Language Learning ISSN 0023-8333

Fluency Training in the ESL Classroom:

An Experimental Study of Fluency
Development and Proceduralization
Nel de Jong
Vrije Universiteit Amsterdam

Charles A. Perfetti
University of Pittsburgh

The present study investigates the role of speech repetition in oral fluency development.
Twenty-four students enrolled in English-as-a-second-language classes performed three
training sessions in which they recorded three speeches, of 4, 3, and 2 min, respectively.
Some students spoke about the same topic three times, whereas others spoke about three
different topics. It was found that fluency improved for both groups during training but
was maintained on posttests only by the students who repeated their speeches. These
students had used more words repeatedly across speeches, most of which were not specif-
ically related to the topic. It is argued that proceduralization of linguistic knowledge
represented a change in underlying cognitive mechanisms, resulting in improvements in
observable fluency.

Keywords classroom research; English as a second language; fluency training; oral

fluency; proceduralization; second language; task repetition; vocabulary

The ultimate goal of many second-language (L2) learners is to be fluent in the

target language—that is, to be able to express their thoughts easily, with more

The data of the pretests and posttests were presented at the AAAL conference in Costa Mesa, CA,
on April 24, 2007. The authors would like to thank Colleen Davy, Jessica Hogan, Rhonda McClain,
and Laura Halderman for their contributions to transcription, coding, and analysis of the data from
the training sessions. Thanks to Laura Halderman, Mary Lou Vercellotti, and Scott Walters for
their feedback to drafts of this article. Funding for this research was provided by the National
Science Foundation, grant No. SBE-0354420 to the Pittsburgh Science of Learning Center (PSLC;
http://www.learnlab.org).
Correspondence concerning this article should be addressed to Nel de Jong, Vrije
Universiteit Amsterdam, De Boelelaan 1105, 1081 HV Amsterdam, Netherlands. Internet:
[email protected]

Language Learning 61:2, June 2011, pp. 533–568 533


C 2011 Language Learning Research Club, University of Michigan

DOI: 10.1111/j.1467-9922.2010.00620.x
de Jong and Perfetti Fluency Training

attention to meaning than form, in any given situation. Communication should

eventually be smooth, with some processes of production relatively fast and
automatic. Although there is not a single agreed-upon definition in the litera-
ture, fluency is often understood to refer to the flow and smoothness of delivery
(Chambers, 1997; Koponen & Riggenbach, 2000). Beyond this core idea are
some distinctions about fluency that are sometimes made. Lennon (1990), for
example, distinguished between broad and narrow fluency. In the broad sense,
fluency is like general proficiency and includes accuracy and complexity of the
output. In the narrow sense, however, fluency is restricted to temporal measures,
such as length and number of pauses and the number of hesitations (e.g., I uh
like sports) and repetitions (e.g., I like I like sports). Another distinction is made
by Segalowitz (2000), who differentiated between cognitive fluency and perfor-
mance fluency. The former concerns “the efficiency of the operation of the cog-
nitive mechanisms underlying performance,” whereas the latter refers to “the
observable speech, fluidity, and accuracy of the original performance” (p. 202).
Although performance fluency is the goal of many language learners, it is highly
dependent on the knowledge and skills of the speaker, which are the bases of
cognitive fluency. More specifically, fluency is related to the extent of a speaker’s
linguistic knowledge as well as the use of that knowledge, the speed of access,
and control over the available linguistic forms and syntactic devices. However,
because speed of access and control of linguistic forms become fluent only with
much practice in speech production, how to support gains in this narrow sense
of fluency within the reduced opportunities for practice provided by language
classrooms is an important question and is the focus of the present research.
In the following sections, we will first review research on oral fluency in
L2 learning settings. Next, we will give an overview of explanations of fluency
and fluency development as well as measures of fluency.

Oral Fluency in the Context of L2 Learning and Teaching

Fluency and fluency development have been studied in a number of different
contexts, including learning contexts, planning, task repetition, and language
instruction. These include studies of immersion, study abroad, and language
instruction in the home country (e.g., Freed, 1995; Freed, Segalowitz, & Dewey,
2004; Segalowitz & Freed, 2004; Towell, 2002; Towell, Hawkin, & Bazegui,
1996). Emphasizing the limitations of classroom opportunities, the results of
such studies favor immersion settings with home-country language courses
producing the least progress in the measures of rate of speech and length of
runs. DeKeyser (2007) concluded that a “majority of [study-abroad] students

Language Learning 61:2, June 2011, pp. 533–568 534

de Jong and Perfetti Fluency Training

make measurable progress in speaking, especially in terms of fluency, at least

in the programs of longer duration” (p. 211). However, he also argued that
many students make less progress than expected, which is often due in part
to the limited quantity and quality of opportunities for proceduralization and
automatization of rule use.
Gains in oral performance, however, can be achieved through procedures
that theoretically affect the processes of fluent production. Both pretask plan-
ning (Foster & Skehan, 1996; Mehnert, 1998; Yuan & Ellis, 2003) and task
repetition (Bygate, 2001; Bygate & Samuda, 2005; Lynch & Maclean, 2000,
2001) support fluency by freeing up attentional resources. Task repetition, for
example, supports speakers’ selection of words, morphemes, and grammati-
cal structures. Bygate (2001) found that repetition affected performance even
when speeches were 10 weeks apart. Complexity was higher in the repeated
task, but there was a trade-off with fluency, which may have been due to a
shift of attention to complexity. Importantly, fluency was measured only as the
number of unfilled pauses per t-unit; results may have been different if different
measures had been used. Finally, no beneficial effect was found on a repeated
task that had a different topic. Immediate repetition, on the other hand, was
shown by Lynch and Maclean (2000, 2001) to improve accuracy in terms of
phonology, vocabulary, semantic precision, and syntax. In addition, there was
some evidence for gains in fluency.
Despite the research on fluency in task-based learning, few studies have
investigated how instructional techniques affect the mechanisms underlying
the longer term development of fluency. A notable exception is the study by
Snellings, Van Gelderen, and De Glopper (2002). They showed that classroom
instruction can increase lexical retrieval speed—a process of language produc-
tion that supports fluency—even with only 10 encounters over a period of 4
weeks. A follow-up study provided evidence for transfer to a narrative writing
task, in that more of the trained words were used (Snellings, Van Gelderen, &
De Glopper, 2004). However, no effect on global quality was found, and writing
fluency was not examined. Gatbonton and Segalowitz (2005) argued that au-
tomaticity can be promoted by inherently repetitive tasks that elicit formulaic
language and are genuinely communicative and thus fit well in a Communica-
tive Language Teaching classroom setting, although they report no data that
support this conclusion.
The instructional technique that is the focus of the present study was specif-
ically designed for oral fluency development: the fluency workshop, or 4/3/2
procedure (Arevart & Nation, 1991; Maurice, 1983; Nation, 1989; Wood, 2001).
Although this technique has been used in classrooms for several decades, few

535 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

studies have empirically investigated its effects so far. In this task, students
speak about a given topic for 4 min and then retell it twice, as close to verbatim
as possible, in 3 and 2 min, respectively. The 4/3/2 involves task time pressure
and repetition, but in contrast to the studies by Bygate and others discussed ear-
lier, the speeches are repeated immediately. In that way, the speakers have the
additional benefit of having used certain vocabulary and grammatical construc-
tions, which can facilitate retrieval through lexical and syntactic priming (e.g.,
Bock & Loebell, 1990; Branigan, Pickering, & Cleland, 2000; McDonough &
Mackey, 2006; Pickering & Branigan, 1999; Youjin & McDonough, 2008).
Nation (1989) investigated the fluency, accuracy, and complexity of
speeches given in the 4/3/2 task, comparing the first and last speeches. He
found an increase in speech rate (words per minute) and a decrease in the num-
ber of false starts, repeated words, and hesitations (such as uh, um). Accuracy
improved only slightly for half of the participants, mostly when grammatical
contexts were repeated but not for errors that involved inflections. The strate-
gies used by the speakers to fit their speeches into less time included omitting
unimportant details and changing grammatical constructions, which in some
cases involved more complex sentences. Arevart and Nation (1991) replicated
the study with a greater number of participants and found that both speaking
rate (words per minute) and hesitations per minute improved significantly on
the retellings. They concluded that the 4/3/2 task gives learners the opportunity
to speak with higher than normal fluency and complexity during their third
delivery. Neither study tried to tease apart the effects of repetition and time
pressure, nor did they include posttests to examine the long-term effects of the
task, in contrast to the present study.
In summary, the study of fluency has mostly focused on short-term effects
instead of longer term development. Whereas short-term effects on fluency
can be explained by planning and repetition, which enable the speaker to
shift attention and to benefit from priming, longer term effects may require
proceduralization and automatization, which will be discussed in more detail
in the following section.

An Information Processing Perspective on Oral Fluency

Fluency development is often explained in terms of procedural knowledge
and automatic processes, as when Schmitt (1992) characterized fluency as
“an automatic procedural skill” (p. 358) and Segalowitz (2000) argued that
observable fluency reflects the balance between automatic and controlled pro-
cesses (p. 214). Procedural knowledge, or “knowledge how,” is different from

Language Learning 61:2, June 2011, pp. 533–568 536

de Jong and Perfetti Fluency Training

declarative knowledge, or “knowledge that.” Declarative knowledge includes

not only knowledge of word forms and encyclopedic knowledge but also of
explicit grammatical or phonological rules, and it is generally slower to use and
requires more attention and cognitive resources than procedural knowledge.
Because procedural knowledge is processed fast and in parallel with other
processes and because it puts less of a burden on the limited resources of work-
ing memory, it is more suitable for fluent speech. The declarative/procedural
distinction is well known in cognitive psychology and neuroscience (e.g.,
Anderson, 1983; Anderson et al., 2004; Anderson & Lebiere, 1998; Squire,
1987, 1992) and has been applied to L2 acquisition (e.g., De Jong, 2005;
DeKeyser, 1997; Ferman, Olshtain, Schechtman, & Karni, 2009; Hilton, 2008;
Towell, 2002; Towell et al., 1996; Ullman, 2001a, 2001b, 2004). It has been
argued that, in fluent L1 and L2 speakers, procedural knowledge is mostly
involved in the encoding stages of language production (e.g., phrase and clause
structure building) and in articulation, whereas declarative knowledge is asso-
ciated with retrieval of lexical items and their syntactic information as well as
conceptualizing and monitoring (Kormos, 2006; Levelt, 1989, 1999; Towell et
al., 1996). Less advanced L2 learners may still rely on declarative knowledge,
which they need to proceduralize.
The description of the process of proceduralization is found in the ACT-R
model (Anderson & Lebiere, 1998). According to this model, declarative
knowledge takes the form of chunks, which are small independent patterns
of information (e.g., 3 + 4 = 7, or the English regular past tense morpheme
is “–ed”). These chunks can stem from encoding an object in the environment,
like remembering a word, or they can be the result of a previously executed
production (e.g., after computing that 3 + 4 equals 7, the information “3 + 4 =
7” is encoded as a chunk; after creating work + –ed, worked is encoded as a
chunk). Procedural knowledge consists of production rules, which correspond
to steps of cognition and have the basic form of “goal condition + chunk re-
trieval => goal transformation.” For instance, if the goal is to add 3 and 4, the
chunk “3 + 4 = 7” can be retrieved, and then the goal can be changed to the next
goal (e.g., the next step in a multicolumn addition). In ACT-R, each produc-
tion rule is triggered by a goal and retrieves one or, at most, a few declarative
chunks. Production rules are the units of skill acquisition and each has its own
learning curve. Declarative chunks and production rules are competing with
other chunks or rules, and the strongest one is triggered. Strength is determined,
among other things, by amount and recency of use.
Anderson et al. (2004) argued that the change in retrieval speed of declara-
tive chunks follows a power law, with initial practice leading to large gains and

537 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

later practice leading to gains that gradually diminish. Eventually, performance

moves toward an asymptote. Proceduralization involves the creation of produc-
tion rules and combining smaller production rules into larger ones. These new
production rules subsequently need to gain strength so as to be able to compete
with other, previously existing, production rules. Strength can be gained by
repeated practice.
In the context of language, we could say that the retrieval speed of words and
phrases increases with repeated practice, with large initial gains that gradually
diminish to smaller gains. The creation and strengthening of new chunks can
lead to the emergence of formulaic sequences. Language use can also lead to
the construction of new production rules and the collapsing of production rules
into larger ones. Repeated practice is necessary for these collapsed production
rules to be able to compete with (and defeat) their “parent” production rules.

Measures of Fluency
The combination of several measures, as used in the present research, can
give evidence of chunking and proceduralization, as explained below. First,
there is the mean length of pauses measured in seconds. The different ways of
determining pauses and setting cutoff points are discussed below. Second, the
phonation/time ratio is calculated as the percentage of time spent speaking as a
proportion of the total time taken to produce the speech sample. This measure is
related to the number of pauses in a speech: If the mean length of pauses is stable
but the number of pauses decreases, phonation/time ratio increases. Third, the
mean length of fluent runs is the mean number of syllables produced between
pauses. Finally, the articulation rate—in syllables per minute—is calculated by
dividing the total number of syllables produced by the amount of time taken
to produce them, excluding pause time. It is slightly different from speech
rate, which includes pause time. Kormos and Dénes (2004) found that the
first three of these measures were good predictors of fluency ratings by native
and nonnative speaker judges, although articulation rate was not. (Two other
measures not included in this study were also good predictors: speech rate and
pace—that is, the number of stressed words per minute.)
Towell et al. (1996) argued that these measures in combination can be used
as indicators of proceduralization. The number and length of pauses by them-
selves are not reliable indicators of proceduralization, as they vary with task
demands, planning opportunities, and speaker characteristics (some speakers
pause more and longer than others). Another measure to consider is the mean
length of fluent runs (i.e., stretches of speech that are spoken without pauses).

Language Learning 61:2, June 2011, pp. 533–568 538

de Jong and Perfetti Fluency Training

In theory, when proceduralization has taken place, learners are able to produce
longer fluent runs, but a speaker can also produce longer runs by taking more
time for planning, which may show up as longer pauses. Therefore, if the mean
length of fluent runs increases while the mean pause length and phonation/time
ratio are stable, more silent planning time was not needed, which indicates
that encoding and sentence building have been proceduralized.1 Hence, mean
length of fluent runs may be used as an indicator of proceduralization when it is
used in combination with the mean length of pauses and phonation/time ratio.
Finally, articulation rate is a measure of the speed of articulatory processes
and is thus not strongly related to the proceduralization of lexical and syntactic
knowledge.

Purpose of the Study

Despite the evidence for improvements due to task repetition discussed earlier,
it is not obvious whether the 4/3/2 procedure would result in a long term effect
and transfer to new topics. Short-term effects can be explained by reduced
cognitive load due to priming. However, long-term effects and transfer to new
topics would only be expected if the processes underlying speech production
were changed as a result of the 4/3/2 task (e.g., through proceduralization of
knowledge). Accordingly, the goal of this study is to examine the effectiveness
of the 4/3/2 task with a focus on long-term effects, transfer, and proceduraliza-
tion.
The 4/3/2 task has two main features: time pressure and speech repetition.
Time pressure may encourage students to express their ideas more quickly and
efficiently, with shorter pauses and more complex language structures. Speech
repetition, which is our focus here, may lead to an increase in fluency because
of advantages at several levels. First, at the semantic level (conceptualization),
students generate the content for their delivery during their planning time and
while they speak. In the second and third deliveries, they can benefit from this,
which removes the need to pause and hesitate to plan new semantic content.
Second, vocabulary and grammatical structures are generated not so much
during pretask planning as during the first delivery of a speech (cf. Bygate
& Samuda, 2005, p. 65). Again, in subsequent deliveries students can benefit
from having generated content. Even though they may not be able to remember
and reproduce their first delivery verbatim, the words and grammar items they
used are still more activated than before and thus more readily available for use
(the priming effect discussed earlier). Pauses related to lexical searches and
hesitations related to monitoring of grammar are likely to be reduced for these

539 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

items. Overall, when students repeat their speech, they do not have to generate
content (semantic, grammatical, lexical), which frees up cognitive resources,
which can be used in several ways. One way is to speak more fluently, with
shorter pauses and fewer hesitations, as Nation (1989) found. Another way is to
get access to different language items, such as more sophisticated, or specific,
vocabulary and more complex grammatical structures, which is also consistent
with Nation’s findings.
Beyond affecting the accessibility of specific content (declarative knowl-
edge), speech repetition may support proceduralization. For example, if a stu-
dent uses a grammatical item, like a relative clause or an embedded question,
which he/she knows but does not use very often, it might take some time to use
it. Monitoring and corrections may be needed to arrive at the correct construc-
tion (e.g., I don’t know what is his name uh what his name is). However, if the
learner can encode the new chunk and reuse the item in subsequent speeches,
this may strengthen its representation and accelerate retrieval. A new produc-
tion rule may be formed to retrieve the chunk. Similarly, a word that may not
be available when there is a heavy burden on working memory during the first
speech may become available during a subsequent delivery, when more cogni-
tive resources are available. For example, a student may first say that shopping
takes a long time and in a repeated delivery be more specific in saying that it
wastes time. Using the word waste will strengthen its representation and make
it more available on subsequent occasions.
The present study, in order to focus on the effect of speech repetition,
compared students who repeated their speeches with students who spoke about
different topics each time. Speech repetition was expected to increase cognitive
fluency during the 4/3/2 procedure, because it would—temporarily—increase
the availability of vocabulary and sentence structures, which would lead to
shorter pauses and a higher phonation/time ratio. This would leave more cog-
nitive resources for other processes, which, in turn, would lead to a longer
mean length of fluent runs. In addition, it was expected that cognitive fluency
would increase long term, due to the repeated practice. Because the longer term
effect of the training was measured in new speeches about different topics, any
indication of proceduralization must be ascribed to something broader than
changes in the processing of topic-related vocabulary only. If no indication of
proceduralization were to be found and instead only speed of articulation were
to increase, this should show up as an increase in articulation rate only.
The overall research question was whether the 4/3/2 task would lead to a
long-term increase in fluency through proceduralization, leading to the follow-
ing hypotheses.

Language Learning 61:2, June 2011, pp. 533–568 540

de Jong and Perfetti Fluency Training

1. Gains in fluency during the 4/3/2 task and from pretest to posttest would
show evidence of proceduralization, more so for students who repeated
their speeches than for those who did not.
2. Any gains in fluency would persist over 1–4 weeks and would transfer to
new topics, more so for students who repeated their speeches than for those
who did not.
The sources of the fluency gains were examined by studying the fluency
measures and vocabulary use across speeches in the 4/3/2 task.
In contrast to Nation’s (1989) study, in the present study students worked
individually on a computer. By not having students work in pairs, the task
was less naturalistic, but because the influence of a conversation partner was
eliminated, control over the training task was increased. An additional benefit
for the students and teachers was that there was more time for each student to
speak, as they only took on the role of speakers, not of listeners.

Method
Participants
This study took place in an institute for English as a second language (ESL)
at a large university in the United States during the fall semester of 2006.2 All
47 students enrolled in Speaking courses at a high intermediate level (level 4,
three sections) agreed to participate. Students are typically placed at this level
when they have a score of 60–79 on the Michigan Test of English Language
Proficiency (MTELP; Corrigan et al., 1979). Placement is also based on an
in-house listening test and a writing sample. Most students are simultaneously
enrolled in reading, writing, listening, and grammar courses at the same insti-
tute. Due to absences, only 24 complete datasets were available for analysis.
Four students chose not to disclose their background information, including
gender, age, and first language. Of these remaining 20 students, 7 were female
and 13 were male. The age range was 19–37, with an average of 25 years.
All tasks—pretests and posttests as well as all training sessions—were part
of the students’ regular class requirements. First languages spoken included
Arabic (8), Chinese (5), French (2) Korean (2), and single speakers of Spanish,
Japanese, and Portuguese.

Materials
In the 4/3/2 task, students spoke about a given topic for 4 min, then for 3
min, and, finally, for 2 min. The pretest and posttests were part of the regular
course curriculum and consisted of 2-min recorded monologues. These tests

541 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

were part of a larger graded activity for which students also transcribed their
speeches and commented on their accuracy, and teachers gave feedback. These
transcriptions, comments, and feedback were not part of the present study. All
topics, in both the training and the test sessions, were of general interest to
the student population at the language institute and included topics such as
“What do you think about pets?” and “Who is your favorite artist?” The topics
were followed by a few additional questions in order to give students more
suggestions for the contents of their speech (see the Appendix). There may
have been variability in the difficulty of the topics, but this had minimal impact
on the analyses that focused on between-subjects comparisons. Moreover, the
topic of the last 4/3/2 fluency training session was identical to the pretest topic
(Pets) in order to track progress on comparable topics.
Procedures
The participants were randomly assigned to one of three conditions. The first
condition was the Repetition condition: students performed the original 4/3/2
task, in which they spoke about one topic three times. In the second condition
(No Repetition), students performed the same task as the Repetition condition
but spoke about three different topics. The third condition was the Repetition-II
condition, in which students performed exactly the same tasks with the same
topics as in the Repetition condition but later in the semester. The Repetition-
II condition therefore served as a control condition (no training) for the first
part of the study (comparing Test 1 and Test 2), whereas in the second part of
the study (between Test 2 and Test 3), this was a Repetition condition. Note
that Test 2 was the immediate posttest for the Repetition and No Repetition
conditions, whereas it was a pretest for the Repetition-II condition. Students
of two sections were randomly assigned to the Repetition and No Repetition
conditions. Students in a third section were all assigned to the Repetition-II
condition. All students performed the 4/3/2 task three times over a period of
2 weeks.
The pretests and posttests were 2-min speeches given 2 or 3 days before
the first training session (pretest), as well as approximately 1 and 4 weeks after
the last training session (immediate and delayed posttests). All training and test
sessions took place during regular class hours. The tests and fluency training
each had their own introductory session to familiarize the students with the
procedures. Table 1 presents the schedule for the tests and training sessions.
Fluency Training Sessions (4/3/2)
A common misconception among the students, according to the course admin-
istrators, was that fluency is nothing more than speaking fast. In order to avoid

Language Learning 61:2, June 2011, pp. 533–568 542

de Jong and Perfetti Fluency Training

Table 1 Schedule of tests and training sessions

Test Training sessions Test Training sessions Test

1 A B C 2 A B C 3

Repetition a 1, 1, 1 4, 4, 4 7, 7, 7 b c
No Repetition a 1, 2, 3 4, 5, 6 7, 8, 9 b c
Repetition-II a b 1, 1, 1 4, 4, 4 7, 7, 7 c
Note. The numbers and letters refer to the topics as listed in the Appendix.

students trying to speak as fast as possible, which may have prevented other
effects of repetition, students were given a short description of fluency at the
beginning of each 4/3/2 training session. The description included reference to
the temporal factors of length and number of pauses as well as planning and
the use of familiar vocabulary and grammar.
In each 4/3/2 training session, students were given a topic and instructed to
“make a few notes about what you want to say” and “don’t write sentences, but
only a few keywords.” Time for note-taking was 3–5 min, after which students
were encouraged to continue. This pretask planning time was provided for
students to generate enough semantic content to fill 4 min and was given for
each new topic (not for repeated topics). Audio recordings of the students’
speeches were made with the software (16 bits, 22 Hz, one-channel sound
quality). During the speeches, the topic and the student’s notes were presented
on the screen, but the notes could no longer be edited. A clock indicated the
elapsed time.
After the first delivery, students were asked a number of evaluation ques-
tions, which were designed to have them reflect on their performance. Students
checked the boxes of statements about what they would do differently next time,
pertaining to the factors mentioned in the description of fluency (e.g., “I should
use only words that I know well” and “I should pause fewer times”). Other
open-answer questions concerned general performance, such as which difficult
words the students wanted to remember and what superfluous information they
would not include in the next speech. Next, students spoke again, this time for
only 3 min. Students in the Repetition and Repetition-II conditions were asked
to repeat their speech, whereas students in the No Repetition condition saw a
new topic and took new notes. After this second delivery, evaluation questions
followed that asked students to compare their performance to the first delivery.
Then another delivery followed, this time for 2 min only. The session ended
with evaluation questions and a brief questionnaire about the students’ general

543 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

performance in the session. In this questionnaire, students gave ratings on a

7-point scale (1 = disagree; 7 = agree) to 13 statements, including “I knew
what words I wanted to say” and “This exercise was useful.”
All training sessions were run by the first author, and the class teacher
assisted with keeping the students on task. Because of the classroom setting,
and at the request of the teachers, feedback was provided to the students after
each session. Individual reports were written by the first author and handed
out in the next session. These reports included the student’s average length of
pauses, the number of short pauses (<1 s) and long pauses (>1 s), as well as
the overall averages of the students in the study.

Test Sessions
The test sessions were part of the regular curriculum and, therefore, run by
the regular teachers and graduate student assistants. The topics for the pretests
and posttests were selected from among the three topics discussed in the week
before the test. Students were not informed which topic would be on the test. In
contrast, none of the topics in the fluency training were discussed beforehand.
The pretests and posttests were run on Apple PowerMac computers, and
the training sessions were run on Dell personal computers. The software for
the tests and training was developed with the Revolution Studio 2.6.1. package
(Shafer, 2006).

Analyses
All speeches were first transcribed with pauses indicated so that the mean
pause length and the phonation/time ratio could be calculated. Syllables were
counted to compute the mean length of fluent runs. The transcriptions were
coded for parts of speech and retracings (e.g., repetitions, corrections), so
that the number of word types repeated across deliveries could be calculated.
Analyses were performed on the data of students who completed all training
and test sessions: 10 students in the Repetition condition, 9 students in the No
Repetition condition, and 5 students in the Repetition-II condition. Because
of the small group size of Repetition-II, this group was collapsed with the
Repetition condition in the analyses of the training data.

Transcription and Pauses

All speeches from the pretests and posttests were transcribed by the first author,
using PRAAT 4.6.06 (Boersma, 2001). The speeches from the 4/3/2 training
sessions were transcribed by two trained research assistants who checked each
other’s work. To find pauses, the beginning and end of each speech segment

Language Learning 61:2, June 2011, pp. 533–568 544

de Jong and Perfetti Fluency Training

was determined first by using the PRAAT function “To textgrid (silences).” All
pause boundaries were checked and adjusted by the transcribers as necessary,
by listening to the recording and visually inspecting the spectrogram and wave-
form. Nonverbal fillers such as “uh,” “ah,” “um,” and “mmm” were transcribed
and treated as pauses.
A pause was defined as silence or a nonverbal filler of 200 ms or longer.
This cutoff point is slightly lower than the 250–400 ms that other researchers
use (e.g., Freed et al., 2004; Goldman-Eisler, 1961; Segalowitz & Freed, 2004;
Towell et al., 1996), but it follows Lennon (1990) because the majority of the
pauses of 200 ms and longer sounded dysfluent. Although the pause length and
phonation/time ratio computed in this study include both fluent and dysfluent
pauses, any decrease in pause length or increase in the phonation/time ratio is
likely to be due mainly to changes in the number and length of dysfluent pauses.
The upper limit to pauses was set to 2.5 standard deviations above the mean
in a student’s particular speech, which is a fairly conservative and commonly
used criterion for eliminating outliers. Pauses that were longer than this were
replaced by the mean plus 2.5 standard deviations. The trimming was necessary
because some students may have been briefly distracted; very infrequently,
students needed to be encouraged to continue speaking. Such long pauses,
usually around 3 or 4 s, would not be an indication of the students’ fluency or
proceduralization. After the pauses were determined, the phonation/time ratio
was computed by dividing the total time filled with speech (not including silent
pauses and nonverbal fillers like “uh”) by the total time spent speaking (time
filled with speech + pauses and nonverbal fillers).

Syllable Counting
In order to calculate the length of fluent runs, syllables were counted by a
research assistant. Where there was doubt about the number of syllables pro-
nounced (e.g., “every” can be pronounced as /εvri/ or /εv´ri/), the original
recording was consulted. False starts were counted as syllables, but fillers such
as “uh,” “um,” and “mmm” were not. To obtain a reliability measure, syllable
counts of 36 min of speech selected randomly from the 4/3/2 training data were
recounted by the first author. The percentage agreement between the two counts
was 96%. Where there were discrepancies, the difference was usually only one
syllable.

Retracings and Parts of Speech Coding

Prior to the lexical analyses, retracings were coded and part-of-speech tags were
added to the transcripts, using the CLAN software (http://childes.psy.cmu.edu/).

545 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

Retraced words and syllables were included in the syllable counts but not in the
vocabulary analysis. Three types of retracings were coded: without correction
(e.g., it’s [/] um it’s [/] it’s like a dog), with correction (e.g., <the fish is> [//]
the fish are swimming), and with reformulation (e.g., all of my friends had [///]
uh we had decided to go home for lunch).3 The MOR and POST programs of
the CLAN software were run to generate part-of-speech tags for each word.
These tags were needed to generate accurate vocabulary lists for the lexical
analysis (e.g., to distinguish between the noun and verb travel). Ambiguities
were resolved by a trained research assistant who had also transcribed the
speeches, and the tags were checked by a second assistant.
Statistical Analyses
General linear model (GLM) analyses with repeated measures were used to
analyze the fluency data of the 4/3/2 speeches and the pretests and posttests.
Separate GLMs were performed for each measure: mean length of fluent runs,
phonation/time ratio, mean length of pauses, and articulation rate. For the
pretest and posttest data, the within-subjects variable was time (Test 1, 2, and
3) and the between-subjects variable was condition (Repetition, No Repetition,
and Repetition-II). For the analyses of the data from the training sessions,
multivariate GLMs were used with the three sessions (Session A, B, and C) as
measures. Planned post hoc univariate GLMs were performed to analyze the
effects within the training sessions. The within-subjects variable was delivery
(Delivery 1, 2, and 3; respectively, the 4-min, 3-min, and 2-min speeches).
The two Repetition conditions (Repetition and Repetition-II) were collapsed
because they involved the same training tasks. The between-subjects variable
for the training tasks, therefore, was condition (Repetition, No Repetition).
Univariate analyses for each session are also reported. The alpha level for all
statistical tests was set at .05.
Lexical Analysis
In order to assess the extent to which the students repeated their speeches—at
least in terms of vocabulary—the amount of overlap in vocabulary between
pairs of speeches (“lexical overlap”) was calculated by computing the number
of words that were used in all three speech deliveries within a session, in
two deliveries, or in only one. Only lexical words (nouns, verbs, adjectives,
and adverbs) were included in this analysis; retracings were not included. In
addition, correlations were computed between the number of repeated words
and gain scores of the four temporal measures of fluency from pretest to
immediate posttest. Finally, the number of repeated topic-related and topic-
unrelated words was compared.

Language Learning 61:2, June 2011, pp. 533–568 546

de Jong and Perfetti Fluency Training

Results
We first present the results of the pretests and posttests to assess evidence for
proceduralization (hypothesis 1) and for long-term retention and transfer to
speeches about different topics (hypothesis 2). Next, three additional analyses
are presented to examine the source of the fluency gains.

Pretest/Posttest Data
To test the proceduralization hypothesis 1, the temporal measures were analyzed
to find evidence for longer fluent runs with stable or improved length of pauses
and phonation/time ratios. In addition, articulation rate was examined to see
if any improvement concerned speed only. To test the long-term retention and
transfer hypothesis 2, we examined whether gains were retained over 4 weeks
and transferred to different topics.

Proceduralization
Table 2 presents for each of the three conditions the three measures of proce-
duralization: mean length of fluent runs (in syllables), mean length of pauses
(in seconds), and phonation/time ratio. Note that the students in the No Repeti-
tion and Repetition conditions performed the 4/3/2 training between Test 1 and

Table 2 Means and standard deviations of the measures of proceduralization in the

pretests and posttests

Test 1 Test 2 Test 3

Mean length of fluent runs (in syllables)

Repetitiona 4.54 (1.34) 4.83 (1.79) 4.70 (1.53)
No Repetitionb 4.42 (1.55) 4.12 (1.01) 4.28 (.80)
Repetition-IIc 4.58 (1.25) 4.80 (.95) 5.44 (1.20)
Mean length of pauses (in seconds)
Repetition 1.12 (.29) 0.94 (.21) 0.97 (.25)
No Repetition .87 (.13) 1.00 (.27) 0.92 (.11)
Repetition-II .94 (.07) 0.92 (.11) 0.84 (.11)
Phonation/time ratio
Repetition 54.75 (12.13) 60.22 (10.51) 58.63 (9.25)
No Repetition 59.34 (8.85) 55.50 (8.82) 57.11 (7.78)
Repetition-II 57.00 (4.24) 58.63 (5.51) 62.43 (7.67)
a
n = 10.
b
n = 9.
c
n = 5.

547 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

Test 2, whereas the Repetition-II condition performed the training between Test
2 and Test 3.
The GLM analyses showed a significant interaction between time and con-
dition for pause length, F(4, 42) = 3.897, p = .009, partial η2 = .271, and
phonation/time ratio, F(4, 42) = 2.563, p = .052, partial η2 = .196, but not
for mean length of fluent runs. A series of post hoc two-tailed t-tests was per-
formed, comparing Test 1 and Test 2, and Test 2 and Test 3, for each measure in
each condition. The Repetition condition showed significant differences only
between Test 1 and Test 2 for mean pause length and phonation/time ratio,
t(9) = 3.647, p = .005; t(9) = 2.932, p = .017, respectively. Although the
small number of students in the Repetition-II condition was a concern, the
difference in mean length of fluent runs between Test 2 and Test 3 still reached
significance, t(4) = 3.189, p = .033, whereas the difference between Test 1
and Test 2 did not. No significant differences were found for the No Repe-
tition condition. These results showed that in the Repetition condition, mean
pause length decreased and the phonation/time ratio increased, whereas for the
Repetition-II condition, mean length of fluent runs increased and pause length
and the phonation/time ratio were stable. In both conditions, performance only
changed over the time interval in which the fluency training had taken place. As
will be discussed in more detail later, we take both of these patterns of results as
evidence for proceduralization and thus support for hypothesis 1. Performance
in the No Repetition condition did not change over time.

Articulation Rate
Articulation rate, presented in Table 3, was measured as the number of syllables
per minute of speech (pauses excluded). It is considered a measure of speed,
unrelated to proceduralization. The main effect of time was significant, F(2,
42) = 7.232, p = .002, partial η2 = .256, but the effect of condition was not.

Table 3 Means and standard deviations of the articulation rate (in syllables per minute)
in the pretests and posttests

Test 1 Test 2 Test 3

a
Repetition 194.34 (25.21) 195.49 (33.35) 204.22 (40.64)
No Repetitionb 193.91 (22.92) 189.82 (22.84) 203.53 (18.89)
Repetition-IIc 206.94 (27.68) 214.44 (35.04) 232.69 (24.03)
a
n = 10.
b
n = 9.
c
n = 5.

Language Learning 61:2, June 2011, pp. 533–568 548

de Jong and Perfetti Fluency Training

There was no interaction between time and condition. Post hoc two-tailed t-tests
comparing Test 1 and Test 2, and Test 2 and Test 3 did not reveal any significant
results, except the comparison between Test 2 and Test 3 under the No Repetition
condition, which showed a trend, t(8) = 2.255, p = .054, indicating an increase
in articulation rate. However, this increase occurred after training had ended
and may thus be due to the students’ regular language classes. The fluency
training therefore did not seem to effect an increase in speed.
To summarize the students’ performance on the tests, the Repetition-II
condition shows the pattern of performance as expected, with an increase in
length of fluent runs in combination with a stable mean length of fluent runs and
stable phonation/time ratios. This is an indication of proceduralization, which
enables learners to produce longer fluent stretches of speech without additional
time for pausing. The Repetition condition shows a similar pattern in that
length of fluent runs is stable while mean pause length and phonation/time
ratios improve. This indicates that students were producing the same length
of fluent stretches of speech but needed less pause time. Again, this could be
considered evidence for proceduralization, contrasting with the results of the
No Repetition condition, which showed no change in the proceduralization
measures. The evidence thus supports both the proceduralization hypothesis 1
and the retention and transfer hypothesis 2, with gains observed over 4 weeks
and for new topics (see the Discussion section for a fuller discussion of the
evidence for both hypotheses).
Training Data
To examine whether gains would be observed within deliveries during train-
ing, the same fluency measures were applied to the training sessions. Mul-
tivariate repeated-measures GLM analyses were performed for each of the
measures of proceduralization. Each session was a separate measure (training
sessions A, B, and C). Delivery (the 4, 3, and 2-min deliveries) was a within-
subjects independent variable and condition (Repetition vs. No Repetition) was
a between-subjects variable. Because the Repetition (n = 10) and Repetition-II
(n = 5) conditions performed exactly the same training—only at a different
time—and because they both showed indications of proceduralization in their
pretest/posttest data, their training data were collapsed (n = 15) for these anal-
yses. However, for comparison with the test data, Tables 4 and 5 show the
measures for the Repetition and Repetition-II condition separately.
Proceduralization
Table 4 shows the three measures of proceduralization per condition. For mean
length of fluent runs, the multivariate analysis revealed a significant main effect

549 Language Learning 61:2, June 2011, pp. 533–568

 Table 4 Means and standard deviations of the measures of proceduralization during the 4/3/2 training sessions

Session A Delivery Session B Delivery Session C Delivery

1 2 3 1 2 3 1 2 3

Mean length of fluent runs (in syllables)

Repetitiona 4.28 4.32 4.98 4.59 5.27 5.30 4.03 4.34 4.70
de Jong and Perfetti

(1.26) (1.15) (1.73) (1.25) (1.77) (1.51) (1.01) (1.08) (1.44)

No Repetitionb 4.00 5.30 4.60 4.50 4.25 4.77 3.52 3.94 4.58
(0.74) (1.83) (1.30) (0.95) (0.75) (1.25) (0.55) (0.71) (1.03)
Repetition-IIc 4.62 4.73 5.55 5.16 5.37 5.51 5.02 5.12 5.42
(0.83) (0.92) (1.52) (0.96) (0.78) (1.13) (1.35) (0.85) (1.72)
Mean length of pauses (in seconds)
Repetition 1.10 1.08 0.91 0.96 0.98 0.95 0.99 0.89 0.82
(0.28) (0.27) (0.21) (0.21) (0.31) (0.19) (0.23) (0.20) (0.22)

Language Learning 61:2, June 2011, pp. 533–568

No Repetition 0.92 0.79 0.81 0.87 0.85 0.89 0.91 0.90 0.79
(0.23) (0.17) (0.14) (0.11) (0.12) (0.15) (0.12) (0.16) (0.16)
Repetition-II .99 0.84 0.78 0.83 0.78 0.82 0.88 0.81 0.76
(0.22) (0.15) (0.07) (0.12) (0.07) (0.10) (0.22) (0.20) (0.10)
Phonation/time ratio
Repetition 52.62 52.48 59.91 54.49 58.04 58.58 54.48 59.39 61.71
(9.64) (10.09) (8.48) (9.24) (12.03) (8.87) (8.83) (9.23) (11.68)
No Repetition 55.07 62.23 60.52 59.04 58.32 59.55 53.12 55.05 61.13
(10.30) (10.42) (8.44) (6.26) (6.68) (7.79) (7.58) (7.11) (8.26)
Repetition-II 53.86 58.99 62.39 60.69 63.10 62.68 59.13 62.81 63.44
(7.52) (5.10) (5.41) (6.19) (5.50) (3.43) (8.70) (8.17) (6.81)
a
n = 10.
b

550
Fluency Training

n = 9.
c
n = 5.
de Jong and Perfetti Fluency Training

of delivery and a significant interaction between delivery and condition, F(6,

86) = 6.039, p = .000, partial η2 = .296; F(6, 86) = 3.628, p = .003, partial
η2 = .202, respectively. The univariate tests show that the main effects and
interactions reached significance for all three sessions, except the interaction
of delivery and condition in Session C. There seems to have been a fairly
steady increase in length of fluent runs for the two Repetition conditions,
whereas performance in the No Repetition condition was more variable, which
is illustrated by the values in Table 4. For mean length of pauses, the multivariate
analysis revealed only a significant effect of delivery, F(6, 86) = 4.329, p =
.001, partial η2 = .232, showing a decrease in pause length for all conditions.
Univariate analyses showed the effect was only found in Sessions A and C, as the
mean length of pauses decreased. For the phonation/time ratio, the multivariate
analysis showed a significant main effect of delivery and a significant interaction
between delivery and condition, F(6, 86) = 6.516, p = .000, partial η2 = .313;
F(6, 86) = 2.633, p = .022, partial η2 = .155, respectively. The univariate tests
showed that the main effect of delivery only reached significance for Session
A and Session C, as the phonation/time ratio increased, and the interaction
between delivery and condition reached significance only in Session A. Again,
as illustrated by the data in Table 4, there appears to be a steady improvement for
the two Repetition conditions (Repetition and Repetition-II) and more variable
performance in the No Repetition condition.
Overall, the data from the training sessions reveal that the performance
of all groups improved on all three measures, but mostly in the first and last
training sessions (A and C). This may be due to increasing time pressure,
as the available time decreased from 4 to 2 min. In addition, improvements
were steadier for the two Repetition conditions than for the No Repetition
condition, whose performance was more variable, probably because of the
changes in topics during the sessions. In sum, the Repetition groups showed
longer fluent runs with improved length of pauses and phonation/time ratios,
but not over and above the improvements found in the No Repetition group.
Interestingly, the questionnaire data showed that the majority of the students in
the Repetition groups felt the second and third delivery were “easier” than the
first, whereas fewer students did so in the No Repetition group. Nevertheless,
the two conditions’ ratings for the usefulness of the activity were very similar.

Articulation Rate
Table 5 shows the articulation rate during the 4/3/2 training session. The multi-
variate analysis revealed a significant main effect of delivery and a significant
interaction between delivery and condition, F(6, 86) = 6.549, p = .000, partial

551 Language Learning 61:2, June 2011, pp. 533–568

 de Jong and Perfetti

Table 5 Means and standard deviations of articulation rate (in syllables per minute) during the 4/3/2 training sessions

Session A Session B Session C

Delivery Delivery Delivery
1 2 3 1 2 3 1 2 3

Repetitiona 202.45 205.08 211.00 214.32 220.11 225.69 196.04 196.26 205.30
(27.96) (29.67) (32.09) (29.66) (32.81) (30.24) (27.31) (29.19) (27.45)

Language Learning 61:2, June 2011, pp. 533–568

No Repetitionb 203.35 223.78 213.33 205.48 206.89 209.35 197.69 208.71 216.85
(14.92) (20.69) (16.77) (13.68) (15.40) (19.87) (23.13) (25.04) (24.29)
Repetition-IIc 231.44 230.13 243.78 226.08 233.93 235.24 226.81 221.04 238.56
(27.47) (20.98) (30.76) (19.30) (22.89) (30.04) (29.31) (24.77) (32.57)
a
n = 10.
b
n = 9.
c
n = 5.

552
Fluency Training
de Jong and Perfetti Fluency Training

η2 = .324; F(6, 86) = 1.943, p = .035, partial η2 = .156, respectively. Univari-

ate analyses showed that the effect of delivery and the interaction of delivery
and condition were significant in all three sessions except the interaction in
Session B. As Table 5 illustrates, there was an overall increase of articulation
rate. Performance in the No Repetition condition was more variable, which
suggests that articulation rate also depends on the topic of the speech, perhaps
due to interest or familiarity.

Lexical Overlap
Having established the general effects of speech repetition on oral fluency de-
velopment, we turn to a preliminary look at the possible role of word repetition
in these effects. We calculated the number of words that were in only one, two,
or all three speeches of a session. The results from Session A are presented in
Table 6. It is clear that the students in the two Repetition conditions repeated
many more word types across all three deliveries than students in the No Rep-
etition condition, who used more word types in only one speech. However,
the No Repetition group used a wider range of word types than the two Rep-
etition groups, as their total number of word types was higher. The number
of word tokens (not included in Table 6), on the other hand, was similar for
the Repetition and No Repetition groups (162 and 170, respectively), whereas
it was slightly higher for the Repetition-II group (190). However, a one-way
ANOVA contrasting the three groups did not reveal any significant differences.
In sum, more topic-related and topic-unrelated word types were used in all three
deliveries under the two Repetition conditions than under the No Repetition
condition.
In order to examine whether repeated use of vocabulary affected fluency
gains, correlations were computed with the number of repeated word types and
simple gain scores from pretest to immediate posttest. These correlations were
computed for each of the four fluency measures (mean length of fluent runs,
phonation/time ratio, mean length of pauses, and articulation rate). Table 7
shows there were moderate but significant correlations between the number of
words used in three deliveries of a training session and pretest to posttest gains in
the phonation/time ratio. Negative correlations were obtained with mean length
of pauses. In Session C, correlations with these two fluency measures were also
significant for words used in only two out of three deliveries. Correlations
in Session B for phonation/time ratio with the number words used in three
deliveries and in one delivery just missed significance but show a trend in the
same direction as found in Sessions A and C.

553 Language Learning 61:2, June 2011, pp. 533–568

 Table 6 Means and standard deviations of words types repeated in the three deliveries of Session A

Types used in Types used in Types used in

de Jong and Perfetti

three deliveries two deliveries one delivery Total types

All Topic-related All Topic-related All All

A. Repetitiona 23.9 (7.2) 7.7 (4.1) 21.7 (5.7) 3.8 (2.0) 47.2 (10.2) 93
B. No Repetitionb 5.3 (2.2) 0.0 (0.0) 18.8 (4.1) 2.0 (2.3) 116.8 (18.6) 141
C. Repetition-IIc 30.8 (3.1) 11.0 (2.1) 22.0 (9.7) 5.2 (3.8) 53.6 (12.1) 106
One-way ANOVA
F(2, 23) 50.365 29.476 .672 2.688 62.662 13.984

Language Learning 61:2, June 2011, pp. 533–568

p .000 .000 .521 .091 .000 .000
Independent t-test
∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗
A–B n.s. n.s.
A–C n.s. n.s. n.s. n.s. n.s. n.s.
∗∗∗ ∗∗∗ ∗∗∗ ∗
B–C n.s. n.s.
a
n = 10.
b
n = 9.
c
n = 5.
∗
p < .05.
∗∗
p < .01.
∗∗∗
p < .001.

554
Fluency Training
de Jong and Perfetti Fluency Training

Table 7 Correlations between the number of words that were used in three, two or one
delivery in a training session and the gain score from pretest to immediate posttest for
each fluency measure

Mean length of Phonation/ Mean length of Articulation

fluent runs time ratio pauses rate

Session A 3 deliveries 0.335 0.413∗ −0.461∗ 0.375

2 deliveries 0.216 0.201 −0.107 0.024
1 delivery −0.365 −0.504∗ 0.521∗ ∗ −0.274
Session B 3 deliveries 0.275 0.372 −0.444∗ 0.395
2 deliveries 0.227 0.243 −0.277 0.312
1 delivery −0.177 −0.396 0.511∗ −0.218
Session C 3 deliveries 0.382 0.545∗∗ −0.532∗ ∗ 0.335
2 deliveries 0.354 0.470∗ −0.424∗ 0.161
1 delivery −0.328 −0.519∗∗ 0.590∗ ∗ −0.307
Note. n = 24.
∗
p < .05.
∗∗
p < .01.

These correlations indicate that students who repeated more words across
all three deliveries showed greater improvement, in that they were able to fill
more time with speech and have shorter pauses on the immediate posttest
than on the pretest. On the other hand, students who used more words in
only one delivery showed a smaller gain in fluency from pretest to posttest
in terms of phonation/time ratio and pause length. Correlations between the
number of repeated (and nonrepeated) words and mean length of fluent runs
and articulation rate were not significant.
It could be expected that the repeated words were those that were specifically
related to the topic. For example, students speaking about sports are likely to
use words like sports, football, and score in all three deliveries. To examine
this, the number of topic-specific words per delivery was counted. We define
topic-specific words as words that have a clear semantic relationship to the
topic, in that they can be expected to be used with mostly that topic and less so
with another topic in this study. It can be expected that students use the words
soccer and play when talking about sports but not shopping. On the other hand,
favorite can be used for both topics (e.g., my favorite sportsman, my favorite
store) and is therefore considered non-topic-specific. Proper names of people
and organizations were included in the analysis. Table 6 shows that some of the
repeated words were specific to the topic, but, more importantly, most were not.

555 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

For example, student #164 from the Repetition condition used the following
words in two or three deliveries in Session A. The words that can be considered
as specific to the topic (Sports) are underlined.
Words used in all three deliveries (student #164)
favorite (Adj), even (Adv), very (Adv), also (Adv), well (Adv), Beckham
(proper name), David (proper name), soccer (N), sport (N), sportsman (N),
TV (N), be (V), know (V), like (V), make (V), play4 (V), prefer (V),
watching (verb) [18 words, of which 4 were topic-specific]
Words used in two out of three deliveries (student #164)
famous (Adj), good (Adj), healthy (Adj), really (Adv), why (Adv), sometime
(Adv), America (proper name), British (proper name), Cup (proper name),
England (proper name), Europe (proper name), World (proper name), day
(N), friend (N), game (N), man (N), partner (N), thing (N), village (N), do
(V), feel (V), go (V), have (V), practicing (verb), remember (V), say (V),
see (V), watch (V) [28 words, of which 5 were topic-specific]
Only 9 out of these 46 repeated words were specifically related to the topic
of Sports. These observations are in stark contrast to the data of student # 286
from the No Repetition condition. Again, words specific to any of the three
topics are underlined (Sports, Learning English, and Travel).
Words used in all three deliveries (student #286)
good (Adj), example (N), be (V), have (V) [4 words, of which none was
topic-specific]
Words used in two out of three deliveries (student #286)
important (Adj), very (Adv), in (Adv), always (Adv), well (Adv), Colombia
(proper noun), States (proper noun), United (proper noun), country (N),
family (N), kind (N), time (N), know (V), make (V), prefer (V), think (V),
travel (V), want (V) [18 words, of which 1 was topic-specific]
Only 1 of the 22 repeated words was specifically related to one of the
topics. In contrast, the student from the Repetition condition was able to repeat
49 words, only 9 of which can be considered specific to the topic; the other
40 words are more general. Similar patterns were found for the other students
and in the other sessions. In sum, these results indicate that students who were
asked to repeat their speeches did so indeed, repeating many words across the
three deliveries, but most of those words were not topic-specific.
At first sight, it appears that the increase in fluency found in the two
Repetition conditions could be attributed to lexical retrieval, but we argue

Language Learning 61:2, June 2011, pp. 533–568 556

de Jong and Perfetti Fluency Training

Table 8 95% confidence intervals of the three measures of proceduralization on Test 1

Mean length Mean length Phonation/time Articulation

of fluent runs of pauses ratio rate

Lower Upper Lower Upper Lower Upper Lower Upper

bound bound bound bound bound bound bound bound

Repetitiona 3.61 5.46 1.00 1.32 48.30 61.21 177.99 211.69

No Repetitionb 3.45 5.39 0.71 1.04 52.53 66.14 176.67 211.15
Repetition-IIc 3.28 5.89 0.72 1.17 48.17 66.42 183.81 230.07
Note. Mean length of runs is given in syllables; mean length of pauses is given in seconds;
phonation/time ratio is given in percentage; articulation rate is given in syllables per
minute.
a
n = 10.
b
n = 9.
c
n = 5.

that this is not the case. Although those words that were used in all three
deliveries may have been retrieved more easily during the posttest, on average,
students in the two Repetition conditions used only six of the repeated words
on the immediate posttest, which is just three more than in the No Repetition
condition. Moreover, many of these words were high-frequency words such as
be, have, good, make, and think. It can be speculated that it is not the words
themselves but the processing of sentence constructions and expressions they
are used in that was proceduralized. This calls for further analysis, but it is
outside the scope of this article.

Progress on Speeches With Identical Topics

In order to track the students’ progress on more comparable speeches, the
topic on Test 1 was the same as the topic of Session C (for the No Repetition
condition, only the first speech in that session was the same as on Test 1).
Ninety-five percent confidence intervals were computed for each of the four
measures on Test 1 (see Table 8). The values of those confidence intervals
were then compared to the values of the fluency measures of the speeches in
Session C.
A number of observations can be made. First, in the No Repetition condition,
the mean length of fluent runs and phonation/time ratio on the first delivery of
Session C (3.52 and 53.12%, respectively) were close to the lower bound of the
confidence interval of Test 1 (3.45 and 52.53%, respectively).5 Thus, there was
a temporary drop in performance. This may be related to the fact that the first
delivery in Session C was 4 min long, compared to 2 min on Test 1. It may be

557 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

difficult for learners to maintain a high level of performance in longer stretches

of speech (cf. Skehan & Foster, 2005). Interestingly, such a drop was not found
for the two Repetition conditions, indicating perhaps that it was canceled out
by the benefit of the two intervening training sessions. A second observation
is that in the Repetition condition, the mean length of pauses on all deliveries
of Session C was below the lower bound confidence level of Test 1 (i.e., 0.99,
0.89, and 0.82 are below 1.00). This indicates an improvement in performance,
possibly due to the two intervening training sessions. A third observation is
that in the Repetition condition, the phonation/time ratio of the last delivery
of Session C (61.71%) was above the upper bound confidence interval of Test
1 (61.21%). Therefore, it seems that the students in this condition were able
to improve their performance during the session while repeating their speech
and ultimately exceed their level of performance on Test 1. Finally, in the No
Repetition and Repetition-II conditions, the articulation rates of the last delivery
of Session C (216.85 and 238.56, respectively) were above the upper bound
confidence interval of Test 1 (211.15 and 230.07, respectively). This indicates
that the students in these two conditions were able to improve their performance
during the session. In the case of the Repetition-II condition, this may be related
to their repeating the speech. In the case of the No Repetition condition, it may
be due to the difference in topic. The topic of the last speech was e-mail, which
more students had a more favorable opinion of than pets. In sum, it seems that
the students in the two Repetition conditions were able to benefit from speech
repetition to improve their performance (e.g., reaching a substantive increase in
the phonation/time ratio by the last delivery of the 4/3/2 task). In general, pets
seemed to be an unpopular topic with the students in this study, as evidenced
by informal feedback from the students and the content of their speeches, in
which they often expressed negative opinions about pets. This may be a partial
cause of the lower levels of performance on the fluency measures compared
to speeches on other topics. Nevertheless, the drop in performance seemed
smaller for the two Repetition conditions than for the No Repetition condition,
which suggests an effect of the preceding 4/3/2 sessions.

Discussion
Hypothesis 1 was supported, in that the increases in fluency found in the pretests
and posttests show evidence of proceduralization, but only in the two Repetition
conditions. The Repetition-II condition showed the pattern as described by
Towell et al. (1996): increased mean length of fluent runs with a stable mean
length of pauses and phonation/time ratio. The Repetition condition showed a

Language Learning 61:2, June 2011, pp. 533–568 558

de Jong and Perfetti Fluency Training

different pattern: decreasing pause length, increasing phonation/time ratio, and

a stable mean length of fluent runs. These two patterns may reflect alternative
manifestations of proceduralization. Speakers may produce longer stretches
of fluent speech without having to pause more for planning, as measured by
the mean length of pauses and phonation/time ratio. Alternatively, speakers
may produce stretches of speech of the same length but with less pausing for
planning. Therefore, it can be concluded that the students in both conditions
had proceduralized some of their L2 knowledge.
Hypothesis 2 was supported, in that fluency improvements were maintained
over 4 weeks and transferred to new topics, but only in the two Repetition con-
ditions. In the Repetition condition, pause length decreased and phonation/time
ratio increased from pretest to immediate posttest, and this was maintained in
the delayed posttest. In the Repetition-II condition, mean length of fluent runs
increased from the pretest (Time 2) to the immediate posttest (Time 3). This
condition did not have a delayed posttest. In contrast, performance under the
No Repetition condition did not change significantly during the semester. Inter-
estingly, articulation rate (in syllables per minutes) increased mostly from Test
2 to Test 3. However, because this group had not received the fluency training
between these two tests, this increase in articulation rate cannot be ascribed
to the training. Instead, it may have been a result of the continuing Speaking
classes. In sum, these results from the pretests and posttests show that cognitive
fluency increased, but only as a result of speech repetition in the training.
It is important to note that the improvements described here took place
mostly between the pretest and immediate posttest and can thus be ascribed to
the 4/3/2 fluency training. In addition, the fluency measures in the Repetition
condition increased between Test 1 and Test 2 (their pretest and immediate
posttest) and were stable between Test 2 and Test 3, whereas in the Repetition-
II condition, the fluency measures were stable between Test 1 and Test 2 (their
two pretests) and increased between Tests 2 and 3 (their second pretest and
immediate posttest); that is, there were improvements only during the period in
which students received the fluency training. In addition, it should be stressed
that these effects were found on posttests that were administered 1 and 4 weeks
after the last training session and involved new topics. The effect of speech
repetition, therefore, went beyond the training sessions themselves.
In examining the source of the gains, it was found that performance in all
three conditions improved on all four measures during the sessions; however,
performance in the Repetition conditions did not improve over and above
that of the No Repetition condition. It may be speculated that the topics of
the speeches were of influence, as the improvements appear to be steadier in

559 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

the two Repetition conditions (Repetition and Repetition-II) than in the No

Repetition condition with varied topics. Alternatively, gains may reflect not
proceduralization, but momentary effects of time pressure.
One explanation for the differences in longer term fluency gains between
the Repetition and No Repetition groups was revealed by an analysis of lex-
ical overlap, which examined the number and types of words repeated across
deliveries in a training session. There were two notable results. First, students
who repeated more words across deliveries showed greater improvement from
pretest to posttest in terms of the phonation/time ratio and length of pauses.
Second, the analysis of lexical overlap showed that not only the number of
repeated topic-related words was higher in the two Repetition conditions but
also, and more importantly, the number of repeated non-topic-related words.
Moreover, only few repeated words were also used in the immediate posttest
(five or eight from each session in the two Repetition conditions and three or
four in the No Repetition condition). It seems likely, therefore, that procedural-
ization was not a specific lexical effect; rather, the effect may have been in the
repeated use of sentence structures with those repeated words, thus leading to
proceduralization of phrase building. Further analysis is needed to find out if
sentence structures are repeated, what types of structures, and if they contribute
to longer fluent runs and shorter pauses.
In sum, performance fluency and cognitive fluency seemed to have im-
proved from pretest to posttest in the two Repetition conditions (i.e., after
training in which students repeated their speeches). In contrast, performance in
the No Repetition condition improved during the training sessions, but this im-
provement was not retained in the posttests. The effect found in the Repetition
conditions may be ascribed to the proceduralization of linguistic knowledge
due to repeated use. The analysis of the overlap between speeches in terms
of vocabulary showed that the improvements were not limited to vocabulary
related to the topics in the training but to more general vocabulary. A similar
effect may be expected to have taken place for morphological and syntactic
structures, but that requires further investigation.
Overall, the present study shows that repeated practice increases fluency.
In itself this is not a new finding, but it is nontrivial for several reasons. First,
transfer and long-term retention were found. The transfer of improvements to
new topics shows that the results cannot be explained by lexical priming, as
much of the same vocabulary could not be used. Indeed, it was found that only
few words used in the 4/3/2 task were also used in the posttests. In addition, there
was a long-term effect. During the 4/3/2 task itself, across deliveries, fluency
may increase due to lexical and structural priming: When vocabulary and

Language Learning 61:2, June 2011, pp. 533–568 560

de Jong and Perfetti Fluency Training

grammatical structures are more readily available, fewer searches are needed,
reducing the number and length of pauses. Furthermore, fluency may increase
due to planning and attentional resources: Because the students know what to
say and how to say it, they have more resources for retrieving vocabulary and
grammatical structures, again reducing the need for frequent and long pauses.
However, in the present study, the long-term effect cannot be explained by
priming or planning, as there was a delay of several weeks and there were no
differences in planning between the conditions. The effect, therefore, must be
attributed to changes in the students’ underlying knowledge and processing.
Such long-term retention and transfer to new topics have not been shown in
previous research. Many studies of fluency, such as Nation’s (1989) and Arevart
and Nation’s (1991) studies of the 4/3/2 task, did not include immediate and
delayed posttests. Bygate (2001) did not find transfer to a new topic. However,
his study did not include repeated practice, so it is unlikely that there were
changes in underlying processing mechanisms that could have had a broader
effect than planning a particular speech. Often, transfer does not take place from
one task to another because proceduralized knowledge is highly specific and a
common component between tasks is missing. In the present study, however,
the tasks were similar in that they all required oral production of monologues,
albeit about different topics. The common component between the training and
the posttests therefore may have mostly been of a morphological or syntactic,
rather than lexical, nature.
A second reason for the importance of our findings is that the effect of
repetition seems to scale up from item and sentence level practice to longer
stretches of speech of up to 4 min. Effects of repetition at the item or sentence
level have been shown in many studies, both for language learning and other
types of learning (e.g., Anderson et al., 2004; Anderson & Lebiere, 1998; De
Jong, 2005; DeKeyser, 1997; Ferman et al., 2009). Gatbonton and Segalowitz
(2005) argued that inherently repetitive but communicative activities can pro-
mote automaticity. This study shows that the 4/3/2 task is one such activity.
Although in this study students spoke to a computer, in the original 4/3/2 task
students spoke to three different classmates, which is more naturalistic and
communicative.
The repetition of speeches in the 4/3/2 task is likely to have led to changes in
the underlying knowledge and processing mechanisms and cannot be explained
as faster retrieval due to the repeated use of specific lexical items. More likely,
changes affected the encoding stages of language production, like phrase and
clause structure building (cf. Kormos, 2006; Levelt, 1999). Students may have
been able to form new production rules and strengthen them by repeated use

561 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

(Anderson et al., 2004; Anderson & Lebiere, 1998). This could show up as
repeated use of certain phrases and phrase structures and perhaps formulaic
sequences (Towell et al., 1996; Wray, 2002). Proceduralization is considered a
slow process that requires many encounters with the same items. However, in the
present study, words and phrases were repeated relatively few times. Therefore,
it seems the improvement in performance might be a reflection of the initial
stages of proceduralization, in which new production rules are formed that
lead to relatively greater gains in performance. Thus, an important question
remains unanswered: Although there was evidence that proceduralization of
language knowledge took place, it is not clear exactly what knowledge was
proceduralized. It was argued that more than just topic-related vocabulary was
involved. A deeper, qualitative analysis of the types of grammatical structures
used and perhaps the emergence of formulaic sequences will give indications
of what knowledge was proceduralized, and a detailed analysis of syntactic
complexity and accuracy can assess if there was a trade-off among accuracy,
complexity, and fluency or if the 4/3/2 procedure in fact led to higher accuracy
and complexity. However, such analyses would deserve an in-depth discussion
that is outside of the scope of the present article; a separate study is currently in
progress. Finally, future studies will need to include focused tests of vocabulary
and grammar before and after training to identify where development takes
place.

Conclusion
This study not only investigated fluency development but also examined un-
derlying changes in the processing of language knowledge. In addition, it
combined data from training tasks and pretests and posttests, in order to study
long-term effects and to clarify the causes of the increase in fluency. It was
shown that fluency increased during the 4/3/2 task, in which students spoke
three times, for 4, 3, and 2 min, respectively. However, this increase in fluency
only transferred to a speech about new topic when the students had repeated
their speeches in the 4/3/2 training. More importantly, their improvement was
most likely due to proceduralization because after the training they were able
to produce fluent runs of similar lengths but filling more time with speech and
pausing less long. This proceduralization may have been due to the repeated
use of particular words and sentence structures because it was found that those
students who repeated more words across the three deliveries showed higher
gains in fluency from pretest to posttests, even though few of these repeated
words were semantically related to the topic. In addition, very few of these

Language Learning 61:2, June 2011, pp. 533–568 562

de Jong and Perfetti Fluency Training

repeated words were used again in the immediate posttest. Proceduralization,

therefore, clearly concerned more than the retrieval of topic-related words. In
conclusion, speech repetition in the 4/3/2 task may cause changes in under-
lying cognitive mechanisms, resulting in a long-term and transferrable effect
on performance fluency. Having established the overall effect of repetition on
fluency development, a deeper, qualitative analysis of the students’ production
is needed.
Revised version accepted 6 November 2009

Notes
1 Proceduralization of sentence building may in part involve the use of formulaic
sequences, such as the point is that and to give an example. At this point it is not
clear how many such formulaic sequences were used in these data, especially
because many of the sequences may be idiosyncratic; few nativelike sequences
seem to have been used.
2 Although the students in this institute are in an immersion setting—in that the target
language is the dominant language and all classes are taught in the target
language—a large part of the students’ language learning takes place in the
classroom. In addition, due to the design of the study, the effect of the training can
be isolated, and results are expected to be generalizable to nonimmersion classroom
settings. Swain (1991) found that students in a French immersion setting in Canada
had limited opportunities to engage in oral production and much of their “public”
talk was not longer than a clause. In contrast, the 4/3/2 task provides the students
with practice in monologues of up to 4 min.
3 The [/], [//], and [///] symbols indicate the type of retracing, and the < and >
symbols indicate stretches of speech that were retraced.
4 Some words can be seen as related to other topics in the study. These were
examined in the other speeches as well. To give an example, the verb play was also
used in Session B by two students and in Session C by six students, all of whom
used it in only one delivery. In addition, in Session C it was only used with the Pets
topic. In comparison, the same verb was used in Session A by 19 students, 10 of
whom used it in all three deliveries and three used it in two deliveries. Therefore, we
can conclude that the verb play is more strongly semantically related to Sports than
to any other topic in this study. Importantly, the verb play was used by only four
students in Test 1 (Pets), two in Test 2 (Important Person), and none in Test 3
(Biggest Problem) and thus confirms that topic-specific words did not have the
strongest effect on the fluency measures in the tests.
5 In fact, in the No Repetition condition the mean of the mean length of fluent runs on
Test 1 (4.42) is above the upper bound of the 95% confidence interval on the first
speech of Session C (4.18), indicating that performance on the 4/3/2 task is below
performance on Test 1.

563 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

References
Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard
University Press.
Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C., & Qin, Y. (2004).
An integrated theory of the mind. Psychological Review, 111(4), 1036–1060.
Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought. Mahwah,
NJ: Erlbaum.
Arevart, S., & Nation, P. (1991). Fluency improvement in a second language. RELC
Journal, 22, 84–94.
Bock, K., & Loebell, H. (1990). Framing sentences. Cognition, 35(1), 1–39.
Boersma, P. (2001). PRAAT, a system for doing phonetics by computer. Glot
International, 5(9/10), 341–345.
Branigan, H. P., Pickering, M. J., & Cleland, A. A. (2000). Syntactic co-ordination in
dialogue. Cognition, 75(2), B13–B25.
Bygate, M. (2001). Effects of task repetition on the structure and control of oral
language. In M. Bygate, P. Skehan, & M. Swain (Eds.), Researching pedagogic
tasks: Second language learning, teaching and testing (pp. 23–48). Harlow, UK:
Pearson Longman.
Bygate, M., & Samuda, V. (2005). Integrative planning through the use of
task-repetition. In R. Ellis (Ed.), Planning and task performance in a second
language (Vol. 11, pp. 37–74). Amsterdam: Benjamins.
Chambers, F. (1997). What do we mean by fluency? System, 25(4), 535–544.
Corrigan, A., Dobson, B., Kellman, E., Spaan, M., Strowe, L., & Tyma, S. (1979).
Michigan Test of English Language Proficiency. Ann Arbor: University of
Michigan.
De Jong, N. (2005). Can second language grammar be learned through listening? An
experimental study. Studies in Second Language Acquisition, 27(2), 205–234.
DeKeyser, R. M. (1997). Beyond explicit rule learning: Automatizing second language
morphosyntax. Studies in Second Language Acquisition, 19, 195–221.
DeKeyser, R. M. (2007). Study abroad as foreign language practice. In R. M. DeKeyser
(Ed.), Practice in a second language: Perspectives from applied linguistics and
cognitive psychology (pp. 208–226). Cambridge: Cambridge University Press.
Ferman, S., Olshtain, E., Schechtman, E., & Karni, A. (2009). The acquisition of a
linguistic skill by adults: Procedural and declarative memory interact in the learning
of an artificial morphological rule. Journal of Neurolinguistics, 22, 384–412.
Foster, P., & Skehan, P. (1996). The influence of planning and task type of second
language performance. Studies in Second Language Acquisition, 18, 299–323.
Freed, B. F. (1995). What makes us think that students who study abroad become
fluent? In B. F. Freed (Ed.), Second language acquisition in a study abroad context
(pp. 123–148). Amsterdam: Benjamins.
Freed, B. F., Segalowitz, N., & Dewey, D. P. (2004). Context of learning and second
language fluency in French: Comparing regular classroom, study abroad, and

Language Learning 61:2, June 2011, pp. 533–568 564

de Jong and Perfetti Fluency Training

intensive domestic immersion programs. Studies in Second Language Acquisition,

26, 275–301.
Gatbonton, E., & Segalowitz, N. (2005). Rethinking communicative language teaching:
A focus on access to fluency. Canadian Modern Language Review, 61(3), 325–
353.
Goldman-Eisler, F. (1961). The distribution of pause durations in speech. Language
and Speech, 4(4), 232–237.
Hilton, H. (2008). Connaissances, procédures et production orale en l2. AILE, 27,
63–91.
Koponen, M., & Riggenbach, H. (2000). Overview: Varying perspectives on fluency.
In H. Riggenbach (Ed.), Perspectives on fluency (pp. 5–24). Ann Arbor: University
of Michigan Press.
Kormos, J. (2006). Speech production and second language acquisition. Mahwah, NJ:
Erlbaum.
Kormos, J., & Dénes, M. (2004). Exploring measures and perceptions of fluency in the
speech of second language learners. System, 32, 145–164.
Lennon, P. (1990). Investigating fluency in EFL: A quantitative approach. Language
Learning, 40(3), 387–417.
Levelt, W. J. M. (1989). Speaking: From intention to articulation. Cambridge, MA:
MIT Press.
Levelt, W. J. M. (1999). Producing spoken language: A blueprint of the speaker. In C.
Brown & P. Hagoort (Eds.), The neurocognition of language (pp. 83–122): Oxford:
Oxford University Press.
Lynch, T., & Maclean, J. (2000). Exploring the benefits of task repetition and recycling
in classroom language learning. Language Teaching Research, 4(3), 221–250.
Lynch, T., & Maclean, J. (2001). “A case of exercising”: Effects of immediate task
repetition on learners’ performance. In M. Bygate, P. Skehan, & M. Swain (Eds.),
Researching pedagogic tasks: Second language learning, teaching and testing
(pp. 141–162). Harlow, UK: Pearson Longman.
Maurice, K. (1983). The fluency workshop. TESOL Newsletter, 17(4), 29.
McDonough, K., & Mackey, A. (2006). Responses to recasts: Repetitions, primed
production, and linguistic development. Language Learning, 56(4), 693–720.
Mehnert, U. (1998). The effects of different lengths of time for planning on second
language performance. Studies in Second Language Acquisition, 20, 83–108.
Nation, P. (1989). Improving speaking fluency. System, 17(3), 377–384.
Pickering, M. J., & Branigan, H. P. (1999). Syntactic priming in language production.
Trends in Cognitive Sciences, 3(4), 136–141.
Schmidt, R. W. (1992). Psychological mechanisms underlying second language
fluency. Studies in Second Language Acquisition, 14, 357–385.
Segalowitz, N. (2000). Automaticity and attentional skill in fluent performance. In H.
Riggenbach (Ed.), Perspectives on fluency (pp. 200–219). Ann Arbor: University of
Michigan Press.

565 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

Segalowitz, N., & Freed, B. F. (2004). Context, contact, and cognition in oral fluency
acquisition. Studies in Second Language Acquisition, 26, 173–
199.
Shafer, D. (2006). Revolution: Software at the speed of thought. Monterey, CA: Shafer
Media.
Skehan, P., & Foster, P. (2005). Strategic and on-line planning: The influence of
surprise information and task time on second language performance. In R. Ellis
(Ed.), Planning and task performance in a second language (pp. 193–216).
Amsterdam: Benjamins.
Snellings, P., Van Gelderen, A., & De Glopper, K. (2002). Lexical retrieval: An aspect
of fluent second language production that can be enhanced. Language Learning,
52(4), 723–754.
Snellings, P., Van Gelderen, A., & De Glopper, K. (2004). The effect of enhanced
lexical retrieval on second language writing: A classroom experiment. Applied
Psycholinguistics, 25, 175–200.
Squire, L. R. (1987). Memory and brain. New York: Oxford University Press.
Squire, L. R. (1992). Memory and the hippocampus: A synthesis from findings with
rats, monkeys, and humans. Psychological Review, 2, 195–231.
Swain, M. (1991). Manipulating and complementing content teaching to maximize
second language learning. In R. Philipson, E. Kellerman, L. Selinker, M. Sharwood
Smith, & M. Swain (Eds.), Foreign/second language pedagogy research
(pp. 234–250). Clevendon, UK: Multilingual Matters.
Towell, R. (2002). Relative degrees of fluency: A comparative case study of advanced
learners of French. IRAL, 40, 117–150.
Towell, R., Hawkins, R., & Bazergui, N. (1996). The development of fluency in
advanced learners of French. Applied Linguistics, 17(1), 84–119.
Ullman, M. T. (2001a). The declarative/procedural model of lexicon and grammar.
Journal of Psycholinguistic Research, 30(1), 37–67.
Ullman, M. T. (2001b). A neurocognitive perspective on language: The
declarative/procedural model. Nature Reviews: Neuroscience, 2, 717–
726.
Ullman, M. T. (2004). Contributions of memory circuits to language: The
declarative/procedural model. Cognition, 92, 231–270.
Wood, D. (2001). In search of fluency: What is it and how can we teach it? Canadian
Modern Language Review, 57(4), 573–589.
Wray, A. (2002). Formulaic language and the lexicon. Cambridge: Cambridge
University Press.
Youjin, K., & McDonough, K. (2008). Learners production of passives during
syntactic priming activities. Applied Linguistics, 29(1), 149.
Yuan, F., & Ellis, R. (2003). The effects of pre-task planning and on-line planning on
fluency, complexity and accuracy in L2 monologic oral production. Applied
Linguistics, 24(1), 1–27.

Language Learning 61:2, June 2011, pp. 533–568 566

de Jong and Perfetti Fluency Training

Appendix

Topics
Tests
Test 1: a. How do you feel about pets? Do many people have pets in your
country? How are they treated, in general? [Note: This topic is the
same as the first topic in Session C]
Test 2: b. Talk about a person who was very important to you in the
past. Who was this person? Why was this person important to
you?
Test 3: c. What is the biggest problem your country is facing today? How
would you change it?

Training (topics in bold were given in both conditions):

Session A

1. Do you like sports? Why? If you do, what is your favorite sport? Why?
Do you prefer watching the sport or doing it yourself? Who is your
favorite sportsman or sports woman? Give an example of a game in
which he or she played well.
2. Do you think it is important to learn English? Why? Give an example of
a situation in which English is important. Are other languages important
for you? Which languages do you speak? What other languages would you
like to learn? Why?
3. When you travel, what kind of transportation do you use? How do you
prefer to travel if you have a choice? Does distance make a difference?
Give an example of transportation you use for short and long distances.

Session B

4. Do you like shopping? Why? What do you think of shops in the U.S.?
Do you like them? Why? Can you buy everything you want? Give an
example of something from your country that you can’t buy in the U.S.
5. What do you think about cell phones? Do you think they are useful? Give
an example of why they are useful or not useful. How are cell phones used
in your country?
6. What do you think about television? What do you like about it? What don’t
you like? Give an example of something you like and something you don’t
like about television.

567 Language Learning 61:2, June 2011, pp. 533–568

de Jong and Perfetti Fluency Training

Session C
7. How do you feel about pets? Do many people have pets in your country?
How are they treated, in general? [Note: This topic is the same as the
topic in Test 1]
8. What kind of clothing do you usually wear? Why do you like it? Is clothing
in the U.S. different from clothing in your own country? How? Give an
example of clothing that is different in your country.
9. What do you think of e-mail? Is it a good way to keep in touch with your
family and friends? Do you prefer e-mail, phone, or letters? Why? Give an
example of when you use e-mail.

Language Learning 61:2, June 2011, pp. 533–568 568