Promoting perception: lexical

segmentation in L2 listening
John Field

This article calls for greater attention to the perceptual processes involved in
second language listening—and particularly to the part they play in
breakdowns of understanding. It suggests employing basic auditory phonetics
as a means of classifying, diagnosing, and predicting problems of lexical
segmentation. Recognition of how and why learners find speech input diªcult
to process can provide a programme of simple practice exercises which
anticipate or rectify listening problems.

The importance of Two major considerations prompt this article. The first is a view that
the signal writers on second-language listening have perhaps concentrated
overmuch on higher-level understanding. It is entirely appropriate that
we should encourage second language learners to bring world knowledge
to bear upon their listening experience; it is also important to provide
adequate training in strategies which compensate for gaps in word
recognition (Field 2000). But we should not lose sight of the primacy of
the signal. We need to concern ourselves more than we do with speech as
a physical phenomenon—with what English sounds like to the non-
native listener and with the features which cause obstacles to
understanding.
Some modicum of perceptual information, even if only a few words, is
clearly needed before contextual knowledge can be brought to bear. But it
is also a fact that many high-level breakdowns of communication
originate in low-level errors. A second-language listener who hears I
won’t go to London as I want to go to London, is making a small mistake
based on phoneme discrimination. But this mistake may impact upon
the interpretation of what comes next, and may even influence
understanding of the text as a whole. Once learners have constructed a
set of expectations for a text, they are notoriously reluctant to revise them,
even if evidence comes in that contradicts them.

Inadequacies of the The second consideration is the dissatisfaction felt by many practitioners
‘comprehension’ with the present approach to the teaching of second-language listening,
approach based entirely upon the achievement of comprehension tasks. The
comprehension approach is open to question for three reasons. Firstly it
assumes that there is one ‘correct’ interpretation of an utterance,
whereas comprehension theory (Brown 1995) tells us that there may be

ELT Journal Volume 57/4 October 2003 © Oxford University Press 325
 more than one, depending upon the perspective of the listener. Secondly,
it tends to home in upon small discrete points rather than checking
global understanding. And, most importantly, it encourages us to focus
upon the product of listening (in the form of answers to questions) but
tells us nothing about the process.
It has often been said that we test listening rather than teaching it
(Sheerin 1987). It is diªcult to disagree with this assertion; but perhaps
it misses the point. The fact is that the listening process is not accessible
to inspection in the way that skills such as speaking and writing are. In
order to establish if understanding has taken place, we have no choice
but to employ what testing specialists term ‘indirect methods’: judging
learners’ listening competence by their success or failure in answering
questions or performing tasks. The problem, I suggest, lies not in the
types of task we employ to check understanding, but in the use we make
of the answers.
Suppose that Students A and B both give the correct answer in a
comprehension exercise. We congratulate them and move on to the next
question. What we have not taken on board is that Student A arrived at
the answer as a result of understanding 96 words out of 100 in the text.
Student B, on the other hand, understood only 20, but managed to
achieve the correct answer by employing appropriate strategies. This is
not to say that Student B is a worse listener than Student A. The point is
simply that that these two students are listening in di¤erent ways, and
need di¤erent types of support from the teacher.
It is essential that we follow up answers to comprehension tasks (both
correct and incorrect ones) to find out how they were derived. Only in
this way can we gain a clear picture of the strengths and weaknesses of
our learners, enabling us to contribute constructively to their
development as listeners. We should view the principal aim of a full-
length listening session as diagnostic. It should provide us with insights
into where understanding has broken down—insights which we can
then follow up with small-scale remedial exercises which aim to prevent
errors of interpretation (especially low-level errors) from occurring
again.

A remedial approach A framework for classifying learners’ problems of understanding is

to listening provided by the levels of processing which feature in simplified
A framework for psycholinguistic accounts of listening. These accounts view meaning as,
analysing breakdown in part, constructed through the ‘bottom-up’ assembly of larger and
of understanding larger units. The signal is processed through several levels: auditory-
phonetic, phonemic, syllabic, lexical, syntactic, semantic, propositional,
pragmatic, interpretive. It is useful to regard these levels not only as steps
in the construction of the message, but also as a series of stages at any
one of which understanding can break down. It is often possible to
determine that an error derives from inadequate processing at, for
example, the phonemic, the lexical, or the syntactic level. To give
examples: a learner might not be able to distinguish the di¤erence
between won’t and want (a phonemic problem), might not recognize the
spoken form of the word vegetable (a lexical problem), or might overlook

326 John Field

 the semantic implications of the phoneme /v/ in I’ve lived (a syntactic
problem).
We need to be rather more persistent than at present in determining why
breakdowns of understanding have occurred. Current practice
encourages us to make easy but unfounded assumptions. Suppose an
error occurs as the result of a learner failing to understand a particular
word. The conclusion usually reached is that the word is not in the
listener’s vocabulary. The teacher explains the meaning of the word, and
feels that the problem has been resolved. Perhaps it has. But it could
equally have arisen from several other causes, perceptual or semantic:
π Learner knows the word but attributed the wrong sense
π Learner failed to recognize a phonetic variation of a known word
π Learner knows the word in reading but not in spoken vocabulary
π Learner was unable to segment the word out of a piece of connected
speech.

Exercise type Once a problem has been identified, a set of simple exercises can be
designed which raise awareness and provide focused practice. The most
e¤ective remedial listening work involves dictation. However, by
‘dictation’ is not meant the notorious dictée which many of us recall from
our schooldays. Instead, the exercise should require that learners write
down short, unpaused sentences, each sharing the feature which caused
the original diªculty; that these sentences are uttered as naturally as
possible or taken from an authentic text; and that attention to errors of
listening does not become side-tracked by attention to errors of
spelling.
The remainder of this article discusses what is arguably the commonest
perceptual cause of breakdown of understanding: namely, lexical
segmentation, the identification of words in connected speech. The aim is
to exemplify how low-level listening problems can be diagnosed by
employing the basic knowledge of phonetics which most ELT
practitioners possess. The di¤erence is that the knowledge has to be
stood on its head, so that we view phonetics not from the perspective of
pronunciation practice but through the ears of the listener. Once
identified, areas of diªculty can be tackled by means of simple 5-minute
exercises; these might be remedial or they might anticipate problems of
listening before they occur.

Lexical segmentation We tend to overlook the fact that pauses in natural speech only occur
every 12 syllables or so, which means that, unlike readers, listeners do
not have regular indications of where words begin and end.¡ It is
remarkable that we manage to separate out words within these 12-
syllable chunks as consistently as we do.
Determining where word boundaries fall is a greater problem for the
non-native listener than is generally recognized. A learner with limited
English or weak listening skills adopts a strategy of scanning continuous
speech for matches between sequences of sounds and items of known

Lexical segmentation in L2 listening 327

 vocabulary. In the anxiety to achieve matches, word boundaries are often
breached:
Speaker: went to assist a passenger. Student 1 extracts sister
Speaker: the standard the hotel achieves. Student 2 extracts: stand at the
hotel
We are all familiar with this phenomenon of grasping at cross-boundary
straws and assuming words to be present which were not intended. (See
Voss 1984 for examples.) The ‘matching’ strategy is a natural and
productive one in the early stages of learning. The danger lies not in the
strategy itself but in the tendency of students to overlook the tentative
nature of the matches they achieve. There is a strong likelihood that
Student 1 above will go on to construct a mental model of the text which
includes somebody’s sister, even to the point of reshaping what comes
next, in order to fit her in somehow.
So a first approach to lexical segmentation should demonstrate to
learners the need for caution in word boundary allocation. One
technique is to dictate ambiguous sequences, then to disambiguate them
by adding additional words:
T dictates: an ice cream … [Ss write] … T continues dictation: a nice
cream dress
T dictates: the boxes of … [Ss write] … T continues dictation: the boxes
have been opened.
(For more examples, see Gimson 1994: 253.) This may seem a trivial
exercise; but it is an e¤ective way of demonstrating to learners that word
boundary location may be a matter of guesswork, and that guesses may
have to be revised in the light of later evidence.

Native segmentation A strategic approach to the lexical segmentation issue asks how it is that
strategies native listeners manage to locate word boundaries so successfully.
Simple matching is not the answer. If it were, we would automatically
begin a new word after, for example, hearing PORT in porter or PORTER
in portable. The research of Anne Cutler and her associates (Cutler 1990)
suggests that native listeners use a strong-syllable strategy, based on the
premise that each stressed syllable marks the beginning of a new word.™
This strategy pays dividends. Using a corpus of spoken English, Cutler
and Carter (1987) calculated that some 85.6% of all content words in
running speech are either monosyllabic or stressed on the first syllable.£
The finding ties in with evidence from Hyman (1977) that lexical stress
often fulfils a demarcative role. Many of the world’s languages have fixed
lexical stress, which occurs on the first, the penultimate, or the final
syllable of a word, and thus serves as a reliable cue to word boundaries.
From this, one might conclude that it is worthwhile to train learners of
English to emulate the segmentation strategy adopted by native listeners.
In fact, Cutler takes the view that it is impossible for learners to develop a
segmentation routine in L2 which di¤ers from the one used in their first
language. However, her reservation refers to responses to the speech
signal which are automatic. It does not rule out the possibility that

328 John Field

 learners make slightly delayed decisions about what they hear, which
resemble those of native listeners.
This, indeed, appears to be what many of them do. Learners show
themselves sensitive to rhythmic regularities in the target language, and
appear to learn from experience the value of inserting word boundaries
before stressed syllables (Field 2001)—without being aware of what they
have learnt. We should not be too surprised at this finding: infants
acquiring English appear to use rhythm in the same serendipitous way to
crack the code of connected speech (Jusczyk 1997).
With appropriate training, learners might acquire the technique much
faster—though it may be necessary to train their ears to recognize lexical
stress if it is marked di¤erently in English from their native language.
A challenging piece of authentic text might be played, and learners asked
to write down stressed syllables and match them to words they know.
Their attention can then be drawn to how many of these syllables initiate
words. A similar awareness-raising exercise might involve playing
recordings on low volume, and asking learners to transcribe the more
salient syllables.
Focusing on stressed syllables in this way not only assists learners to
locate boundaries but also draws attention to the fact that, in connected
speech, such syllables are ‘islands of reliability’: louder and longer than
unstressed ones. Some researchers (notably, Grosjean and Gee 1987)
have even suggested that it is stressed syllables which serve to identify
words for native listeners and that weak ones are accorded a di¤erent
type of attention. On this analysis, our representation of the word appear
is triggered by the sequence /pə/, our representation of indestructible by
/strk/.

Specific problems If one major cause of segmentation problems is the lack of between-word
in the signal pauses, a second and equally important one is the way in which the
standard citation forms of words are modified when they occur in
connected speech. Several di¤erent aspects will briefly be considered
(reduction, assimilation, elision, resyllabification, and cliticization) and
suggestions made for practice. For a detailed account of these
phenomena, see Brown 1990.

Reduced forms Words, and even entire phrases, often appear in connected speech in a
reduced form. One reason is that speakers economize on e¤ort: for
example, they avoid diªcult consonant sequences by eliding sounds.
Another reason is rhythmic: the patterns of English prosody dictate that
certain closed class words such as prepositions, pronouns, and
conjunctions are rarely stressed, and indeed that some may appear in a
weak form (usually featuring schwa) in these unstressed contexts.
Unstressed syllables are shorter in duration, and less salient than
stressed. They are also much less informative, because only two vowels,
/ə/ and a shortened form of //, predominate.¢ Small wonder
therefore that they pose perceptual problems for the foreign-language
listener.

Lexical segmentation in L2 listening 329

 Three main types of reduction give rise to segmentation problems:
contraction, weak forms, and the chunking of formulaic phrases. It is
relatively easy to design a structured programme which introduces the
second-language learner to these features as a listener. The proposal here
is for 5-minute dictation sessions in which the sentences for
transcription contain examples of particular types of reduced form.
A start can be made with the relatively simple area of contractions. Here,
one goal is to get learners to recognize that the contracted verb is present
at all. Assume that a class interprets I’ve lived in London for 3 years to
imply that I no longer live there. It may well be that they have not
understood the implications of this use of the Present Perfect. But the
truth may lie at a much lower level: it may be that they have not noticed
the presence of the /v/.
Fifty-one function words in English possess alternative weak forms, most
of them of high frequency. A second step in the kind of micro-listening
programme proposed aims to ensure that the learner is able to recognize
these words when they occur in connected speech. Very sensibly, many
teachers choose to treat the weak form as the standard one, and the full
form as the exception. This encourages the listener to construct a
phonological representation which matches what is by far the more
frequent form in connected speech. However, learners’ expectations of
what they will hear are sometimes unduly influenced by exposure to the
written language.
It is worthwhile covering the weak forms as comprehensively as possible
(perhaps exemplifying four or five at a time). They can be presented
again in read-aloud naturalistic sentences for transcription, where they
should be given the kind of low prominence that they receive in
spontaneous speech. For reference, a list of all 51 forms, mainly based on
Gimson 1994, is provided in the Appendix.
When dealing with weak forms, it is important not to lose sight of the
fact that we are presenting the material from the point-of-view of the
listener, and not of the speaker. We also need to bear in mind that several
of these forms are homophonous; the listener (whether native or non-
native) needs to use syntactic context to distinguish them from each
other.

ə a əv of ən an əz as jə you
are have and has your
table 1 of
Homophonous weak
er
forms
Finally, we should recognize that native speakers often produce high-
frequency sequences of words as chunks (Pawley and Syder 1983). These
sequences may become very reduced, with phonemes and even whole
syllables elided. They are only recognizable as a unit—and, indeed, it
seems likely that native listeners store them as a single semantic and
phonetic entity. It is good practice to dictate the most common of these
formulaic phrases to learners, so that they can process them holistically

330 John Field

 when they encounter them. Favourites from my own collection are:
/ mɔmɔ/ /nɑpmim/∞
Penny Ur (1984: 46) provides a useful set of reduced sequences which
can be used as material for exercises, as does Gimson (1994: 261–2).

Assimilation and It is not just the lack of pauses that makes it diªcult to identify words in
elision connected speech. Accommodatory phonological processes a¤ect
precisely the points at which the listener needs unambiguous
information—namely word beginnings and endings. The most familiar
of these processes are assimilation and elision. We tend to think of
these phenomena as random, or at least as very complex. However, as
Table 2 shows, assimilation is restricted in its operation, and quite
systematic.
It is worth noting that assimilation in English is usually anticipatory,
adjusting the ends of words in expectation of the sound that follows. The
message for the learner is: trust the beginnings of English words rather
than the ends. The sounds which are most subject to assimilation and
elision are final /t/, /d/, and /s/. These, of course, provide many of the
inflectional endings in English. Hence the irony of the grammar teacher
telling learners to listen out for such endings, when they may be absent
in spontaneous speech.
How to deal with the assimilation problem? Again by using dictation.
The nine types of assimilation distinguished in Table 2 can provide the
basis for a programme in which examples are either dictated as two-word
sequences, or embedded in simple sentences.

/n/ " [m] before [p, b, m] ten people " tem people
" [ŋ] before [k, g] ten cars " teng cars
/t/ " [p] or a glottal stop before [p, b, m] that boy " thap boy
" [k] or a glottal stop before [k, g] that girl " thak girl
/d/ " [b] or a glottal stop before [p, b, m] good play " goob play
" [g] or a glottal stop before [k, g] good cause " goog cause
/s/ " [ʃ] or omitted before [ʃ] this shirt " thi shirt
table 2 /z/ " [] or omitted before [ʃ] those shoes " tho shoes
Assimilation
/t, d, s, z/ " [tʃ, d, ʃ, ] before [j] Right you are " rye chew are
(drawn from Gimson
Did you go? " di due go
1994: 257–60)
Elision, unfortunately, follows a less consistent pattern than
assimilation; but frequent examples such as didn’t " [dnt] should
certainly be practised for recognition in a connected-speech context or
pointed out when they occur in a listening passage. We also need to pay
special heed to the way complex clusters of consonants are elided:
next spring " [nek´sprŋ]
Awareness of this kind of feature can aid learners in producing these
clusters, as well as recognizing what has been omitted.

Lexical segmentation in L2 listening 331

Resyllabification Over the years, our ears become habituated to the vagaries of English.
and cliticization We rarely notice the e¤ects which rhythm imposes upon words—e¤ects
which, for the learner, can considerably heighten the diªculty of
recognition. Firstly, there is the process of resyllabification, where, in
certain circumstances, a syllable-final consonant attaches itself to the
following syllable:
went in " when tin made out " may doubt (can’t) help it " tell pit
What complicates the situation for the listener is that, after
resyllabification, words sometimes acquire false boundary cues. Thus, in
the ‘went in’ example, the /t/ may well be lightly aspirated, suggesting that
it is word-initial. Similarly, in made out, the removal of the /d/ from the
first word is likely to be accompanied by a lengthening of the diphthong,
so that it sounds to all intents and purposes like the open syllable may.
Secondly, there is cliticization: an e¤ect which results from the way in
which natural English speech tends towards a regular stressed-
unstressed pattern. The preference of English speakers for the basic SW
(strong-weak) foot means that they often attach two words for no reason
other than a rhythmic one. This can happen in defiance of syntactic
structure (Example 1). It can even lead to prefixes getting dislodged and
being produced as if they were suªxes (Example 2).

figure 1
E¤ects of cliticization
How to handle these e¤ects? They do not really lend themselves, like
other segmentation problems, to short 5-minute dictation slots. The best
advice is simply to be aware that they exist—and, when you encounter
them in a listening text, to play and replay the relevant section to see if
learners can puzzle out for themselves the correct distribution of
phonemes and/or syllables.
Indeed, that is the message for all the perceptual diªculties described
here. The important thing is to be aware of them, and to be prepared to
practise them intensively if there are signs that they are preventing
learners from identifying familiar words because of the special
conditions of connected speech. The value of a signal-based approach of
the kind described is that it draws our attention to problems of both
perception and comprehension that would otherwise pass unnoticed.
Revised version received June 2002

Notes 2 This is somewhat of a simplification. The

1 Though accounts in the phonology literature criterion used by Cutler is not stress but the
often refer to allophonic cues as markers of presence of a full-quality vowel.
word onsets, such cues are often absent in 3 They were unable to adjust this for frequency,
medium-paced spontaneous speech. (Gimson but frequency-adjusted figures from a larger
1994: 265). written corpus were very similar.

332 John Field

4 The shortened form of /υ/ is much rarer. Accent: Southern California Occasional Papers in
5 more and more; Do you know what I mean? Linguistics 4. Los Angeles: University of Southern
California.
References Jusczyk, P. 1997. The Discovery of Spoken Language.
Brown, G. 1990. Listening to Spoken English (2nd Cambridge, MA : Bradford/MIT .
edn.). Harlow: Longman. Pawley, A. and Syder, F. H. 1983. ‘Two puzzles for
Brown, G. 1995. Speakers, Listeners and linguistic theory’ in J. Richards and R. Schmidt
Communication. Cambridge: Cambridge (eds.). Language and Communication. London:
University Press. Longman.
Cutler, A. 1990. ‘Exploiting prosodic possibilities Sheerin, S. 1987. ‘Listening comprehension:
in speech segmentation’ in G. T. M. Altmann (ed.). teaching or testing?’. ELT Journal 41/2: 126–31.
Cognitive Models of Speech Processing. Cambridge, Ur, P. 1984. Teaching Listening Comprehension.
Mass.: MIT . Cambridge: Cambridge University Press.
Cutler, A. and D. M. Carter. 1987. ‘The Voss, B. 1984. Slips of the Ear. Tubingen: Gunter
predominance of strong initial syllables in the Narr Verlag.
English vocabulary’. Computer Speech and
Language 2: 133–42. The author
Field, J. 2000. ‘Not waving but drowning: a reply to John Field researches L1 and L2 listening. He also
Tony Ridgway’. ELT Journal 54/2: 186–95. writes and lectures on psychological processes in
Field, J. 2001. Lexical Segmentation in First and language, and this year has published two books
Foreign Language Listening. Unpublished PhD on psycholinguistics (both with Routledge). He
Dissertation, Cambridge University. has experience as a teacher trainer and materials
Gimson, A. C. 1994. Gimson’s Pronunciation of writer in many parts of the world, including
English (5th edn., revised by A. Cruttenden). Belgium, China, Hong Kong, and Tanzania, and
London: Edward Arnold. has written English programmes for both radio
Grosjean, F. and J. P. Gee. 1987. ‘Prosodic and television. He currently teaches for the
structure and spoken word recognition’. Cognition University of Leeds and for King’s College London.
25: 135–55. He is joint Co-ordinator of IATEFL Research SIG .
Hyman, L. M. 1977. ‘On the nature of linguistic Email: [email protected]
stress’ in L. M. Hyman (ed.). Studies in Stress and

Lexical segmentation in L2 listening 333

Appendix Grouped by word class (after Gimson 1994)
Weak forms
word weak form word weak form
a ə am əm
an ən are ə
any n be b
some səm sm been bn
the ðə was wəz
at ət were wə
for fə can kən kn
from frəm could kəd kd
of əv ə do dυ də
to tə does dəz dz
and ənd nd n had həd əd
but bət has həz əz
as əz have həv əv
than ðən ðn must məst
that ðət shall ʃəI ʃI
who υ hυ should ʃəd ʃd
there ðe ðə (+r) will wəl əI I
he  h would wυd wəd d
her ə hə -n’t n
him m Saint sənt snt
his z Sir sə
I 
me m
we w
she ʃ
them ðəm ðm
us əs
you jə
your jə
our ɑ: 

334 John Field