Fostering fluency in second language learning

From LearnLab
Revision as of 15:41, 4 December 2008 by (talk) (Abstract)
Jump to: navigation, search
Project title Fostering fluency in second language learning:

Testing two types of instruction

PI Dr. N. de Jong (faculty, Queens College of CUNY)
Co-PIs Dr. L.K. Halderman (post-doc, University of Pittsburgh)
Dr. C.A. Perfetti (faculty, University of Pittsburgh)
Others with > 160 hours Claire Siskin, Jessica Hogan, John laPlante, Mary Lou Vercelotti
Study start and end dates Study 1: September - November 2006
Study 2: January - March 2007
Study 3: January - March 2007
Study 4: January - March 2008
Study 5: September - November 2008
Learnlab ESL, Speaking courses (levels 3, 4, 5)
Number of participants 200
Total Participant Hours 600 hours
Datashop Recordings of Studies 1, 2 and 3 available


Many studies have investigated the effect of exposure to language on fluency. It has been established, for instance, that fluency increases after a period of immersion or study abroad (Freed et al., 2004; Segalowitz & Freed, 2004). However, few types of instruction have been designed to increase oral fluency, and even fewer have been tested.

One such type of instruction is Nation’s 4/3/2 procedure, in which learners prepare a four-minute talk and repeat it twice to different partners, first in three minutes, then in two minutes (Nation, 1989). He found that the number of hesitations decreased in the retellings, and that sentences were more complex. We may characterize such an outcome as resulting from the fluency pressure exerted by the 4/3/2 procedure. It was not investigated, however, whether the effect transferred to new speeches, which is what we showed in Study 1. Another task that may increase fluency is shadowing, in which student talk along with a recording of a short speech by a native speaker. Shadowing may also increase the feature strength of formulaic sequences, resulting in faster access to them in subsequent production tasks.

This project is transformative in the sense that it moves research on fluency away from single- or multiple-case studies, using technology to collect and analyze larger amounts of oral production data (30 to 40 students per study), and to generate multiple measures of fluency, accuracy, and complexity. For example, not only articulation rate is measured, but also pause length, length of fluent run, and phonation/time ratio.

Study 1 investigated what characteristics of fluency are affected by the 4/3/2 procedure. Measures included the number of syllables per second (speech rate); mean length of fluent runs between pauses; phonation/time ratio; number of interphrasal and intraphrasal pauses; morphosyntactic accuracy; and number of embedded clauses (syntactic complexity). The posttest tested transfer to a different topic.

In Study 2 we investigated whether fluency is further enhanced by a pretraining of formulaic sequences, like the point is that, what I’m saying is that, and and so on). Fast and effortless access to these sequences frees up cognitive headroom which can then be used to construct sentences. This results in fewer and shorter pauses, and/or greater lexical and structural complexity.

Study 3 investigated whether shadowing leads to increased use of formulaic sequences (chunking) and native-like pauses in subsequent production tasks.

In studies 4 and 5 are investigating in further detail how the characteristics of the 4/3/2 task lead to fluency development. In Study 4a, we investigated how time pressure affects the benefits of repetition in terms of fluency, accuracy and complexity. We examined recordings both from the 4/3/2 task itself, and from long-term retention tests. In addition, we investigated the role of specific knowledge components in fluency development.

In Study 4b we tracked how the control and retrieval of specific vocabulary items and morphosyntactic structures develop as a result of the 4/3/2 training. Next, in Study 5, we examined whether priming these same items leads to greater accuracy and fluency during training and later. Data collection for Studies 4a, 4b and 5 took place in the Spring and Fall 2008 semesters, in the English as a Second Language (ESL) learnlab (level 4, higher intermediate). Data analysis is currently in progress.


4/3/2 procedure
A teaching method in which students talk about a topic for four minutes. Then they repeat their speech in three minutes, and again in two minutes.
Repeating speech while it is being spoken.
Formulaic sequence
A sequence, continuous or discontinuous, of words or other elements, which is, or appears to be, prefabricated (see Wray, 2002, p. 9), e.g., The point is that, What I'm trying to say is that, and Take something like.
Articulation rate
Number of syllables per second
Phonation/time ratio
The percentage of time spent speaking as a percentage proportion to the time taken to produce the speech sample
Morphosyntactic accuracy
In this study we will investigate subject-verb agreement, tense errors, definite/indefinite articles
Syntactic complexity
In this study we will investigate the number of embedded finite and non-finite clauses

Research questions

  • Study 1:
    • a. What characteristics of fluency are affected by repetition of a short speech under increasing time pressure (the 4/3/2 procedure)?
    • b. Does knowledge refinement take place during the 4/3/2 training, in terms of morphosyntactic accuracy and syntactic complexity?
  • Study 2:
    • a. Does pretraining of formulaic sequences lead to an increase in their use in the subsequent 4/3/2 procedure and posttest? If so, does this increase overall fluency?
    • b. Does proficiency level affect fluency development during the 4/3/2 procedure?
  • Study 3:
    • a. What characteristics of fluency are affected by shadowing a text with formulaic sequences and a pausing pattern characteristic of spontaneous speech?
    • b. Does shadowing texts with formulaic sequences lead to an increase in their use in the posttest? If so, does this increase overall fluency?

For studies 2 and 3, questionnaire data were collected about the students' contact with the second language (English) and their first language, in terms of types of contact (e.g., listening to the radio, talking to friends, talking to strangers) and amount of contact (number of days per week, number of hours per day). We will explore whether these individual differences affect pretest performance and fluency development.

  • Study 4:
    • a. How do time pressure and repetition affect fluency, accuracy and complexity in the 4/3/2 task?
    • b. Which knowledge components contribute to fluency development in the 4/3/2 task?
  • Study 5:
    • Does priming lead to an immediate and a long-term increase in fluency, accuracy and complexity?

Background and significance

Many studies in the field of second language acquisition that have studied fluency have investigated the effect of study abroad, immersion and regular classroom practice on fluency (Freed, Segalowitz, and Dewey, 2004; Segalowitz & Freed, 2004). Very few studies, however, have investigated specific activities that lead to fluency, which can be done in classrooms. Two such activities are tested in this project.

The first activity that is tested is the 4/3/2 procedure as proposed by Nation (1989). He investigated the development of fluency during this task, but used a limited number of measures and did not test the long-term effect: he only analyzed fluency during the task itself, not during the following weeks. This project will test the long-term effect and will include more measures, such as length and location of pauses. An attempt will be made to link these measures to cognitive mechanisms.

A general effect of the 4/3/2 task on fluency development was found in Study 1. The effect will be investigated in more detail in Study 4a, focusing on the two main characteristics of the task: repetition and increasing time pressure.

The contribution of knowledge components is tested in Studies 2, 4b and 5. In Study 2, students received a pretraining of a set of formulaic sequences before the first 4/3/2 fluency training session. In Study 5 students will receive a pretraining at the start of each 4/3/2 fluency training session. The knowledge components in this pretraining will be selected based on the results of Study 4b, which investigates the role of vocabulary breadth, vocabulary depth and grammatical knowledge in oral fluency.

Study 3 investigated whether shadowing affected fluency development. In addition, it was tested whether the presence of formulaic sequences in the model speeches increased use of those sequences in later speaking tasks, and whether such an increase affected fluency measures.

Dependent variables

  • Temporal measures of fluency:
    • Articulation rate: number of syllables per second
    • Pauses:
      • mean length of fluent runs between pauses
      • mean length of pauses
      • phonation/time ratio
      • number of interphrasal and intraphrasal pauses
    • Formulaic sequences: number of appropriate formulaic sequences repeated from training
  • Accuracy: morphosyntactic accuracy (target-like use of several structures, including subject-verb agreement, tense errors, and definite/indefinite articles; see Mizera, 2006: 71)
  • Complexity: number of embedded finite and non-finite clauses (cf. Nation, 1989); lexical variety as measured by the Mean Segmental Type-Token Ratio (Towell, Hawkins & Bazergui, 1996)
  • Near transfer, immediate and delayed, normal post-test: After completing the last training session, students performed a similar task (spontaneous speech about a given topic), to test whether any gains in fluency during the training task were maintained in a new instance of the same task. This test was given one week and four weeks after the last training session, each time with a different topic. These recordings were made as part of the Recorded Speaking Activities (RSAs) from the project "The self-correction of speech errors (McCormick, O’Neill & Siskin)".
  • Far transfer, delayed: The delayed posttests in Study 4 will include measures of vocabulary breadth, vocabulary depth and grammatical knowledge.

Independent variables

(For screenshots exemplifying some of the independent variables, see Screenshots below.)

  • Studies 1-3: Pretest vs. immediate posttest vs. long-term retention posttest
  • Study 1: Repetition vs. No Repetition
In the Repetition condition students talk about one topic three times. In the No Repetition condition, students talk about three different topics.
  • Study 2:
    • a. Pretraining vs. no pretraining of formulaic sequences
In the Formulaic Sequences condition, students receive a short training of a number of formulaic sequences before they start the fluency training (4/3/2 task). In the No Formulaic Sequences condition, students do not receive this pretraining, and only do the 4/3/2 task.
  • b. Low intermediate vs. high intermediate proficiency level
Low intermediate students are enrolled in ELI Speaking courses at level 3, high intermediate at level 4.
  • Study 3: Shadowing text with formulaic sequences vs. without formulaic sequences
In the Formulaic Sequences condition, students shadow texts that contain formulaic sequences. In the No Formulaic Sequences condition, students shadow the same texts, from which the formulaic sequences that are being studied have been removed.
  • Study 4:
    • a. Repetition vs. No Repetition
The 4/3/2 task will include either 1 topic (repeated) or 3 topics (not repeated)
    • b. Increasing Time Pressure vs. No Increasing Time Pressure
The 4/3/2 task will include recordings of either 4, 3 and 2 minutes, or 3, 3, and 3 minutes
  • Study 5: Pretraining vs. no pretraining of vocabulary and grammar knowledge components
In the Pretraining condition, students perform short tasks before the 4/3/2 training sessions to prime their vocabulary and grammar knowledge. Students in the No Pretraining condition do not receive this pretraining.


  • Study 1: It is hypothesized that repetition of a short speech (independent variable) under increasing time pressure (fluency pressure) increases articulation rate and sentence complexity (dependent variables), and decreases the number and length of pauses (dependent variables). The reason is that repetition will--temporarily--increase the availability of vocabulary and sentence structures (leading to increase speech rate, short and fewer pauses), leaving more cognitive headroom for other processes (higher accuracy and syntactic complexity).
  • Study 2: It is hypothesized that the presence of a pretraining of formulaic sequences (independent variable) leads to an increase in their use in subsequent spontaneous speech (dependent variable). Effortless use of these sequences will free up cognitive headroom for sentence structure planning, which may lead to overall more fluent performance, in terms of speed and pausing patterns (dependent variables). Thus, the training of formulaic sequences may accelerate future learning.
  • Study 2: Students at different proficiency levels may benefit in different ways from the 4/3/2 training. At lower proficiency levels, repetition may facilitate the use of particular words and grammar, leading to more instances of correct usage of vocabulary, morphosyntax and syntax. At higher proficiency levels, on the other hand, repetition may lead to a greater number of reformulations resulting in higher complexity.
  • Study 3: It is hypothesized that shadowing a speech that contains formulaic sequences (independent variable) leads to an increase in their use in subsequent spontaneous speech (dependent variable). Since effortless use of these sequences will free up headroom for sentence structure planning, performance may become more fluent overall, in terms of speed and pausing patterns (dependent variables). Thus, shadowing may accelerate future learning. In addition, shadowing a text with target-language pausing patterns is expected to lead to a more native-like pausing pattern in subsequent spontaneous speech, mainly in terms of position (dependent variables: interphrasal and intraphrasal pauses).
  • Study 4a: We hypothesize that time pressure in combination with speech repetition encourages a strategy of retrieval (shorter pauses, high lexical overlap between two subsequent recordings), while repetition without time pressure encourages computation, leading to higher accuracy and complexity with lower fluency. We expect that the strategy of computation may result in higher fluency in the longer term, because it leads to more refinement and strengthening of knowledge components, thus reducing hesitations and accelerating retrieval in future speeches (cf. the assistance dilemma).
  • Study 4b: We hypothesize that students with a broader vocabulary—-since they have more words to choose from—-will be able to find an appropriate word more often and more quickly, and therefore speak more fluently (shorter and fewer pauses, fewer hesitations). This is a measure of individual differences, and of general vocabulary knowledge. Greater vocabulary depth will also increase fluency, because students have more control over vocabulary items. In addition, it is predicted that retrieval speed for words used in the fluency training will show a increase in retrieval speed compared to items that were used in the fluency training. Finally, we expect to find a positive correlation between fluency in speech production and the accuracy scores on a test of morphosyntactic knowledge.
  • Study 5: We hypothesize that the pretraining will lead to more fluent speech production, as well as more accurate and more complex output. We expect that this effect will be retained on the posttest and delayed posttest. This would be an example of accelerated future learning through feature focusing.
  • Near transfer, immediate: In all studies, a posttest is administered about a week after the last training session. This will be a similar task—a 2-minute monologue—with new content—a new topic.
  • Near transfer, retention: In Studies 1 and 2, another posttest is administered two to three weeks after the immediate posttest (three to four weeks after the last training session). Again, this will be a similar task—a 2-minute monologue—with new content—a new topic.
  • Acceleration of future learning: In Study 5, the students in the experimental condition first receive a pretraining of a number of formulaic sequences. It will be tested whether their fluency, accuracy and syntactic complexity increases more during subsequent training, than of students who do not receive this pre-training.


Data collection for Study 1 was completed in November, 2006. Data for Studies 2 and 3 were collected in the Spring semester of 2007.

Results of Study 1 show that, during each training session, the mean length of fluent runs increased mostly for the two conditions in which speeches were repeated (Repetition and Control/Repetition). In addition, for all three groups, pauses, on average, became shorter, and phonation/time ratio increased (i.e., students were able to fill more time with speech).

Figures 1, 2 and 3 show the mean length of fluent runs for each recording in each of the sessions. It is clear that in sessions 1 and 2 the two conditions in which speeches were repeated pattern more closely together than the condition in which speeches were not repeated. The improvement in performance of the two repetition conditions seems more stable, whereas the performance of the No Repetition condition seems to be influenced by the topic of a particular speech.

FluencyStudy1 SPR-Session1.JPG

FluencyStudy1 SPR-Session2.JPG

FluencyStudy1 SPR-Session3.JPG

Transfer to new topics

On the immediate posttest, students in the Repetition condition are able to produce the same length of fluent runs with shorter pauses. Also, they fill relatively more time with speech (increased phonation/time ratio). It seems, therefore, that they speak more fluently than students in the No Repetition condition. However, on the delayed posttest, the No Repetition condition seems to have caught up with the Repetition condition, also having shorter pause lengths, with stable lengths of fluent runs.

Both groups reach a higher articulation rate, measured in syllables per minutes, on the delayed posttest. This may have been due to their continued Speaking classes in the English Language Institute, and may not have been related to this study.

It should be noted that the posttests were administered one and four weeks after the last session of the fluency training, and involved a new topic, which the students had not talked about during the 4/3/2 training.

  No Repetition (n=9)   Repetition (n=10)   Control&Repetition (n=5)
  Pretest Immediate Delayed   Pretest Immediate Delayed   Pre-pretest Pretest Immediate
    Posttest Posttest     Posttest Posttest2       Posttest
Length of fluent runs (in syllables) * 4.42 4.11 4.27   4.42 4.82 4.69   4.58 4.80 5.44
Pause length (in sec.) * 0.92 1.08 0.96   1.18 0.96 0.99   0.99 0.97 0.84
Phonation/time ratio * 0.59 0.55 0.57   0.54 0.60 0.59   0.57 0.59 0.62
Syllables per minute 194 190 204   196 196 204   209 214 232

* Significant interaction Condition x Time

The results of Study 1 suggest that knowledge becomes more easily accessible in subsequent speeches, since phonation/time ratio increases while pauses on average become shorter and the length of fluent runs is at least stable. In the two conditions in which speeches are repeated, the length of fluent runs increases, indicating an advantage over the no-repetition condition, in which this length is stable. It seems, therefore, that repeating a speech about a particular topic enables students to produce longer fluent runs. The overall advantage of these two conditions over the no-repetition condition parallels patterns in the pre- and posttest data.

Overall, it seems that the performance of the no-repetition condition was more variable across speeches. This may be due to an effect of topic, which may be more or less familiar, complex, or linguistically difficult (e.g., eliciting present vs. past tense).


This project is part of the Refinement and Fluency cluster. The studies in this cluster concern the design and organization of instructional activities to facilitate the acquisition, refinement, and fluent control of critical knowledge components. The general hypothesis is that the structure of instructional activities affects learning. This project addresses the core issues of task analysis, fluency from basics, in vivo evaluation, and scheduling of practice. The 4/3/2 task has been analysed into its components. In Study 1, the effect of the component of repetition is investigated. Practice with the basic skills of using vocabulary and grammar is expected to increase fluency. This will be the case in the Repetition condition, where students have the opportunity to re-use the words, formulaic sequences and grammar in subsequent recordings. In Study 2, students are encouraged to use formulaic sequences that have been taught before training. In Study 3 it is investigated whether shadowing promotes the use of formulaic sequences in spontaneous speech. All three studies take place in an in vivo setting.

Further information

For a summer intern project in June and July 2007, Kara Schultz did a multiple case study of six students from Study 1. The project was a first step towards more in-depth analyses of the data of all three studies in the ESL fluency project, addressing the following research questions:

  • Does the absence of the need to generate new semantic content in the two retellings during the 4/3/2 task free up headroom, resulting in changes in fluency, morphosyntactic accuracy, and complexity?
  • If so, what types of changes occur, and what are the causes for these changes?
  • Is there long-term retention of the changes (one week)?

In September 2007 the PSLC executive committee approved our new project plan, in which we proposed follow-up studies that investigate the effect of time pressure and the role of specific knowledge components (vocabulary, grammar) in oral fluency. These studies will be run in the Spring and Fall semesters of 2008.

Peer-reviewed presentations

De Jong, N., McCormick, D., O'Neill, C., and Bradin Siskin, C., Self-correction and fluency in ESL speaking development. Paper presented at the American Association for Applied Linguistics 2007 Conference in Costa Mesa, California, April 2007.

Other presentations

De Jong, N., Approaches to the study of second language acquisition. Guest lecture at the CUNY Graduate Center (invited by Prof. Den Dikken and Prof. Otheguy), December 2007

De Jong, N., Oral fluency development in ESL classrooms. Guest lecture at the CUNY Graduate Center (invited by Prof. Klein), November 2007

De Jong, N., Presentation of the software component at the Multimedia Showcase sponsored by the Robert Henderson Media Center at the University of Pittsburgh, September 2006


Fluency Summer Intern Project

Fluency Summer Intern Project 2008

Annotated bibliography

Nation, I.S.P. (1989). Improving speaking fluency. System, 17, 377-384.

Freed, B. F., Dewey, D. P., Segalowitz, N., & Halter, R. (2004). The language contact profile. Studes in Second Language Acquisition, 26, 349-356.

Freed, B. F., Segalowitz, N., & Dewey, D. P. (2004). Context of learning and second language fluency in French: Comparing regular classroom, study abroad, and intensive domestic immersion programs. Studies in Second Language Acquisition, 26, 275-301.

Mizera, G. J. (2006). Working memory and L2 oral fluency. Unpublished doctoral dissertation. University of Pittsburgh, Pittsburgh.

Segalowitz, N., & Freed, B. F. (2004). Context, contact, and cognition in oral fluency acquisition. Studies in Second Language Acquisition, 26, 173-199.

Towell, R., Hawkins, R., & Bazergui, N. (1996). The development of fluency in advanced learners of French. Applied Linguistics, 17, 84-119.

Screen shots

Fluency screenshot-notes.jpg

Screenshot of the screen where students take notes before they start speaking.

Fluency screenshot-speech.jpg

Screenshot of the screen where students record their speeches

Fluency screenshot-questions.jpg

Screenshot of some of the questions after each speech has been recorded.