Applied Linguistics 21/3: 354±375 # Oxford University Press 2000

Measuring Spoken Language: A Unit for

All Reasons
1
PAULINE FOSTER, 2ALAN TONKYN, AND
3
GILLIAN WIGGLESWORTH
1
St Mary's University College, 2University of Reading, 3Macquarie University

The analysis of spoken language requires a principled way of dividing

transcribed data into units in order to assess features such as accuracy and
complexity. If such analyses are to be comparable across di€erent studies, there
must be agreement on the nature of the unit, and it must be possible to apply
this unit reliably to a range of di€erent types of speech data. There are a number
of di€erent units in use, the various merits of which have been discussed by
Crookes (1990). However, while these have been used to facilitate the analysis
of spoken language data, there is presently no comprehensive, accessible
de®nition of any of them, nor are detailed guides available on how to identify
such units in data sets. Research reports tend to provide simplistic two-line
de®nitions of units exempli®ed, if at all, by unproblematic written examples.
These are inadequate when applied to transcriptions of complex oral data,
which tend not to lend themselves easily to a clear division into units. This
paper was motivated by the need each of the three authors felt for a reliable and
comprehensively de®ned unit to assist with the analysis of a variety of
recordings of native and non-native speakers of English. We ®rst discuss in
very general terms the criteria according to which such a unit might be selected.
Next, we examine the main categories of unit which have been adopted
previously and provide a justi®cation for the particular type of unit that we
have chosen. Focusing on this unit, we identify a number of problems which
are associated with the de®nition and exempli®cation of units of this type, and
give examples of the awkward cases found in actual data. Finally we o€er a
de®nition of our unit, the Analysis of Speech Unit (AS-unit), providing
adequate detail to address the problematic data analyses we have illustrated.

1. APPROACHES TO SPOKEN LANGUAGE ANALYSIS

Researchers dealing with spoken ®rst or second language data are often
seeking to measure the frequency of certain discourse features (such as
con®rmation checks, clari®cation requests, self-corrections) or the frequency
of certain grammatical features (such as morphemes marking number or case)
or else they are seeking to measure quantitatively such dimensions as the
relative grammatical accuracy, syntactic complexity and ¯uency of the
language in their data. In order to do this, they need ®rst to segment the
data into units against which frequencies and ratios can be calculated.
Accordingly, a unit for the segmentation of oral data is an essential tool in
 PAULINE FOSTER et al. 355

applied linguistics, without which much quantitative analysis of language

development would not be possible.
In the assessment of spoken language performance by division into
segments, more has often meant better. The `more' can be seen in two
ways: productivity and complexity. In the assessment of young children's or
aphasics' ®rst language performance, productivity has been encapsulated in
such measures as the Mean Length of Utterance (MLU) devised by Roger
Brown (1973), or in the Mean number of Sentences per Turn of the original
LARSP1 procedure (Crystal et al. 1976). Complexity in the case of such
performances has been measured by a range of indices, including the LARSP
structural analysis, the Index of Productive Syntax (Scarborough 1989), and
the Type±Token ratio (e.g. Wagner 1985).
Such measures have not gone uncriticized. Snow (1996) has identi®ed a
range of problems with the MLU which arise from its focus on product rather
than developmental process and its failure to distinguish di€erences in
morphological sophistication among children. Similarly, Richards (1987) has
pointed out that although type±token ratios have been widely used to assess
children's language development, the measure is unsatisfactory because the
ratio inevitably decreases as sample size increases.
In the measurement of older children's ®rst language performance, and in
second language measurement, it is dicult to work with the productive
dimension alone. The relative cognitive maturity of such performers means
that they can be highly productive despite limited resources. They can fashion
lengthy continuous turns using a range of communication strategies, such as
paraphrases, gestures, and ®lled pauses. However, such performances are not
usually valued as highly as those where high productivity is supported by
relatively high complexity, or a wide repertoire.
Other approaches have focused on units or `chunks' of spoken language
within lengthy turns. These can be linked to psycholinguistic processesÐmore
speci®cally, planning processes. A number of researchers and commentators
(e.g. Butterworth 1980; Chafe 1980; Garman 1990) have noted that both
macro- and micro-planning processes are involved in composing speech. The
former may cover quite long (e.g. multi-`sentence') stretches of speech, while
the latter are associated with shorter units similar to the clause or sentence.
Regarding the shorter units, from the planning perspective, the more pro®cient
speaker will be the person who can keep track of more complex micro-units.
This ability allows the speaker to produce a more complex message in a shorter
time span. This may improve information ¯ow and free working memory
capacity for other communicative tasks, such as monitoring an interlocutor's
reactions, and adjusting the style and/or sequence of information. Keeping
track of more complex micro-units will, in turn, involve being aware of the
syntactic requirements and constraints of a construction which has been
started, and being able to cope with these as the full plan unfolds. Assessment of
the performance will involve, amongst other things, charting the relationships
between ¯uency, accuracy, and complexity within the micro-planning units.
356 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

Recently, an area of research has opened up into the way speakers draw on
memorized sequences of language (see for example Ellis 1996; Weinert 1995)
rather than constructing streams of speech word by word. Such lexicalized
sequences range from the invariable or formulaic (you can't teach an old dog
new tricks; be that as it may) to the variable (NP be two NP short of a NP; NP be
the sort of person who would go around V-ing NP) but all share the important
characteristic that their surface complexity seems to owe less to syntactic
processing than it does to simple selection from a memory store. Though this
suggests that pro®ciency is not necessarily re¯ected in surface complexity of
language, it is still valid to suppose that more pro®cient speakers are those
who are able to keep track of where they are, syntactically, as they
incorporate fully or partially ®xed sequences with language freshly minted
for the occasion.2 Foster (in press) has shown that adult native speakers of
English are adept at this and seem to rely more on memorized sequences of
language when under some degree of communicative pressure than they do
when the pressure is o€. Non-native speakers on the other hand, seem to rely
far more on word-by-word processing, increasing signi®cantly in their
complexity, ¯uency, and accuracy when not under communicative pressure
(see also Foster and Skehan 1996). It is perhaps useful, therefore, in
determining what characterizes the pro®le of a pro®cient speaker, whether
of L1 or L2, to consider both the extent to which a person is able to exploit a
store of native-like memorized sequences, and the extent to which a person is
able to construct complex and extended syntactic turns out of smaller
segments. In this paper we are concerned with measuring the second of these,
acknowledging the importance of the ®rst but leaving that problem of how to
measure and investigate memorized language to others (see Foster in press;
Willis 2000).
In deciding upon the most appropriate unit for segmenting oral speech
samples, we must be governed by the well-established methodological criteria
of reliability and validity. If the unit cannot be reliably identi®ed, the
measurements made will be misleading. Crookes (1990), in the ®rst paper to
address this issue in detail, noted that if the unit has little or no relationship to
the psycholinguistic planning process, the measurements made will have little
or no value. They will have no advantage over, say, merely chopping up the
transcript into groups of ten words or twenty morphemes, or whatever. To put
it simply, what we need to know is what the performer can achieve in a single
chunk of micro-planning activity, and how particular types of plan may a€ect
the complexity, accuracy, and ¯uency of the language that is produced.
It follows that if speech researchers can agree on a unit which is
psycholinguistically valid and reliable in its application to speech samples,
then it should, in theory, be possible to establish an international standard
which would enable comparisons to be made across data sets, and ideally,
across di€erent languages. In practice, however, this would probably be
dicult to achieve. We cannot enter into a prolonged discussion of the issues
here, including that of the limitations on the applicability across languages of
 PAULINE FOSTER et al. 357

the unit we identify in this paper, nonetheless its use should allow more
reliable comparisons with research in other languages whose syntax is similar
to that of English.
Two problems characterize the units of segmentation already in use by
researchers working with spoken discourse and bedevil any attempt to make
fruitful comparisons or replications. These are:
1. De®nitions: ostensibly identical units are either de®ned in di€erent ways,
or not de®ned at all, or de®ned in a way which is too simple to be used
with real spoken data.
2. Applications: if exempli®ed at all, de®nitions are accompanied by one or
two citation examples which bear little resemblance to the messy reality
of speech transcripts.
In the sections that follow we examine both of these problems in more detail.

2. DEFINITIONS: SIMILAR OR DIFFERENT?

Recently we looked at a signi®cant and substantial sample of research
published over the last seventeen years3 in which either written or spoken
language had been segmented in some way for quantitative analysis. Four
leading journals in the applied linguistic and second language ®eld were
surveyed: Applied Linguistics, Language Learning, Studies in Second Language
Acquisition, and TESOL Quarterly. Our survey revealed a plethora of de®nitions
of units of analysis, a paucity of examples, and a worrying tendency for
researchers to avoid the issue altogether by drawing a veil of silence over their
methods. A total of 87 studies were identi®ed which used some kind of unit to
segment language for analysis. Of these 87, 44 provided no de®nition at all of
the unit they chose to employ. The worrying implication in this omission is
that there is no need for a linguist to de®ne an `utterance' or a `T-unit', or any
other unit of analysis chosen as a research tool: it is enough merely to label it.
However, anyone involved in the analysis of spoken data knows that this is
not in fact the case. In order to achieve both accurate and comparable
analyses based on a quantitative evaluation of the data, you need more than a
simple label to work from.
The following extract serves to illustrate how dicult it is to work with the
fragmentary and elliptical data which is typical of oral language samples,
particularly from second language learners. This is a typical piece of non-
native oral interaction, taken from a research study in which participants had
to discuss prison sentences for a list of o€enders (Foster and Skehan 1996):
A: which which what is your opinion?
B: (1.0) maybe er (5.0) he (7.0)
A: long time? or it's for for you it's a major mistake or a small
mistake?
B: maybe three months
A: three months for this one okay for me it's ten
358 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

B: ten?
A: ten years
B: yeah ten years oh very long
Without providing at least a de®nition of the unit employed to segment such
data, is it possible for the reader to know how it should be dealt with? Is B's
second turn counted as a unit even though it lacks a verb? Is there a unit in
B's third turn, (which appears to be a clari®cation request) even though it is
only one word? How many units are there in A's third turn? Or B's ®nal turn?
Can B's ®rst turn be included in any way? In sum, how much of this is usable,
and in what terms?
We found that 43 of the 87 studies in our survey did at least provide a
de®nition of the unit they used. However, these varied considerably in the
amount of detail provided, and studies which used the same unit often did not
work from the same de®nition of it. In examining the kinds of de®nitions
given, we would argue that they can be grouped into three broad categories:
semantic, intonational, and syntactic.4 We shall look at each of these in turn,
giving examples of the de®nitions provided by the researchers who used them
and discussing brie¯y the issues they raise.

2.1 Mainly semantic units

. Proposition: `a semantic unit consisting of at least one major argument and
one or more predications about this argument' (Sato 1988: 375).
. C-unit (semantic focus): `utterances, for example, words, phrases and
sentences, grammatical and ungrammatical, which provide referential or
pragmatic meaning' (Pica et al. 1989: 72).
. Idea unit (semantic focus): `a chunk of information which is viewed by the
speaker/writer cohesively as it is given a surface form . . . related . . . to
psychological reality for the encoder' (Kroll 1977: 85).
The identi®cation of semantic units based on information/meaning chunks
might initially appear appealing. However, it is clear that the extent of an
idea, or an argument, is never easy and often impossible to establish with
certainty. Thus de®nitions which rely exclusively on semantic criteria will
tend to be extremely hard for the analyst to work with reliably. It is therefore
understandable that in the literature these semantic criteria seldom stand
alone: they are usually supported by grammatical and intonational ones.

2.2 Mainly intonational units

. Tone unit/phonemic clause: `a distinctive con®guration of pitches, with a clear
centre, or nucleus. . . . The nucleus is the syllable (or, in some cases, series
of syllables) which carries the greatest prominence within the tone-unit'
(Crystal and Davy 1975: 16).
. Idea unit (intonation focus): `The `tone-unit' of Crystal . . . is essentially the
same. . . . Most idea units end with an intonation contour that might
 PAULINE FOSTER et al. 359

appropriately be called clause-®nal: usually either a rise in pitch, . . . or a

fall. . . . A second factor is pausing. Idea units are typically separated by at
least a brief pause' (Chafe 1980: 13±14).
. Utterance: `a stream of speech with at least one of the following
characteristics: 1. under one intonational contour; 2. bounded by pauses;
3. constituting a single semantic unit' (Crookes and Rulon 1985, cited in
Crookes 1990: 187).
The central focus of each of these de®nitions is intonational. Chafe and
Crookes both incorporate pausing as a secondary feature, and both also
incorporate semantic criteria. However, for the second language researcher,
units which rely purely or mainly on pausing and intonational features are
particularly problematic because of the vagaries of such features in the speech
of non-native-speakers. Pauses in L2 performance are not necessarily at unit
boundaries, and it can be dicult to distinguish between pauses that result
from message formulation or a lexical search. Also, the tone unit is likely to
prove especially unhelpful in that dys¯uent speakers are likely to produce
more nuclei than ¯uent ones. Even ¯uent native speakers tend to produce a
fairly large number of sub-clausal tone units. Such units are unlikely to reveal
much about planning abilities and pro®ciency. Proponents of the third
intonational unit, the utterance, such as Crookes (1990), do not seem always
to distinguish it clearly from the idea unit, as the de®nition makes clear; in
terms of intonational patterns, will the `intonation contour' be a short mono-
nuclear one, similar to that of the tone unit, or will it cover a longer stretch of
speech, and be identi®ed by a gradual declination of fundamental frequency?
Intonational criteria are then, in our view, rather unstable.
However, although intonation may be an insecure single foundation for the
unit de®nition, it may in addition, as Loban (1976) pointed out, be a useful
complement to other criteria, especially syntactic ones. It is to syntactic units
that we now turn, as we consider that these promise greater reliability and
validity. Below we examine in rather more detail the range of syntactically
based units which have been used in recent research. Before we do, however,
it will be useful to de®ne the subordinate unit common to each.
. Clause and s-node: `either a simple independent ®nite clause, or a dependent
®nite or non-®nite clause' (Foster and Skehan 1996: 310); `S-nodes are
indicated by tensed or untensed verbs' (Ellis et al. 1994: 483).
The terms `clause' and `s-node' are interchangeable. They provide a relatively
unproblematic measure to subdivide units into smaller segments. As we shall
see, the problems arise when de®ning how speakers organize these smaller
segments in longer and more complex syntactic units.
360 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

2.3 Mainly syntactic units

. Sentence. The sentence, as is widely recognized (see e.g. Quirk et al. 1985:
47) is problematic for spoken (and written) data, and will be ignored in this
paper.
. Idea unit (structurally de®ned): Kroll gives a lengthy de®nition, which can
be summarized as: An Idea Unit is a clause with its pre- and post-V clause
elements. Also counted as IU's are non-®nite subordinate clauses, and ®nite
relative clauses where the relative pronoun is present (Kroll 1977: 90).
. T-unit: our survey revealed the T-unit to be clearly the most popular unit
for the analysis of both written and spoken data. Hunt (1965, 1966, 1970)
de®nes the T-unit as essentially a main clause plus any other clauses which
are dependent upon it. Researchers using the T-unit and wishing to de®ne
it use one of Hunt's four versions.
`one main clause with all subordinate clauses attached to it' (Hunt 1965:
20);
`one main clause plus whatever subordinate clauses happen to be attached
to or embedded within it' (Hunt 1966: 735);
`the shortest units into which a piece of discourse can be cut without
leaving any sentence fragments as residue' (Hunt 1970: 189);
`a main clause plus all subordinate clauses and non-clausal structures
attached to or embedded in it' (Hunt 1970: 4).
These four versions suggest that `non-clausal structures' and `sentence
fragments' may be included or excluded from an analysis depending upon
which T-unit de®nition the researcher chooses to adopt.
As can be seen below, other de®nitions repeat Hunt's (1965, 1966, 1970)
basic criteria, although Schneider and Connor (1990) allow a `non-
independent clause' to stand as a T-unit in written data if it is punctuated
as a sentence.
Schneider and Connor (1990: 427): `a T-unit is 1. Any independent
clause plus all its required modi®ers, 2. Any non-independent clause
punctuated as a sentence, 3. Any imperative' (for written data).

Young and Milanovic (1992: 409): `an independent clause and any
associated dependent clauses' (for oral data).

Santos (1988:73): `a single independent clause together with all of its

modifying subordinate clauses' (for written data).

There are indications in the literature that the T-unit de®nition is inadequate
to deal with a full analysis of spoken discourse.5 Tarone (1985) remarks that
she was unable to analyse some of her recorded speech because it was so
dys¯uent, there were so few complete sentences (sic) and so much hesitation
and repetition. Young, also working with oral data, and using the T-unit for
analysis, de®nes this unit in the text unremarkably as `one clause plus any
subordinate clauses' (1995: 19). However, he adds a substantial endnote to
 PAULINE FOSTER et al. 361

this simple de®nition, presumably because he found it insucient for the task
of segmenting his oral data:

The following elements were counted as one t-unit each: a single

clause, a matrix plus subordinate clause, two or more phrases in
apposition, and fragments of clauses produced by ellipsis. Co-ordinate
clauses were counted as two t-units. Elements not counted as t-units
include back channel cues such as mhm and yeah, and discourse
boundary markers such as okay, thanks or good. False starts were
integrated into the following t-unit (Young 1995: 38).

These modi®cations move some way beyond the original idea of a unit as a
main clause plus any subordinate clauses and one might wonder why these
signi®cant modi®cations are footnoted rather than included in the working
de®nition in the text. With this footnote, Young is describing a unit very like
the unit which came to be called a communication unit, or c-unit. This
measure, ®rst described by Loban in 1966, is an attempt to make the T-unit
work with the elliptical nature of the spoken language. Loban ®rst de®ned it
in these terms:
. C-unit: `grammatical independent predication(s) or . . . answers to questions
which lack only the repetition of the question elements to satisfy the
criterion of independent predication. . . . `Yes' can be admitted as a whole
unit of communication when it is an answer to a question such as `Have
you ever been sick?' (Loban 1966: 5±6).
Many, but by no means all, researchers investigating spoken language data
have used the c-unit, though not necessarily Loban's de®nition of it. Pica et al.
(1989: 72) extended Loban's de®nition into the de®nition we have classi®ed
as semantically-based aboveÐa rather vague catch-all term which is not
particularly illuminating or helpful. Lin and Hedgecock (1996), also using the
c-unit for oral data, adopted Chaudron's (1988: 45) de®nition of a c-unit: `an
independent grammatical predication, the same as a T-unit except that in oral
language elliptical answers to questions also constitute predication'. A major
problem with these Loban-based de®nitions is their seeming exclusion of
elliptical constructions which arise within a speaker's turn rather than link to
an interlocutor's question.6
Brock (1986: 52±3) refers to Loban (1966) but surprisingly, she also adopts
Freed's (1978) de®nition that a c-unit `may have several sentence nodes as a
consequence of having several sentences, several clauses, or being a run-on or
compound sentence' (1978: 43). This de®nition, unlike any other for the c-
unit, appears to describe a very long unit capable of encompassing several
`sentences' and is thus outside clear syntactic criteria.
The main choice amongst these units lies between those which are
essentially clause-based (the clause itself, the S-node and the Idea Unit) and
those which are potentially supra-clausal (the c-unit and the T-unit). While
the clause-based units enable easier analysis in manyÐthough not allÐcases,
362 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

the potentially supra-clausal units o€er greater validity. They allow the
analyst to give credit to performers who can embed clauses and hence
construct chunks of speech which re¯ect more sophisticated planning
processes. However, coping with the di€erent degrees and types of embedding
and attachment in these longer units raises a range of problems. It is to these
which we now turn.

3. APPLICATIONS: SILENCE AND SIMPLIFICATION

Occasionally, but not often, researchers exemplify the units they have
employed with models of what they look like. For the most popular, the T-
unit, the following examples have been provided.
When the boy woke up he noticed the frog was gone (Cohen 1989:
148).
He goes to the bookmaker and gets some money (one T-unit) (Lennon
1990: 406).
When you make a milk shake, you mix it in a blender (one T-unit and
2 s-nodes) (Ellis et al. 1994: 483).
Schneider and Connor (1990) are exceptional in providing a whole section of
their data with T-unit boundaries included, a part of which is given below:
There are many di€erent contributions between artists and scientists to
society. || First artists contribute to society for entertainment. || Many
people need it for relax after hard work. || Artists contribute to society
as ®lm artists, singers and so on. || Furthermore artists contribute to
society with make new-work ®elds which are related with kind of
activity. || (Schneider and Connor 1990: 415)
This is straightforward enough, and certainly the common basis for all these
de®nitions (that a T-unit is a main clause with all its dependent/subordinate/
modifying/associated clauses) works well here.
However, the reason it works well is, of course, that the data is written
language. Transcribed spoken data is not nearly so tidy and clear cut.
Application of T-units to the segmentation of oral data requires considerable
modi®cation of the unit. But our survey of recent research clearly showed
many studies of speech have happily used T-units without admitting to any
problems, raising the interesting question of what proportion of the data
ended up, so to speak, on the cutting-room ¯oor. Remaining silent on the
subject, or only providing simple and unproblematic examples of how the T-
unit was employed, shows researchers dodging some dicult but important
questions.
To date, we have been unable to unearth any published examples of
segmentation into c-units of an extended oral text. The emergence of a
segment designed for oral data certainly seems to acknowledge the short-
comings of the classic T-unit, and is, therefore, to be applauded. However, the
 PAULINE FOSTER et al. 363

lack of examples to illustrate what a c-unit might look like, especially one
`with several sentences', means we cannot be altogether certain what a c-unit
analysis might involve. To address this problem, we now examine some
typical features of oral performance to illustrate how dicult it is in practice to
apply these rather spare c-unit de®nitions to the segmentation of oral data.

3.1 `Because' adverbial clauses

Optional adverbial clauses introduced by `because', are frequently problematic
in oral language. This is due to the fact that the relationship between the
`because' clause and its putative main clause is often uncertain, and the word
appears to function as much cataphorically as anaphorically. Additionally,
`because' often performs a discourse marker function. In examples 1, 2, and 3
below `because' could be paraphrased as `I say this because . . .'. This discourse
function is frequently signalled by pause and intonation phenomena.7
1 A: it's usual in this age to get in love with an older 1 person because I'm
talking about what happened to me (1.0) because you see experience
in that older person.
2. A: I think I would ask er some- no er a 1 judge.
B: Yes!
A: because she is quite in dangerous position.
3. B: . . . I would like to to see now a Europe a strong Europe like er maybe
in the eighteenth century (1.0) because er I think we're going to have
a lot of more more troubles erm in the next er decade (1.0) because er
Japan for example or Russia and America are the two countries two
strong countries and Europe is between the two these two countries.
The question here is whether the apparently dependent `because' clauses
should be accorded independent status.

3.2 Co-ordination
T-unit analysis, while it decrees that coordinated main clauses should be
treated as separate units, also rules that coordinated verb phrases with the
same subject should be treated as part of a single unit, as shown in the Lennon
(1990: 406) example quoted above. In the following data however it is clear
from pause and intonation phenomena, that the subsequent verb phrase
constitutes a new start:
4. and the woman, um, (0.5) speaks um, um, go telephone, speaks
telephone, (1.0) and look in in the other direction,
5. sometimes maybe bring some powder milk (1.5) and put it to some some
er water and some nutrients protein or something like that this may be
also sort of arti®cial milking.
364 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

6. The other woman is very happy now (0.5) and (3.0) just walking away
with a gr great smile.
In many instances, non-native speakers will drop subjects in this way,
creating apparent coordinated phrases which are almost certainly not
intended as such.

3.3 `Topical' noun phrases

As Bygate (1988) has noted, independent noun phrase satellite units are
common in speech, and particularly so in the case of second language learners
whose ®rst language is typologically a topic-comment language. Neither the
T-unit, nor the c-unit, as previously de®ned, provides a satisfactory answer to
the analysis of such phenomena. The problem is exempli®ed in the
underlined sections of 7, 8, and 9 below.
7. It's just a matter of passing (exams) And especially the basic education
(0.5) they have to pass automatically from one grade to another
8. Apples are grown in Pakistan These fruits are same as like yours but not
gooseberries and strawberries these are not grown in my country
9. and on the pool three people maybe they are children three people they
are swimming now

3.4 Sca€olding and interruption

Co-operative building up of a conversation, with sharing of units, is common
in highly interactive conversations. However, the analysis of such interactions
raises quite complex questions about how the resulting units are to be divided
up and assigned. Examples 10, 11, and 12 are typical:
10. B: I think um (3.0) yeah it's hard maybe he like her but no like a
girlfriend, just a
A: Just like a sister.
B: Yeah.
11. A: They both look fairly sort of er
B: miserable
A: yeah
B: for a wedding day
12. A: This guy looks like he is some kind of a (1.0)
B: priest?
A: priest. He might be an ayatollah
Given the breadth and range of articles in which the detailed analysis of oral
data has been required (as noted above, we found many in our survey of just
four journals), it is clear that other researchers must have faced similar
 PAULINE FOSTER et al. 365

problems to those outlined above. The dearth of published information about

how their analyses were achieved is frustrating. Oral data is particularly
messy; second language oral data is generally messier than ®rst language data.
Researchers who face the dicult task of segmenting it for analysis are
presently doing so largely without the bene®t of the shared experience of
their fellows. What is needed is an accessible standard unit of analysis, explicit
and exempli®ed, which is psycholinguistically valid, and which can be applied
reliably to a wide range of oral data. In the next section therefore we present a
detailed de®nition of such a unit, illustrated with examples from authentic
oral data, with justi®cation for our choices.

4. THE ANALYSIS OF SPEECH UNIT (AS-UNIT)

The unit we have adopted, and termed the AS±unit, is a mainly syntactic unit.
There are a number of reasons we consider this to be a valid unit for analysing
spoken language. First, the studies of pausing in native-speaker speech suggest
that syntactic units are genuine units of planning, since many pauses occur at
syntactic unit boundaries, and especially clause boundaries (e.g. Raupach
1980; Garman 1990). Secondly, our de®nition allows analysis of speech units
which are greater than a single clause since there is evidence from intonation
and pause features that speakers may plan multi-clause units (Beattie 1980).
We propose that the ability to plan at the multi-clause level is important for
establishing a speaker's level of pro®ciency, and evaluating the complexity of
a particular performance by that speaker. There is a good prima facie case for
interest in such a unit since the ability to produce units with more than one
clause seems to be associated with planned speech (e.g. Foster and Skehan
1996), and with L2 development in instructed students (Tonkyn 1996).
Our unit takes Hunt's T-unit as its starting point and then elaborates this to
deal with the features characteristic of spoken data.

4.1 De®nition of the unit

An AS-unit is a single speaker's utterance8 consisting of an independent clause,
or sub-clausal unit, together with any subordinate clause(s) associated with
either. In the examples that follow below an AS-unit boundary is marked by
an upright slash . . . | . . . A clause boundary within an AS-unit is marked by a
double colon (::). False starts, functionless repetitions, and self-corrections are
put inside brackets {..}.
An independent clause will be minimally a clause including a ®nite verb.
13. | That's right |
14. | Turn left |
15. | I take a di€erent way |
16. | You go to the main street of Twickenham |
366 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

However, unlike the T-unit, our de®nition allows for the inclusion of
independent sub-clausal units, which are common in speech, and speci®es
the nature of these more clearly than has previously been the case where the
c-unit has been used as the unit of analysis (see (b) below). Our unit is still
primarily syntactic because this o€ers an acceptable level of reliability, given
that syntactic units are easier to identify than intonational and semantic ones.
However, in addition to this we argue for the principled use of intonation and
pause phenomena to deal with some awkward cases, and we discuss these
below, together with examples from native speaker and non-native speaker
data which we have been collecting and analysing.
An independent sub-clausal unit will consist of: either one or more phrases
which can be elaborated to a full clause by means of recovery of ellipted
elements from the context of the discourse or situation:
17. A: | how long you stay here |
B: | three months. |
or a minor utterance, which will be de®ned as one of the class of `Irregular
sentences' or `Nonsentences' identi®ed by Quirk et al. (1985: 838±53.)
18. | Oh poor woman |
19. | Thank you very much |
20. | Yes |
A subordinate clause will consist minimally of a ®nite or non-®nite Verb
element plus at least one other clause element (Subject, Object, Complement
or Adverbial).
21. | I serves in in a organization government organization in Bangladesh ::
which is called er department of agricultural extension | (2 clauses, 1 AS-
unit)
22. | I have no opportunity to visit | (1 clause, 1 AS-unit)
23. | it is my hope :: to study crop protection | (2 clauses, 1 AS-unit)
24. | and you you be surprise :: how he can work | (2 clauses, 1 AS-unit)
Since subordination is frequently used as a measure of complexity (see, e.g.,
Crookes 1989; Foster and Skehan 1996; Wigglesworth 1997) it is crucial to
clarify the issue of what constitutes a subordinate clause. In particular, the
classi®cation of non-®nite clauses can be dicult. For example, in the
utterance: `I like reading', `reading' could be analysed as a minimal non-®nite
clause, or simply as an NP. In this analysis, we require at least one additional
clause element to establish clausal status, as in (23) above.
Clause coordination and subordination require additional clear speci®ca-
tion. First, there is the problem of coordinated Verb Phrases, as exempli®ed in
examples 4, 5, and 6 above. We argued that in many cases, the second VP,
 PAULINE FOSTER et al. 367

even if the subject is omitted, constitutes a fresh start for the speaker as shown
in example 6. To address this problem we have identi®ed conditions for
determining where the unit boundary lies.
In cases where coordination of verb phrases occurs, the coordinated phrases
will normally be considered to belong to the same AS-unit, unless the ®rst
phrase is marked by falling or rising intonation and is followed by a pause of
at least 0.5 seconds. This represents a clearly noticable pause which can be
reliably measured without access to specialist equipment.
25. | and they pinned er a notice to his front :: telling everybody :: what he
had done (0.5) | and marched him around the streets with a gun at his
back | (2 AS-units)
Subordinate clauses within an AS-unit can realize the following functions:
(a) subject (initial or postponed)
26. | sometimes it creates problems :: that he knows nothing |
(b) verb complementation (object, complement, or catenative verb comple-
mentation)
27. | and er they told :: that there there was no food crisis |
28. | I wish :: to er visited other areas of England |
29. | I would like :: to ask you :: if you can give me three weeks leave now |
30. | the main object of this organisation is :: to raise up the people's attitudes |
(c) phrasal post-modi®er or complement
31. | still in our country the school and er college students learned the English
:: which were er taught to the students before thirty years. |
Under certain conditions, the subordinate clause within an AS-unit can
realize an adverbial function. Initial and medial examples do not usually cause
a problem: it is clear where they belong as shown in examples 32 and 33:
32. | when I was in the university :: er I have specialized in this er subject |
33. | very early in my university when I was study :: I I did my thesis my
underground thesis in pasture grassland |
However, the loose concatenations of optional subordinate adverbial clauses,
particularly in ®nal position, do cause problems as discussed in examples 1 to
3 above. We have, therefore, established a fairly tight condition for allowing
such clauses to be included in the preceding AS-unit.
Speci®cally, such a ®nal adverbial clause should be within the same tone
unit as at least one of the other preceding clause elements of the AS-unit.
We see this linkage as establishing that the adverbial clause is de®nitely part
of the plan which produced the initial main clause.
368 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

34. | and I can bring him tomorrow together :: where you can talk with him |
(1 AS-unit)
35. | I can under'stand :: when I 'read scien 1 ti®c English | (1 AS-unit)9
36. | specially for 'reading scien'ti®c 1 papers | because er all the 'papers that er
ar'rived to the 'library in 'Chile are 1 English paper | (2 AS-units)

4.2 False starts, repetitions, and self-corrections

Under certain circumstances, it may also be necessary to consider how certain
dys¯uency features, such as false starts, repetitions, and corrections, will be
handled in relation to the unit. These are frequent phenomena in oral
language data, and particularly so in second language data. Di€erent
researchers may wish to deal with actual linguistic material within the false
starts and corrections in di€erent ways, depending on their interests. For
example, if words per unit are being calculated as a measure of complexity it is
necessary to have a principled way of excluding such phenomena from the
total word count. We de®ne a false start as an utterance which is begun and
then either abandoned altogether or reformulated in some way:
37. | {That's cos } you're saying that :: cos you're a man |
38. | {What about } can you give me a credit slip? |
In cases where an AS-unit is produced before the message is abandoned, that
part of the utterance which meets the AS-unit criteria will be counted as an
AS-unit, with the remainder being recorded as a false start.
39. | I have to do this project { because in Brazil } |
| is completely di€erent here |
A repetition is where the speaker repeats previously produced speech. This is a
device which may be used to hold the ¯oor, or to allow time for planning on
line. However, it is necessary to distinguish between those repetitions which
indicate dys¯uency, and those which are used for rhetorical e€ect.
40. | { and one person } er one person enjoy the canoe |
41. | it's a very very bad man |
A self-correction occurs when the speaker identi®es an error either during or
immediately following production and stops and reformulates the speech; self-
corrections will therefore include an element of structural change:
42. | I think ::{ they're a very } they have good time |
43. | I'll work out :: {what should I } what can we do for you |
The following example includes both a self-correction, and a repetition:
44. | so he { decides }{ decided } decided :: to go ®shing |
 PAULINE FOSTER et al. 369

Where a self-correction occurs, the ®nal version is counted as an AS-unit,

with previous versions excluded.10

4.3 Topicalization
Topicalized noun phrases generally belong to the unit of which they are the
topic; thus the following comment includes the (generally pronominalized)
noun phrase repetition of the topic.
45. | but modern oce { they has } they had er many convenient machine |
46. | and some children they are playing the ball |
However, noun phrase satellite units which are separated from the following
AS-unit by falling intonation and a pause (equal to or greater than 0.5 sec)
will be treated as separate AS-units. Thus the underlined section of example 7
above would be treated as a separate AS-unit because the word `education' is
marked by falling intonation and followed by a pause of 0.5 seconds.

4.4 Interruption and sca€olding

In highly interactive discourse, interruption and sca€olding are common and
can present considerable problems for analysis as indicated above. A number
of these problems can be addressed by adopting one of the di€erent levels of
analysis outlined in the next section. Others, however, require speci®c
discussion. In example 47, the subordinate clause is latched to the preceding
utterance by the same speaker, but speaker B has interrupted. In this case, A's
utterances would be analysed as one AS-unit with 3 clauses provided that the
previously stated criteria for including a ®nal adverbial in the preceding AS-
unit are met.
47. A: | oh that's a big problem
B: oh no!
A: :: because my shop's policy is only :: to give the credits for the return
goods |
In the following example, the ®rst speaker (who is a seven-year old child
relating a story from a picture book) cannot access the collective term for bees,
which is provided by her mother (B). In this case, speaker A is credited with
two AS-units in the following example, with the second including a repair
which is concluded after the interjection by speaker B:
48 A: | the dog pushed the tree |
| { the the herd of bees fell } the um
B: | hive the hive of bees|
A: hive of bees fell down |
In example 10, above, `just like a sister' completes the AS-unit of B, and is also
credited to speaker A, as a complete AS-unit.
370 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

Finally, we return to the extract we began with, now coded for AS-units:
A: | { which which }what is your opinion? |
B: (1.0) { maybe er } (5.0) { he } (7.0)
A: | long time? |
| { or it's for } for you it's a major mistake or a small mistake? |
B: | maybe three months |
A: | three months for this one |
| okay |
| for me it's ten |
B: | ten? |
A: | ten years |
B: | yeah |
| ten years |
| oh very long |

4.5 Three levels of application

Up to now, researchers have often felt the need to exclude certain data from
their analyses, but this has frequently been done on what appears to be an ad
hoc basis, with no explanation. The following is an attempt to provide criteria
for the principled exclusion of data where this is necessary for the purposes of
coherent analysis. Thus the three di€erent levels of inclusiveness are proposed
to cope with a range of research purposes and di€erent types of spoken
language data. To clarify further how these levels are intended to work we
have included an appendix with an example of a transcript analysed under
the three di€erent levels.
Level One: to be used for a full analysis of all the data.
. Include everything except untranscribable data, although single inaudible
words of identi®able word class should be included.
Level Two: to be used for highly interactional data. This is for researchers
who are working with interactional data which can yield a high proportion of
minimal units (e.g. one-word minor utterances and echoic responses) whose
inclusion in an analysis could distort the perception of the nature of the
performance.
. Exclude one-word minor utterances:
48. Yes; No; Okay; Uhuh; Right
. Exclude echo responses which are verbatim:
49. A: I think two years
B: Two years
Level Three: This is for use in special cases where analysis of non-
fragmentary AS-units is required. For example, in cases where performances
on di€ering types or sections of OPIs11 need to be standardized. This level
 PAULINE FOSTER et al. 371

takes the principle behind level 2 a stage further. It is principally for

researchers who are primarily interested in what the performer can do in the
production of relatively `complete' units. In this case, largely transactional
data may be required, but interactional interludes may produce, for some
subjects, uncharacteristically short units whose inclusion in the analysis
could distort any comparison of subjects. This level introduces a greater
degree of standardization of the unit, without eliminating elliptical units
altogether.
. Exclude items in level 2 above. In addition:
. Exclude V-less elliptical AS-units involving ellipsis of elements of the
interlocutor's speech.
50. I: what is your mother tongue then
A: Arabic Arabic
. Exclude AS-units involving substitution of clause, predicate, or predica-
tion level units of interlocutor's speech.
51. I: yes but do you ®nd that there are periods say over a period of ten
years when the weather is drier and then over the next ten years it
may be wetter do you ®nd that as well
L: yes I think so
. One or two word greetings and closures can also be excluded in this level
of analysis.

5. CONCLUSION
Clearly, neither the de®nition, exempli®cation, or the employment of units of
segmentation for oral data is a straightforward matter. Anyone embarking on
an analysis of oral data must quickly realize that numerous decisions have to
be made as to how to deal with problematic data which will not easily ®t the
simple de®nitions of units currently available. While researchers spend
considerable time, toiling in the dark, making decisions about how to segment
their data using the unit they have chosen, such detail cannot usually be
included in the research report due to the restriction on space to which
published articles are subject, and thus other researchers may not be able to
bene®t from their insights.
The purpose of this paper is therefore twofold. In the ®rst place we have
sought to identify and discuss the shortcomings of the units of segmentation
currently in widespread use in oral data analysis. We have tried to show that
even where de®nitions are provided and examples are given these are
inadequate to deal with the fragmentary nature of oral data.
In the second place we have sought to provide a useful resource for the
researcher. We have presented a comprehensive de®nition of a unit which is
accessible, clearly de®ned and easily appliedÐcriteria which the AS-unit
372 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

meets. In addition, it is capable of the ¯exibility required to suit a variety of

research purposes, hence the `unit for all reasons' of our title. We have argued
that the unit is valid, and sensitive to genuine di€erences in performance. We
have also provided extensive examples over a wide range of data to assist
researchers in the segmentation of their data and we hope that this detail in
our de®nition will mean that other researchers will be able to use the unit
reliably. As a consequence perhaps, second language researchers with an
interest in oral language data can start to talk the same language.
Revised version received January 2000

ACKNOWLEDGEMENTS
We are grateful to three anonymous Applied Linguistics reviewers for their very helpful and
insightful comments.

APPENDIX
The following example transcripts are analysed into AS-units, with each unit
beginning on a new line. All units would be included in a ®rst level analysis. Units
which would be excluded from a second level analysis are indicated in bold plain type.
Additional units which would be excluded from a third level analysis are indicated in
bold italic type. False starts are inside brackets { . . . } To summarize:
Level 1 = all units
Level 2 = excludes units in bold plain type
Level 3 = excludes units in bold plain type and bold italic type
A: | oh yes |
| hello |
B: | hello |
| can I help you? |
A: | yes I hope so |
| I bought this cassette recorder here last weekend |
B: | yes |
A: | and when I took it home :: I couldn't record with it |
B: | oh |
A: | so er I'm really upset |
| and I want my money back please. |
B: | oh yeah |
| oh very sorry |
| { I think } I apologise for the faulty goods |
A: | oh right |
| well |
| yes |
B: | this shops { is prol is er only give } can't give er the money back |
A: | you can't give my money back? |
B: | yeah |
| very sorry |
 PAULINE FOSTER et al. 373

A: | but it's no good |

| I don't want another one really |
B: | oh { this er } have you got a receipt? |
A: | { well actually } I'm sorry |
| no I can't ®nd the receipt |
| I've looked everywhere |
B: | Ah ah I don't know er |
| I thinks :: er need to go :: to er ask the manager |
A: | oh well yes perhaps I should yes |
B: | yes because you haven't got a receipt |
A: | mm yes |
B: | { I ca } I can't you give money back |
A: | you can't |
A: | oh I see |
A: | oh right |
B: | I'm really sorry |
A: | { what about } can you give me a credit slip? |
B: | er credit slip? |
| er you haven't got { a s } a receipt |
A: | receipt no |
B: | er ask the manager ®rst |
| I think :: it's better |
A: | oh okay |
| alright |
| I suppose so |
| you're only doing your job |
B: | yeah |
A: | alright |
| when can I see the manager please? |
B: | yes so |
A: | now? |
B: | yeah |
A: | oh good |
| alright |
B: | thanks |
A: | good |
| okay |

NOTES
1 LARSP: Language Assessment, Remediation blind eye, or sticking your head in the sand, but
and Screening Procedure. for standing up to be counted'.
2 Memorized sequences of words cannot be 3 A period of time corresponding to the
merely threaded together, like beads on a resources available in the libraries we
string, in order to produce an extended consulted.
utterance. The syntax of a sequence is often 4 It should be stressed that these three
governed by what precedes it, or follows it. categories are broadly de®ned and features
Consider the following, in which the ®rst of the other categories may be taken into
sequence constrains the verb form of the account when evaluating or de®ning units
other three: `This is not the time for turning a in a particular category. We acknowledge
374 MEASURING SPOKEN LANGUAGE: A UNIT FOR ALL REASONS

that it is impossible completely to separate 8 The common phenomenon of sca€olding is

semantic, intonational, and syntactic cri- discussed below in example 47.
teria. 9 : = pitch `hitched up'; ' = level stress; 1 =
5 This is scarcely surprising as it was devel- falling nuclear tone; again, intonational
oped originally by Hunt to measure syn- diacritics are included only in examples
tactic development in the written work of where intonational criteria are being
schoolchildren. What is surprising is that it appealed to.
should ever have been adopted for analys- 10 If the length of AS-units is being counted,
ing interactive oral data. false starts, or items which are replaced for
6 We shall deal with one common form of grammatical or lexical reasons, will typi-
these, the topical noun phrase, in some cally not be included in the count. The
detail below. same would be true of functionless repeti-
7 In these and the following examples the tions. However, the instances of reformula-
number of seconds in a pause is given in tion and repetition may be noted, to
brackets. 1 indicates a falling intonation. establish, say, the ¯uency characteristics of
Intonational diacritics are included only in the speech.
examples where intonational criteria are 11 OPI = Oral Pro®ciency Interview.
being appealed to.

REFERENCES
Beattie, G. 1980. `The role of language pro- variation.' Studies in Second Language Acquisi-
duction processes in the organisation of tion 11: 367±83.
behaviour in face-to-face interaction.' in Crookes, G. 1990. `The utterance and other
B. Butterworth (ed.): Language Production: basic units for second language discourse
Vol 1: Speech and Talk, pp. 69±109. London: analysis.' Applied Linguistics 11: 183±99.
Academic Press. Crookes, G. V. and K. Rulon. 1985. Incorpora-
Brock, C. 1986. `The effects of referential tion of Corrective Feedback in Native Speaker/
questions on ESL classroom discourse.' Non-Native Speaker Conversation. Technical
TESOL Quarterly 20/1: 47±59. Report No. 3. Center for Second Language
Brown, R. 1973. A First Language: The Early Research, Social Science Research Institute,
Stages. London: Allen & Unwin. University of Hawaii: Honolulu.
Butterworth, B. 1980. `Evidence from pauses Crystal, D. C. and D. Davy. 1975. Advanced
in speech' in B. Butterworth (ed.): Language Conversational English. London: Longman.
Production: Vol 1: Speech and Talk, pp. 155±77. Crystal, D., P. Fletcher, and M. Garman. 1976.
London: Academic Press. The Grammatical Analysis of Language Disabil-
Bygate, M. 1988. `Units of oral expression and ity. London: Edward Arnold.
language learning in small group inter- Ellis, N. 1996. `Sequencing in SLA: phonolo-
action.' Applied Linguistics 9/1: 59±82. gical memory, chunking, and points of
Chafe, W.L. 1980. `The deployment of con- order.' Studies in Second Language Acquisition
sciousness in the production of narrative' 18: 91±126.
in W.L. Chafe (ed.): The Pear Stories: Ellis, R., Y. Tanaka, and A. Yamazaki. 1994.
Cognitive, Cultural and Linguistic Aspects of `Classroom interaction, comprehension, and
Narrative Production. Norwood, NJ: Ablex, the acquisition of L2 word meanings.'
pp. 9±50. Language Learning 44: 449±91.
Chaudron, C. 1988. Second Language Classrooms: Foster, P. in press. `Rules and routines: a
Research on Teaching and Learning. Cambridge: consideration of their role on the task-
Cambridge University Press. based language production of native and
Cohen, A. 1989. `Attrition in the productive non-native speakers' in M. Bygate,
lexicon of two Portuguese third language P. Skehan, and M. Swain (eds): Language
speakers.' Studies in Second Language Acquisi- Tasks: Teaching, Learning and Testing, London:
tion 11/2: 135±49. Longman.
Crookes, G. 1989. `Planning and interlanguage Foster, P. and P. Skehan. 1996. `The influence
 PAULINE FOSTER et al. 375

of planning on performance in task based Variables in Speech: Studies in Honor of Freida

learning.' Studies in Second Language Acquisi- Goldman-Eissler. The Hague: Mouton.
tion 18: 299±324. Richards, B. 1987. `Type-token ratios: what do
Freed, B. F. 1978. Foreigner talk: A study of they really tell us?' Journal of Child Language
speech adjustments made by native speakers 14: 201±9.
of English in conversation with non-native Santos, T. 1988. `Professors' reactions to
speakers. Unpublished doctoral dissertation, academic writing of NNS students.' TESOL
University of Pennsylvania. Quarterly 22/1: 69±90.
Garman, M. 1990. Psycholinguistics. Cambridge: Sato, C. 1988. `Origins of complex syntax.'
Cambridge University Press. Studies in Second Language Acquisition 10/3:
Hunt, K. 1965. Grammatical Structures Written at 371±95.
Three Grade Levels. Champaign, Ill: National Scarborough, H. 1989. `Index of productive
Council of Teachers of English. syntax.' Applied Psycholinguistics 11: 1±22.
Hunt, K. 1966. `Recent measures in syntactic Schneider, M. and U. Connor. 1990. `Analys-
development.' Elementary English 43: 732±9. ing topical structure in ESL essays.' Studies in
Hunt, K. 1970. Syntactic Maturity in School- Second Language Acquisition 12/3: 411±27.
children and Adults. Monograph of the Society Snow, C. 1996. `Change in child language and
for Research into Child Development. child linguists' in H. Coleman and
Kroll, B. 1977. `Combining ideas in written L. Cameron (eds.): Change and Language.
and spoken English: a look at subordination Clevedon, Avon: Multilingual Matters.
and co-ordination' in E. Ochs and T. Bennett Tarone, E. 1985. `Variability in interlanguage
(eds.): Discourse across Time and Space. SCOPIL use: a study of style-shifting in morphology
(Southern California Occasional Papers in and syntax.' Language Learning 35/3: 373±405.
Linguistics) no. 5. Tonkyn, A. 1996. `The oral language develop-
Lennon, P. 1990. `Investigating fluency in EFL: ment of instructed second language learners:
a quantitative approach.' Language Learning the quest for a progress-sensitive proficiency
40/3: 387±417. measure' in H. Coleman and L. Cameron
Lin, Y.-H. and J. Hedgecock. 1996. `Negative (eds.): Change and Language. Clevedon:
feedback incorporation among high- Multilingual Matters, pp. 116±30.
proficiency and low-proficiency Chinese- Wagner, K.R. 1985. `How much do children
speaking learners of Spanish.' Language- say in a day?' Journal of Child Language 12:
Learning 46/4: 567±611. 475±87.
Loban, W. 1966. The Language of Elementary Weinert, R. 1995. `The role of formulaic
School Children. (Research Report no. 1). language in second language acquisition: A
Champaign, Ill: National Council of Teachers review.' Applied Linguistics 16/2: 181±205.
of English. Wigglesworth, G. 1997. `An investigation of
Loban, W. 1976. Language Development: Kinder- planning time and proficiency level on oral
garten through Grade Twelve. Urbana, Ill: test discourse.' Language Testing 14/1: 101±22.
National Council of Teachers of English Willis, J. 2000. `Lexical chunks: identifying
Pica, T., L. Halliday, N. Lewis, and frequent phrases, classifying and teaching
L. Morgenthaler. 1989. `Comprehensible them.' Paper presented at Association for
outputs as an outcome of linguistic demands Language Awareness 2000 conference, Uni-
on the learner.' Studies in Second Language versity of Leicester 26±29 June.
Acquisition 11/1: 63±90. Young, R. 1995. `Conversation style in lan-
Quirk, R., S. Greenbaum, G. Leech, and guage proficiency interviews.' Language
J. Svartvik. 1985. A Comprehensive Grammar Learning 45/1: 3±45.
of the English Language. Harlow: Longman. Young, R. and N. Milanovic 1992. `Discourse
Raupach, M. 1980 `Temporal variables in first variation in oral proficiency interviews.'
and second language production' in H. W. Studies in Second Language Acquisition 14/4:
Dechert and M. Raupach, (eds.): Temporal 403±24.