Key concepts in ELT

‘Bottom-up’ and ‘top-down’

Assist. Lect. Huda Abd Alkareem Zhgair

In accounts of foreign-language listening and reading, perceptual information is often described as ‘bottom-up’, while information provided by context is said to be ‘top-down’. The terms have been borrowed from cognitive psychology, but derive originally from computer science, where they distinguish processes that are data-driven from those that are knowledge-driven.

Underlying the metaphors ‘top’ and ‘bottom’ is a hierarchical view of the stages through which listening or reading proceeds. In listening, the lowest level (i.e. the smallest unit) is the phonetic feature. A simple analysis might present the listener as combining groups of features into phonemes, phonemes into syllables, syllables into words, words into clauses, and clauses into propositions. At the ‘top’ is the overall meaning of the utterance, into which new information is integrated as it emerges. Drawing on this concept of levels of processing, many ELT commentators present a picture of listening and reading in which bottom-up information from the signal is assembled step by step, and is influenced throughout by top-down information from context.

The truth is rather more complex. First, it is not certain that bottom-up processing involves all the levels described. Some psychologists believe that we process speech into syllables without passing through a phonemic level; others that we construct words directly from phonetic features. Nor does bottom-up processing deal with one level at a time. There is evidence that in listening it takes place at a delay of only a quarter of a second behind the speaker—which implies that the tasks of analysing the phonetic signal, identifying words, and assembling sentences must all be going on in parallel.

A quarter of a second is roughly the length of an English syllable—so the listener often begins the processing of a word before the speaker has finished saying it. The listener forms hypotheses as to the identity of the word being uttered, which are said to be activated to different degrees according to how closely they match the signal. The candidates compete with each other until, when the evidence is complete, one of them outstrips the rest.

Like ‘bottom-up’ processing, ‘top-down’ is more complex than is sometimes suggested. Contextual information can come from many different sources: from knowledge of the speaker/writer or from knowledge of the world; from analogy with a previous situation or from the meaning that has been built up so far. It can be derived from a schema, an expectation set up before reading or listening; it can take the form of spreading activation, where one word sparks off associations with others; or it can be based upon the probability of one word following another. It is important to specify which of these cues is intended when the expression ‘top-down’ is employed. Also unspecified in many accounts of L2 reading and listening is the way in which bottom-up and top-down processes interact. Does one occur before the other, or do they operate simultaneously? The evidence from LI research is contradictory. Some findings suggest that contextual information is invoked before perception, helping us to anticipate words; others, that it becomes available during the perceptual process; others, that it is only employed after a word has been identified. Goodman’s much-quoted view (1970) that successful readers guess ahead using current context has not been conclusively demonstrated.

Some researchers argue for completely interactive models of listening and reading, in which top-down and bottom-up processes extend simultaneously through all levels. In support of such models, they cite evidence of word superiority effects, where knowledge of complete words influences the way we perceive sounds or letters. This kind of effect is appropriately described as ‘top-down’ since it involves knowledge at a higher level affecting processing at a lower. So note that the term ‘top-down’ is not always synonymous with ‘contextual’.

Finally, the vexed question of the use of bottom-up and top-down information by foreign-language learners. A truism of ELT has it that low-level listeners and readers become fixated at word level, and do not have enough spare attentional capacity to construct global meaning. In truth, learners appear to make considerable use of top-down information: employing it compensatory to plug gaps where their understanding of a text is incomplete. The best account of this process is mechanism (1980), originally formulated for LI reading. Stanovich suggests that we use contextual information to make up for unreliability in the signal (bad handwriting, for example, or ambient noise). The more flawed the bottom-up information, the more we draw upon cues from top-down sources. This seems to describe accurately the way in which L2 learners resort to top-down inferencing when understanding is impaired by limited vocabulary or syntax. The strategy may be more common in listening than in reading: see Lund (1991).