Models of the Reading Process 

By Murray Peglar B.A., B.Ed


To be able to teach reading, it is important to understand what happens when we read. 

     Reading basically involves transforming a text, which is a graphic representation, into thought, or meaning.  It used to be thought that this was simply a matter of combining letters into words, words into sentences and sentences into meanings.  However, over the last thirty years, psychologists and linguists, using a variety of experimental techniques, have discovered that things are much more complex.  Several models of the reading process have been put forward to account for the experimental findings.  A key element in explaining reading is the amount to which what the brain already knows affects perception of what is being read (top-down processing).  This idea was initially thought to be in contrast to earlier ideas that reading was a linear progression from page to understanding (bottom-up processing), but newer research seems to indicate that both elements play important parts in reading.

 The following sections outline some very important research and ideas in our understanding of reading:

Kenneth Goodman

     In the early 1960s Kenneth S. Goodman began studying the reading of authentic texts by urban and rural young people. His earliest miscue research, published in 1965, is probably the most widely replicated study in reading research history. But it was his article, "Reading: a Psycholinguistic Guessing Game" (1967), that began a revolution moving away from a view of reading as rapid accurate sequential word recognition to an understanding of reading as a process of constructing meaning - making sense - of print. That research is part of the basis for the whole language movement and disagreements over his conclusions about the nature of reading fuel the current "reading wars." (Stenhouse Publishers, 2003)

     Goodman defined reading as: “a receptive psycholinguistic process wherein the actor uses strategies to create meaning from text” (Goodman, 1988).  Basically, the study of reading looks at translating a linguistic surface representation (text) into thought.  Goodman based much of his theory on analysing miscues (mistakes) in texts being read-aloud.  He believed that efficient readers minimize dependence on visual detail, but focused his theories on the interactions of reader and text.  Basic physical sensory information (the physiological process) is cycled into deeper levels of cognitive processes.

Cycles– readers move from text to understanding through cycles of deeper processing, moving from optical, to perceptual, to syntactic, to meaning

Cognitive Processes of the brain used in reading are:

  • recognition / initiation – the brain must recognise text and initiate reading
  • prediction – anticipates and predicts as it seeks order and significance of input
  • confirmation – verification of predictions or disconfirmation
  • correction – reprocessing when it finds inconsistencies or disconfirmations
  • termination – formal ending of reading act
N.B.:  Goodman treats these processes as sequential, whereas later models may not

      This limited view, however, was still an improvement upon Noam Chomsky’s ‘generative grammar’, which lacked explanation of top-down processing.  Goodman also promoted the use of ‘natural texts’, believing that language must be studied in context.  This follows from his postulated three sources of linguistic information: symbols (characters), language structure (syntax), and semantic (meaning).


David Rumelhart

     Rumelhart helped develop the field of cognitive science in the 1970’s with his work on long term memory and semantic mapping in the mind. He improved upon Goodman’s model by creating a non-sequential model that relies heavily on the use of schemata and top-down processing for explaining understanding.


“[can delineate] in a general manner, without limitation to any single determinate figure as experience, or any possible image that I can represent in concreto” (Kant, 1781).
“abstract structure of information” (Anderson, 1984)
“meanings [encoded] in memory in terms of the typical or normal situations or events that instantiate the concept” (Rumelhart, 1980)
     A schema filled in with default values is called a prototype. Whereas a schema is an organized abstract framework of objects and relations, a prototype consists of a specified set of expectations. A prototype is a highly typical instantiation or instance of a schema (Langacker, 1987).  If the instantiation (example) matches our schema (idea), we comprehend.  If understanding does not occur, we can infer that the text does not have enough clues, or that the reader does not have the appropriate schema.  Learning involves creating or changing schemata through:
  • accretion – filling in variables in general and specific schemata
  • tuning – changing the constraints on one variable
  • restructuring –building new schemata based off old models
     There is room for flux in a perceived schema, as variables can compensate for missing or altered factors.  However, pre-reading to activate a schema may not really help because schemata are still relatively fixed and solid, especially in common or familiar areas.  Quick introductions may not undo years of solidifying schemata.
We can therefore think of schemata in terms of our:
  • Play schemata: with a script (schema) that is interpreted (instantiations)Theory schemata: a predictable and useful reality is represented and continually recreated by the sum of our schemata
  • Procedure schemata: with (limitless?) subsets of meanings and processes
  • Parse schemata: they determine how “legal” a situation is (whether it fits with meaning) 
     Though schema theory would seem to explain only top-down and internal processing, it also operates at lower levels, using ‘feature-detectors’ that confirm attributes to interpret sensory data.  These sensory schemata then activate higher and higher level schemata, eliminating erroneous possibilities, and narrowing understanding to the appropriate meaning.


Rayner and Pollatsek

     In their foundational work on reading psychology, Raynor and Pollatsek used experimental evidence to show that neither top-down, nor bottom-up theories in isolation can fully account for reading data.  This continues to be the main issue in modern research: how to intersect bottom-up and top-down models.  Though Raynor and Pollatsek, and Clarke (see Short Circuits), have used experimental research to show these two theories pointing to an intersection that is ‘reading’, how these models can work together is still unknown.  They have, however, created a detailed model of sentence reading that takes into account the interactions of initial encoding, long term memory / knowledge, and the active processes of working memory and parsing.

     In this model, the parser (the part of the brain that analyses sentences for structure) is seen as a purely syntactic device.  It uses input from the lexicon (personal vocabulary of language and morphemes) to produce a structural representation for the sentence.  The parser uses the principles of minimal attachment and late closure.

     An example of minimal attachment is illustrated by Rayner and Pollatsek (1989) in the sentences, "The girl knew the answer by heart" and "The girl knew the answer was wrong". The minimal attachment principle leads to a grammatical structure in which "the answer" is regarded as the direct object of the verb "knew". This works for the first sentence, but not the second, illustrating the effect of late closure having a bearing on the grammatical structure.

     They also assume that the nature of temporary storage in the working memory is phonological.  Therefore, if comprehension fails, the inner speech module can replay the message.  There is little mention of details about how meaning is represented.

     Though there is a detailed mapping of cognitive processes during reading, Raynor and Pollatsek also found that good readers are able to recognise lexical forms at a processing speed faster than the time required to activate context effects and conscious predicting.  Thus, their theories present a more integrated approach, involving both bottom-up and top-down processing, as “the interactive models, attempting to be more comprehensive, rigorous and coherent, give emphasis to the interrelations between the graphic display in the text, various levels of linguistic knowledge and processes, and various cognitive activities” (Weber, 1984).


Short Circuits and Reading

     Though one would assume that good readers would use larger chunks of text, and rely on semantic (meaning) cues rather than syntactical (grammar) ones, and that these differences would hold for L1 and L2 reading, some surprising evidence has been found.  Strong L1 readers did rely more on semantic cues, and weak readers more on syntactical, however, both used syntactic cues equally in L2.  During oral reading miscue tests, differences between strong and weak readers also diminished, though the types of mistakes made were different, with strong readers making more semantically acceptable miscues.  What this means is that good L1 readers appeared less able to use their reading strategies in L2.  It is hypothesised that “limited control over the language ‘short circuits’ the good reader’s system causing him/her to revert to poor reader strategies” (Clarke, 1980), in difficult L2 tasks.  This creates short circuits (gaps in reading understanding) in the following situations:
  • good L1 strategies + poor L2 competency = poor L2 reading
  • good L1 strategies + good L2 competency = good L2 reading
     Thus, it appears that strategies and behaviours, not necessarily knowledge, have a large effect on reading abilities.  Strategies that Clarke indicates as being useful are: ‘concentration on passage-level semantic cues; the formulation of hypotheses about the text before reading, then reading to confirm, refine, or reject those hypotheses; the de-emphasis of graphophonic and syntactic accuracy, that is, developing a tolerance for inexactness, a willingness to take chances and make mistakes’.  This being said, the importance of language skills for effective reading should not be undermined.  Especially for weaker L1 readers, explicit teaching of strategies, as well as language, would be appropriate, whereas stronger L1 readers may only need reminders of effective reading strategies.

Short Circuit  – any reading that does not end with meaning

  • letter naming – spelling out words
  • recoding – print is matched to another code (ie: sound) with no meaning
  • syntactic nonsense – approximating understanding when the load is too great
  • partial structures – alternating periods of productive reading creating partial understanding
(Goodman, 1984)


Teaching Implications

     The balance between top-down, and bottom-up processing, though identified as complimentary, is still somewhat nebulous.  Therefore, much of the recommended teaching practice based on these theories still centre around exercises that isolate and improve top-down and bottom-up skills.  Patricia Carrell (1987) has categorised some such exercises:

Bottom-Up Exercises:

  • Grammatical Skills - basic grammar awareness will, of course, help in reading comprehension, but decoding skills should also include learning cohesive devices (substitution, elipsis, conjunction, and lexical cohesion)
  • Vocabulary Development - with the introduction of schema theory, vocabulary acquisition is now seen to involve deeper understanding of words and their contexts, and should thus be taught with an eye to quality, not quantity of learned words.
Top-Down Exercises: 
  • Schema Activation - by building background knowledge, we can increase students' understanding of texts.  Cultural and experiential knowledge gaps can create the impression of a language barrier, when it is simply that the student lacks the appropriate schema.  Pre-reading exercises, realia in the classroom, bit-by-bit exposure to text, visual representations, semantic mapping, sub/superordinating, and comparisons with previous knowledge are all ways to create understanding of the concept before the language.  For specific approaches, see also:


References - Literacy Strategy - Teaching Reading - Whole Language Reading Instruction - ERIC Digest - Reference Reviews on Machine Learning, Stenhouse Publishers “About the Authors” (1997-2003)
Rayner, K., & Pollatsek, A. (1989) The Psychology of Reading. Hillsdale, NJ: Lawrence
Erlbaum Associates.
Goodman, Kenneth S. (1988) in Carrell et al. Interactive Approaches to L2 Reading Cambridge, CUP

All webpages last visited on March 26 / 03

This website last modified April 03 / 03