Language Learning 49:4, December 1999, pp.

549–581

The Effects of Task Repetition

on Linguistic Output

Susan Gass
Michigan State University

Alison Mackey
Georgetown University

María José Alvarez-Torres and Marisol Fernández-García

Michigan State University

This article explores form/meaning relationships,

focussing on the use learners make of their internal L2
linguistic resources as a function of focus on meaning.
Native speakers of English watched video segments 4
times while recording their own on-line rendition in Span-
ish. One group watched the same video 3 times and the

Susan Gass, and María José Alvarez-Torres, Department of English; Alison

Mackey, Department of Linguistics; Marisol Fernández, Department of Ro-
mance and Classical Languages.
Partial funding for this project was provided by a federal grant to
establish a National Foreign Language Resource Center at Michigan State
University—Grant #P229A60012. A preliminary version of this article was
presented at the Pacific Second Language Research Forum (PacSLRF),
Tokyo, 1998. We are grateful for the comments of the participants in the
symposium on Task Design and Interlanguage, organized by Jonathan
Newton. We also acknowledge the important contributions of India Plough
and Charlene Polio who played a significant role with various parts of this
research. We also thank Jenefer Philp and David Yarowsky for their help.
Finally, we express our gratitude for the astute comments of Language
Learning reviewers and the editor. All remaining errors are ours.
Correspondence concerning this article may be sent to Susan Gass,
English Language Center, A-714 Wells Hall, Michigan State University,
E. Lansing, MI 48824. Internet: gass@[Link]

549
550 Language Learning Vol. 49, No. 4

other group watched different videos each time. At Time 4

both experimental groups saw a new video. A control group
saw videos only at Time 1 and Time 4. Analyses were
conducted on the basis of overall proficiency, morphosyn-
tax, and lexical sophistication. The results provide limited
support for the prediction of improvement over time for the
group that saw the same video, but no support for a
“carryover” effect when the content changed.

The recent literature on second language acquisition (SLA),

and in particular, SLA within a classroom context, has witnessed
a growing interest in focus on form and focus on meaning. This
article explores the form/meaning relationship from the point of
view of the use that learners may or may not make of their internal
L2 linguistic resources as a function of greater or lesser focus on
meaning. That is, how do learners extend and refine their L2
linguistic knowledge?
One area of central concern to the study of form-meaning
processing is known as input processing. Input processing focusses
on the conversion of input to intake during comprehension and on
the form-meaning connections that occur while learners are pro-
cessing the input (VanPatten, 1995, 1996; VanPatten & Cadierno,
1993; VanPatten & Sanz, 1995). In a series of experiments, Van-
Patten and his colleagues presented a model for instructional
intervention that relies heavily on the notion of attention to form
and the crucial role it plays as a learner moves from input to intake
and finally to output. Within this approach, there are certain
operating assumptions: (a) For acquisition to take place learners
need to attend to form, (b) humans are limited in their processing
capacity, (c) attention has a limited capacity—not everything can
be attended to at once, and (d) form competes with meaning for
attention.
VanPatten and his colleagues compared two instructional
models: a traditional instruction model and a processing instruc-
tion model. In the case of the traditional instruction model, gram-
mar was taught via a form of output manipulation; in other words,
Gass, Mackey, Alvarez-Torres, and Fernández-García 551

information was presented to the learner to then be practiced. In

the second model, the processing instruction model, there was an
attempt to change the way input was perceived and processed.
Rather than allow an internalized system to (begin to) develop,
the input processing model attempts to influence the way that
input is processed and hence the way the system develops. The
results of these experiments (both sentence level and discourse
related) suggest a positive effect for processing instruction. Par-
ticipants in the processing instruction group were better able to
understand and produce target structures than participants in the
traditional instruction group (VanPatten & Cadierno, 1993; Van-
Patten & Sanz, 1995).
In another study, VanPatten (1990a) explicitly considered the
question of whether learners can consciously attend to both form
and meaning as they process input. The results of this study
suggest that attending to morphosyntactic form and attending to
meaning are often in a competitive relationship.
Another series of studies that considers the role of teacher-
induced attention to form is the “garden path” research by
Tomasello and Herron (1988, 1989) in which learners (a) are
guided down the garden path into making errors of overgenerali-
zation, or (b) receive explicit rule instruction (in the VanPatten
framework, this could be considered akin to the input processing
mode in which the focus is on processing input before internaliza-
tion of that input). They found that corrective feedback was more
meaningful after learners had been induced to produce an over-
generalization.1
Within the second language literature, much focus in recent
years has been on the concept of attention and the extent to which
it is a necessary condition for learning (cf. Schmidt, 1995, for a
review; Ellis, 1994 [implicit/explicit learning]; Leow, 1997, 1998;
Robinson, 1994, 1996; Schachter, Rounds, Wright & Smith, in
press; Tomlin & Villa, 1994). Within the second/foreign language
classroom there are many ways of drawing learners’ attention to
form, including explicit instruction, structure-based tasks that
target specific predetermined structures (cf. Gass, 1997; Mackey,
552 Language Learning Vol. 49, No. 4

1994; McDonough & Mackey, in press), and interaction (Gass,

1997; Long, 1996—the latter including well-discussed topics of
negotiation and recasts).
In this article, we approach the issue of attention from a
different but related perspective. Our study focusses on the ability
learners have to utilize their L2 knowledge2 in production. In
particular, we investigate whether there is evidence of greater
targetlike production when the need to focus on meaning3 has
been minimized through task repetition, thereby freeing learners
to attend to form, not from input, but from their own internal
system.
Two studies have direct relevance for our research, both
relating to the repetition of tasks and the opportunity for allocat-
ing attention to meaning/form. In the first, Bygate (1996), using
Levelt’s model of processing, suggested that variation in task
selection and use can impact oral processing. He argued that the
effects of task repetition may be relevant for learning, teaching,
and testing. In Bygate’s study, 11 participants (no control) orally
retold a video story and then retold the same story 10 weeks later.
Video segments were about a minute and a half in length. Using
various measures of linguistic complexity, Bygate reported (a) an
increase in number of arguments per t-unit by half of the group,
(b) reuse of certain phrases, and (c) an increase in comments
“about the story” or metalinguistic comments, as he called them.
In reviewing these comments, he found that learners increased
their provision of scene-setting information, providing motiva-
tions and intentions for video characters, and their “abstracts” of
what was about to happen.
Another relevant study was carried out by Skehan and Foster
(1997, 1999), who proposed that more structured tasks would
result in greater fluency and/or accuracy, but that linguistic com-
plexity4 was unrelated to task structure. Using a complex meth-
odology, similar in one respect to the study reported here, as well
as the same oral story–retelling tasks used here, Skehan and
Foster differentiated Mr. Bean video segments (the prompt for the
oral retelling) on the basis of story complexity, categorizing
Gass, Mackey, Alvarez-Torres, and Fernández-García 553

unstructured tasks as ones without a predictable story line and

structured tasks as ones with predictable story lines. In their
study, for example, a restaurant scene was classified as a struc-
tured task because the story line could be anticipated in terms of
the courses served and the waiter and customer’s expectations
about the progression of the meal. A “crazy golf” scene was classi-
fied as unstructured because the story line was unanticipated. The
protagonist, in an effort not to disobey the “rules” of the game by
handling the ball, hit the ball all around town. Their study con-
sisted of four conditions: (a) watch and tell on-line (this was the
condition most related to our study), (b) watch one time and then
watch again while telling on-line, (c) watch and retell immediately
afterward, and (d) watch and retell later (delayed). Skehan and
Foster found that fluency, but not accuracy or complexity, in-
creased when there was predictable structure.5 Our study is
similar to theirs in that we too consider familiarity of content; in
our case we induced familiarity, whereas in theirs they assumed
it by virtue of world knowledge.
In this study, we are focussing on the effects of task repetition
on the linguistic output of second language learners. Task repeti-
tion in this study was designed to allow learners to familiarize
themselves with the content of a short video extract, thus freeing
up their attentional resources so they could focus on form rather
than on content, which, as discussed above, we equate with mean-
ing. In other words, we deal with the use that learners make of
existing knowledge, particularly the conditions surrounding this
use. The question of how knowledge comes to be has been ad-
dressed extensively in the SLA literature (see the numerous works
on Universal Grammar, e.g., White, 1998, for a recent example;
work on attention by Schmidt, 1995, and others; and work by Gass,
1997) and is beyond the scope of this article.
In the SLA literature of the past decade, there have been
arguments for the role of output in language learning. Swain
(1995, p. 127) argued that through output “learners can ‘stretch’
their interlanguage to meet communicative goals. They might
work towards solving their linguistic limitations by using their
554 Language Learning Vol. 49, No. 4

own internalized knowledge, or by cueing themselves to listen for a

solution in future input” (see also Swain, 1985; Swain & Lapkin,
1995, 1998). We further assume that among the aspects of language
to be accounted for is the control of knowledge that learners have.
Bialystok and Sharwood Smith (1985), Bialystok (1994), and
McLaughlin (1987) argued for the dual role of (a) knowledge repre-
sentation and (b) language control (i.e., the control that one has
over language knowledge). The latter includes the speed and
efficiency with which language information can be accessed. In
other words, learners may have knowledge of certain features, but
they may not have acquired control over that knowledge. As
Lightbown (1998, p. 183) pointed out, “opportunities to use the
utterances in discourse-appropriate contexts help learners get
this control.” It is control that we are concerned with in this article,
and the conditions that influence control.
We further explore whether increased targetlike production
in second language form carries over to a new context. In other
words, if production of a new linguistic form emerges or if there is
greater accuracy in production of an existing form, there is reason
to believe that there may be generalization from one context to
another (cf. Tarone & Liu, 1995). Our research questions are as
follows:
Research question 1: Does task repetition yield more
sophisticated language use?
Research question 2: Will more accurate and/or sophisti-
cated language use carry over to a new context?
Our general hypothesis was that familiarity with the content
of the story (i.e., having watched and retold a story several times)
results in more resources being available for issues of language
form. Our specific hypotheses are as follows:
Hypothesis 1: Task repetition results in greater “overall”
proficiency.
Hypothesis 2: Task repetition yields greater accuracy in
morphosyntax.
Gass, Mackey, Alvarez-Torres, and Fernández-García 555

Hypothesis 3: Task repetition results in greater lexical

sophistication.
Hypothesis 4: For proficiency and morphosyntactic accu-
racy, change will carry over to a new context.

Method

In this study, students watched snippets from Mr. Bean

videos. Mr. Bean videos involve short vignettes with a protagonist
who gets himself into unusual and comical situations. We selected
video snippets (approximately 6–7 min in length) in which there
was no audio or in which the audio could be removed without
interfering with the overall continuity or comprehension of the
episode. The videos were played in a language laboratory so that
participants could watch the video while simultaneously recording
their own on-line rendition in Spanish of what was happening.
(See Appendix A for instructions.)

Participants

Participants in the study were 103 students6 in their fourth

semester of Spanish at a large university in the United States.

Procedure

There were two experimental groups and one control group.

The two experimental groups watched a Mr. Bean video four times
and the control group saw a Mr. Bean video twice. For the experi-
mental groups, each viewing was separated by 2 or 3 days, with the
exception of the fourth and final viewing, which took place 1 week
after the third viewing. Approximately 2 weeks separated the two
control group viewings. Sections of fourth-semester Spanish par-
ticipants were randomly assigned into one of three groups.7
1. Experimental Group 1. This group saw the same Mr. Bean
episode three times followed by a fourth Mr. Bean video that
556 Language Learning Vol. 49, No. 4

was a different episode. We refer to this as our Same Content

group (n = 32).
2. Experimental Group 2. This group saw a different Mr. Bean
episode each of the 4 viewing days. Their first video was the
same as the Same Content group’s first video and their fourth
video was the same as the Same Content group’s fourth video.
This second group we refer to as the Different Content group
(n = 33).
3. Control Group. The control group saw the first video when
the other groups saw the first video, and they saw the fourth
video when the other groups saw the fourth video (n = 38). They
saw nothing in between.
The design is schematized in Table 1, which includes the names
of the video episodes.
Our criteria for selection of the particular episodes were
similar to those used by Skehan and Foster (1997, 1999). First, we
isolated approximately equal-length snippets from the full version
of the episode. Second, we chose the episodes to be of such a nature
that participants would find them amusing and understandable
and without a particular British cultural bias. Our selection

Table 1

Design of Present Study

Time 1 Time 2 Time 3 Time 4

Group 2–3 days later 2–3 days later 1 week later

Same Content Library Library Library Packing

(n = 32)
Different Content Library The guard Lunch Packing
(n = 33)
Control Library Packing
(n = 38)
Note. “Library,” “Packing,” and so on refer to the names of the Mr. Bean
episodes used in the present study. Average number of minutes per trial =
6.8; hours of data = 38.53.
Gass, Mackey, Alvarez-Torres, and Fernández-García 557

differed from Skehan and Foster’s in that we were not considering

structured versus unstructured tasks. All of our episodes would
probably fit into their unstructured category because most aspects
of the sequence of events were not predictable.

Analytic Procedures

Holistic change: Magnitude estimation. There are a number

of measures used for analysis in this study; the first measure
(utilized to test Hypothesis 1, relating to improvement in overall
proficiency) deals with holistic change across testing sessions
(Time 1 to Time 3 and Time 1 to Time 4). These holistic assess-
ments were made by native speakers of Spanish who were asked
to judge Time 1, Time 3, and Time 4 on-line renditions to see if
they would judge the later ones as superior to Time 1.
The methodology used to ascertain holistic change was mag-
nitude estimation. This is a well-established research tool used in
a variety of disciplines (e.g., psychophysics: Stevens, 1956; linguis-
tics: Bard, Robertson, & Sorace, 1996; Fucci, Ellis, & Petrosino,
1990; Green, 1987; Grosjean, 1977; Pavlovic, Rossi, & Espesser,
1990; Takefuta, Guberina, Pizzamiglio, & Black, 1986; Toner &
Emanuel, 1989) when one not only wants to rank items as one
better than another, but also wants to know how much better X is
than Y. That is, we can easily rank something, in this case speak-
ers, into an order of 1–9, but we want to know if each of the
rankings is equidistant from the others, and if not, the magnitude
of the ranking differences. Magnitude estimation, the reliability
and validity of which are widely accepted (see references above),
was “developed by psychophysicists to make maximal use of
participants’ ability to make fine judgments about physical stim-
uli” (Bard et al., p. 32) and in recent years has been extended
to other disciplines, including language research. As Bard et
al. (p. 41) point out, there are a number of positive aspects to
magnitude estimation, among which are: (a) Researchers do not
set the number of values that are used to measure the particular
property of concern and (b) one can observe meaningful differences
558 Language Learning Vol. 49, No. 4

that directly reflect differences in impressions of the property being

investigated. This latter point was particularly germane to our study.
We first describe what it was that raters were asked to judge
and then discuss the way magnitude estimation judging works. As
mentioned, native speakers of Spanish were asked to judge
whether Time 3 and Time 4 episodes demonstrated “better Span-
ish” than Time 1 episodes. Thirty participants, 10 from each of the
two experimental groups and 10 from the control group, were
selected on the basis of having produced the most speech in each
of their respective groups.8 In Table 2 we present the number of
words used for the 30 students about whom judgments were made.
Each rater listened to six or nine excerpts, three different trials of
three different individuals (two trials from individuals in the
control group). These excerpts were randomly ordered. Each par-
ticipant was rated by three different raters.
In a magnitude estimation procedure, each rater determines
his or her own scale for a particular stimulus. Each subsequent
stimulus is rated according to the basis established from the
previous stimulus. This is perhaps best understood by showing
how it works with a physical stimulus. In this study, as in other
studies in linguistics using magnitude estimation (cf. Bard et al.,
1996), raters were trained on the physical stimulus of line length.
To begin, each rater was shown a horizontal line and was asked
to assign a number to it. They were then shown another horizontal
line and were asked to assign a number to it in comparison to the
length of the previous line. This continued for another 10 lines.
The 12 lines were between 26 and 152 mm. The first line shown
was in the middle range—77 mm. It was suggested that they use
a scale larger than 10. This makes subsequent ratings easier to
work with because raters are dealing with multiples of a previous
number. In other words, it would be difficult if raters were to start
with a number such as 555 and then wanted to view the next
stimulus as 1.5 times better. (For a lengthy discussion of the
procedure, particularly as it relates to language-related issues, see
Bard et al.)
Table 2

Word Count for Participants in Magnitude Estimation Analysis

Participant M SD Participant M SD Participant M SD

1 SC 436.75 58.59 1 DC 457.75 59.04 1 C 532.50 45.96

2 SC 368.25 101.71 2 DC 380.50 54.02 2 C 405.50 31.82
3 SC 358.50 36.83 3 DC 351.75 47.93 3 C 358.00 14.14
4 SC 343.75 63.96 4 DC 328.25 26.02 4 C 343.50 26.16
5 SC 343.50 58.19 5 DC 310.00 24.37 5 C 336.00 52.33
6 SC 327.25 56.42 6 DC 272.25 49.28 6 C 330.00 63.64
7 SC 303.25 64.07 7 DC 268.50 95.44 7 C 326.50 12.02
8 SC 259.00 103.95 8 DC 266.00 23.76 8 C 278.50 10.61
9 SC 259.00 38.58 9 DC 262.75 24.42 9 C 278.00 1.41
10 SC 257.50 73.31 10 DC 261.25 33.01 10 C 277.50 70.00
M = 325.68 M = 315.90 M = 346.60
Note. SC = Same Content; DC = Different Content; C = control.
560 Language Learning Vol. 49, No. 4

After working with line length and before listening to the

actual tapes, raters9 listened to a training tape. They heard three
samples of 1 min each which they rated using the magnitude
estimation methodology.10 For the actual rating, raters listened to
the first 2.5 min11 of each participant’s tape. (See Appendix B for
rater instructions.) For purposes of analysis, to compare the mag-
nitude of improvement judged by each rater it is necessary to
convert the unique scales created by individual raters into a
logarithmic scale.12
Morphosyntax: Ser and estar. The question of change in the
area of morphosyntax is rather complex because we did not focus
participants’ attention on any specific areas of the grammar. Each
participant was free to focus on whatever s/he wanted to. Hence,
group data may not reveal differences; one student may have
focussed on agreement and another on verb tenses, for example.
We did, however, select one area for reporting here on the basis of
the frequent and pervasive difficulty learners have. To determine
change in the area of morphosyntax (Hypothesis 2), we considered
two copula verbs in Spanish, ser/estar. The difference between
these two verbs is notoriously problematic for English learners at
the level of instruction of our participants given that English
collapses the two senses into one copular verb to be. We hypothe-
sized that because of the nature of the task, we could expect
frequent use of these verbs to describe characters in the video or
situations and locations of these characters. The verbs in question
are ser and estar, both with rough translations of to be. In Figures
1 and 2 we present a summary of the uses of each of these verbs
with examples from our data.13
To be included in the analysis used to determine whether
there was positive change from Time 1 to Time 3 in the use of ser
and estar, a participant had to have produced at least two tokens
of the verb under consideration. If not, that participant was
eliminated from consideration for this part of the analysis.
Lexical sophistication. Hypothesis 3 deals with lexical so-
phistication. Popular measures used in describing lexical change
include lexical originality, lexical density, lexical variation, lexical
Gass, Mackey, Alvarez-Torres, and Fernández-García 561

Ser identifies a subject or an object

Es un lápiz para pintar

it is a pencil for to draw
‘It is a pencil for drawing.’

Ser, combined with an adjective, expresses inherent characteristics that define

a subject or an object

el señor Bean no es muy inteligente

the Mr. Bean not is very intelligent
‘Mr. Bean isn’t very intelligent.’

Ser refers to the time and location

pero es la hora y el guardia va a una posición nueva

but it is the time and the guard goes to a position new
‘But it is time and the guard goes to a new post.’

Ser is also used with the preposition de (‘s, from) to indicate origin, and
possession

Las camisas son de Hawaii

the shirts are from Hawaii
‘The shirts are from Hawaii.’

Figure 1. Summary of uses of ser. (See also Cubillos, 1996, and Gutiérrez,
Martínez-Lage, & Rosse, 1995.)

frequency profiles, and lexical sophistication. In what follows we

briefly define each one and then specify which measure we used
and why.14
Lexical originality has usually been operationalized as the
percentage of words used in one particular person’s writing, but
no one else’s in the same group. Lexical density has generally been
defined as the percentage of lexical words in the text (as opposed
to function words). These terms generally correspond to open-
versus closed-class words. According to this measure, a text is
considered dense if there are many lexical words relative to
562 Language Learning Vol. 49, No. 4

Estar is used with adjectives or adverbs of manner to describe the state or

condition of the subject which are true at a given moment, but not necessarily
permanent

el soldado no está contento

the soldier not is happy
‘The soldier isn't happy.’

Estar indicates location of persons/things

el señor Bean está en el parque

the Mr. Bean is in the park
‘Mr. Bean is in the park.’

Some adjectives have different meanings when used with ser and estar

la biblioteca es muy aburrida

the library is very boring
‘The library is very boring.’

me parece que está un poco aburrido

to me it appears that he is a little bored
‘It seems to me that he is a little bored.’

Figure 2. Summary of uses of estar. (See also Cubillos, 1996, and Gutiérrez,
Martínez-Lage, & Rosse, 1995.)

function words. Criticisms of such measures (see, e.g., Laufer &

Nation, 1993, 1995) point out that in written texts, fewer function
words in a composition may reflect subordinate clauses or ellipsis,
reflecting structural, not lexical, characteristics.
Lexical variation has been operationalized as the type-token
ratio between the different words in the text and the total number
of all words. Lexical variation measures are often sensitive to the
length of a text, and sometimes fixed numbers of words—for
example, the last 200 words in the text—have been used as the
basis for the measure. Other problems include the fact that if
Gass, Mackey, Alvarez-Torres, and Fernández-García 563

derivations are counted separately rather than as a “word family,”

counts may be inflated, depending on what the researcher is
looking at. If production of different derivations are to be consid-
ered indicative of development, then, of course, there will be no
problem.
Lexical frequency profiles proposed by Laufer and Nation
(1993, 1995) are claimed to detail increases in the size of the active
lexicon. A lexical profile is constructed by specifying where on the
list each lexical item would fall, and determining the ratio of use
of each entry of the list. Thus, in this system, a learner’s profile of
production at one time could read: 75%-10%-10%-5%. This would
mean that on a composition consisting of 200 word types, 150
belong to the first 1,000 most frequent, 20 to the second 1,000, and
so on. Lexical frequency profile results to date, although promising
in terms of validity and reliability, are preliminary in terms of their
accounts of profile stability and length and type of text (written).
Lexical sophistication, the measure with which we are con-
cerned in this article, has been defined as the number of more
advanced or sophisticated words expressed as a ratio of the total
words produced. Our measure, based on a list of 200 most com-
monly used words in Spanish, is similar to the lexical frequency
profile. In this study, “advanced” was operationalized as words not
on the list of the 200 most commonly used words.15 To summarize
very simply, however, our measure is more basic in nature than
the lexical frequency profile, in that we classified lexical items only
in terms of whether they did or did not appear on a list of 200 most
commonly used words.
In this consideration of the lexicon, our intention was to
explore whether the Same Content group would have a more
extensive use of vocabulary by Time 4 than would the other two
groups. As noted earlier, at Times 1 and 4 all groups saw the same
video, and thus the contexts for vocabulary were equivalent and
could be compared. We predicted, following Ard and Homburg
(1992), that because there would be little need to focus on vocabu-
lary already practiced or known, more resources could be freed up
to focus on little-used vocabulary. To summarize, for this analysis
564 Language Learning Vol. 49, No. 4

we explored Time 1 to Time 4 comparisons and compared the

words used with a standard list of 200 most commonly used words.
We considered both open-class words and closed-class words. In
the data set as a whole, there were 76,207 words. We used a
Unix-based natural language processing tool to classify, tag, and
count these words. We looked at increases in production broken
down by four classes of counts:
1. increases in open-class word production
2. increases in closed-class word production
3. increases in production of all words
4. increases in production of medium- to low-frequency words
(measure of lexical sophistication)
We quantified increase in production as an increase in the
number of unique word types per fixed number of tokens. These
type-token ratios were created in order to explore any changes in
production of different types of words across the different groups.
We carried out an analysis of type-token ratios rather than simple
word counts because these reflect learners’ increase in lexical
diversity. Simple word counts do not take into account the fact that
continued use of the same word is not illustrative of a wide range
of word choice. For example, many learners produced large quan-
tities of “El Señor Bean” (Mr. Bean).

Results

As noted above, three measures were used to address the

research questions and our specific hypotheses.
1. A holistic measure (magnitude estimation) was carried out
between Times 1 and 3 and Times 1 and 4 (Hypotheses 1 and 4).
2. An analysis of morphosyntax between Times 1 and 3 and
between Times 1 and 4 was carried out to assess whether more
accurate production of morphosyntax could be observed, and if
so, whether it generalized to new contexts (Hypotheses 2 and 4).
Gass, Mackey, Alvarez-Torres, and Fernández-García 565

3. An analysis of lexical sophistication was carried out be-

tween Times 1 and 4 to ascertain whether the learners’ lexi-
cons improved across generalizable contexts. Because lexical
use is highly context dependent, the lexical comparison was
not made between Times 1 and 3, because at Time 3, the two
experimental groups saw different videos (Hypothesis 3).

Holistic Judgments

For holistic ratings, we first examine Time 1–Time 3

differences.
Figure 3 presents the percentage of improvement for the
10 participants of the two experimental groups at Times 1 and
3. There are two important points to note. First, the Different
Content group started out higher than the Same Content group.
Second, and perhaps more importantly for our purposes, the
Same Content group showed greater improvement than the
Different Content group. Their scores at the end were higher
than those of the Different Content group, despite the fact that
they started lower. Differences on a Mann-Whitney U, however,
were nonsignificant.16
We next wanted to explore the extent to which the overall
trends seen for grouped data reflected individual evaluations.
These can be seen in Figure 4, which presents the results for both
the Same Content and Different Content groups. As can be seen,
overall, the magnitude of improvement was greater for individuals
in the Same Content group (sum of differences = 4.56) than for
individuals in the Different Content group (sum of differences =
3.39). This is the same trend shown in the grouped data reported
above.
As noted earlier, one of our hypotheses (Hypothesis 4) relates
to the extent to which improvement that was evidenced during
treatment would carry over to a new context. Figure 5 shows a
comparison of grouped data for Times 1, 3, and 4 and Times 1 and
4 for the control group. As can be seen, the Same Content group
decreased considerably and the Different Content group decreased
566 Language Learning Vol. 49, No. 4

51

50
Log. Magnitude estimation

49

48

47
Same content
46
Different content
45

44

43

42

41
Time 1 Time 3

Figure 3. Change on holistic judgments (magnitude estimation) from Time 1

to Time 3 for the Same Content and Different Content groups

Figure 4. Comparative results between Time 1 and Time 3 for the 10 greatest
producers in the Same Content and Different Content groups
Gass, Mackey, Alvarez-Torres, and Fernández-García 567

slightly in the holistic ratings.17 There was a slight increase for

the control group from Time 1 to Time 4.
What we have done thus far is examine holistic judgments of
improvement. The next results relate to specific areas of improve-
ment in morphosyntax and the lexicon.

Ser/Estar

Table 3 (ser) and Table 4 (estar) show that there was a ceiling
effect for both groups in their accurate use of ser. Because there
was so little opportunity for improvement, it is unclear how to
interpret the positive change that did occur. However, the situation
with estar is different.18 Here there was a greater percentage of
participants who improved in the Same Content group as op-
posed to those who improved in the Different Content group
(44% versus 30%). No statistical analyses were performed due
to the fact that upon close examination, the starting points of
the two groups who showed improvement (13 participants in
the Same Content group and 9 in the Different Content group)
51

50
Total log. Magnitude estimation

49

48

47
Time 1
46 Time 3
Time 4
45

44

43

42

41
Same Different Control
content content

Figure 5. Comparative results over time for the Same Content, Different
Content, and control groups
568 Language Learning Vol. 49, No. 4

were not equal. The average percentage correct for the 13 Same
Content participants at Time 1 was 40%, whereas the average
percentage correct for the 9 Different Content participants at
Time 1 was 12%. This discrepancy and the ceiling of 100% made
any reasonable statistical comparison impossible.
Tables 5 (ser) and 6 (estar) show the percentage of partici-
pants who improved from Time 1 to Time 4. The purpose of this
comparison was to determine whether the improvement trend
that we saw from Time 1 to Time 3 for the Same Content group
carried over to Time 4 (Hypothesis 4). Even though the Same
Content group did show a slight improvement for ser (28% versus
24% for the Different Content group), with only 8% improvement
for the control group, it is a complex issue to decide how to
interpret these results given the high accuracy rates at Time 1.
The results for estar indicate a greater improvement for the
Different Content group as opposed to either the Same Content
group or the control group. So, whatever benefit there may have
been at Time 3 stemming from less need for participants to focus
on the content of the video, that benefit did not carry over to a time
when a different video was seen, thus addressing the second
research question and Hypothesis 4: Improvement did not gener-
alize to a new context on the morphosyntactic measure.

Table 3

Percentage of Participants Who Improved From Time 1 to Time 3 (Ser)

% correct (of % correct (all

Subset who subset) participants) at
Groups improved (%) Time 1 Time 3 Time 1

Same Content 28 50 94 83
(n = 32)
Different Content 21 42 100 74
(n = 33)
Note. Criterion for improvement: at least two contexts at Time 3.
Gass, Mackey, Alvarez-Torres, and Fernández-García 569

Table 4

Percentage of Participants Who Improved From Time 1 to Time 3

(Estar)

% correct (of % correct (all

Subset who subset) participants) at
Groups improved (%) Time 1 Time 3 Time 1

Same Content 44 40 75 46
(n = 32)
Different Content 30 12 74 38
(n = 33)
Note. Criterion for improvement: at least two contexts at Time 3.

Table 5

Percentage of Participants Who Improved From Time 1 to Time 4 (Ser)

% correct (of % correct (all

Subset who subset) participants) at
Groups improved (%) Time 1 Time 4 Time 1

Same Content 28 48 89 83
(n = 32)
Different Content 24 42 92 74
(n = 33)
Control 8 17 85 88
(n = 38)
Note. Criterion for improvement: at least two contexts at Time 4.

Lexical Analysis

Of the total words in the database, 51% were open-class

words (39,028), 44% of the total words were closed-class words
(33,167), and 5% were English words (3,672). The learners often
used English words where they presumably did not have the
Spanish lexical item. For example, white-out was consistently
referred to as such in English by the learners. We also measured
570 Language Learning Vol. 49, No. 4

Table 6

Percentage of Participants Who Improved From Time 1 to Time 4

(Estar)

% correct (of % correct (all

Subset who subset) participants) at
Groups improved (%) Time 1 Time 4 Time 1

Same Content 31 43 88 46
(n = 32)
Different Content 46 19 79 38
(n = 33)
Control 34 29 84 43
(n = 38)
Note. Criterion for improvement: at least two contexts at Time 4.

and categorized production of mid- to low-frequency words. These

were words that were not on the list of 200 most commonly used
words in Spanish.19
Table 7 shows the increase from Time 1 to Time 4 (we
focussed on this comparison because at Time 4 both experimental
and control groups watched the same video snippets), in open-
class, closed-class, and all words. In terms of open-class words
(type-token ratios), there was an increase of 11.6% for the Same
Content group, whereas the Different Content group only in-
creased by 5.6%. The control group, however, increased by 9.6%,
so this finding was less interesting than the results for closed-class
words. The control group increased in production of closed-class
words by only 5.9%, the Different Content group by 12.3%, and the
Same Content group by 26.8%. These findings for closed-class
words are primarily responsible for the overall increase in all words.
Table 8 shows the mid- to low-frequency words that were not
on the list of 200 most common words in Spanish, and again the
results were interesting. The Same Content group increased by
12.3%, the Different Content group by 2.7%, and the control by
1.6%.
Gass, Mackey, Alvarez-Torres, and Fernández-García 571

Discussion

This study tested the following four hypotheses:

1. Task repetition results in greater “overall” proficiency.
2. Task repetition yields greater accuracy in morphosyntax.
3. Task repetition results in greater lexical sophistication.
4. For holistic judgments and morphosyntactic accuracy,
change will carry over to a new context.

Table 7

Percentage Increase in Type-Token Ratio From Time 1 to Time 4

Open-class words Closed-class words All words (%)

Groups (%) (%) (OC + CC + L1)

Same Content 11.6 26.8 17.6

(n = 32)
Different Content 5.6 12.3 10.4
(n = 33)
Control 9.6 5.9 6.8
(n = 38)

Table 8

Percentage Increase in Type-Token Ratio of Mid- to Low-Frequency

Words From Time 1 to Time 4

Time 1 Time 4 % increase

Groups M SD M SD Time 1 → Time 4

Same Content 53.2 12.6 59.7 27.5 12.3

n = 32
Different Content 52.0 10.8 53.4 9.9 2.7
n = 33
Control 53.2 10.9 54.1 6.7 1.6
n = 38
572 Language Learning Vol. 49, No. 4

Our results suggest that there is some validity to each of these

hypotheses. In terms of Hypothesis 1, greater overall proficiency,
the findings of better evaluations of the Same Content group at
Time 3 (when the content was the same) correspond to the findings
of Skehan and Foster (1997, 1999), who dealt with predictable
sequences of events. In our case the predictable sequences came
from having seen a story at an earlier point; in the Skehan and
Foster study, it came from world knowledge, that is, events that
were predictable on the basis of one’s knowledge of particular
situations. In general, the mere fact of repeating the task yielded
improvement. In both the experimental groups as well as in the
control group, Time 4 was better than Time 1. Content familiarity
resulted in a greater increase, as can be seen by the large increase
for the Time 3 Same Content group, although this greater increase
was not sustained, as can be seen by the Time 4 results (Hypothe-
sis 4). This is not surprising given that fluency is likely to be
partially context dependent (Riggenbach, 1989, 1991).
In terms of Hypothesis 2, greater accuracy in morphosyntax,
again there was evidence that task repetition led to improvement.
More participants in the Same Content group showed a trend
toward more targetlike production of estar at Time 3 than did
participants in the Different Content group (44% versus 30%). In
terms of Hypothesis 4, this improvement did not generalize to
Time 4. One possible explanation for the lack of generalization of
improvement at Time 4 has to do with the notion of task repetition.
Plough and Gass (1993) suggested that when carrying out task-
based work in a classroom, learners can easily become somewhat
disinterested in the tasks given to them when those tasks have
been carried out repeatedly. At a certain point in this study, the
novelty of the task may have ended and disinterest settled in.20
In terms of lexical sophistication (Hypothesis 3), the findings
show clearly that less common words (type-token ratios) were
used in much larger numbers by the Same Content group. Know-
ing content thus also appears to have positively affected lexical
sophistication in this study. This increase, which mostly took place
in closed-class words, may be due to an increase in structured,
Gass, Mackey, Alvarez-Torres, and Fernández-García 573

ordered storytelling devices, representing a more controlled retell-

ing. This was also a finding reported by Bygate (1997).

Conclusion

In summary, using a range of measures, we found some

evidence that task repetition resulted in improvement in overall
proficiency, selected morphosyntax, and lexical sophistication.
There does appear, then, to be limited support for the claim that
freeing up attention to meaning allows learners to gain greater
control over their linguistic knowledge. However, where we were
able to measure generalization, using judgments of overall profi-
ciency and morphosyntax, these findings of improvement did not
generalize to a new context. Our approach to the study of the
linguistic effects of task repetition was not to constrain partici-
pants to focus on any particular part of their grammars in the task
repetition exercise, but to allow them complete freedom in carry-
ing out the tasks. We thus cast a wide net in looking for any
interlanguage change, and we found clear indications that such
change was taking place. Our study suggests that future research
might now benefit from focussing more closely on specific areas of
the grammar in task repetition,21 in order to further explore the
cognitive processes of attention to meaning, attention to form, and
L2 acquisition.
Revised version accepted 04 January 1999

Notes
1
It should be noted that criticisms have been allayed against the Tomasello
and Herron studies in the form of both the methodology and the analysis (see
Beck & Eubank, 1991, and the response by Tomasello & Herron, 1991).
2
In terms of the L2 knowledge of these learners, we made the assumption in
this study that because all learners had similar amounts and types of
instruction and had been placed into the same level, L2 knowledge would be
broadly similar. For example, one of the features that we examine in this
study is the Spanish copula (ser and estar). All classes had had significant
amounts of instruction on these verbs prior to the onset of this study.
574 Language Learning Vol. 49, No. 4

3
In this study we operationalized meaning as knowledge of content. Hence,
throughout, we use the terms content and meaning interchangeably.
4
In the analysis presented in this article, we will not be specifically consid-
ering measures of complexity or fluency (although these may have entered
into our holistic judgments of improvement, particularly in the case of
fluency). There were two reasons for not considering complexity. First, the
level of the students’ knowledge of Spanish was low and thus the number of
complex sentences was low from the outset. Second, given the on-line nature
of the task, learners tended to produce short simple sentences, and very often
incomplete ones, as they abandoned what they were saying to keep up with
the events on the screen.
5
Crookes (1989) also found that learners, as a result of planning, improved
on measures of complexity, but not on measures of accuracy. As will be seen
from the description of our study, our study differs from the planning
literature in that planning was not a variable, and, in fact, until the beginning
of the task, learners were unaware of what the task content would be,
although they could predict the task procedure (watch and retell).
6
The study began with a larger data pool. However, because we required that
all students participate in every data collection session, there was attri-
tion from beginning to end. In addition, participants who produced fewer
than 50 words in two sessions were not included.
7
Steps were taken to ensure that, for example, not all of the 8:00 A.M. sections
or not all of the night sections were assigned to the same group.
8
We selected the top 10 in each of the two experimental groups and the control
group as a way of limiting the number of tapes each rater had to listen to.
Listening to more than nine segments (at 2.5 min each) might have reduced
the reliability of the judgments. In addition, 10 tapes from each group were
deemed sufficient to provide information on change over time.
9
There were seven raters in all: Five were university-level instructors of
Spanish, one was a graduate student, and one was a bilingual secretary.
10
Because only three samples were rated by seven raters, interrater reliabil-
ity between pairs of raters could not be calculated. Instead, the seven raters’
rankings of pairs of samples were correlated. The reason for this was that if
all seven raters ranked Sample 1 first, for example, and Sample 2 second,
there would be a high correlation between them (i.e., 1.0). As it was, the
Spearman correlations between pairs of the samples were .93, .96, and .98.
11
We selected the first 2.5 min of the tape (as opposed to other parts of the
tape) because the beginning was the only place where we could be assured of
common ground; all participants began talking at the beginning of the film.
Had we selected any later portions, we were not guaranteed equivalent
uniformity.
12
Conversion to a logarithmic scale is standard procedure when using mag-
nitude estimation. Because the methodology allows for unique scales to be
created by each rater, there must be a way to standardize the scales across
raters to obtain a meaningful comparison.
Gass, Mackey, Alvarez-Torres, and Fernández-García 575

13
Two native speakers of Spanish coded these data. For both verbs, interrater
reliability was .97.
14
Most of these measures come out of the L2 writing literature and all have
their critics. Criticisms have focussed on issues of reliability, validity, or both
(Arnaud, 1984; Laufer, 1995; Laufer & Nation, 1993; Read, 1988).
15
After advertising on Linguist list, we obtained from an Internet site a list
of 2,000 most commonly used words, broken down incrementally. For the
analysis presented in this article, we utilized 200 most commonly used words.
The patterns reported in this article also held for increments at higher levels
with minor variations. See ([Link]
WWW7 and [Link] The first
address is for the actual list and the second is for the instructions.
16
The issue of meaningfulness versus significance (in the statistical sense) is
an important one in this and in other L2 studies, particularly those studies
that are based on classroom treatments. Ideally, in an experiment such as
the one reported here, one would want a longer period to pass in order to
allow learners to demonstrate change in language use. Language learning is
not instantaneous; it takes time for information to become integrated into
one’s grammar (Gass, 1997; Lightbown, 1998). However, the realities of many
experimental contexts do not allow a more protracted period given the
constraints placed on our use of class time (see papers in Schachter & Gass,
1996, for discussions of difficulties and constraints in conducting classroom
research). This returns us to the issue of significance. From our perspective,
we have trends that are meaningful. However, these trends are not significant
in the statistical sense with a probability level of <.05, the generally accepted
standard in the field.
The need to have all results adhere to a .05 standard may be question-
able. Shavelson (1988) noted that the convention of using .05 or .01
grew out of experimental settings in which the error of rejecting
a true H0 was very serious. For example, in medical research, the
null hypothesis might be that a particular drug produces unde-
sirable effects. Deciding that the medicine is safe (i.e., rejecting
H0) can have serious consequences. Hence, conservatism is de-
sired. (p. 248)

Shavelson went on to say that “often in behavioral research . . . the

consequences are not so dire” (p. 248). He suggested that a level of .25 might
be appropriate in some cases. He concluded his discussion by saying that
“some wisdom, then, should be exercised in setting the level of significance”
(p. 248) and pointed out that there is a trade-off between the level of
significance that one sets and the power of one’s conclusions.
Given the essential arbitrariness in setting significance levels and
given the constraints in conducting classroom research, we feel that trends
are important and at least point to the notion that experiments should be
replicated, particularly when it is impractical or impossible for experiments
576 Language Learning Vol. 49, No. 4

to cover a long period. We also believe that trends may at times be as

meaningful as statistical significance.
17
Given our instructions to raters to evaluate holistically, we cannot ascertain
to what extent the holistic evaluations reflect fluency, accuracy, or complexity.
Our goal was not to differentiate among these possibilities; rather, these were
taken to represent an overall improvement in “proficiency.”
18
VanPatten (1985, 1987, 1990b) showed that ser is learned before estar. He
provided three reasons for this learning order: (a) Ser is more frequent in the
input, (b) simplification—learners adopt a “one copula” system in the early
stages of acquisition, and (c) language transfer. The greater frequency of ser
and the natural learning mechanism of one form/one meaning leads learners
to rely on the native language “one copula” system.
19
For learners in this developmental range, words not on the list of 200 most
commonly used words can be classified as mid- to low-frequency in terms of
use, as opposed to native speakers for whom, obviously, they would not be
mid to low. Examples of these words include dentro (within), mitad (half),
próximo (next), lleno (full), durante (during, for), también (also, as well as),
bien (good, well), grande (big), and comenzar (to start).
20
As pointed out in Note 6, some students were eliminated from the study if
they produced very little (i.e., were not on task). It is also possible that not
being on task was evidence of boredom.
21
One way of achieving more robust results in a study of task repetition would
be to use tasks that produce contexts for particular structures (cf. Mackey,
1995, 1999).

References

Ard, J., & Homburg, T. (1992). Verification of language transfer. In S. Gass &
L. Selinker (Eds.), Language transfer in language learning (pp. 47–70).
Amsterdam: John Benjamins.
Arnaud, P. J. L. (1984). The lexical richness of L2 written productions and the
validity of vocabulary tests. In T. Culhane, C. Klein-Braley, & D. K.
Stevenson (Eds.), Practice and problems in language testing. Colchester,
UK: University of Essex.
Bard, E., Robertson, D., & Sorace, A. (1996). Magnitude estimation of linguis-
tic acceptability. Language, 72, 32–68.
Beck, M., & Eubank, L. (1991). Acquisition theory and experimental design:
A critique of Tomasello & Herron. Studies in Second Language Acquisition,
13, 73–76.
Bialystok, E. (1994). Analysis and control in the development of a second
language. Studies in Second Language Acquisition, 16, 157–168.
Gass, Mackey, Alvarez-Torres, and Fernández-García 577

Bialystok, E., & Sharwood Smith, M. (1985). Interlanguage is not a state of

mind: An evaluation of the construct for second-language acquisition.
Applied Linguistics, 6, 101–117.
Bygate, M. (1996). Effects of task repetition: Appraising the developing
language of learners. In J. Willis & D. Willis (Eds.), Challenge and change
in language teaching (pp. 136–146). London: Heinemann.
Bygate, M. (1997, March). The effect of task repetition on language structure
and control. Paper presented at the 1997 American Association for Applied
Linguistics Convention, Orlando, FL.
Crookes, G. (1989). Planning and interlanguage variation. Studies in Second
Language Acquisition, 11, 367–383.
Cubillos, J. H. (1996). Siempre adelante [Onwards and Upwards]. Boston:
Heinle & Heinle.
Ellis, N. (1994). Implicit and explicit language learning—An overview. In
N. Ellis (Ed.), Implicit and explicit learning of languages (pp. 1–32).
London: Academic Press.
Fucci, D., Ellis, L., & Petrosino, L. (1990). Speech clarity/intelligibility: Test-
retest reliability of magnitude estimation scaling. Perceptual and Motor
Skills, 70, 232–234.
Gass, S. (1997). Input, interaction and the second language learner. Mahwah,
NJ: Lawrence Erlbaum Associates.
Green, K. (1987). The perception of speaking rate using visual information
from a talker’s face. Perception and Psychophysics, 42, 587–593.
Grosjean, F. (1977). The perception of rate in spoken and sign languages.
Perception and Psychophysics, 22, 408–413.
Gutiérrez, J. R., Martínez-Lage, A., & Rosse, H. L. (1995). ¡Tú dirás! [Tell me!].
Boston: Heinle & Heinle.
Laufer, B. (1995). Beyond 2000: A measure of productive lexicon in a second
language. In L. Eubank, L. Selinker, & M. Sharwood Smith (Eds.), The
current state of interlanguage: Studies in honor of William E. Rutherford
(pp. 265–272). Amsterdam: John Benjamins.
Laufer, B., & Nation, P. (1993, August). Lexical richness in L2 written produc-
tion: Can it be measured? Paper presented at the 10th Congress of the
International Association of Applied Linguistics, Amsterdam.
Laufer, B., & Nation, P. (1995). Vocabulary size and use: Lexical richness in
L2 written production. Applied Linguistics, 16, 307–322.
Leow, R. (1997). Attention, awareness and foreign language behavior. Lan-
guage Learning, 47, 467–506.
Leow, R. (1998). The effects of amount and type of exposure on adult learners’
L2 development in SLA. The Modern Language Journal, 82, 49–68.
578 Language Learning Vol. 49, No. 4

Lightbown, P. (1998). The importance of timing in focus on form. In C. Doughty

& J. Williams (Eds.), Focus on form in classroom second language acquisi-
tion (pp. 177–196). Cambridge: Cambridge University Press.
Long, M. (1996). The role of the linguistic environment in second language
acquisition. In W. Ritchie & T. Bhatia (Eds.), Handbook of second language
acquisition (pp. 413–468). San Diego, CA: Academic Press.
Mackey, A. (1994). Targeting morpho-syntax in children’s ESL: An empirical
study of the use of interactive goal-based tasks. Working Papers in Edu-
cational Linguistics, 10, 67–88.
Mackey, A. (1995). Stepping up the pace: Input, interaction and second
language development. Unpublished doctoral dissertation, University of
Sydney, Australia.
Mackey, A. (1999). Input, interaction and second language development: An
empirical study of question formation in ESL. Studies in Second Language
Acquisition, 21, 557–587.
McDonough, K., & Mackey, A. (in press). Form and meaning: Designing
communicative tasks to target grammar in Thai classrooms. Foreign
Language Annals.
McLaughlin, B. (1987). Theories of second language acquisition. London:
Edward Arnold.
Pavlovic, C., Rossi, M., & Espesser, R. (1990). Use of the magnitude estimation
technique for assessing the performance of text-to-speech synthesis sys-
tems. Journal of the Acoustical Society of America, 87, 373–382.
Plough, I., & Gass, S. (1993). Interlocutor and task familiarity: Effects on
interactional structure. In G. Crookes & S. Gass (Eds.), Tasks and language
learning: Integrating theory and practice (pp. 35–56). Clevedon, UK:
Multilingual Matters.
Read, J. (1988). Measuring the vocabulary knowledge of second language
learners. RELC Journal, 19, 12–25.
Riggenbach, H. (1989). Nonnative fluency in dialogue versus monologue
speech: A microanalytic approach. Unpublished doctoral dissertation, Uni-
versity of California, Los Angeles.
Riggenbach, H. (1991). Toward an understanding of fluency—A microanalysis
of nonnative speaker conversations. Discourse Processes, 14, 423–441.
Robinson, P. (1994). Implicit knowledge, second language learning and sylla-
bus construction. TESOL Quarterly, 28, 160–166.
Robinson, P. (1996). Learning simple and complex second language rules
under implicit, incidental, rule-search and instructed conditions. Studies
in Second Language Acquisition, 18, 27–67.
Schachter, J., & Gass, S. (Eds.). (1996). Second language classroom research:
Issues and opportunities. Mahwah, NJ: Lawrence Erlbaum Associates.
Gass, Mackey, Alvarez-Torres, and Fernández-García 579

Schachter J., Rounds, P., Wright, S., & Smith, T. (in press). Comparing
conditions for learning syntactic patterns: Attentional, nonattentional,
and awareness. Applied Linguistics.
Schmidt, R. (1995). Consciousness and foreign language learning: A tutorial
on the role of attention and awareness in learning. In R. Schmidt (Ed.),
Attention and awareness in foreign language learning (pp. 1–64). Hono-
lulu: University of Hawai’i Press.
Shavelson, R. (1988). Statistical reasoning for the behavioral sciences. Boston:
Allyn and Bacon.
Skehan, P., & Foster, P. (1997, March). Language and content planning, task
structure, and task performance. Paper presented at the 1997 American
Association for Applied Linguistics Convention, Orlando, FL.
Skehan, P., & Foster, P. (1999). The influence of task structure and processing
conditions on narritive retellings. Language Learning, 49, 93–120.
Stevens, S. (1956). The direct estimation of sensory magnitudes—Loudness.
American Journal of Psychology, 69, 1–25.
Swain, M. (1985). Communicative competence: Some roles of comprehen-
sible input and comprehensive output in its development. In S. Gass &
C. Madden (Eds.), Input in second language acquisition (pp. 235–253).
Rowley, MA: Newbury House.
Swain, M. (1995). Three functions of output in second language learning. In
G. Cook & B. Seidlhofer (Eds.), Principle and practice in applied linguistics:
Studies in honour of H. G. Widdowson (pp. 125–144). Oxford: Oxford
University Press.
Swain, M., & Lapkin, S. (1995). Problems in output and the cognitive pro-
cesses they generate: A step towards second language learning. Applied
Linguistics, 16, 371–391.
Swain, M., & Lapkin, S. (1998). Interaction and second language learning:
Two adolescent French immersion students working together. Modern
Language Journal, 82, 320–337.
Takefuta, Y., Guberina, P., Pizzamiglio, L., & Black, J. (1986). Cross-lingual
measurements of interconsonantal differences. Journal of Psycholinguis-
tic Research, 15, 489–507.
Tarone, E., & Liu, G. (1995). Situational context, variation, and second
language acquisition theory. In G. Cook & B. Seidlhofer (Eds.), Principle
and practice in applied linguistics: Studies in honour of H. G. Widdowson
(pp. 107–124). Oxford: Oxford University Press.
Tomasello, M., & Herron, C. (1988). Down the garden path: Inducing and
correcting overgeneralization errors in the foreign language classroom.
Applied Psycholinguistics, 9, 237–246.
580 Language Learning Vol. 49, No. 4

Tomasello, M., & Herron, C. (1989). Feedback for language transfer errors:
The garden path technique. Studies in Second Language Acquisition, 11,
385–395.
Tomasello, M., & Herron, C. (1991). Experiments in the real world: A reply to
Beck & Eubank. Studies in Second Language Acquisition, 13, 513–517.
Tomlin, R. S., & Villa, V. (1994). Attention in cognitive science and second
language acquisition. Studies in Second Language Acquisition, 16,
183–203.
Toner, M., & Emanuel, F. (1989). Direct magnitude estimation and equal
appearing interval scaling of vowel roughness. Journal of Speech and
Hearing Research, 31, 78–82.
VanPatten, B. (1985). The acquisition of ser and estar by adult classroom
learners: A preliminary investigation of transitional stages of competence.
Hispania, 68, 399–406.
VanPatten, B. (1987). Classroom learners’ acquisition of ser and estar: Ac-
counting for developmental patterns. In B. VanPatten, T. Dvorak, & J. Lee
(Eds.), Foreign language learning: A research perspective (pp. 61–75).
Cambridge, MA: Newbury House.
VanPatten, B. (1990a). Attending to content and form in the input: An
experiment in consciousness. Studies in Second Language Acquisition, 12,
287–301.
VanPatten, B. (1990b). Theory and research in second language acquisition
and foreign language learning: On producers and consumers. In B. Van-
Patten & J. Lee (Eds.), Second language acquisition—Foreign language
learning (pp. 17–26). Clevedon, UK: Multilingual Matters.
VanPatten, B. (1995). Input processing and second language acquisition:
On the relationship between form and meaning. In P. Hashemipour,
R. Maldonado, & M. van Naerssen (Eds.), Festschrift in honor of Tracy D.
Terrell (pp. 170–183). New York: McGraw Hill.
VanPatten, B. (1996). Input processing and grammar instruction: Theory and
research. Norwood, NJ: Ablex.
VanPatten, B., & Cadierno, T. (1993). Explicit instruction and input process-
ing. Studies in Second Language Acquisition, 15, 225–243.
VanPatten, B., & Sanz, C. (1995). From input to output: Processing instruction
and communicative tasks. In F. Eckman, D. Highland, P. Lee, J. Mileham,
& R. Weber (Eds.), Second language acquisition theory and pedagogy
(pp. 169–186). Hillsdale, NJ: Lawrence Erlbaum Associates.
White, L. (Ed.). (1998). The implications of divergent outcomes in L2 acqui-
sition [Special issue]. Second Language Research, 14(4).
Gass, Mackey, Alvarez-Torres, and Fernández-García 581

Appendix A

Instructions for Participants

You will watch a story about a man named El Señor Bean.

This is approximately a 6-minute silent video called xxxx. When
the video begins, your teacher will begin recording your voice.
While watching the video, tell the story in Spanish with as many
details as possible. That is, talk about what you see as it is
happening. Do not wait until the video is over to talk. Try to say
as much as you can.

Appendix B

Instructions for Raters

You will hear nine tapes of different non-native speakers of

Spanish doing an on-line description in Spanish of a video they
were watching. Your task is to rate their Spanish. Assign any
number that seems appropriate to you to the first speech sample.
This number will be your “base.” Then assign successive numbers
in such a way that they reflect your subjective impression (use a
range wider than 10). For example, if a speech sample seems 20
times as good, assign a number 20 times as large as the first. If
it seems one-fifth as good, assign a number one-fifth as large, and
so forth. Use fractions, whole numbers, or decimals, but make each
assignment proportional to how good you perceive the person’s
Spanish to be.