saccade overshoots or due to processing difficulty of a particular word. Longer regressive saccades are more likely to occur when a reader has not understood a larger piece of text (Rayner, 1998). These show a tendency to be directed to the end of a previous word. Rayner (1979) found that there are more interword than intraword regressions.
There are also specific differences between fixation locations depending on where they occur within a line of text or a sentence. At the start or end of a line, fixations tend to fall 4-8 letter spaces from the extreme character (Rayner, 1979), with the first saccade therefore not following the usual pattern and the final word being skipped more frequently. The final saccade lands further from the end of the sentence but the last word is still within the perceptual span. The perceptual span defines the space within which information can be accessed for one fixation. It is assumed to stretch asymmetrically from 3-4 characters to the left of the fixation point to 14-15 characters to the right (Rayner, 1998). However, the span for actual identification of words is slightly smaller, at roughly 7-8 spaces to the right. Importantly, this may be individually determined by complexity of a word or an overall piece of text, among other factors.
Influences of word length and space information:
Because the perceptual span usually extends outside of the currently fixated word boundary, it is possible for some words to be processed and identified without being fixated. These words are then skipped, receiving no fixations. This tends to occur with short function words. Content words are rarely skipped, being fixated 85% of the time, while function words which are more redundant and easy to identify, are only fixated 35% of the time (Carpenter and Just, 1983, as cited in Rayner, 1998). This is most likely to be because function words such as ‘and’ tend to be very short. Word length is a particularly crucial factor in determining fixation locations. The length of a word to the right of the currently fixated word can be ascertained before the saccade is programmed. The number of fixations a word receives depends highly on its length. Rayner and McConkie (1976), investigated basic word length effects; looking at the mean number of fixations on words of different lengths and the probability of a word being fixated. They found that words over 8 characters long were likely to be fixated more than once and showed that the probability of fixating a word increased as word length increased. A 12 letter word had a probability of 0.973 for being fixated, while for a 3 letter word it was only 0.318.
As well as determining whether a word is fixated, word length information may be involved in the programming of a successive saccade so that it lands far enough into the word (O’Regan, 1975, as cited in Rayner and McConkie, 1976). Specific placement of a saccade is assumed to be at least partly determined by the length of the word to be fixated. It therefore seems likely that removing word length information, such as taking out spaces between words, will disrupt the mechanism which programs saccades. This view is presented by Rayner and Pollatsek (1996), in a response to a study by Epelboim, Booth and Steinman (1994). They argue that previous claims that “unspaced text is relatively easy to read” are only founded on results from a very small number of subjects and that evidence from alternative studies is more convincing. They refer to a study by Spragins, Lefton and Fisher (1976) which found that reading rate was halved for unspaced text, and also to Pollatsek and Rayner (1982), whose results showed much slowed reading rate for text with spaces filled with either random letters or digits.
Rayner and Pollatsek (1996) report findings of the Pollatsek and Rayner (1982) experiment which determined more about how space information is processed. Using digit space fillers did not disrupt reading as much as using letter fillers, but did still slow reading rate by about 40%. The authors use this to assume that readers require actual space information rather than information about where words end, as the digits should have had a similar effect in this respect. The experiment varied the time delay before the space fillers appeared, finding that reading performance improved the longer the space information was available. It also seemed that preserving information about the first space to the right of fixation improved reading ability even if all spaces after that point were obscured, although there was still some cost to reading time.
With specific respect to fixation locations, Rayner and Pollatsek (1996) conclude that the most likely reason why Epelboim et al. (1994) found very little difference between fixation locations in unspaced and normal text is that readers have to adopt a different strategy. They suggest that when space information is not available readers will programme saccades to land roughly the average length of one word away from the current fixation point. This would lead to very similar landing sites to normal text if words were not particularly long or short.
Within the article by Rayner and Pollatsek (1996), a response from Epelboim, Booth and Steinman (1996) refuted the criticisms of Rayner and Pollatsek, emphasising their claim that it is word recognition that primarily guides saccadic programming rather than space information. They do, however, state that word length information does serve to facilitate word recognition and point out that adding spaces in inappropriate locations does disrupt reading. A more recent paper is also referred to in which they found a larger decrease in reading rate when spaces were removed from incoherent text formulated of unrelated words than when normal, coherent text had spaces removed. They use this as evidence for word recognition driving eye movements rather than visual information such as spaces. Epelboim et al. (1996) dismiss Rayner and Pollatsek’s criticisms, suggesting that their experiments are unreliable due to the different paradigms used, where fillers were introduced instead of merely taking spaces out. There do appear to be many differences between these two sets of experiments, particularly with Epelboim et al. concentrating on reading aloud rather than silently.
A later study by Epelboim, Booth, Ashkenazy, Taleghani and Steinman (1997), considered the effects of using fillers to examine effects of space information, as applied by Rayner and Pollatsek (1996). They showed that even just surrounding words with foreign characters and leaving space information available slowed reading. The more similar the characters were to real letters, the harder reading became. These effects, due to the nature of the manipulation, may possibly be related to orthographic effects, which are discussed later. Epelboim et al. (1997) used these results to back an argument that word identification processing is disrupted by the fillers rather than the lack of space information in the paper by Rayner and Pollatsek (1996). Importantly, in this study, Epelboim et al. did not specifically look at fixation location effects other than to speculate on how saccadic programming might be affected in terms of the optimal viewing position. They claim that if saccadic programming had been disrupted the pattern of reading times found would have differed; however, this does not seem well argued and is not reliable as a speculation.
This issue was addressed again by Rayner, Fischer and Pollatsek (1998), who conducted similar experiments to those of Epelboim et al. (1994; 1997). Firstly, for just removing spaces in text, they found that the mean saccade size in spaced text in single lines was 7.3 characters, while for unspaced text, it was 4.4. They also noted a much larger percentage of regressions in unspaced text, which had not been commented on by Epelboim et al., who only considered reading times. In more detailed examinations involving the different space-removing conditions, saccades into and out of words in which space information was not available were much shorter than for other conditions. Within word saccade sizes were relatively unaffected between conditions. Landing positions were closer to the beginning of the word when there was no space information available.
In terms of causes of these effects, Rayner et al. (1998) make several suggestions as to why space information is important. Firstly, removing word length information may interfere with word identification because it cannot be determined where words begin and end. Secondly, beginning and end letters might be particularly important and are more obscured by not having spaces adjacent. Thirdly, the lack of knowledge about where words end and begin may interfere with saccadic programming. Rayner et al. conclude that it is likely that both word identification and saccadic programming are interrupted.
In conclusion, it seems that space information is accessed as an aid to reading and in particular to help program saccades. Removing this information slows readers down although they are still capable of reading the text. The reasons why these effects occur are slightly less certain, although Rayner et al. (1998) suggest that processing is affected on at least two different levels. Interesting effects found by Kohsom and Gobet (1997) seem to support these findings, as they discovered that adding spaces in Thai, which usually has no spaces, actually assisted reading slightly. They also compared this with removing spaces in English and examined reading by native speakers of Thai. The removal of spaces affected reading in English more so than in Thai. They also considered coherence effects, testing results found by colleagues of Epelboim, and found no interactions for coherence and spacing, therefore challenging Epelboim et al (1994, 1997) on the preference of word recognition effects over space effects. Kohsom and Gobet conclude that spacing is a “cue that is universal across languages” (pg 393).
Considering Wrap-up effects:
Having considered the effect of some visual, lower-level information on fixation locations, I now intend to examine whether more lexically based information can be accessed in order to aid saccadic programming. One of the first indications that at least some semantic information is used in determining saccadic movement patterns is the apparent existence of wrap-up effects. Rayner, Kambe and Duffy (2000) investigated these effects, finding that when a reader reaches the end of a clause, fixations tend to be longer. The ends of clauses also seem to have effects on the landing position of saccades. Firstly, components of clauses were skipped less often if they were clause-final. Secondly, saccades were longer following the end of a clause, when entering a new constituent, than saccades from the same word if not at the end of a clause. Hence initial landing positions in new clauses were placed further into the clause. There were no differences in launch site in these conditions, although there were more regressions from clause end words as well.
Wrap-up effects, in which processing of a clause is finalised, seem to affect both when and where decisions of eye movement control. Larger numbers of regressions indicate that comprehension problems are resolved before leaving a clause, presumably meaning there is a high processing difficulty for words closing a clause. Despite this, it seems that more parafoveal information is available because saccades land further into the next section. In this instance it seems where and when decisions are related, as Rayner et al. (2000) suggest that wrap-up processes “ensure that the information from the clause is fully integrated” (pg 1075) and thus clear working memory for the clause and so more parafoveal information can be gathered from the next clause. This seems to agree with results referred to by Rayner et al. (2000) from Rayner (1975), who found similar results for the ends of whole sentences.
Influences of lexical information:
At this point I have considered low-level visual effects on where saccades land and also introduced a notion that lexical factors can determine to some extent how long saccades are, through processing stages like wrapping-up clauses. On a processing basis, it also seems relatively obvious that simple typographical factors will influence the way that saccades are programmed, with font differences, quality of print and text organisation having basic effects, slowing down processing and making overall saccadic amplitudes smaller. The more contested and investigated issue is that of whether any higher level linguistic factors are capable of inducing differences in online, word-by-word saccadic programming.
A study by Ehrlich and Rayner (1981) did provide evidence for an effect of lexical constraint on saccade programming. This was related to whether highly predictable words are more often skipped than a less predictable target. The experiments looked at passages of text where constraint was built up over a whole passage making a target word highly predictable. It was found that readers were less likely to fixate a target word when it was highly predictable. Misspelling was also looked at, to test the theory that high constraint levels influence how sensitive readers are to features of words in central vision. There was a higher probability of fixating a word if it contained a misspelling, although this was a much stronger finding in low constraint passages and subjects were less likely to report misspellings of the target in high constraint conditions. Ehrlich and Rayner used these results to support the hypothesis that context effects are capable of controlling eye movements. This suggests that readers are able to predict words before fully processing them. However, Ehrlich and Rayner stress that their belief is that there is no extra information gathered from parafoveal vision, but that the high constraint simply means that information “can be utilised more efficiently” (pg 654).
Reversely, a large scale effect of lexical information can be observed by the fact that saccade lengths decrease over a whole passage of text when conceptual difficulty increases (Rayner and Pollatsek, 1989, as cited in Rayner, 1998). More difficult text also increases the number of regressions that occur and this in turn could produce a difference in observed mean saccade length, as longer saccades may follow a regression, to “place the eyes ahead of where they were prior to making the regression” (Rayner, 1998, pg 376). The size of the perceptual span appears to be decreased in difficult text, causing these effects on eye movements. These two corresponding findings indicate that processing difficulty has the capacity to alter saccadic amplitude but it does not determine whether this can be on a moment-to-moment online basis. These effects are only shown for whole passages of text which influence lexical factors. It cannot be stated from this that lexical constraint or text difficulty either do or do not influence saccadic programming for individual words.
Effects of Informativeness:
Early experimentation attempting to gain insight in this area focused on the informativeness of words and whether this variable had the ability to influence the saccadic length when first entering a word. Informativeness is generally considered by calculating the number of words that begin with or end with a particular sequence of letters. This was presumed to indicate how many candidate words there were available from the information at that part of the word. Hyönä, Niemi and Underwood (1989) manipulated informativeness in different regions of words and investigated the effects on eye movements using artificial display situations and also placing the words in sentences. This aimed to find similar results to previous studies by O’Regan (1984, as cited in Hyönä et al., 1989), who observed that if an informative-beginning word was read normally, with the initial fixation at the start of the word, more words were read with only one fixation than if the word had a non-informative beginning. This meant that non-informative end parts of words were more often skipped. However, if the reader had to fixate the end of the word first, initial fixations were always short, independent of whether the end was informative or not.
The experiment by Hyönä et al. (1989) was conducted using Finnish words, attempting to confirm predictions that uninformative parts of words (the first or last 6 letters) would be left unfixated more often than informative parts. They looked at normal reading using eye-tracking techniques, controlling for overall word frequency but using a mixture of compound words and adjectives. In a repeat of the experiment by O’Regan (1984), they found similar results, also noting that when regressions were made from a forced initial fixation on the end of the word, the saccade was longer if the informative region was at the beginning of the word. This seemed to indicate that lengths of saccades are sensitive to location of information and are drawn further toward the beginning of an informative region than a non-informative one. However, their second experiment, using more natural reading, showed less conclusive evidence.
The results from the Hyönä et al. (1989) second experiment found that the first half of the target word was always more likely to be fixated, regardless of informativeness. However, there was a trend to skip non-informative endings, although an interaction between the probability of fixating a word and information location was not significant. This suggests that saccades out of words with non-informative endings may have larger amplitudes. First fixations were located closer to the centre of a word if the informative region was at the end, which was used to suggest that the preferred landing position was altered by informativeness information. This attraction towards informativeness suggested that pre-processing of a word to the right of fixation occurs; the word is “not identified, but unconscious processing of the meaning influences fixation location” (Rayner and Morris, 1992, pg 165). An extra part of the experiment, however, found this attraction to be insignificant after words appearing at the end of line were excluded from analysis.
This Hyönä et al. experiment was criticised by Rayner and Morris (1992), who argued against any semantic pre-processing of lexical information and failed to replicate the results. They claimed that the explanation for why the effect found by Hyönä et al. should occur was not clear. If readers do determine that the informative part of a second word is at the end while still fixating and processing the previous word, it seems to be a very complex and difficult task and there is very little time in which one can assume this occurs. Rayner and Morris’ experiment used more accurate eye tracking software and larger numbers of subjects to carry out a very similar experiment in English which found no tendency for fixations to fall further into the word depending on informativeness.
After this challenge, another investigation by Hyönä (1995) considered irregular letter clusters as an indicator of orthographic regularity and looked at the influences of these on saccadic programming, rather than an effect of informativeness, which was this time also not replicated. This followed suggestion of a ‘pull assumption’ by Hyönä (1993) in which “optimal landing is due to some features of the word pulling or attracting the eyes towards the more informative part whether it be orthographic, lexical-semantic or some other type of information”. He also proposed that “orthographic saliency is perhaps the most plausible feature”. Hyönä (1993) therefore suggested that a ‘pop-out’ effect of infrequent letter combinations could be responsible. Within this hypothesis Hyönä also reasoned that increased processing demands decreases the amount of information available to parafoveal processing in some circumstances, which could explain the variable results concerning landing position effects.
In Hyönä (1995), orthographic regularity was varied using words that had an uncommon letter combination in the first half of the word. The prediction was that, for these words, saccades would land nearer to the beginning of the word. Other stimuli had identical beginnings and either infrequent or frequent orthographic endings. The initial fixations were influenced by regularity of the letter string at the beginning of a word, but not by regularity of the end of the word. Second fixations also seemed to be drawn closer to the beginning for irregular beginning words. Saccades within the words were also shorter for infrequent beginning words but incoming saccades showed no difference.
The results therefore indicated that orthographical differences at the ends of words had very little influence over initial fixations. However, the infrequent orthography at the beginning of words was capable of causing earlier fixations, particularly on the space preceding the target word. A difficulty which has been acknowledged with these findings is that Hyönä used extremely unusual letters for the infrequent orthographical word beginnings because they involved letters almost never observed in Finnish which only appear in borrowed words from other languages (for example ‘b’). It is less clear whether these effects will be produced with less obvious infrequent letter strings, as this effect in Finnish could merely be due to the visual unfamiliarity of the particular letters.
From this evidence, the pull assumption hypothesis was not supported. However, Hyönä (1995) interpreted the data to produce a processing difficulty hypothesis. This suggests that when unusual letter clusters are perceived in the parafoveal region, processing difficulty causes the saccade to fall short of the preferred viewing location, particularly seeming to aim for the preceding space instead. Processing difficulty of the unusual section then also affects second saccades within the word, causing them to be shorter than if the orthographical information at the beginning had been more regular. These findings led Hyönä and Pollatsek (2000) to examine fixations in compound words, also in Finnish. Using the processing difficulty hypothesis, they predicted that frequency of compound constituents could influence the programming of saccades.
Hyönä and Pollatsek (2000) found that the length of the first constituent influenced the length of the first within-word saccade. If the first constituent was long, the within-word saccade was directed further towards the end of the word. On the other hand, initial landing positions were unaffected by constituent length. This seemed to point to an effect of morphological structure on fixation landing sites, and to the fact that compound words are treated similarly to two separate words, with exiting saccades being longer for longer constituents. Frequency effects were also found, with incoming saccades landing closer to the beginning of a low frequency first constituent and the second fixation falling less far into the word. This supports the processing difficulty hypothesis for both parafoveal processing and active processing. For words with a low frequency second constituent, opposite results were found, with initial saccades being unaffected and second saccades being shorter.
Predictions that were not confirmed involved decreased exit saccade length when second constituent frequency was low and an overall effect of word frequency. The authors suggest that for compound words “frequency effects associated with the fixated word would not spill over to the parafoveal processing of the next word” in relation to saccade length (Hyönä and Pollatsek, 2000, pg 80). This means that, while fixating the word, the processing difficulty is capable of altering programming of saccadic landing points, but the word is sufficiently processed before attention is moved to a succeeding word and the exit saccade is programmed. They explain that this encompasses why initial fixations are affected as shown, because attention shifts from the preceding word before a saccade is programmed but preceding word frequency effects are not carried forward.
This work analyses the way that Finnish compound words are fixated, providing evidence for decomposition into constituents of these words during processing. The constituents then appear to be processed like normal separate words except their frequency effects do not carry over to affect the following word. The size of constituents can be used like word length information once the compound has been processed as split into its two parts. The results argue for a processing difficulty system which is capable of altering saccadic amplitudes online as a function of characteristics of individual context and length factors in words made up of two words joined together. However, some of these effects were noted as being small effects and it is unknown how much this can relate to processing of English words.
Following the collective findings of Hyönä (1995) and Hyönä and Pollatsek (2000), experiments by White and Liversedge (in prep) have tried to examine whether similar results are replicable in English. They conducted separate experiments to distinguish between informativeness and orthographic factors, using misspelled words to introduce particularly infrequent orthographic strings. As opposed to Hyönä’s use of pre-test familiarity judgements to represent ‘informativeness,’ these experiments used a measurement more like the number of words in a dictionary beginning with a certain letter string. Four misspelled conditions of varying degrees of orthographic frequency were compared with an un-misspelled condition. Predictions that initial fixations would occur closer to the start of misspelled words were upheld. This lends evidence to the argument that words are pre-processed while in parafoveal vision and this affects saccade programming.
Additional evidence from these experiments also showed that the more irregular the misspelling, the more regressions and refixations were observed. Differences between a correctly spelled condition and a high frequency misspelling seem to indicate that at least 4 letters at the beginning of a word are pre-processed for the recognition of irregular strings. A second experiment used correctly spelled and misspelled words with uninformative initial trigrams and a misspelled word with an informative initial trigram to eliminate confounding effects with informativeness. All three conditions had equal orthographical frequency for the first three letters. No effects of word type were found on initial fixation location, meaning these results differed from those of the first experiment. Neither informativeness nor misspelling showed initial saccades landing closer to the start of the word. However, misspellings were shown to influence regressions and refixations, with all misspelled conditions gaining more refixations than a correctly spelled condition and the informative misspelled condition in particular having more regressions.
Subsequent experiments (White and Liversedge, personal communication) have shown similar effects of
orthography with correctly spelled words in English with different initial letter sequence frequencies.
Fixations were shown to fall closer to the beginning of words with less frequent letter sequences and these
effects were shown to be unchanged for upper case versus lower case text, demonstrating that
“orthographic familiarity influences where words are first fixated even with visually less distinctive and
less familiar text”.
Overall conclusions based on these studies regarding the parafoveal access of information of a lexical nature such as orthography are mixed. The studies by Hyönä (1995) and White and Liversedge (in prep) provide suggestion that word initial orthography can influence first fixations, with unusual orthography drawing saccades to land closer to the beginning of the word. This indicates that some information beyond word length can be extracted parafoveally. However, whether informativeness, or any more lexical information regarding frequency of parts of words, can influence fixation locations, is less certain. The extent to which saccadic amplitude can be influenced varies; with long words in Finnish being capable of shortening saccades within words and into words due to frequency and length effects; while in English only an effect of word length has been found. The evidence so far does seem to indicate some use of a processing difficulty system, in which less parafoveal information can be acquired if processing difficulty is high within foveal or parafoveal words (Hyönä, 1995; Hyönä and Pollatsek, 2000).
An average initial saccade is 7-9 characters in amplitude and lands somewhere between the beginning and middle of the fixated word. The programming system for this landing point seems to be subject to influence from word length information; where within a clause or sentence the word appears; passage complexity and lexical constraint variables; and orthographic regularity of beginning letter strings. However, frequency or informativeness of words or word sections, does not appear to have a reliable online word to word effect. Some of these factors seem to operate mostly on a processing difficulty type system, where higher difficulty of a number of text elements can increase or decrease probability of fixation or specific landing position of fixation. However, this explanation does not cover every angle of these results that are reviewed and not all are without dispute.
Ehrlich, S.F. and Rayner, K. (1981). Contextual effects on word perception and eye movements during reading. Journal of Verbal Learning and Verbal Behaviour, 20, 641-655
Epelboim, J., Booth, J.R., Ashkenazy, R., Taleghani, A. and Steinman, R.M. (1997). Fillers and spaces in text: the importance of word recognition during reading. Vision Research 37, 2899-2914.
Epelboim, J., Booth, J.R. and Steinman, R.M. (1996). Much ado about nothing: the place of space in text. Vision Research, 36, 465-470
Hyönä, J. (1993). Eye movements during reading and discourse processing. Psychological Research
Reports 65. University of Turku: Turku.
Hyönä, J. (1995). Do irregular letter combinations attract reader’s attention? Evidence from fixation locations in words. Journal of Experimental Psychology: Human Perception and Performance, 21, 68 – 81.
Hyönä, J., Niemi, P. and Underwood, G. (1989). Reading long words embedded in sentences: informativeness of word halves affects eye movements. Journal of Experimental Psychology: Human Perception and Performance, 15, 142 – 152.
Hyönä, J. and Pollatsek, A. (2000). Processing of Finnish compound words in reading. In: Kennedy, A., Radach, R., Heller, D. & Pynte, J. (Eds). Reading as a Perceptual Process. Elsevier. Oxford, UK.
Kohsom, C. and Gobet, F. (1997). Adding spaces to Thai and English: effects on reading. Proceedings of the Cognitive Science Society, 19, 388-393.
Rayner, K. (1979). Eye guidance in reading: fixation locations within words. Perception, 8, 21-30.
Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124, 372-422.
Rayner, K., Fischer, M.H. and Pollatsek, A. (1998). Unspaced text interferes with both word identification and eye movement control. Vision Research, 38, 1129-1144
Rayner, K., Kambe, G. and Duffy, S.A. (2000). The effect of clause wrap-up on eye movements during reading. Quarterly Journal of Experimental Psychology, 53A, 1061-1080.
Rayner, K. and McConkie, G.W. (1976). What guides a reader’s eye movements? Vision Research, 16, 829-837.
Rayner, K. and Morris, R.K. (1992). Eye movement control in reading: Evidence against semantic pre-processing. Journal of Experimental Psychology: Human Perception and Performance, 18, 163-172.
Rayner, K. and Pollatsek, A. (1996). Reading unspaced text is not easy: comments on the implications of Epelboim et al.’s (1994) study for models of eye movement control in reading. Vision Research, 36, 461-465.
White, S. J. and Liversedge, S.P. (in prep). Orthographic familiarity influences initial eye fixations in reading.
White, S.J. and Liversedge, S.P. (personal communication). Orthographic regularity influences where the
eyes land during reading.