Reading Models: Bottom-Up and Top-Down

There has been extended controversy among reading authorities about the approach to use in teaching beginning reading, an either/or kind of controversy, the essence of which centres upon whether the first emphasis in word recognition instruction should be phonics- or meaning-based. In the one view of reading, learners are perceived as being almost passive decoders of visual stimuli, while in the other, learners are viewed as active participants who construct their own encodings.

It is useful to think of these two views of teaching reading as a dichotomy, as falling into two main camps. One camp consists of authorities who contend that the process of reading begins with letters and their sounds (phonics). These experts support what is termed a code-emphasis, text-driven or "bottom-up" model to explain the reading process. The other camp, in opposition, consists of those authorities who perceive reading as chiefly "externally guided", who subscribe to a hypothesis-test, or "top-down" model of the reading process. In the media, the implications of these two opposing views are often polarized in terms of what they mean for instruction, represented by: (1) those who advocate a phonics approach to teaching beginning reading and (2) those who would prefer meaning as the base, often critics maintain, fostering a "look-say" approach.

To elaborate, Gough (1972) proposes what may be classified as a phonics-based or "bottom-up" model of the reading process which portrays processing in reading as proceeding in serial fashion, from letters to sounds, to words, to meaning, in the progression suggested in the accompanying figure.

Taken from this perspective, the implications for reading instruction are that students need to begin reading by learning the letter names, associating the letter names with their sounds, and then be shown how to blend these sounds together into words. Stated in Gough's more technical terms the reading system, from a "bottom-up" perspective, functions in sequence as follows. First, the graphemic information enters through the visual system and is transformed at the first level from a letter character to a sound, that is from a graphemic representation to a phonemic representation. Second, the phonemic representation is converted, at level two, into a word. The meaning units or words then pass on to the third level - TPWSGWTAU (the place where sentences go when they are understood) and meaning is assimilated into the knowledge system. Input is thus transformed from low-level sensory information to meaning through a series of successively higher-level encodings, with information flow that is entirely "bottom-up", no higher level processing having influence on any lower level processing (Rummelhart, 1977). This process is also referred to as "data-driven" (Bobrow and Norman, 1975).

But other theorists disagree. For them, efficient reading does not result from the precise perception and identification of all the elements in a word, but from skill in selecting the fewest, most productive cues necessary (Goodman, 1970). These authorities contend that readers have a prior sense of what could be meaningful in the text, based upon their previous experiences and their knowledge about language. Readers are not, in their view, confined only to one source of information - the letters before their eyes, but have at their disposal two other important kinds of information which are available at the same time: semantic cues (meaning), and syntactic cues (grammatical or sentence sense).

Thus according to theorists in this camp, what readers bring to the text separately in terms of both their prior knowledge of the topic and their knowledge about language, assists them in predicting what the upcoming words will be. Readers sample the print, assign a tentative hypothesis about the identity of the upcoming word and use meaning to confirm their prediction. If meaning is not constructed, the reader resamples the text and forms a new hypothesis. Thus readers need only briefly sample the "flutters" on the page in order to confirm word identity.

In this model it is evident that the flow of information proceeds from the top downward so that the process of word identification is dependent upon meaning first. Thus the higher level processes embodied in past experience (semantics) and the reader's knowledge of the language pattern (syntax) interact with and direct the flow of information (Stanovich, 1980), just as listeners may anticipate what the upcoming words of speakers might be. The figure depicts fluent readers as actively engaged in predicting or hypothesis-testing when progressing through text. This view identifies reading as a kind of "psycholinguistic guessing game" (Goodman, 1970).

A representation of the "top-down" process is depicted in the following figure.

Summary

In general, it may be said that there are two opposing theories regarding what is involved in the reading process. One theory envisions reading as a data-driven process ("Bottom-up") in which: (1) letters are transformed into phonemic representations; (2) phonemic representations are then transformed into word representations; (3) words are next assigned meaning; (4) words are combined into meaning-bearing sentences; (5) meaningful associations are formed, and (6) information is finally stored. The contrasting theory views reading as a "top-down" process in which higher level conceptual processes direct word recognition and the reader: (1) samples the print; (2) makes predictions as to what the word might be based upon prior knowledge of the topic and sentence sense; (3) reads to confirm the hypothesis; (4) constructs meaning; and (5) assimilates new knowledge.