Information processing theory attempts to articulate the unobservable through models that illustrate a cognitive-processing perspective on the processing, storage and retrieval of information (Tracey and Morrow, 2012). It is vital to the understanding of the process of learning to read and being taught to read because implicit in the theory is the understanding that any cognitive activity has a process associated with it. The implication for the teaching of reading and is profound: it is not an instinctive or natural process that develops through immersion in language and text; it is a cognitive, developmental and progressive process that can be broken into clear and teachable stages. Each of these stages is part of an information-processing construct that requires knowledge to be embedded in the long-term memory for manipulation into schema prior to regular application in the development towards fluency.
The theory is marked by a discrete, stage-by-stage, conceptual orientation that describes the processing, storage and retrieval of knowledge to and from the mind (Stanovich, 2000). Atkinson and Shiffrin’s (1968) model identifies the cognitive processing of information from stimulus and perception held briefly in the sensory memory, its development in the working memory and the relationship with working memory and encryption in the long-term memory. Only information that receives sufficient attention in the working memory will be successfully encoded into the long-term memory (Slavin, 2003) and then constructed into schemata thus ensuring a structure for future information storage, retrieval and application.
Reading as an information processing system is initiated through the human eye and at the core of this is retina and its construction into rods and cones with their associated acuity advantages and limitations. The concentration of cones at the fovea enables the discrimination of the fine detail required for identifying letters on a page but has the restriction of a constrained area of focal acuteness limited to a few letters. Once fixated on by the parafovea and the fovea, the pattern recognition process can be initiated thus enabling the recognition of words. The lowest level of the information processing system is the sensory store which for visual data is referred to as the iconic memory (Sperling, 1960) where information is held for a fraction of a second for later processing. Because of the transient nature of the iconic memory, a more permanent structure is required to store the data if the information is not to be lost. The evidence then passes to the short-term or working memory which has the capacity to store between five and nine items (Miller, 1956). These items can be chunks of information, but the short-term memory is quickly overloaded (Sweller, Ayres and Kalyuga, 2011) and strategies, such as rehearsal, that do not overwhelm its capability, are required to enable it to become the flexible workspace for processing that enables reading. It is here that words must be integrated and where the comprehension process is initialised (Rayner and Pollatsek, 1989).
The short-term memory’s capacity and flexibility is greatly enhanced by the ability to retrieve information stored in the long-term memory thus relieving the cognitive demands and preventing overload (Sweller, Ayres and Kalyuga, 2011). Information stored in the long-term memory takes the form of episodic memories of sequences and events and semantic memory of facts and knowledge (Donaldson and Tulving, 1972). It is the semantic memory that is most utilised when reading. Recognition of letters, letter patterns, sound-to-letter correspondences, words, meanings, idioms, metaphors and contextual knowledge are all accessed from the long-term memory and thus release the load on the working memory enabling it to concentrate on meaning and comprehension. This has implications not only for competent readers but also for emergent readers who are learning to decode words using phonics strategies.
Whilst at the decoding stage of reading development, the working memory of a learner is recognising letters and letter combinations but accessing the attendant sounds associated with those letters from long-term memory thus enabling the working memory to then combine and blend those sounds. The word may then be retrieved from long-term memory or, if not present in the brain’s lexicon, merely be read without meaning. As the speed of decoding increases, the letter patterns are accessed and recognised more rapidly from long-term memory, the word superiority effect (Reicher, 1969) becomes apparent and reading automaticity emerges. At this point in a reader’s development, the long-term memory is privileging legitimate sound to letter pattern correspondence to attempt valid pronunciation. Access to the lexical pathway (Dehaene, 2011) will be restricted by vocabulary knowledge, which for early readers will be necessarily constrained by chronological development. This may manifest itself in the regularly observed (and unfairly criticised) ‘barking at print’ (2012, michaelrosenblogspot.com) phenomenon with the resultant absence of speed, accuracy and prosody associated with reading fluency.
Once decoding has become automatic and the decoding of words swift, the long-term memory supports reading through rapid deciphering along with lexical recognition and identification. This releases working memory to attend to the prosody, rate and accuracy associated with fluent reading as well as emerging comprehension and the use of contextual cues to decipher the likely meaning of unknown words (not the decoding of these word). When encountering longer unknown words with more complex phonetic patterns, the cognitive process will revert to its decoding paradigm evidenced by the slower sounding out of phonemes and associated blending of syllables to decipher the word. This phenomenon is evident even in very experienced readers (Dehaene, 2011).
The assumption of information processing theory is that there exist discrete processing stages, and only when processing is complete within one stage is the information transferred to the contiguous stage. The validity of the assumption is based on Sternberg’s (1969) study on mental processing measuring response times to stimuli and conclusion that a clear ‘stage’ prior to short-term memory search had been identified and the inference that other stages were thus implicit within an information processing model. Sternberg’s research has been widely retested (Jacobs, Hwang, Curran, and Kahana, 2006; Borst, Ghuman and Anderson, 2016; Anderson et al., 2016)and remains robustly accepted within the field of cognitive psychology.
In line with the growing advances in neuroscientific research, the top-down processing models of reading espoused by Goodman (1970) and Smith (1971) have become increasingly unconvincing. These models suggest that the part of the system that constructs meaning (the top) controls the flow of data and that the passive flow of information through the processing system is sluggish and often hindered by bottlenecks which force the reader to retard the reading progress. This retardation is eased by reference to information stored in long-term memory which enables the formulation of hypotheses. These hypotheses are formed by reference to the long-term memory for relevant syntactic, semantic, phonological and contextual cues enabling the reader to make a tentative choice as to the word. The hypothesised word is then tested against prior context of the text. If the reader is successful in ‘guessing’ the word, it is held in ‘medium-term memory’ (Goodman, 1970); if the reader is unsuccessful, previous text is fixated once more and a further guess is made.
The model is critically flawed in four main areas. Firstly, reading a word becomes reliant on context and thus constrained by what has been previously analysed. It is thus not clear as to how the process is initiated before sufficient context is established to proceed fluently and is unclear and unspecific on the analysis of graphic cues. Secondly, skilled readers do not utilise hypothesis testing as a strategy or have a reliance on contextual information (Stanovich, 1980) but exploit the rapid and efficient analysis of text. Guessing words is almost always the reserve of poor readers (Stanovich, 1980). Thirdly, a central feature of the model is that that the process of reading is similar for both emergent readers and established readers. The establishing of the word superiority effect (Reicher, 1969) as a fundamental difference between early readers and skilled readers undermines and discredits this premise. Finally, and crucially, the model fails at the point the word is identified. In the Goodman (1970) model the word is identified after parsing and interpreting the sentence to date, guessing the word and then checking whether the hypothesised word is compatible with the prior context. Only at this late point in the process is the word established as ‘read’. There is little evidential support for this supposition (Rayner and Pollatsek, 1989) and it is undermined by Dehaene’s (2011) MRI scanning research which indicates that for this to be the point of word recognition is incompatible with the speed of processing required for fluent reading along with Gough and Hillinger’s (1980) assertion that a linguistic guessing strategy is only employed at the very earliest stages of text deciphering and is not efficient enough for reading fluency.
The recognition that information processing during reading is rapid (Dehaene, 2011) has discredited top-down models and favours the bottom-up models of reading which argue that processing is fast and that information flows through a series of stages without the influence of contextual hypothesising or higher order processing strategies until inferential comprehension is required. The most influential model, proposed by Gough (1972), built upon Atkinson and Shiffrin’s (1968) work. It recognised that at the first fixation point text enters the iconic memory which acts as a buffer whilst the letters in the word are recognised in a series of saccades (Williams,1979) and letter pattern recognition routines are consulted in long-term memory. These characters are mapped onto a string of systematic phonemes and analysed against grapheme-phoneme coding knowledge with the character of meaning (the word) held in inner speech (Rayner and Pollatsek, 1989) until identified as a plausible option after reference to the brain’s lexicon. Fixated words are held in short-term memory until the sentence is parsed and comprehension initiated. Gough’s model is exemplary due to its clarity in explaining the stage by stage cognitive processing perspective of reading (Tracey and Morrow, 2012). Because the model made clear predictions it could be tested and refined and remains the most robust model of information processing of written words, highlighting the limited role played by context in lexical access.
LaBerge and Samuels (1974) developed a more advanced model, the Automatic Information Processing Model, which added, after letter recognition by the iconic memory, a reference to the phonological memory which enabled sounds to be associated with the visual image before receiving the attention of the episodic and semantic memory resulting in correct word identification. They developed the concepts of external and internal attention. External attention being observable evidence of reading and internal attention being the unobservable elements of cognitive alertness (how cognitively vigilant the reader is and how much effort is being applied), selectivity (what experiences the reader is drawing on to comprehend the text) and capacity (how cognitively attentive the reader actually is). Labergere and Samuels (1974) also introduced the concept of automaticity: the ability to perform a task whilst devoting little attention to the commission. In the case of reading, this relates specifically to the swift and accurate decoding of text with almost no imposition on working memory with the resultant benefit that almost all attention is available for comprehension. The implication that there is a crucial stage buffering decoding mastery and reading fluency is central to the understanding of development in readers for whom phonics mastery has not been achieved.
Attempts have been made to integrate both top-down and bottom-up models (Rumelhart, 1977; Just and Carpenter, 1980) into an interactive model of reading that suggests that readers employ information from multiple sources simultaneously during the reading process (Tracey and Morrow, 2012). Syntactic, orthographic and lexical information is concurrently processed and fused in a mutual interpretative interaction that is neither top-down or bottom-up. Stanovich (1980) extended the interactive model to integrate the concept that not only were text processors interactive and non-linear but that they were also compensatory.
Stanovich (1980) concluded that if one processor was not working or had insufficient data, then another processor would compensate for the failure. He elucidated this through the example of a partially blurred text where the orthographic processor had insufficient data and relied upon the syntactic processor to garner suggested meaning. Although interactive models rationalised the multi-cuing approach to the teaching of reading, these models were undermined by research into eye movements (Fisher and Shebilske, 1984) that highlighted the inadequacies of fixation times required to support the efficacy of the models. Nonetheless, the models hold true for poor readers who are unable to decode new words and privilege (or have been taught) logographic, semantic and contextual strategies at the expense of phonic strategies. Stanovich’s (1980) concept of compensatory processing greatly enlightens the implications of non-mastery of phonics decoding in early primary schooling and the increasing evidence of reliance of word guessing in the upper primary stages of education. The results of the UK Department for Education’s Year 3 Phonics Pilot (DfE, 2017) suggest that the problem may have considerable implications for future literacy rates.
This blog is number 14 in a series of blogs.
You may be interested in reading:
Follow the reading ape on twitter @thereadingape
Follow the reading ape on twitter - @thereadingape