SYSTEM

www.elsevier.com/locate/system

System 32 (2004) 363–377

An insight into listeners’ problems: toomuch bottom-up or too much top-down?

John Field *

Universities of Leeds and Reading, UK

Received in revised form 27 April 2004; accepted 12 May 2004

Abstract

Difficulty in the early stages of second language listening is sometimes said to derive from

too heavy a reliance upon bottom-up information. Less experienced listeners supposedly focus

so much attention upon identifying sounds and words that they have no time or mental ca-

pacity left for building higher-level units of meaning. However, there is contrary evidence

which indicates that non-native listeners make considerable use of top-down processes. This

paper suggests that listening to a foreign language may be assisted by an interactive-com-

pensatory mechanism already available in L1, which compensates for gaps in understanding.

Two major questions are then raised: If top-down and bottom-up information are in apparent

conflict, which one prevails? And how do learners deal with new items of vocabulary when

they crop up in a listening passage? Three experiments attempted to find answers.

� 2004 Elsevier Ltd. All rights reserved.

Keywords: Listening; Bottom up; Top down; Language processing; Context; Speech signal; Lexis

1. Background

1.1. ‘Bottom up’ and ‘top down’

The terms ‘bottom up’ and ‘top down’ occur frequently in the literature on second

language listening and reading. They are often used to mark a distinction between

* Tel.: +44-20-7483-4568.

E-mail address: jcf1000@dircon.co.uk (J. Field).

0346-251X/$ - see front matter � 2004 Elsevier Ltd. All rights reserved.

doi:10.1016/j.system.2004.05.002

364 J. Field / System 32 (2004) 363–377

information derived from perceptual sources and information derived from con-

textual ones. 1 Strictly speaking, however, the terms refer not to particular levels of

processing but to directions of processing. In a ‘bottom-up’ process, small (‘lower

level’) units are progressively reshaped into larger ones; in a top-down process, larger

units exercise an influence over the way in which smaller ones are perceived. Con-

sider, for example, the vocabulary effects which potentially occur in both first andsecond language listening, where the listener’s interpretation of a string of phonemes

is constrained by the knowledge that a particular word exists. They qualify as a top-

down process, since information from one level (the word) shapes the interpretation

of information at a lower level (the phoneme). I shall return to this point in due

course.

The term ‘contextual’ as used in relation to ‘top-down’ processing is also some-

what misleading. Writers use it to refer to the impact of world knowledge upon

processing; but they also use it to refer to the impact of information gleaned fromearlier content in the conversation or reading passage (what Brown and Yule, 1983:

46 term ‘co-text’). This paper considers both types of context but attempts to keep

the two distinct.

A further confusion arises when ‘bottom up’ and ‘top down’ processing are

represented as if they were alternatives. This is clearly not the case. The kind of co-

textual information that might be used ‘top down’ to assist interpretation can only

be achieved by, in the first instance, processing what is actually there in the input. To

give a simple example of the power of perceptual data, assume that, early in aconversation, an L2 listener mishears I won’t go to London as I want to go to London.

He/she is likely to construct incorrect expectations as to the direction that the

conversation will take, and may even be at some pains to reshape what comes next to

make it conform to this early misunderstanding.

The relationship between ‘bottom-up’ and ‘top-down’ processing is a complex one

based upon a considerable degree of interdependence, a fact that has been recognised

by a number of recent commentators (e.g., Tsui and Fullilove, 1998). What is at issue

when investigating this aspect of second language listening and reading is not whichpath is chosen but which of the two processing routes is preferred over the other. To

put it another way: If top-down evidence conflicts with bottom-up, which of the two

is the second language learner most likely to trust? This is the central issue addressed

in the series of pilot experiments reported here.

1.2. ‘Bottom up dependency’

One established view of the problems faced by the second language listener orreader takes the following form: Weaker second language learners worry about not

1 Greatly preferable are the terms lower-level processing (for decoding what is in the speech stream) and

higher-level processing (for the building of meaning). However, they have to be used with some

circumspection in discussions of L2 listening since they easily become confused with references to lower

and higher levels of learner (indicating degree of competence in the target language).

J. Field / System 32 (2004) 363–377 365

understanding each word of the input. They focus their attention at word level and this

occupies much working memory capacity, preventing them from building the words into

higher-level meaning.

A slightly different take on this ‘bottom-up dependency’ view is provided by

Gernsbacher’s (1990) Structure Building theory, based on findings in first-language

reading. Gernsbacher suggests that it is a characteristic of less skilled readers thatthey build small-scale units of meaning and are unable to integrate these units into

larger ones. It is interesting to consider whether this might also apply to the per-

formance of L2 listeners with a limited knowledge of the target language.

Certainly, there is ample evidence of learners with limited L2 competence drawing

heavily upon perceptual data. After testing 235 learners for both detailed and global

understanding of academic material, Hansen and Jensen (1994, p. 265) interpreted

their findings as ‘indirect evidence that low proficiency students rely heavily on

bottom-up processing skills’.However, there is also a contrary line of research, represented in (e.g.) Long

(1989), which takes it as axiomatic that top-down information plays an important

role in the final interpretation that is derived by the second language listener, and

maintains that such information supports the weaker listener as well as the more

advanced.

This paper attempts to establish more precisely the relationship between the raw

evidence extracted from the speech signal by an L2 listener and external information

drawn from (a) the listener’s world knowledge and (b) the listener’s recall of whathas been said in the conversation so far.

1.3. Evidence from studies of L2 listening

A number of researchers have studied the relationship between higher and

lower level processes in L2 listeners. Conrad (1983) concludes that non-native

listeners direct more attention to syntactic information in the speech stream than

do native listeners and less attention to semantic. However, her results should betreated with some caution. Her criteria for distinguishing between ‘syntactic’ and

‘semantic’ are subjective; and she does not consider factors such as the relative

saliency of different words in her spoken texts or the vocabulary knowledge of

her L2 subjects.

Mack (1988) found that non-native listeners made fewer syntactic errors of per-

ception than did native listeners, but more semantic ones. However, this was not the

main point of the research. The experiment featured anomalous sentences (A painted

shoulder thawed the misty sill.) – depriving her subjects of the top-down informationnormally provided by background knowledge or by co-text. Her non-native subjects

were, overall, extremely inaccurate, and Mack suggests that, if they had been ac-

customed to processing word-by-word, they would not have been nearly as disad-

vantaged by the absence of contextual information as they proved to be. A similar

conclusion was reached by Wolff (1987), who found that his subjects were more,

not less, inclined to use top-down strategies when they were given a harder text to

understand.

366 J. Field / System 32 (2004) 363–377

Some well-designed research by Koster (1987) investigated the extent to which

listeners rely on ‘top-down’ information provided by lexical association. One ex-

periment explored whether subjects were able to identify a word more easily when it

was preceded by one with which it collocated closely. Koster found, not surprisingly,

that predictability aided recognition to a significant degree. But the most interesting

feature of the experiment was that, of three groups of subjects (intermediate non-native listeners, advanced non-native listeners and native listeners) it was the first

whose recognition of the second word improved most when the co-text provided

assistance. Koster replicated this result with a more on-line task, checking Reaction

Time in a word/non-word decision. He reached a conclusion which runs counter to

the idea of bottom-up dependence: ‘non-native subjects appear to make use of

context in much the same way as native subjects, with one important difference;

when enough time is available for processing, the effect of context is potentially

greater for non-natives than for natives’. (p. 56)Further evidence of L2 listeners’ use of contextual information comes from a

study by Mueller (1980) which indicated that visual support enhanced the compre-

hension of lower-level students much more than it did that of higher-level students.

Voss’s series of transcription tasks (1984) also indicated a strong dependence on

top-down processing, though Voss concluded that over-reliance on bottom-up in-

formation underlies much breakdown of understanding. Long (1990) took as a

theoretical assumption the fact that non-native listeners did indeed use top-down

information, and her results indicated an important effect of background knowledgeon ability to comprehend listening texts.

In sum, much of the evidence from listening appears to favour the view that

low-level second-language users rely heavily upon contextual and co-textual in-

formation. Their reasons for doing so are suggested by the most extensive in-

vestigation of the ‘bottom up’/top down’ issue to date. It was undertaken by Tsui

and Fullilove (1998), who analysed answers given by 20,000 Hong Kong exam-

ination candidates to different types of listening question. The researchers dis-

tinguished items where top-down schematic information might be used to supportdecoding from those where it could not. They traced a correlation between level

of listening skill and success in answering the items which were not schematically

supported. This suggested that it was the less skilled listener who relied most

heavily upon top-down processes, and that he/she did so in order to compensate

for problems of perception. Tsui et al. found evidence that weaker listeners were

sometimes misled by false assumptions based upon contextual cues (compare the

similar findings of Field, 1997).

This finding on the compensatory use of ‘top-down’ information carries aninteresting echo of current perspectives on L1 reading. Researchers such as

Perfetti (1985) have demonstrated that weaker readers often fall back on con-

textual and co-textual evidence because their decoding skills are insufficiently

developed.

Evidence that phoneme and word recognition are indeed a major source of dif-

ficulty for low-level L2 listeners comes from a major study by Goh (2000). Of 10

problems reported by second-language listeners in interviews, five were connected

J. Field / System 32 (2004) 363–377 367

with perceptual processing. Low-level learners were found to have markedly more

difficulties of this kind than more advanced ones.

1.4. Evidence from studies of reading

The evidence from second language reading supports the ‘bottom-up dependency’view rather more strongly that that from listening. Indeed, the notion that language

learners focus too heavily upon details of the signal may have originated in reading

studies. The two skills are clearly very different at the perceptual level, but, given the

parallels between them at the conceptual level, it is worthwhile reviewing the evi-

dence briefly.

Clarke (1980) takes the view that the attention of lower-level readers is so focused

upon decoding that they are unable to transfer into L2 the kind of higher-level

processing that comes naturally to them in the native language. Cummins (1979)advances the similar ‘threshold’ theory that a minimum language competence is

necessary before effective use of higher-level processes can be made. Further ‘bottom

up’ evidence comes from Cziko (1980) who analysed the errors made by learners of

French at two levels when they read aloud. He concluded that those less competent

in the language made less use of contextual information and showed signs of a de-

pendence upon the words of the text. One has, nonetheless, to exercise some caution

in drawing hard-and-fast conclusions about general reading skills from a perceptual

task such as reading aloud.Not all reading researchers are in agreement. Stanovich (1980) adopts a very

different view (in this case, of L1 reading 2). He proposes that the relationship be-

tween top-down and bottom-up information is regulated by an interactive-compen-

satory mechanism. He argues that, when a message is degraded in some way (for

example, by bad handwriting), readers automatically compensate by relying more

heavily than normal upon contextual clues. In other words, the relationship between

input and context is not a constant one but can be varied according to the listener’s

confidence as to the reliability of each. It seems reasonable to suppose that theprinciple can be extended to the case of an individual who is listening in conditions of

noise. This, as we shall see, can provide a useful perspective on L2 listening.

The interactive-compensatory view receives support from the work of Perfetti

(1985), referred to above. Perfetti believes that weaker L1 readers make considerable

use of contextual cues and that their problems derive chiefly from slower access to

word meanings because of their inability to decode accurately. He, like Stanovich,

argues that higher-level information is used compensatorily – not to reinforce

meaning that has already been derived from a text but to restore those parts whichhave not fully been understood.

2 Writers such as Paran (1996) have drawn parallels between the weak L1 readers of Perfetti’s studies

and novice L2 readers.

368 J. Field / System 32 (2004) 363–377

1.5. Conflicting evidence; possible solutions

There would, then, appear to be conflicting views as to how much use second

language learners make of top-down information. Two possible reasons might ac-

count for this lack of consensus.

1.5.1. Reading vs listening

It may be significant that much of the evidence favouring ‘bottom-up dependency’

appears to come from reading rather from listening. Lund (1991) compared the two

experimentally, and found support for the idea that ‘while the available compre-

hension processes are the same, the strategies for their application may vary sig-

nificantly with the modality’. Specifically, he found that readers recalled more detail

than did listeners; but that listeners focused more on main ideas and were more

willing to construct a plausible context to compensate for insufficient understandingof the text. This confirmed his hypothesis that ‘listeners rely more on top-down,

schema-based processing than readers’. His view is echoed by Forster in an impor-

tant paper (1989, p. 93) which distinguishes a ‘bottom-up’ and a ‘crossword’ mode of

processing, the latter using top-down information compensatorily. Forster specu-

lates: ‘With speech inputs, it may be much more likely that the crossword mode will

prove to be the superior system...’.

1.5.2. The legacy of scripted materials

The ‘bottom-up dependency’ view developed at a time when most L2 listening

materials were scripted and graded to reflect the level of the learner. Learners de-

veloped an expectation of understanding nearly everything in a text, and a belief that

comprehension was incomplete unless every word had been recognised. This, of

course, did not reflect the kind of listening that took place outside the classroom,

where learners by no means understood everything and had to adopt a much more

strategic ‘top down’ approach, based upon identifying fragments of what was said

and constructing inferences as to the ideas which linked them. The increased use ofauthentic texts has led to a realisation that much of L2 listening involves only partial

recognition at word-level, and to a more enlightened attitude by teachers, who now

reassure learners that understanding every word is not a prerequisite. On this ar-

gument, then, bottom-up dependency may have been a product of instructional

methodology and may never have been characteristic of real-life listening.

The foregoing comments support the view that top-down information is likely to

be used more rather than less by lower level learners. The important point is that it is

used compensatorily.

This behaviour can be satisfactorily accounted for in terms of the interactive-

compensatory model already outlined in relation to reading. A trading relation

would seem to exist which enables the first-language listener to compensate auto-

matically for noise in the environment by relying more heavily than usual upon top-

down information. Such a mechanism can, it is suggested, be adapted to the very

different circumstances of second-language listening, where perceptual problems

arise not from degradation due to noise, but from inability to recognise words in the

J. Field / System 32 (2004) 363–377 369

speech-stream or from limitations of vocabulary or syntax. The response could be

expected to be the same – namely, the invoking of cues provided by background

knowledge, co-text, analogy or knowledge of speaker in order to compensate for

inadequate bottom-up data.

2. Top-down evidence overruling bottom-up

Field (1997) cites evidence from learner-focused video which suggests that L2 lis-

teners tend to construct a schema relating to the topic of a listening text and to use this

to guide their processing of incomplete bottom-up information. The most striking

finding, however, was that some learners seem to place more confidence in their pre-

formed schema than in incoming data from the speech-stream. Instead of modifying the

former to fit the evidence, they appear inclined to do the opposite: to alter their versionof what they hear to fit it to preconceived ideas of what the text covers. Thus, in a text

about travel, one student converted the word mat into map and another chose not to

identify the word ledge as an unknown vocabulary item but interpreted it as bridge.

Two conclusions can be drawn from these results:

(a) that, in some cases, learners may place greater trust in top-down evidence than in

bottom-up. This may reflect an underlying lack of confidence in their ability to

process the sounds of the target language accurately.

(b) that, instead of always assuming that unrecognised words represent new items ofvocabulary, some learners prefer to match them very approximately to known

words which are supported by top-down evidence.

On point (a), other commentators have recognised the extent to which learners

can be misled by dependence on top-down information. Long (1990, p. 72) com-

ments that ‘schemata can... have dysfunctional effects on L2 listening comprehen-

sion’. Similarly, Lund (1991, p. 202) asserts that ‘The construction of inappropriate

contexts and schemas seriously interferes with comprehension of the actual text’.

3. The experiments

3.1. Design

Three experiments were designed to test the extent to which second-language

learners are inclined to place their trust in top-down rather than bottom-up

information.Experiment 1: Groups of four to six words were composed, all within a vocab-

ulary of about 1000 words and therefore likely to be known by subjects. Sometimes

all the items in a set belonged to the same lexical field

wet – cloudy – dry – cold – hot

and sometimes only the last two words were associated

big – new – empty – cold – hot

370 J. Field / System 32 (2004) 363–377

In the target items, the onset of the last word was then changed to turn it into an

similar word which did not belong to the set (hotfi ‘got’).

Foils, where the last word had not been changed, were mixed in with the target

items. Subjects were asked to listen to each group of words and to write down the

last word in each. The purpose was to establish whether top-down influences (here

based on vocabulary sets) would so constrain the subjects that they would overrulethe ‘bottom-up’ evidence of their ears and substitute a semantically more appro-

priate item (HOT for got). If so, it would provide strong evidence of top-down de-

pendency – of an underlying view that inference was perhaps more dependable than

the learner’s ability to identify sounds and words accurately in the target language.

Experiment 2. Here, a semantically constraining sentence was provided in place of

a list of words. A highly predictable word at the end of the sentence was replaced by

one which differed from it by one phoneme. This substitute word was much less

predictable but nonetheless acceptable in the context. Both original word and sub-stitute were of high enough frequency to be within the subjects’ vocabulary. Examples

I couldn’t listen to the radio because of the boys. [NOISE]

The people at the party were Germans, Italians, Spanish and some friends [FRENCH]

The sentences were played to subjects, who were asked to write down the last

word in each. The purpose was again to see to what extent the context (this time, the

propositional content of the sentence) encouraged them to write down a different

word from the one that they had heard.

Experiment 3. Low frequency words were chosen which were unlikely to fall

within the vocabulary of the learner but which phonologically resembled high-fre-quency words they were likely to know. Sentences were then designed which pro-

vided a meaningful context for the low-frequency item but a contradictory one for

the high-frequency alternative. Examples:

They’re lazy in that office; they like to shirk. [not WORK]

When the plane didn’t arrive, the passengers were in a terrible plight. [not FLIGHT]

Inmost of the items, the targetword occurred at the endof the sentence, and learners

were asked to write down the last word they heard. In seven items, the word was within

the sentence, and subjects were asked to write down the first word after a signal.

Here, the purpose was to see whether subjects opted for a known, frequent and

phonologically similar word despite the fact that it was inappropriate in the context,or whether they were prepared to accept the presence of a new vocabulary item.

3.2. Subjects

The subjects for the three experiments were students at Eurocentre Cambridge, a

leading British EFL school. They were in four different classes, of which two

(N ¼ 31) had been graded as Lower Intermediate and two (N ¼ 17) were a high

Elementary. They were chosen because they represented a block of learners whosefamiliarity with English could be regarded as limited; all had similar scores of

J. Field / System 32 (2004) 363–377 371

between 25 and 36 out of 80 on the school’s entry test. The experimental material fell

into two versions; each was presented to a group of 24 subjects in all, comprising one

class from each level. One of the Elementary students submitted a blank paper and

was omitted from consideration, leaving a total population of 47 subjects.

A range of first languages was represented. These were: German (9), Italian (3),

Korean (5), Arabic (6), Spanish (5), Portuguese (6), Cantonese (1), French (2),Japanese (3), Russian (1) and Thai (6). At the time of the experiments, most subjects

had been in the UK for about eight weeks. Most had about 5-6 years of secondary

school English but none had resided in an English-speaking country before.

3.3. Procedure

The texts for each experiment were played once only on a cassette, using high-

quality equipment in a purpose-designed classroom with good acoustics. Answerswere recorded by subjects on an answer sheet. Each group of subjects was presented

with one version of Experiments 1 and 3 and with the 10 items in Experiment 2.

3.4. Results

The results from Experiments 1 and 2 are described briefly before going on to a

more detailed discussion of the results for Experiment 3.

Experiment 1. No evidence was obtained of subjects reinterpreting what they hadheard in order to fit it to the lexical set. Only one sequence showed any such effect. In

items ending summer – string, a minority (17.02%) of subjects recorded SPRING

instead of string. However, their reasons for doing so cannot be attributed with

certainty to the constraining influence of the lexical set; indeed nearly 30% of re-

sponses involved alternatives other than SPRING.

Closer examination of all mistranscriptions suggested that it had been an error of

experimental design to adjust the onset of the target word rather than the offset. In-

correct words recorded by subjects had a correctly transcribed onset in all but two cases(2.8%). Where words had been misheard, it was the offset or the vowel which had been

changed. This result suggests that, like native listeners (Brown and McNeill, 1966;

Forster and Davis, 1991), non-native ones place great importance upon word onsets.

Experiment 2. More positive results were achieved than in Experiment 1. Though

the results were not consistent across items, words in seven of the 20 items were

substituted. The level of substitution ranged from 15% to 62% of responses.

Some of the changes were not those that had been expected. This provides further

evidence of the robustness of the word onset, to which even in a foreign language,listeners seem to pay particular attention. Thus, in Item 1:

I couldn’t listen to the radio because of the boys. (Predicted: NOISE)

subjects preferred to substitute the word VOICE, whose onset shares labiality

with boys, even though this meant ignoring the voicing of the offset by substituting

/s/ for /z/.

Table 1

Mean percentage responses in transcribing new vocabulary items

Target Non-word Substituted word Blank

All responses

Mean 24.83 20.57 33.31 21.29

SD 22.79 13.94 17.48 11.04

Filled responses

Mean 30.23 27.38 42.39

SD 24.98 19.14 20.87

372 J. Field / System 32 (2004) 363–377

Furthermore, though subjects did sometimes allow top-down evidence to prevail

over bottom-up, the effect only occurred with items where the context was extremely

constraining. For example:

3 Th

failed

that ha

recogn

The people at the party were Germans, Italians, Spanish and some friends.

where 42.1% substituted FRENCH or FRANCE for friends.

Experiment 3. Here, the results were striking. Responses were categorised ac-

cording to whether subjects had recorded a blank, had reproduced the target word

(allowances were made for erratic orthography), had written a non-word or had

substituted a word phonologically similar to the target. The substituted words were

often not those which had been predicted in designing the experiment.

Mean percentage responses across 20 items are shown in Table 1. It will be seen

that a mean of 33.31% of all responses involved the substitution of another, usuallymore frequent, word for the target, even where that word did not fit the context

semantically. When blank responses are removed from consideration, the mean for

substituted words was 42.39%. Averaged over the items of the experiment, 57.61% of

filled responses thus appear to have identified the target word as a new item of

vocabulary, and to have transcribed it either correctly or as a non-word. But a

striking mean of over 40% did not accept phonetic-acoustic evidence that the item

was unknown and instead matched it, very approximately, to a known one. The

40:60 ratio probably understates the true position. The ‘target word’ responses canbe assumed to include not only subjects who identified a word as a new piece of

vocabulary and succeeded in transcribing it, but also subjects who got the word right

because they already knew it. 3

Confronted by an unknown lexical item, subjects thus chose one of three

expedients:

• a phonological one (columns 1 and 2): attempting to transcribe the item

accurately;

e majority of blank responses were assumed to be indicative of a situation where the subject had

to achieve an auditory match, but remained unclear as to whether the target word was a known one

d not been identified or a new one. However, some may admittedly have occurred when a subject

ised the presence of a new word but felt unable to transcribe it.

J. Field / System 32 (2004) 363–377 373

• a zero one: submitting a blank response;

• a lexical one: matching the item, extremely approximately, to a known word.

It is not possible to quantify statistically the significance of the decision to sub-

stitute a known word. One might postulate that the three choices open to subjects

(transcription, zero and substitution) indicate an expected frequency of 33.33%

against which the figures in Table 1 could be analysed for probability. A calculationon this basis would simply confirm that the ‘substitution’ figure falls exactly at the

level of chance. However, it would not take account of the most striking fact about

the figure – that it was achieved despite the evidence of the listener’s ears and, in many

cases, contrary to the evidence of the co-text.

The substitution of a known word occurred in answers to all 20 items in the test;

there was no item where the stratagem was not in evidence. However, the number of

responses involving substitution varied considerably from item to item: ranging from

18 out of a possible 23 or 24 down to 2. A chi square test explored the impact ofindividual items on the choices that subjects made. It examined the variation be-

tween observed frequencies for responses that identified a word as new (columns 1

and 2 in Table 1) and those for responses that substituted a known word (column 3).

The result, v ¼ 63:18 (df ¼ 19), was highly significant at p < 0:001, and indicated a

strong interaction between the item and the type of response adopted. A similar test

on subjects produced a non-significant value of v ¼ 38:08 (df ¼ 46), n.s. This re-

flected the extent to which the strategy of substituting known words was shared

across subjects rather than peculiar to a few individuals. Only one subject showed nosigns of using it.

It was striking that many of the known words that were proposed were chosen

without any regard to appropriacy. Not only were many of them semantically in-

appropriate; around 50% were not even in the correct word class. One concludes that

there is evidence here of a strategy which is neither bottom-up nor top-down but is

lexical – a rough attempt at a one-to-one match with a known item which potentially

overrules contextual information and modifies perceptual. This is a type of process

that is often overlooked in the ‘bottom-up’/‘top-down’ controversy. As noted earlier,it qualifies as ‘top down’, but not in the way in which the term is usually interpreted

in the SLA literature.

4. Conclusions

The three experiments presented here attempted, in different ways and with dif-

ferent degrees of success, to explore the relationship between top-down and bottom-up information in the processing of second-language listeners.

Experiment 2 provided some evidence of top-down expectations overruling the

evidence of the listeners’ ears – but only when the sentential context was highly

constraining. It can be suggested, of course, that the sentences in question might

have the same effect upon native listeners. Indeed, the sentence which caused the

highest substitution of a contextually more predictable word was one which featured

in the author’s own corpus of native-speaker Slips of the Ear. But this is not a

374 J. Field / System 32 (2004) 363–377

counter-argument: it indicates the extent to which the use of top-down information

by non-native listeners (even at lower levels of English) approximates to that by

native listeners and possibly exceeds it.

The conditions in question were exceptional in that they involved the replacement

of a word for which there was perceptual evidence by one which resembled it pho-

nologically. The results in Experiment 2 thus do not represent extreme cases of top-down effects overriding all bottom up data: they simply testify to a willingness to

substitute part of a word when there is a contextually appropriate and perceptually

similar alternative.

The evidence seems to be that the onsets of words are remarkably robust for non-

native as for native listeners. Trusting the onset of the word rather than the coda is a

sensible strategy for a native listener. This is because assimilation in English is

mainly regressive (Gimson, 1992, p. 255, making the ends of words subject to change

and therefore less reliable. 4 But it is remarkable to note the extent to which non-native listeners appear to have picked up the same strategy. Appropriately, subjects

in Experiments 1 and 2 placed less confidence in their perception of vowels and of

word offsets and were more prepared to alter their interpretation of what they heard

at these points.

Experiment 3 explored the way in which non-native listeners respond to unknown

vocabulary items. A marked tendency was noted to associate a new word form with

a known one, even if this meant ignoring both top-down and bottom up evidence.

Again, word onsets seemed especially robust but subjects seemed less confident of theidentity of vowels and offsets and more prepared to change them in the interests of

making a lexical match.

This finding on the processing of new words gives an additional insight into the

bottom-up/ top-down relationship which is the main topic of this paper. It appears

that, when a salient word is unfamiliar, learners do not consistently adopt a tech-

nique of visualising the orthographic form of the word and inferring its meaning

from context. Instead, they frequently choose to match what they hear with a known

word which is approximately similar. The match(a) may be regardless of context and even regardless of word-class (Here frequency

may play a role); or

(b) may draw upon top-down expectations.

The degree to which a chosen word is permitted to vary from the available

bottom-up data will reflect the listener’s judgements as to how reliable that data is,

given their difficulties in recognising with certainty the phonemes of the second

language.

The suggestion here is that when circumstance (a) arises, fuel is provided for thosewho argue for bottom-up dependence (though in truth what is at issue is a kind of

lexical effect). When (b) arises, the outcome can be cited as evidence of top-down

4 Marslen-Wilson’s Cohort Theory (1987) and other sequential models of auditory processing would

also suggest that onsets are critical because it is at this point that a group of word candidates is opened up.

J. Field / System 32 (2004) 363–377 375

dependence. The truth may well lie between the two – and may, like most strategic

activity, be highly variable from one learner to another, from one text to another and

from one task to another.

Appendix A. Experimental material

Experiment 1

Set 1

1. jump big pen give take2. wet cloudy dry cold got (hot)

3. tired end live cheap cupboard shelf

4. walk earn read night (write)

5. look shirt heavy hands meat (feet)

6. quiet bag push phone short tall

7. shoe broke angry car train

8. orange black red blue clean (green)

9. knife earth child dog hat (cat)

Set 2

10. thin name catch cup easy hard

11. friend ill lake buy tell (sell)12. aunt man same hole drive fly

13. plate cup knife talk (fork)

14. high sorry small near wrong quite (right)

15. light time new key eat think (drink)

16. June March summer string (spring)

17. ten hurry sharp bag case

18. old young early wait (late)

Experiment 2

1. I couldn’t listen to the radio because of the boys. (noise)

2. The people at the party were Germans, Italians, Spanish and some friends.

(French)

3. We arrived at the airport on time, then we had to wait two hours for the train.

(plane)

4. You can go into the town when it’s day and when it’s light. (night)

5. He’s good at football, tennis and running ; you often see him in shorts. (sports)6. I thought of the husband I had just buried. (married)

7. Number 7 ran very slowly, but Number 3 was last. (fast)

8. Do you know what books the children need? (read)

9. I’ve lived in the north and the east, but this place is best. (west)

10. I saw him climb on to the roof, then I heard him call. (fall)

Appendix A (continued)

Experiment 3

Set 1

1. We can’t go skating because there’s no ice in the rink. (drink)

2. They’re lazy in that office ; they like to shirk. (work)

3. There aren’t many children in the town; in fact, there’s quite a dearth. (birth)

4. When you’ve cut up the meat, add some spices. (slices)

5. The money disappeared to Switzerland as the result of a fraud. (afford)

6. We rode along the river to its source. (horse)7. He argues a lot, but I like to hear his views. (lose)

8. When the plane didn’t arrive, the passengers were in a terrible plight. (flight)

9. He hardly ever smiles. I’d describe him as grave. (brave/great)

10. I’m sorry but the cheque is blank. (bank)

Set 2

11. After ten minutes in the rain the cigarettes were completely soaked. (smoked)

12. The water ran off the platform into a drain. (train)

13. Going to hospital fills me with dread. (bed)

14. More information about the soldier was never sought. (fought)

15. The ship’s carrying a freight that’s dangerous. (afraid)

16. I don’t know how he copes with all his problems. (hopes)

17. They travel at such a pace that they see very little. (place)18. We need some wooden stools for the children. (schools)

19. He stood there and spat on the pavement. (sat)

20. The office workers had left litter all over the grass. (letter)

Bold typeface indicates target words; words in brackets are possible substitutes. Italic typeface indicates

words that potentially provide false co-textual cues.

376 J. Field / System 32 (2004) 363–377

References

Brown, G., Yule, G., 1983. Discourse Analysis. Cambridge University Press, Cambridge.

Brown, R., McNeill, D., 1966. The ‘tip of the tongue’ phenomenon. Journal of Verbal Learning and

Verbal Behaviour 5, 325–337.

Clarke, M., 1980. The short circuit hypothesis of ESL reading – or when language competence interferes

with reading performance. Modern Language Journal 64, 203–209.

Conrad, L., 1983. Semantic versus syntactic cues in listening comprehension. Studies in Second Language

Acquisition 7, 59–69.

Cummins, J., 1979. Cognitive/academic language proficiency, linguistic interdependence, the optimum age

question and some other matters. Working Papers on Bilingualism 19, 197–205.

Cziko, G.A., 1980. Language competence and reading strategies: a comparison of first- and second-

language oral reading errors. Language Learning 30 (1), 101–114.

Field, J., 1997. Monitoring for breakdown of understanding. In: Field, J., Graham, A., Griffiths, E.,

Head, K. (Eds.), Teachers Develop Teachers Research 2. Whitstable, IATEFL.

Forster, K.I., 1989. Basic issues in lexical processing. In: Marslen-Wilson, W. (Ed.), Lexical Represen-

tation and Process. MIT, Cambridge, MA.

Forster, K.I., Davis, C., 1991. Repetition priming and frequency attenuation in lexical access. Journal of

Experimental Psychology: Learning, Memory and Cognition 10, 680–698.

Gernsbacher, M.A., 1990. Language Comprehension as Structure Building. Erlbaum, Hillsdale, NJ.

J. Field / System 32 (2004) 363–377 377

Gimson, A.C., 1992. Gimson’s Pronunciation of English. 5th ed. A Cruttenden. Edward Arnold, London.

Goh, C.C.M., 2000. A cognitive perspective on language learners’ listening comprehension problems.

System 28, 55–75.

Hansen, C., Jensen, C., 1994. Evaluating lecture comprehension. In: Flowerdew, J. (Ed.), Academic

Listening. Cambridge University Press, Cambridge, pp. 241–268.

Koster, C.J., 1987. Word Recognition in Foreign and Native Language: Effects of Context and

Assimilation. Foris Publications, Dordrecht.

Long, D.R., 1989. Second language listening comprehension: A schema-theoretic perspective. Modern

Language Journal 73, 32–40.

Long, D.R., 1990. What you don’t know can’t help you. Studies in Second Language Acquisition 12,

65–80.

Lund, R.J., 1991. A comparison of second language listening and reading comprehension. Modern

Language Journal 75, 196–204.

Mack, M., 1988. Sentence processing by non-native speaker of English: evidence from the perception of

natural and computer-generated anomalous L2 sentences. Journal of Neurolinguistics 3, 293–316.

Marslen-Wilson, W.D., 1987. Functional parallelism in spoken word recognition. Cognition 25, 71–103.

Mueller, G., 1980. Visual contextual cues and listening comprehension: an experiment. Modern Language

Journal 64, 335–340.

Paran, A., 1996. Reading in EFL: fact and fiction. ELT Journal 50 (1).

Perfetti, C.A., 1985. Reading Ability. Oxford University Press, Oxford.

Stanovich, K.E., 1980. Toward an interactive-compensatory model of individual differences in the

development of reading fluency. Reading Research Quarterly 16, 32–71.

Tsui, A., Fullilove, J., 1998. Bottom-up or top-down processing as a discriminator of L2 listening

performance. Applied Linguistics 19, 432–451.

Voss, B., 1984. Slips of the Ear. Gunter Narr Verlag, T€ubingen.

Wolff, D., 1987. Some assumptions about second language text comprehension. Studies in Second

Language Acquisition 9, 307–326.