We may never know how we learn to read and what the cognitive processes of a skilled, fluent reader actually involve. Although fMRI scanning (more here) allows us to identify what areas of the brain are activated when reading, we can only hypothesise as to what the process actually is through the creation of models and, as Rayner et al (2012) suggest, models always lack a certain cognitive veracity. For example, the dual route and triangle models of reading both explain what may be happening but are at best inconsistent and at worst contradictory. In the same way both the geocentric and heliocentric models of planetary motion explained how the sun rises and sets in apparently different locations until Galileo provided proof; and, boy, was that one complex and bloody paradigm shift (Kuhn, 1963).
However, although we can’t see and measure the processes of the brain, we do have a crucial part of the reading process that we can record and measure and that is often ignored in the pedagogy surrounding reading: the movement of the eyes.
When we look at a page of text, we cannot see all of the words equally well because of the nature of acuity limitations associated with sight; when we look at a line of text falling on the retina it is divided into three regions. The foveal area is the most highly focused area but subtends about 2 degrees of the visual angle – about 7 characters. The parafoveal area subtends about 10 degrees of visual angle (five degrees either side of the fovea) and has far less acuity than the fovea. Everything else fall into the peripheral area which has no useful acuity for reading. Thus, the main element for reading is the fovea, the tiny area of intense focus consisting almost entirely of cones.
When we read, we have the impression that our eyes sweep continuously across the page feeding our mind the information from the text in a smooth, effortless flow that fluently enables the extraction of meaning. This is an illusion, not dissimilar to the illusion of fluid animation created by seeing 24 frames per second in cinematic animation.
When reading, the progress of the eye is not continuous. The eyes come to rest for periods of between 150 and 500 ms: fixations. Between these fixations are periods where the eyes are moving. These movements are ballistic (once started they cannot stop) and are called saccades and our eyes move between seven to nine character spaces in each of these jumps which take between 20 and 35 ms. Saccade suppression (Matin, 1974) ensures that we have no sense of the blurring that must occur during these jumps but that also ensures that the reader takes in no information during the saccade. All visual information is extracted during the fixation. Information extraction during reading is therefore the equivalent of a slide show, with each slide appearing for a quarter of a second followed by a brief delay before the next slide appears with no sensation of blurring as the new slide enters the display. However, the eyes do not move as relentlessly as this suggests. This is not unusual and is the way the eyes work for the perception of any static display as when looking at a picture (one’s perception is of seeing the whole picture’s detail at once which is not possible) – the fixation and saccade lengths are, however, different.
However, while most saccades move forward, 10-15% are regressive saccades: they move backwards. This means we are making a regressive saccade every two seconds. Although we may be aware of regressive eye movements when we are confused by text and meaning and we wish to check or reread, most regressive saccades only go back a few characters and we remain unaware of them.
The nature of text organisation and its necessary spatial constriction requires a further type of eye movement: the return sweep. These occur when the eyes move from near the end of one line to the beginning of the next and although they are right to left, they are not considered regressions as they are moving the reader forward. This large movement of the eyes raises some complications as they usually start a few characters from the end of the line and end five characters in from the next line. This is often followed by a corrective regressive saccade. However, the leftmost fixation on the next line is usually on the second word of the line - we’re not quite sure how we identify the first word on the line, but it is assumed that it is taken in as part of the sweep. Thus only 80% of the line falls between the extreme fixations. Thus, it would appear, that we do not fixate, or need to fixate on every word during reading.
It should, at this point be noted that most testing of eye movements have been made during silent reading and there are differences when readers read aloud (Inhoff et al., 2011). Fixations during oral reading are longer, saccades shorter and there are more regressions which suggests that the reader does not want their eyes to get too far ahead of their voice and eyes are kept in a holding pattern to ensure that this does not happen. In oral reading, eyes are generally two words ahead of the voice. Indeed, this variation in fixation and saccade length is an integral part of reading. For instance, when reading more complex texts, readers reduce saccade length, increase fixation time and regressions. This is also evident when fonts are more convoluted and less transparent and type smaller and constricted (Tinker, 1963) and lines shorter. Interestingly, for readers of Hebrew, saccade length is shorter than for English readers. Although Hebrew is orthographically more transparent (closer to one-to-one letter-sound correspondence), it has fewer words that contain more densely packed information which demand more attention (Pollatsek et al., 1981). Direction of text seemingly makes little difference to eye-movements (Kolers, 1972) with Hebrew/English readers able to switch between the two directions of the two languages with few problems and little variation in eye-movement which performs similarly but inversely. As readers develop their expertise, saccade length increases, and fixations shorten.
Given the small size of the foveal area, the amount of information that can be extracted when reading is limited, thus requiring the eyes to move four or five times every second. But what information is gleaned in a fixation? It would seem sensible to assume that every fixation determines a specific word, and certainly if we read according to the whole word theorists, this must be the case. However, if the fovea is only large enough for seven characters then that is not sufficient for all words. How then are we able to read with such speed and accuracy?
The size of the perceptual span and the work of the parafovea are crucial in determining just what the eyes are doing during reading and how we read. The perceptual span is formed of the intense foveal area but also the parafoveal area to the right and left. However, the left and right boundaries of the perceptual span appear to be constituted differently (remember we read left to right in English). On the left side, information is extracted from a much smaller area relating almost entirely to the fixated word. On the right side, however, the span is greater, and the boundary is not dependent on word boundaries for speed of reading (Rayner et al., 1980). Partial word information beyond the fixated word is being obtained from parafoveal vision. In fact, with parafoveal vision alone (foveal vision supressed) reading is still possible at 12 wpm (Fine and Rubin, 1999).
Crucially, Rayner et al. (1982) showed that when the perceptual span was restricted to only single words, participants read at 200 wpm, but when the perceptual span was not restricted to individual words and included letters from following words the reading speed rose to 330 wpm. Readers were thus doing something more complex than extracting words as visual units. The initial letters from following words (on the right of the perceptual span) were also aiding fluency significantly.
So, during a fixation, the information to left of the fixation (in English that is - it’s the other way round in right to left encoded languages like Hebrew) is largely irrelevant. Information further than about 14 to 15 character spaces to the right of the fixation is not used due to acuity limitations – where the parafoveal area meets the peripheral area. Thus, during a fixation, a word in the critical word location is read by utilising the acuity of the foveal area. If the word is longer than 7 characters, information from the parafovea assists the identification. If this is not possible then a further fixation is required – this is why books with longer words take longer to read. Shorter words are identified in the foveal area at the critical word location and shorter more common words may be identified in the parafoveal area thus not requiring any further fixation. Up to two short words may be skipped (processed by identification in the parafovea) in one fixation when the target word is short (Brysbaert et al., 2003). Three letter words are skipped 67% of the time, whereas 7 letter words are seldom skipped (Rayner and McCronkie, 1976).
Perhaps more important and signalling the death knell for the word method of reading, was Rayner’s (2012) discovery that words to the right of the critical word location did not have to be recognised as wholes and are often not. The first few letters of a word recognised in the parafovea resulted in a far shorter next fixation to identify the whole word resulting in a preview benefit from one fixation to the next. It was the letter sequence that was important and not the shape of the word. Whole word theorists argued that it may have been the initial shape that was being previewed but Rayner (2012) found that for words with the same preview shape but different letters the preview benefit was still manifest. It was the letter combination that was therefore assisting the reading and not the word shape. There is now very strong evidence that sound codes are used to integrate information across eye movements and are part of the benefit readers obtained from preview information (Polatsek et al., 1992). Thus, parafoveal previews assist in two ways. Firstly, the word may be fully identified and skipped. Second, it may be partially activated which speeds up later identification of the word in the next fixation. The fact that this preview phenomenon still occurs with mixed case words further emphasises the importance of letter patterns rather than word shape. The evidence from eye movements suggest, therefore, that phonemical knowledge and mastery of the alphabetic code is a requirement for fast and fluent reading.
That phonological information is extracted early in the word identification process has been evidenced by numerous studies (Ashby and Rayner, 2004; Pollatsek et al., 2005; Ashby et al., 2009); early enough for it to affect the decision for the eyes to move on the next fixation. It also appears that meaning extraction occurs quite early. Fixation times are affected by word frequency and predictability from prior text (nb. predictability for fluent readers is different from the word guessing phenomenon theorised by the whole language proponents) with both phenomenon reducing the fixation times and thus speeding up reading. Meaning extraction research with the use of homophones (Duffy et al., 1988) where meaning is dependent on context indicates that the prior context actually guides the reader to the meaning. That all of this processing occurs in the fraction of a second before a decision is made to fixate the next word is rather amazing, but the speed of processing is clearly enhanced by the phenomenon that the processing of most words starts before they are actually fixated. What this tells us in terms of pedagogy is that vocabulary, language and exposure to words are crucial for the enhancement of reading speed and thus fluency. This seemingly supports Share’s (2002) postulation that orthographic processing and automaticity is developed by regular, frequent exposure to text – reading promotes reading skill and enhances reading.
Finally, several studies (Drieghe et al., 2005; Rayner and Slattery et al., 2011) have demonstrated that words are fixated for shorter periods of time and skipped more frequently when the reader is extracting meaning from a text from a context with which they are familiar. This suggests that high levels of contextual knowledge can affect lexical access during reading and further suggests that in such situations the extraction of information from the parafovea is enhanced. This appears to support the growing evidence (see here) that extracting meaning from text is enhanced by familiarity of the subject implying that reading comprehension is not a skill but a factor of vocabulary and knowledge.
In summary, the extensive research into eye movements when reading seem to suggest that:
· mastery of the alphabetic code and sound codes will enhance reading fluency
· reading fluency is enhanced by extensive and regular reading that builds language and vocabulary to assist predictability and enhance preview
· knowledge (discrete, global, cultural and subject specific) supports fluency and meaning extraction.