Research With A Primary Focus

Reading and the brain

In perceiving the world, the brain appears to operate through a process of prediction and confirmation.
(Destexhe, 2000; Rosenberger and Rottenberg, 2002; Hawkins, 2004; Sherman and Guillery, 2006; Gilbert and Sigman, 2007) 

Over recent decades, functional magnetic resonance imaging and positron emission tomography have enabled neuro-scientists to investigate the workings of the brain as never before.  In the US in particular, vast sums of money have been spent on neuro-anatomical studies of the brain in operation.  Investigators have challenged the classical conception of how the brain operates, in ways that have huge implications for how we think reading proceeds.

Classical neurology presents perception as taking a bottom-up route.  In this classical view, all perception is seen as proceeding from simple registration of sense data, to progressively more complex representations as the 'information' moves up the brain's hierarchy.  Its path and form are determined solely by feed-forward mechanisms: higher levels in the hierarchy arise only from activity immediately below them.  If this is how the brain works, then reading must inevitably be a bottom-up process.

But this uni-directional conception of perception has recently been called into question.  Firstly, in terms of computation, it is inherently unlikely that feed-forward mechanisms can, on their own, lead to the flexible and invariant pattern recognition that allows us to recognise a cup, for example, in a complex and rapidly changing environment (Destexhe, 2000; Sherman and Guillery, 2006; Gilbert and Sigman, 2007).  We would need to supplement the up-coming sense-data information with a conception of the object in view. 

Secondly, recent studies of the traffic between the cortex and the thalamus have revealed that messages proceeding from the thalamus (which, in the classical view acts as a transit station for sensory data from visual, aural and touch receptors) upwards to the cortex (where higher mental activity takes place) are outnumbered ten to one by messages in the opposite direction (Destexhe, 2000; Sherman and Guillery, 2006; Gilbert and Sigman, 2007).  Destexhe claims that the cortical connections may predict the sensory information.  Feed-forward is thus complemented by feed-back; bottom-up is complemented by top-down.  Expectations complement incoming sense-data. 

Indeed such top-down processes are seen by some as central to the brain's operation.  Drawing on a number of recent studies of neuro-anatomy and neuro-physiology, Hawkins argues that the brain uses a vast amount of memory to create a model of the world, and uses this memory-based model to make continuous predictions of future events.  He sees the ability to make such predictions about the future to be the crux of intelligence (Hawkins, 2004).

So perhaps the term feedback is misleading, since it implies that the information from the cortex to the thalamus is subordinate to the information going in the other direction.  However, as Strauss et al. argue (2009), the volume of information coming from the cortex, and the extent of its control over the thalamus makes it clear that neither direction is primary.  Unexpected sense data can be transmitted from the thalamus to the cortex, but expectations originating in the cortex can influence the operation of the thalamus, and even initiate searches for sensory information in other areas of the brain.

Implication
In reading as in other perceptual activities, perception is not principally a passive matter of registering sense data, but operates through prediction and confirmation.

Reading appears to proceed not through a simple hierarchical system of sequential modules, but, like other brain activities, through a process of prediction and confirmation.
(Hawkins, 2004; Strauss et al, 2009)

Not many investigators have applied these new ideas about perception to studies of the reading process.  Indeed, where the processes of reading are involved, the technical advances of functional magnetic resonance imaging and positron emission tomography have not always been used in ways that illumine our understanding. 

A number of studies (e.g. Shaywitz et al., 1996) have claimed to show reading as a bottom-up process, in which the brain first identifies the visual input as a sequence of letters, then matches these to the phonemes they represent.  These studies have been used to justify intensive phonics programmes as 'brain-based learning'.  Yet they do not examine the brains of subjects reading connected text: instead they expose them to displays of letters grouped to make nonsense words.  Such studies are inevitably self-confirming.  In presenting the target 'words' to their subjects without any sort of context, the design eliminates the possibility of the reader drawing on semantic or syntactic cues, or indeed operating in any other way than bottom-up. 

However, Hawkins applies his brain-based model of prediction and confirmation to the act of reading, showing how the brain can predict and confirm letters and sounds.  Strauss et al. go much further than this, applying the idea of prediction and confirmation to large stretches of meaningful language as well as to letters and words.  

This work supports the conceptions of the reading process developed by both Kenneth Goodman and David Rumelhart in the 1960s and 1970s (Goodman, 1967; Rumelhart, 1976).  In this early article Goodman characterised reading as a 'psycho-linguistic guessing game', a top-down process in which the reader's expectations, shaped by the reader's knowledge of language and of the subject matter of the text, guide the perception of the letters on the page.  Decades of studies of readers. miscues, that is of deviations from expected responses in their oral reading of the text, show the influence of readers' expectations (Goodman and Goodman, 1977; Goodman et al., 2005).

Inspired by the new generation of computers, less than a decade later Rumelhart presented reading as 'simultaneous, multi-level, interactive processing', in which word identification is achieved through a two-way process, as the recognition of some letters generates bottom-up hypotheses about words, while linguistic and subject matter knowledge generate top-down hypotheses about the words.  Where the two sets of hypotheses are congruent, reading can proceed smoothly; where they conflict, the reader is jolted into a correction. 

In the many decades in which the bottom-up view of perception held sway, both authors were regarded as unscientific, despite the substantial evidence of reading behaviour that supported their views.  Now, it seems, more than ever, these views should be taken seriously.
Implication
We should all be very wary of approaches to reading that claim to be brain-based, while representing a limited and out-dated view of the brain.s operation.

Recent work on eye movements supports the view of reading as a complex, non-linear process.
(Krauzlis, 2005; Paulson, 2005; Strauss et al., 2009).

More recent studies of the eye movements of both skilled and apprentice readers show that readers do not proceed through a text word by word.  Indeed, they do not look at every word and do not necessarily look at the words in order (Paulson, 2005).  The movements of the eyes 'reflect neural decisions about where crucial information is to be found'. (Strauss et al., 2009).  The same authors write:

The brain is not dependent on the eyes to provide all the possible textual information to the brain.  Rather, the eyes are in the service of the brain while the reader is constructing meaning.
(Strauss et al., 2009, p. 27).

Even more specifically they write:

Patterns of eye movements are selective and purposeful, organized around the construction of meaning, not letter identification.
(Strauss et al., 2009, p. 27).

This view is supported by recent work in the neuro-biology of eye movement, which has called into question the idea of eye movements as driven automatically by low-level visual inputs, replacing this with a view of eye movements as regulated by a process of target selection involving a basic process of decision making.  The selection process itself is guided by a variety of complex processes, including attention, perception, memory and expectation (Krauzlis, 2005).

Implication
The entire process of reading is essentially purposeful, not a mechanical response to visual information.

Different languages make different demands on the brains of readers.
(Paulesu et al., 2000)

Using positron emission tomography, English and Italian investigators studied reading in English and Italian university students.  The differences in consistency of phoneme representation between the two languages were reflected in differences in the students' performance in reading both words and non-words.  The Italian students were faster on both types of task, even when the words were derived from English.  Scans showed that, even with simple and regularly spelled words, different areas of the brain were activated in the two groups studied.  The inferior basal temporal area, an area related to naming and semantic processing, was more strongly activated in the English readers, whereas the Italian readers showed stronger activation of the left planum temporale, a brain region linked to phonological processing.

Implication   
Learning to read English involves more than relating letters to phonemes.

Learning to read appears not to proceed through sequentially ordered modules.
(Freppon, 1991; Lieberman, 2000; Donald, 2001; Coles, 2003; Wolf, 2008)

One assumption of a 'brain-based' approach to learning to read has been that the fundamental reading modules of the brain, those that connect written letters to spoken sounds, must be activated before any other kind of learning can take place.  There is certainly evidence that the parts of the brain involved in making such connections, the occipital lobes, where visual and aural information are associated, are more widely activated in the brains of children in the early stages of learning to read than in those of more experienced readers (Wolf, 2008).

However, learning written language has recently been seen to involve not a pre-determined network of the areas regarded as specific to establishing the phonic connection, but as an evolving and much more extensive network, composed of activity in neuro-anatomical structures distributed throughout the brain (Lieberman, 2000; Coles, 2003).  Learning and experience shape the brain's circuits and how they are used - in learning to read, as in other domains (Donald, 2001).  A connectivity pattern emerges as children learn to read.  This view of the brain's role in learning is very different from the view of a step-by-step progression from module to module. The marked activation of the occipital lobes observed by Wolf (2008) in those learning to read may well be a product of the teaching they have received.

So, the brain areas centrally involved in grasping the sound/symbol principle do not have to be primed first.  Indeed, the very functioning of these areas depends on connections within the entire pattern (Coles, 2003).  Both adults and children can more readily identify written words that are already familiar aurally (Wolf, 2008).  This makes sense of what we know from observational studies of young readers, for example that a beginning reader reads the second half of a text with greater accuracy than the first, being in a better position to develop expectations about the words to be identified (Bussis et al., 1985).

Certainly there are parts of the brain that are specialised in dealing with language.  But learning written language is not solely detemined by these.  Instead it is based more generally on inferential thinking through more extensive neural networks (Coles, 2003).  There is more than one route to learning to read. 

As well as an outdated view of the areas of the brain involved in word identification, the 'phonics first' assumption rests also on the view that the beginning reader has limited working memory.  Such a view has been rejected by many neuroscience researchers.  Donald (2001) claims that most ideas about the limitation of working memory derive from laboratory studies using a methodology that does not mimic the conditions encountered in the world outside, but imposes a brief time-frame, into which 'short-term memory, visual imagery, perceptual illusions and the allocation of attention must be crammed' (Donald, 2001, p.47). But, in real life, 'the width and depth of working memory in such situations are much larger than those suggested by laboratory techniques' (Donald, 2001, p.51).  The memory system is composed of both short-term and intermediate-term awareness that constantly update working memory, allowing it to incorporate more complex mechanisms. 

Classroom-based studies have supported the view that there is no essential phonic gateway through which all those learning to read must pass.  For example, in a study carried out nearly two decades ago, Freppon compared reading outcomes for two first grade groups of children, one taught with skills-based instruction and one through a literature-based, whole language approach (Freppon, 1991).  Freppon found that both groups of children achieved similar test results and, even though the literature-based instruction did not teach these skills, both were knowledgeable about the decoding process.  But the skills-based group used decoding as a primary strategy, whereas for the whole language group, it was one of a wide range of strategies.  However, although the whole language group used 'sounding out' less often than the skills based group, when they did use it, they had a greater success rate than the other group.  Such findings cast doubt on the assumptions that learning to read operates through sequentially ordered modules and that young children have a limited working memory, that dictates an exclusive focus on one strategy as they begin to learn to read.

Implication
We need to focus on ways of teaching reading that encourage children to develop multiple strategies of word identification, while keeping a focus on the construction of meaning.