SSLA, 16,283-302. Printed in the United States of America.

INPUT, INTERACTION, AND

SECOND LANGUAGE
PRODUCTION

Susan M. Gass
Michigan State University

Evangeline Marios Varonis

University of Akron

The role of conversational interactions in the development of a second

language has been central in the recent second language acquisition
literature. While a great deal is now known about the way in which
nonnative speakers interact with native speakers and other nonnative
speakers, little is known about the lasting effects of these interactions
on a nonnative's linguistic development. This paper specifically
investigates the relationship among input, interaction, and second
language production. Through data from native-nonnative speaker
interactions in a direction-giving task, we show that both modified input
and interaction affect task performance. However, only interaction has
an effect on subsequent task performance.

The question of the role of conversational interactions in the development of a

second language has been central in the recent second language acquisition litera-
ture. While this question has been addressed from a variety of perspectives, it has
been difficult to ascertain precisely the role of input and interaction in terms of
actual language development. This, in fact, is the thrust of an article by Schachter

Earlier versions of this paper were presented at the Second Language Research Forum, Pittsburgh, 1993, and
Association International de Linguistique Appliquee (AILA), Amsterdam, 1993. We are grateful to Gary Cook
for his statistical advice. We are also indebted to India Plough for assistance with transcriptions. The students
in English 841 at Michigan State University are appreciated for feedback on the description of the study.
Shona Whyte of the SSLA staff made numerous suggestions for improvement on the final manuscript. The
SSLA readers provided suggestions, advice, and admonition which we took into consideration as we revised
this paper. We are grateful to them for their careful reading of our manuscript. We wish we could say that all
errors are theirs, but, alas, we cannot and must confess that any errors are our own.
e
1994 Cambridge University Press 0272-2631/94 $5.00 + .00 283
284 Susan M. Gass and Evangeline Marios Varonis

(1986) in which she argued for the importance of showing the effect of native speaker
input on the language learning process. The purpose of this paper is to see if and in
what ways it is possible to determine a direct relationship between input, particularly
interactional input, and subsequent language production.
An early view of acquisition held that learners learned grammatical rules and
then applied those rules and practiced them within a conversational setting. Thus,
classroom drills, classroom interactions, and daily interactions with native speakers
were viewed only as a means of reinforcing the grammatical rules acquired by a
learner.
In 1975 Wagner-Gough and Hatch suggested a different role for conversation in
second language development. They argued that conversational interaction forms
the basis for the development of syntax rather than being only a forum for practice
of grammatical structures. Syntax, they claimed, develops out of conversation rather
than the reverse. Example 1 illustrates the way language development can take
place within a conversational setting, as the child learner in this case uses the
conversation to further her syntactic development.

1. From Ellis 0985, pp. 79-80)

NS: Do you want to look at the next picture? Yeah?
NNS: Man.
NS: A man. And do you know what this is? A wall.
NNS: A wall.
NS: Like that one there. A wall.
NNS: A wall, a wall.
NS: Yes. Now, can you see what the man is doing?
NNS: A man wall.
NS: He's going into the wall.

Prior to this point in time, there were no examples of two-word utterances in this
child's discourse. As can be seen, the conversation itself provides the framework or,
as Ellis states, "the breakthrough points" for a two-word utterance to develop. The
teacher in this case broke the task into parts and helped with the crucial vocabulary,
which finally enabled the child to connect man and wall in her final utterance.

CONVERSATION IN NATIVE-NONNATIVE DISCOURSE

From this interactional perspective stem a number of studies in which second lan-
guage conversational interactions and, more broadly, issues of input have become a
primary focus in second language research. Long (1980) made an important distinc-
tion between modified input, or foreigner talk, and modified interaction, differentiat-
ing between the modified talk directed to the learner and the modified structure
of the conversation itself. By interactional features, he included such aspects of
conversation as comprehension checks, topic shifts, and clarification requests. In his
work, Long showed that conversations involving nonnative speakers (NNSs) have
more of these kinds of modification than do conversations between two native
speakers (NSs). He argued that this is so for two reasons: First, these devices aid in
Input, Interaction, and Production 285

avoiding conversational trouble and, second, they serve the function of repairing the
discourse when trouble does occur. Varonis and Gass (1985) extended this line of
research by operationalizing the concept of negotiation of meaning as a central
factor in second language acquisition. An entire volume (Day, 1986) is devoted to
precisely the relationship between conversation and acquisition. A number of other
studies have considered the effect on nonnative speech of such variables as male/
female differences (Gass & Varonis, 1986; Pica, Holliday, Lewis, Berducci, & New-
man, 1991), ethnic differences (Scarcella, 1983, 1992), proficiency (Varonis & Gass,
1985), status and expertise differences (Woken & Swales, 1989; Zuengler, 1989), and
task differences (Duff, 1986; Long, 1980; Pica, 1987; Pica & Doughty, 1985; Plough
& Gass, 1993; Samuda & Rounds, 1993). While these studies have significantly con-
tributed to our growing understanding of the factors that influence the nature of
speech, they only indirectly touch upon the issue of subsequent L2 production and
acquisition. For example, Long (1983) made a deductive argument: Linguistic and/or
conversational adjustments lead to better comprehension; comprehension promotes
acquisition. Therefore, adjustments facilitate acquisition. An underlying assumption
in this body of research is that negotiation increases the possibility that the language
used in the negotiation will be of benefit to the learner in the development of the L2
(Ellis, 1991; Gass & Varonis, 1989; Long, 1992; Pica, 1987).
Despite the promising results of such research, the effect of interaction on acquisi-
tion remains controversial. Sato (1986) questioned a direct positive relationship. She
examined the English of two Vietnamese boys, finding that neither the NS input to
the boys nor the naturalistic interaction between them and their native speaker
interlocutor was reflected in increased language proficiency. Her study, focusing on
the marking of past time reference, did not suggest that grammatical encoding of
such reference increased as a function of proficiency. Instead, the nonnative speak-
ers relied on the situational and/or discourse context to establish a time frame. Since
pastness was for the most part recoverable from context, there was an insignificant
interactional burden on the part of participants. Furthermore, in the case of past
tense marking in English, the feature in question is often not phonologically salient,
reducing the learner's opportunities to utilize relevant information. Thus, at least in
the case of past tense marking, there is little necessity and little opportunity to obtain
or provide linguistic information in the conversation. However, given their relatively
rudimentary knowledge of English, one wonders whether the situation might not be
different if the learners were at a different stage of development, a stage at which
they were "ready" to learn past tense forms (Pienemann, 1992).
Sato's findings have been corroborated by other studies that also focus on NS-NNS
naturalistic conversations. Chun, Day, Chenoweth, and Luppescu (1982) reported
relatively little explicit feedback in free conversation between native and nonnative
speakers and therefore questioned the value of correction as an integral part of
successful acquisition. In another study, Day, Chenoweth, Chun, and Luppescu (1984)
further questioned the role of error correction in L2 acquisition, noting that out of
1,595 student errors in the corpus, only 119 (7.3%) were singled out for corrective
feedback by NS interlocutors.
A follow-up study by Brock, Crookes, Day, and Long (1986) suggested that the
286 Susan M. Gass and Evangeline Marios Varonis

effect of conversational interactions on acquisition may be influenced in part by

task. They investigated a broader range of negative input in native-nonnative free
conversations, examining short-term effects of such input on the nonnative's lan-
guage development. They found surprisingly little change in learner forms, with
only 26 out of 152 instances (17.1%) in which learners clearly responded by incorpo-
rating the native speaker's negative input into their next turn. However, they
pointed out anecdotally the possibility of an effect for task: They did observe NNSs
incorporating examples of native speaker corrective feedback following errors when
communication took place in the context of communication games. In other words,
learners' grammars may be quickly destabilized if they give sufficient attention to
the area in question, with the assumption being that they would pay more attention
in the context of a game as opposed to free conversation. Gass (1988,1991) and Long
(1992) have argued for the importance of selective attention in second language
development, claiming that it is a prerequisite to grammatical development, a point
we return to later. Similarly, Schmidt (1990) argued that conscious awareness is a
necessary condition for language development. In addition, it is important to keep in
mind that the absence of short-term effects does not exclude the possibility of long-
term effects when the learner has had sufficient time to process and incorporate the
feedback.
The importance of task is further supported by Crookes and Rulon (1985), who
examined native-nonnative dyads, considering the issue of the incorporation of
corrective feedback in three situations: One free conversation and two two-way
communication tasks. Feedback was defined as the correct usage by a native
speaker of a word or construction immediately following a nonnative utterance.
They found significantly more feedback in task-related conversation than in free
conversation. They suggested that for maximum grammatical destabilization, lin-
guistic material should be slightly unfamiliar to the nonnative speaker, and the
structure of the task should require the maximum use of this material by both
parties. Pica, Young, and Doughty (1987), in a picture arrangement task in which
input to the learner was either premodified or interactionally modified, found that
comprehension (as measured by task completion) was superior when the negotiation
was allowed as opposed to when it was not. Knox (1992) extended this observation
to naturalistic conversation, suggesting that form-focusing and subsequent NNS self-
modification occurs in certain types of constrained settings, such as a structured
interview or a service encounter.
The relationship between comprehension and acquisition is further called into
question by Doughty (1991). In a study of relativization, she compared three groups
of subjects engaged in a computer-assisted language learning project. The groups
differed in the format of presentation of the language material. Besides a control
group, there were two experimental groups: a meaning-oriented treatment group
and a rule-oriented treatment group. As the names suggest, in the latter group,
the rule-oriented treatment group, explicit metalinguistic statements about relative
clauses were provided, whereas in the meaning-oriented treatment group there
were no such explicit statements. The meaning group had higher comprehension
scores than the rule-oriented group. However, in terms of pretest/posttest scores
Input, Interaction, and Production 287

measuring gains on relative clauses, the two experimental groups improved more or
less equally. Thus, at first glance, it appears that there is no direct relationship
between comprehension and acquisition. However, a closer examination of the
experimental materials brings us back to the question of attention. That is, how can
a learner's attention be brought to language forms? It is beyond the scope of this
paper to discuss this concept in any detail (cf. Gass, 1988; Schmidt, 1990); however,
three aspects are important: form-focused instruction, frequency, and saliency. If we
return to Doughty's study, we see that both saliency and redundancy (i.e., frequency)
were built into the tasks of the meaning-oriented treatment group. In the experimen-
tal material, the meaning-oriented treatment group saw reading passages with cer-
tain features, namely, head nouns and relative clause markers, highlighted on the
screen. Additionally, the juxtaposed head noun and relative clause marker were
capitalized, thereby visually making this part of the reading passage salient to the
learner. Thus, Doughty's results (given her particular methodology) suggest that
what is important for acquisition is not so much immediate comprehension, but the
necessity of drawing learners' attention to particular forms.
Taken as a whole, these studies suggest that NS input is most likely to affect
subsequent NNS language production when the interaction is focused and task-
oriented.
One difficulty in this area of research is the determination of the learner's knowl-
edge at any particular time. An operating assumption is that we need to assess
immediate destabilization of grammatical forms in order to determine the effect of
conversational interaction. However, it may be more appropriate (albeit method-
ologically difficult) to consider longer range effects. Bruton and Samuda (1980) dis-
cuss something similar when they refer to "correction by permeation," with correct
forms gradually becoming incorporated into the learner's grammar over time.

CONVERSATION IN NONNATIVE-NONNATIVE DISCOURSE

Up until now we have dealt primarily with research that deals with native-nonnative
interactions. There is another area in which nonnative speakers frequently find
themselves and that is in conversations with other nonnative speakers. There is
evidence that the changes they make as a result of the interaction are in the direction
of the target language. Gass and Varonis (1989) presented data from nonnative-
nonnative interactions that show that a correctly modeled form by a nonnative
frequently resulted in changes by the other nonnative in the dyad, although the
changes often occurred much later in the discourse. To illustrate such a change as a
result of input from another nonnative, we present an example of a grammatical
modification in a nonnative's speech that appears to have taken place. In example 2,
two nonnatives of different language backgrounds were given the task of going out
onto the street with a tape recorder and asking for directions to the train station.
These students left the tape recorder on during the entire time they were engaged
in this task so that the totality of the conversation between them was also recorded,
even when they were not specifically engaged in alternately stopping passersby to
ask for directions.
288 Susan M. Gass and Evangeline Marios Varonis

2. a. Ana: Can you tell me where is the train station?

b. Keiko: Can you tell me where the train station is?
c. Ana: Can you tell me where is the train station?
d. Keiko: Can you tell me where the train station is?
e. Ana: Can you tell me where is the train station?
f. Keiko: Can you tell me where the train station is?
g. Ana: Can you tell me where the train station is?
h. Keiko: Can you tell me where the train station is?
i. Ana: Can you tell me where the train station is?

In understanding the significance of this example, it is important to note that

nowhere in the entire conversation between requests for directions did the students
discuss the correct grammatical form of English indirect questions. Nonetheless, Ana
made an unprompted change in the form of her utterance, from incorrect to correct,
while Keiko made no change. This can be seen by Ana's incorrect forms in lines a, c,
and e ("Can you tell me where is the train station?") and correct forms in lines g and
i ("Can you tell me where the train station is?"), with Keiko using the standard
English form all the way through. What is particularly impressive in this example is
that the change was made in the direction of the target language and not from a
correct target language form to an incorrect one. In a similar vein, in 10 hr of
taped conversations between nonnatives, Bruton and Samuda (1980) found only one
example of a change from correct to incorrect. In other words, errors of a nonnative
speaker peer are generally not incorporated, while one can find numerous examples
of modifications in the direction of the correct target language forms. Similarly, Gass
and Varonis (1989) found that 89% of all modifications made as a consequence of
an interaction were made in the direction of the target language. This included
incorporated changes that occur immediately as well as those that occur after a
period of time. Furthermore, we noted additional examples of what we call "incor-
rections," in which one of the NNSs offered an incorrect repair. In all of these cases,
the NNSs did not accept the repair and maintained the form that they had originally
used. For instance, in example 3 Hiroko says in his knee and Izumi responds with
the incorrect form in him knee. Interestingly, Hiroko maintains the original form in
terms of the pronominal case (his knee) but changes the preposition (from in to on),
thus clearly recognizing that something was not correct in her original utterance but
not incorporating the incorrection. Both finally end up with the correct form, on his
knee.

3. From Gass and Varonis (1989, p. 81)

Hiroko: A man is uh. drinking c-coffee or tea uh with uh the saucer of the uh uh coffee set
is uh in his uh knee
Izumi: In him knee
Hiroko: uh on his knee
Izumi: yeah
Hiroko: on his knee
Izumi: so sorry, on his knee

In example 4, on the other hand, after the incorrection mouth are open, which
Midori utters, Yoshi uses an entirely different construction.
Input, Interaction, and Production 289

4. From Gass and Varonis (1989, p. 81)

Yoshi: and uhm will she's uhm mouth is open
Midori: mouth are open
Yoshi: She has a rather wide jaw

Thus, although both NS and NNS interlocutors provide corrective feedback in

conversation with NNSs, most, but not all of it, is correct, and not all of it is incorpo-
rated. Another example of incorporated feedback is offered by Macdonald (1993),
who presented evaluations of pronunciation data from learners engaged in interac-
tions. NSs evaluated the pronunciation of NNSs in interactions with other NSs both
before and after such interaction. She found that subjects whose interactional strate-
gies indicated an awareness of pronunciation difficulty were more often rated higher
after the interaction than those who did not. Her study therefore provides a clear
indication of the value of feedback.
The importance of corrective feedback, or negative evidence, has been further
argued by White (1987), who suggested that what is necessary is not comprehensible
input, but incomprehensible input. It is incomprehensible input that may trigger
learners' recognition of the inadequacy of their own rule system. In essence, this is
the crux of the interaction argument: Comprehension difficulties, or "instances of
non-understanding" (Varonis & Gass, 1985), are what allow a learner to notice that
linguistic modification is necessary.
Through the data we examine in this paper, we attempt to further investigate the
relationship among input, interaction, and second language production. The specific
hypotheses discussed are the following:

1. Modified input yields better NNS comprehension than unmodified input.

2. Interaction yields better NNS comprehension.
3. Interaction yields better NS comprehension.
4. Prior interaction yields better L2 production.
5. Prior input modification yields better L2 production.
6. Real-world expectations affect comprehension.

METHOD

Subjects

Conversations from 16 native-nonnative dyads form the data base for this study. All
of the nonnative speakers were enrolled in an intensive language program at a large
U.S. university; all of the native speakers were undergraduate students at the same
university. All NNSs were at the high intermediate level and were of different LI
backgrounds (Chinese, Turkish, Japanese, German, and Korean). Pairing of native-
nonnatives was done by convenience (i.e., according to the times when they were
available to participate in the study).
 290 Susan M. Gass and Evangeline Marios Varonis

a. Script Modified Input Unmodified Input

(8 dyads) (8 dyads)

b. Trial 1 Interactive Noninteractive Interactive Noninteractive

(4 dyads) (4 dyads) (4 dyads) (4 dyads)

c. Trial 2 Inter. Noninter. Inter. Noninter. Inter. Noninter. Inter. Noninter

(2 dyads) (2 dyads) (2 dyads) (2 dyads) (2 dyads) (2 dyads) (2 dyads) (2 dyads)

Figure 1. Diagram of experimental design.

Procedure
Each member of the dyad performed a task in which he or she had to describe to a
partner where to place objects on a board. The boards were depictions of an outdoor
scene, in the first trial a beach scene and in the second a farm scene. The objects
included human figures, animals, and inanimate objects. The describers, who were
visually separated from their partners, had a board on which 20 objects were glued.
The partner had an identical board with the identical 20 objects placed to one side.
The task, then, was to describe to one's partner where to place these objects (Trial
1). In every case the native speaker first read from a script provided by the research-
ers. After this description, a second board, with the identical 20 objects, was given to
the nonnative to describe (Trial 2). Each trial took approximately 20 min, with
roughly 5 min separating the two trials.

Design
Prior to the start of data collection, a native-native pair and a native-nonnative pair
performed the first task described above, using the board for Trial 1. In neither case
was interaction allowed. None of these four individuals participated in the actual
study. The descriptions were recorded and then transcribed. The transcripts of both
pairs were used as the script for Trial 1. The script taken from the description of one
native speaker to the other native speaker was designated unmodified input; the
script taken from the description of the native speaker to the nonnative speaker was
designated modified input.
The 16 NS-NNS dyads in the study were divided into two subgroups: a modified
input group and an unmodified input group, illustrated in Figure 1, part a.
The groups were differentiated by the kind of input that the native speaker (the
NS member of one of the 16 NS-NNS dyads) gave to the nonnative speaker. In both
groups the NS followed one of the two scripts, transcribed from the data gathered
prior to the study.
Each of these two subgroups was further subdivided into two more subgroups
according to whether or not normal interaction (including requests for repetition,
clarification, comprehension checks, etc.) was allowed during the first description
Input, Interaction, and Production 291

Table 1. Number wrong on first trial: Nonnative speaker placement

of objects based on native speaker directions
Input Type
Modified Unmodified Total
(N = 160) (N = 160) (N = 320)

Noninteractive (fe = 160) 36 47 83

Interactive (k = 160) 24 37 61
Total (k = 320) 60 84 144

Note: Total opportunities for error = 320 (16 subjects x 20 figures) or 80 in each of four cells.

(see Figure 1, part b). Finally, there was one additional subdivision, illustrated in part
c of Figure 1, depending on whether interaction was allowed during the second
description (i.e., when the nonnative speaker was describing). The nonnative de-
scription was not scripted, as it was our intent to see which of the input and interac-
tion conditions led to more successful descriptions by the nonnative.
Our dependent variable was comprehension on the part of the NNS (Trial 1) and
the NS (Trial 2), as measured by the degree to which subjects were able to under-
stand instructions and hence accurately place objects on the board.'

RESULTS
The results are presented in terms of number wrong, determined by the accurate
versus inaccurate placement of the objects on the board. On the first trial, we take
this to be a measure of the NNS's ability to comprehend the instructions. On the
second trial, we take this to be the measure of the NNS's success at using the
language to give appropriate instructions.

Nonnative Speaker Performance Based on

Native Speaker Descriptions

We first consider the results of Trial 1 in terms of both of our experimental condi-
tions: (a) modified versus unmodified input (Hypothesis 1) and (b) interaction versus
noninteraction (Hypothesis 2). As can be seen in Table 1, nonnative speakers made
fewer errors (60/160) when they received modified input than when they received
unmodified input (84/160) from the native speaker direction-giver. This difference
was significant (Mann-Whitney U, p = .0087).2 Similarly, nonnative speakers made
fewer errors (61/160) when interaction was allowed than when it was not allowed
(83/160). Here, too, the difference was significant (p = .0209). Thus, both the condi-
tion of modified input and the opportunity for interaction resulted in significantly
fewer errors, confirming our first two hypotheses. In fact, the most successful condi-
tion (24/80 possible errors, or 30%) was that in which NNSs received modified
292 Susan M. Gass and Evangeline Marios Varonis

Table 2. Number wrong on second trial:

Native speaker placement of objects based on
nonnative speaker directions
Interactive Group Noninteractive Group

26 30
Note: Mann-Whitney U, p = .6744, n.s.

Table 3. Number wrong on second trial as a

function of interaction on first trial

First Trial Interactive First Trial Noninteractive

8 48
Note: Mann-Whitney U, p = .0039.

input and had the opportunity to interact with the direction-giver, whereas the least
successful condition (47/80 possible errors, or 59%) was that in which the input was
unmodified and interaction was not allowed (a combination of Hypotheses 1 and 2).

Native Speaker Performance Based on

Nonnative Speaker Descriptions

We next look at the results of the second trial, in which the NNSs switched roles with
the NS interlocutors and gave rather than received directions. We first examine the
effect of interaction on the second trial on object placement (Hypothesis 3). There
were 26 out of 160 possible errors when interaction was allowed, as opposed to 30
out of 160 possible errors when interaction was not allowed (see Table 2) (Mann-
Whitney U, p = .6744, n.s.). Thus, unlike the results in the first trial, on this trial
there is no clear indication that interaction had an effect on the outcome, disconfirm-
ing Hypothesis 3. The opportunity for interaction in the discourse did not affect the
NNSs' ability to provide more accurate and comprehensible instructions to their NS
partners.

Effects of Interaction
Our main research agenda was to determine whether there are linguistic effects
(evident in the NNS's productive trial) as a result of prior interaction. In Table 3 we
isolate the effect that interaction on the first trial has on success on the second trial
(Hypothesis 4). Here, the results are striking. Clearly, the NNSs who had the opportu-
nity to interact on the first trial, when they were receiving directions, were better
able to give directions on the second trial, as indicated by the number of times the
NSs receiving directions incorrectly placed objects on the board (Mann-Whitney U, p
- .0039). When the NNS had had the opportunity to interact on the first trial, the NS
made only 8 out of a possible 160 errors (5%) on the second trial; in contrast, when
Input, Interaction, and Production 293

Table 4. Number wrong on second trial as a

function of modification on first trial
First Trial Modified First Trial Unmodified

41 15

Note: Mann-Whitney U, p = .0587 (approaching significance).

the NNS had not had the opportunity to interact on the first trial, the NSs made 48
errors (30%) on the second trial. Thus, the NNSs performed better or gave more
accurate descriptions when the first trial had allowed them to interact with the NSs
who were describing.
A second variable that could influence performance on the second trial was the
presence of modified or unmodified input on the first trial (Hypothesis 5). Would
NNSs who had received modified input on the first trial, resulting in better perfor-
mance on Trial 1, be better prepared to give directions in Trial 2? This turned out
not to be the case. There were more errors on the second trial in those conditions in
which there had been modified input on the first trial (41/160 [26%] vs. 15/160 [9%])
(Mann-Whitney U, p = .0587 [approaching significance]). These findings are shown
in Table 4. This result is particularly interesting in light of the fact that modified
input resulted in greater success on the first trial, when the NNS was following
rather than giving directions. Thus, Hypothesis 5 is disconfirmed, at a level closely
approaching significance. It appears that receiving modified input may help a nonna-
tive speaker's comprehension in the short run, but may be a deterrent in subsequent
language production.

Real-World Expectations

We also considered whether or not accuracy was at all dependent on expectations

subjects might have about where the objects should go, given real-world knowledge
(Hypothesis 6). This is one way of measuring whether interlocutors rely more on
context than on grammar as an aid in comprehending speech. We had designed the
board so that half of the objects were placed appropriately, as, for example, a frog in
a pond, or a little girl with a watering can watering flowers. The other half were
placed in an unusual manner, as, for example, an automobile in a cloud, or a cat
sleeping on a lake. When we looked at the number wrong in relation to this dimen-
sion, we found that whether or not an object was in an expected place had little
effect on a subject's ability to describe or place the object. On Trial 1 there were 68
errors when the objects were in an expected location and 64 when they were in an
unexpected location.3 Similar results were noted on Trial 2: When the objects were
in an expected location, there were 29 errors as opposed to 24 when they were in an
unexpected location (see Table 5). Thus, at least in this controlled study, real-world
expectations are not an important factor in comprehension.4
294 Susan M. Gass and Evangeline Marios Varonis

Table 5. Number wrong as a function of

real-world expectations
Usual Unusual

First trial 68 64
Second trial 29 24

Summary of Results

In sum, the results indicate that both modified input and interaction affect immediate
task performance. When the NS is the speaker, both of these conditions lead to
greater NNS comprehension, as measured by task performance. Furthermore, the
opportunity to interact when receiving directions in turn enables the NNS to provide
better directions on a subsequent trial. This is not the case, however, when the initial
directions include modified input. The NNS who receives modified input is better
able to understand directions but subsequently less able to offer clear directions than
the NNS who receives unmodified input. In terms of our original six hypotheses, 1,
2, and 4 were confirmed and 3, 5, and 6 were disconfirmed.

DISCUSSION

The Effect of Modified Versus Unmodified Input

In discussing these results, the first question we seek to address relates to the input
conditions. Consistent with previous research (cf. Parker & Chaudron, 1987, for a
review of this literature5), our results show that modification significantly and posi-
tively affects comprehension. The NNSs were better able to comprehend when
listening to a modified script as opposed to an unmodified script regardless of
whether or not interaction was allowed.
However, the modified condition had a negative effect on the subsequent trial
(when NNSs gave directions), suggesting that the linguistic information contained in
the unmodified version was better suited for later language production. In 5 are
examples from the modified condition and in 6 are examples from the unmodified
condition.

5. Modified condition
a. On the other two stones of that path, put the dog. There's a little brown dog. So he's on
the stones close to the stairs.
b. At the top of those stairs, the three brown stairs, going up to the house, at the top of
those stairs, put the yellow cat.
c. Above the sun, place the squirrel. He's right on top of the sun.
6. Unmodified condition
a. Also along that path is the dog. So the dog is on the first two stones of that path. In
other words, he's right next to the stairs of the house.
Input, Interaction, and Production 295

b. The cat is at the top of the stairs. And she's against the house, against the far right side
of the front part of the house.
c. Now we're going to go to the sun, and the squirrel is right on the sun.

In considering what differentiated these two conditions, we note that in the unmodi-
fied utterances, the object to be placed tends to be overtly stated as both the topic
and grammatical subject at the beginning of the discourse, while in modified utter-
ances, the object to be placed appears as the grammatical object of the sentence
following phrases that locate that object in space. For example, in the modified
condition, the cat (example b) is not invoked until the end of the utterance (in the
position of new information); until this time, the hearer's linguistic attention may be
"on hold." In contrast, in the unmodified condition, the object to be placed, the cat,
appears as both topic and subject at the beginning of the utterance. This reduces the
burden on learners since once they identify the object to be placed, all of their
attention can be directed toward the descriptions involved in actual placement.
The unmodified condition also contained greater elaboration than the modified
condition. The unmodified script evidenced more types of expressions than the
modified script, which is characterized by a more limited vocabulary. This is seen in
the unmodified directions, which include such prepositions as along, next to, and
against, which are lacking in modified directions. In addition, the unmodified tran-
script, unlike the modified transcript, evidences the use of cohesive devices such as
also, so, in other words, and now. Our findings are consistent with work by Parker
and Chaudron (1987), who, in a review of the literature on modification, note that
elaborations are more likely to increase comprehension (cf. also Long & Ross, 1992,
for the relationship between retention and elaboration, and Ross, Long & Yano,
1992, for a discussion of the relationship between text type and comprehension). In
general, it appears that the unmodified condition, with its greater elaboration, lin-
guistic redundancy, and more explicit topic nominations allows for comprehended
input (Gass, 1988). It is comprehended input that is necessary for acquisition (or, at
least, later language production), because such input affords the learner the opportu-
nity for analysis, a prerequisite for new or restructured grammatical knowledge (cf.
Sharwood Smith's 1991 notion of enhanced input).

The Effect of Interaction

The second issue we address is the effect of interaction on subsequent performance.

We had hypothesized (Hypothesis 4) that the second trial would yield more accurate
results when the first trial was interactive. In fact, as we have shown, this was the
case. However, we had thought that the major linguistic area where this would
surface would be in the area of the lexicon. An interactive situation is one in which
unknown lexical items could be clarified; learners could use the first trial to gain
information about unknown vocabulary and then use these words to become more
precise in their descriptions. This turned out not to be the case, as we were unable
to find any examples where unknown lexical items (as evidenced by the need to
296 Susan M. Gass and Evangeline Marios Varonis

negotiate their meaning on the first trial) were used correctly on the second trial (see
also Loschky, 1994).
What learners did seem to take from the first trial were various descriptive
devices that NSs had used to explain unknown lexical items. An example is given in
7 below.

7. First trial
Jane: All right now, above the sun place the squirrel. He's right on top of the sun.
Hiroshi: What i s . . . the word?
Jane: OK. The sun.
Hiroshi: Yeah, sun, b u t . . .
Jane: Do you know what the sun is?
Hiroshi: Yeah, of course. Wh-what's the
Jane: Squirrel. Do you know what a squirrel is?
Hiroshi: No.
Jane: OK. You've seen them running around on campus. They're little furry animals.
They're short and brown and they eat nuts like crazy.

Second trial
Hiroshi: The second thing will be . . . put here. This place is . . . small animal which eat
nuts.
Jane: Oh, squirrel?
Hiroshi: Yeah (laughter).

In the first trial Jane is trying to describe to Hiroshi where to place the squirrel, a
word that Hiroshi did not know. Jane's description includes a physical description of
the animal as well as a description of what it eats. In the second trial Hiroshi does
not use the specific lexical item, squirrel, using instead the same descriptive device
Jane had offered on the previous trial. It is likely that the NNS's use of eat nuts to
describe what a squirrel does comes as a result of input during the first trial.6
Turning to data from nonnative speakers whose first trial was noninteractive, a
slightly different situation obtains. In the second trial they also took linguistic infor-
mation provided for them in the first trial but did not necessarily use it in an
appropriate manner. For example, in the native speaker script there were numerous
examples of phrases such as "you will find" and "you will see" (examples are given
in 8a). Additionally, there are statements and/or questions that do not necessitate
the future marker (as in example 8b).

8. Instructions from NSs to NNSs

a. You'll find the beach ball is to the left of the red chair.
You'll see there's a green light above . . .
b. You see the little point that comes up from the apex of the triangle.
c. You see where the two sides intersect?

In the native speaker's description, the use or lack of use of the future marker
depended on whether the speaker was giving instructions (in which case the future
was used) or attempting to determine whether the interlocutor's attention was fo-
Input, Interaction, and Production 297

cused on the relevant object (a form of questioning). In 9a and 9b are examples from
a nonnative speaker's description in which the future marker was used; but nonna-
tive speakers also use the verb form without the future marker (9c) for instruction
giving, a form-function relationship not found in native speaker usage.

9. Instructions from NNSs to NSs (noninteractive first trial)

a. Then, you will see a . . . is a wooden fan.
b. There . . . there's a uh the far, far, right, farrest right farrest right you will see the tree.
c. And uh then you see the girl with uh blue . . . skirt.

In descriptions by native speakers there was frequent use of the word put, or
place, but never in conjunction with will, unless there had been prior "scene-setting"
information. In 10 is an example of an extension of a form to an environment not
present in the native speaker discourse. This nonnative speaker appears to adopt
the native speaker mode for instruction giving but generalizes to inappropriate
environments. Examples such as 10 in which the description begins abruptly with
you will put the cat did not occur in the data from nonnative speakers in which the
first trial was interactive. In other words, NSs did not use this device as a way of
introducing new information, as this NNS did.

10. You will put the cat uh under that two stone . . .

Thus, it appears that the opportunity for prior interaction allows the NNS to spend
more time considering the linguistic input, even if the interaction is task oriented
and not language oriented. This opportunity for latent linguistic processing when the
learner's attention is focused on another task may have an important effect on the
reorganization of linguistic knowledge and subsequent language production.

CONCLUSIONS
Arguments concerning the importance of interaction and/or modified input for
second language acquisition have been primarily indirect. For example, Long (1985)
makes a three-step argument for the existence of a causal relationship. First, one
shows that linguistic/conversational adjustments promote comprehension of the
input. Second, one argues that comprehension and acquisition are directly related.
As comprehension increases, so does acquisition. Third, one must deduce that adjust-
ments promote acquisition.
A second type of argument has been made by Scarcella and Higa (1981) and
Varonis and Gass (1985). This argument has to do with the amount of work that
participants put into the conversation and how much effort they make to gain
comprehensible input. However, this cannot be sufficient, for if it were only a matter
of negotiation, then the most successful cases would be those in which there was
interaction. In the current study, however, only the opportunity for interaction on
the first trial yielded significant results. There were fewer errors in placement on the
298 Susan M. Gass and Evangeline Marios Varonis

first and second trials when the first trial had been interactive, while interaction on
the second trial did not have a significant effect. Thus, the greatest effect on the
successful completion of the task was the NNS's opportunity for preceding interac-
tion.
It appears, then, that the results of interaction are not necessarily immediate.
That is, they may not affect the conversation in which the interaction takes place, as
much as they do subsequent conversations.
How can we account for these results? One possible explanation is that the first
trial, when interactive, provided a longer amount of time for using language. This
would suggest that time on task is the crucial variable. While time on task may be a
partial contributor to an explanation, it is not a sufficient one, because there appear
to be other issues involved as well. We turn our attention to language-related issues
and in particular to native speaker- and nonnative speaker-initiated negotiations
and argue that this part of the discourse focuses a learner's attention on linguistic
form, on ways of creating discourse, and, in particular, on ways of describing objects.
In 11 are instances of native speaker-initiated input clarifications (i.e., feedback)
and in 12 instances of nonnative speaker-initiated negotiated forms.

11. Focus on form—native speaker initiated

a. Correction
NNS: Is he in the . . . pool? Is he in the sea?
-» NS: Yes. In the ocean. He has a hoe in his hand.
b. Unsolicited explanation
NS: OK. Remember we weren't sure which flower it was? It's the one that has a
stem. Do you know what a stem . . . ?
NNS: Astern.
-> NS: It's a line.
NNS: Oh, I see. I see.
NS: OK. Did you pick the right flower?
NNS: Uh.
NS: It's the one y'know y', the flower and you have the stem.
c. Comprehension check
NS: Now, let's go to the dock.
NNS: Dock.
NS: We're moving to the left of the board, but we're still in the water. Above the
water is a brown dock. Now . . .
NNS: Dog 11 have to move this dog.
NS: OK, that's a dog. This is a dock.
NNS: Duck.
-> NS: Dock, do you know what a dock is?
NNS: Yeah.
NS: OK.
NNS: (softly) What's the dock?
12. NNS-initiated negotiations
a. Targeted question
NS: You will see a path that has four stones on it next to the house.
-> NNS: Pat?
NS: A path.
NNS: OK.Uhhuh.
Input, Interaction, and Production 299

b. Request for repetition/definition

NS: Now underneath the second tree, follow the tree all the way down in the
brown part, in the dirt, you'll put the saw . . . in the dirt you'll put the saw.
-» NNS: What's saw? Could you repeat it please?

It is clear that negotiations are what differentiate the interactive from the nonin-
teractive conditions. We suggest that they crucially focus the learner's attention on
the parts of the discourse that are problematic, either from a productive or a re-
ceptive point of view. Attention, in turn, is what allows learners to notice a gap
between what they produce/know and what is produced by speakers of the L2. The
perception of a gap or mismatch may lead to grammar restructuring.
The importance ascribed to attention can also be seen in the general literature of
education. Osborne and Wittrock (1983) stated,

The pathway to the construction of meaning from an experience does not begin
with that experience. Rather, it begins with selective attention [italics added] of
that experience where selective attention is influenced by a variety of aspects of
long-term memory and cognitive processes, (p. 494)

The concept of attention is not new in the second language acquisition literature.
Hakuta (1976) noted that overtly marked forms in a language penetrate a learner's
attention and are learned earlier than forms without overt marking. More recently,
Nagle and Sanders (1986) discussed the essential role of attending in the process of
comprehending. In their terms, "attending involves the application of mental energy
to processing tasks and may range from focusing on specific features of input to
controlled processing for retrieval" (p. 17). There are other similar proposals: Bar-
dovi-Harlig (1987) proposed the concept of "salience" to account for acquisition facts
of second language learners, facts that are difficult to account for in any existing
models of second language acquisition, and Schmidt (1990) made the strong claim
that learning cannot take place without conscious awareness. Similarly, Long (1992)
has argued for the importance of attention-focusing devices in second language
learning. (For a recent review of the literature on attention as it relates to second
language acquisition, see Tomlin & Villa, 1994).
What we claim is that interactional input provides a forum for learners to readily
detect a discrepancy between their learner language and the target language and
that the awareness of the mismatch serves the function of triggering a modification
of existing second language knowledge, the results of which may show up at some
later point in time. Destabilization, then, is crucial if learning is to progress to
higher levels. The data presented here show that this interactive effect is not just an
immediate one (cf. also Gass & Varonis, 1989).
In sum, we have attempted to establish a methodology for directly investigating
the relationship between interactively modified input and second language produc-
tion. We have demonstrated the impact of interaction on later language use. We
utter a cautionary note in that we are only in a position to talk about later language
300 Susan M. Gass and Evangeline Marios Varonis

use; we are not yet in a position to talk about the effect of interaction on actual
incorporation of forms. What this study shows is the potential effect of interaction.
The notion of retention, a concept strongly allied to memory, must be addressed in
future research. We have argued that interaction serves to focus learners' attention
on form in instances where there is some perceived difficulty in communicating,
raising to a level of awareness that area of a learner's grammar that deviates (either
productively or receptively) from native speaker usage. Finally, we have argued that
this raising to awareness is a prerequisite for the restructuring of a learner's linguistic
knowledge.

(Received 9 December 1993)

NOTES

1. One could argue that this measure provides only an indirect measure of actual comprehension and
production. This is clearly the case, but it is to be noted that we were not necessarily interested in a
description of the two different transcripts as the literature is replete with descriptions of foreigner talk. What
we were interested in was a global assessment of comprehension on the part of the NNS (Trial 1) and of the
NS (Trial 2)—This latter we assumed to be a function of the NNS's facility with the L2. Since 16 different NSs
were involved in the study, it is unlikely that the differences found were artifacts of individual differences.
2. A Mann-Whitney test is similar to a t test, testing medians rather than means. It is, therefore, a more
conservative measure. It is often used when the n size is small, since it does not assume a normal distribution.
3. These results do not include all instances of incorrect placement since, for the first two dyads tested,
only the number wrong was recorded rather than a listing of the actual objects that had been incorrectly
placed.
4. This may in part be due to the nature of the experimental paradigm. Since participants were in a
situation of "play" (i.e., placing objects on a board), expectations of reality were suspended. In other words,
in this particular context, "anything goes."
5. The literature on the effect of linguistic adjustments is not completely clear-cut. For example, elabora-
tions appear to better facilitate comprehension than do linguistic adjustments, although some evidence
indicates that linguistic adjustments are beneficial.
6. An interesting possibility suggested by a reviewer is that Hiroshi is involved in "transitional incorpora-
tion." That is, the first trial led Hiroshi to "notice the gap" (i.e., his lack of lexical knowledge), and he is now
using a descriptive device as a strategy for compensating for his lack of knowledge.

REFERENCES

Bardovi-Harlig, K. (1987). Markedness and salience in second-language acquisition. Language Learning, 37,
385-407.
Brock, C, Crookes, G., Day, R., & Long, M. (1986). The differential effects of corrective feedback in native-
speaker conversation. In R. Day (Ed.), Talking to learn: Conversation in second language acquisition (pp.
229-236). Cambridge, MA: Newbury House.
Bruton, A., & Samuda, V. (1980). Learner and teacher roles in the treatment of error in group work. RELC
Journal, 11, 49-63.
Chun, A., Day, R., Chenoweth, A., & Luppescu, S. (1982). Errors, interaction, and correction: A study of
native-nonnative conversations. TESOL Quarterly, 16, 537-547.
Crookes, G., & Rulon, K. (1985). Incorporation of corrective feedback in native speaker/non-native speaker
conversation. TESOL Quarterly, 19, 607.
Day, R. (Ed.). (1986). Talking to learn: Conversation in second language acquisition. Cambridge, MA: Newbury
House.
Day, R., Chenoweth, A., Chun, A., & Luppescu, S. (1984). Corrective feedback in native-nonnative discourse.
Language Learning, 34, 19-45.
Doughty, C. (1991). Second language instruction does make a difference: Evidence from an empirical study
of SL relativization. Studies in Second Language Acquisition, 13, 431-469.
Input, Interaction, and Production 301

Duff, P. (1986). Another look at interlanguage talk: Taking task to task. In R. Day (Ed.), Talking to learn:
Conversation in second language acquisition (pp. 147-181). Cambridge, MA: Newbury House.
Ellis, R. (1985). Understanding second language acquisition. Oxford: Oxford University Press.
Ellis, R. (1991, April). The interactional hypothesis: A critical examination. Paper presented at the RELC
Conference, Singapore.
Gass, S. (1988). Integrating research areas: A framework for second language studies. Applied Linguistics, 9,
198-217.
Gass, S. (1991, December). ELT methodology from a learning perspective. Paper presented at Second Chula-
lonkorn University Conference on Language Teaching, Bangkok, Thailand.
Gass, S., & Varonis, E. (1986). Sex differences in NNS/NNS interactions. In R. Day (Ed.), Talking to learn:
Conversation in second language acquisition (pp. 327-351). Cambridge, MA: Newbury House.
Gass, S., & Varonis, E. (1989). Incorporated repairs in nonnative discourse. In M. Eisenstein (Ed.), The dynamic
interlanguage: Empirical studies in second language variation (pp. 71-86). New York: Plenum Press.
Hakuta, K. (1976). Becoming bilingual: A case study of a Japanese child learning English. Language Learning,
26, 321-351.
Knox, L. (1992, April). Cooperative fellow speakers and the enrichment of input: An application of relevance
theory. Paper delivered at the Second Language Research Forum 1992, Michigan State University, East
Lansing.
Long, M. (1980). Input, interaction and second language acquisition. Unpublished doctoral dissertation, Uni-
versity of California, Los Angeles.
Long, M. (1983). Linguistic and conversational adjustments to non-native speakers. Studies in Second Lan-
guage Acquisition, 5, 177-193.
Long, M. (1985). Input and second language acquisition theory. In S. Gass & C. Madden (Eds.), Input in second
language acquisition (pp. 377-393). Rowley, MA: Newbury House.
Long, M. (1992, March). Input, focus on form, and second language acquisition. Paper presented at the
American Association of Applied Linguistics annual meeting, Seattle, WA.
Long, M., & Ross, S. (1992). Modifications that preserve language and meaning. In M. Tickoo (Ed.), Simplifica-
tion. Singapore: Regional Language Center.
Loschky, L. (1994). Comprehensible input and second language acquisition: What is the relationship? Studies
in Second Language Acquisition, 16, 303-324.
Macdonald, D. (1993, February). Learners and modified/negotiated interactions: What works for pronuncia-
tion. Paper presented at Second Language Acquisition-Foreign Language Learning III conference,
Purdue University, West Lafayette, IN.
Nagle, S., & Sanders, S. (1986). Comprehension theory and second language pedagogy. TESOL Quarterly, 20,
9-26.
Osborne, R. J., & Wittrock, M. C. (1983). Learning science: A generative process. Science Education, 67, 489-
508.
Parker, K., & Chaudron, C. (1987). The effects of linguistic simplifications and elaborative modifications on L2
comprehension. University of Hawaii Working Papers in English as a Second Language, 6, 107-133.
Pica, T. (1987). Second-language acquisition, social interaction, and the classroom. Applied Linguistics, 8, 3 -
21.
Pica, T., & Doughty, C. (1985). Input and interaction in the communicative classroom: A comparison of
teacher-fronted and group activities. In S. Gass & C. Madden (Eds.), Input in Second Language Acquisi-
tion (pp. 115-132). Rowley, MA: Newbury House.
Pica, T., Holliday, L., Lewis, N., Berducci, D., & Newman, J. (1991). Language learning through interaction:
What role does gender play? Studies in Second Language Acquisition, 13, 343-376.
Pica, T., Young, R., & Doughty, C. (1987). The impact of interaction on comprehension. TESOL Quarterly, 21,
737-758.
Pienemann, M. (1992). Teachability theory. Unpublished manuscript, Sydney, Australia.
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.
Ross, S., Long, M., & Yano, Y. (1992, July). Simplification or elaboration? The effects of two types of text
modifications on foreign language reading comprehension. Paper presented at the Pacific Second Lan-
guage Research Forum, University of Sydney, Australia.
Samuda, V., & Rounds, P. (1993). Critical episodes: Reference points for analyzing a task in action. In G.
Crookes & S. Gass (Eds.), Tasks from a pedagogical perspective: Integrating theory and practice (pp.
125-138). Clevedon, UK: Multilingual Matters.
Sato, C. (1986). Conversation and interlanguage development: Rethinking the connection. In R. Day (Ed.),
302 Susan M. Gass and Evangeline Marios Varonis

Talking to learn: Conversation in second language acquisition (pp. 23-45). Cambridge, MA: Newbury
House.
Scarcella, R. (1983). Discourse accent in second language performance. In S. Gass & L. Selinker (Eds.),
Language transfer in language learning (pp. 306-326). Rowley, MA: Newbury House.
Scarcella, R. (1992). Interethnic conversation and second language acquisition: Discourse accent revisited. In
S. Gass & L. Selinker (Eds.), Language transfer in language learning (pp. 109-137). Amsterdam: John
Benjamins.
Scarcella, R., & Higa, C. (1981). Input, negotiation and age differences in second language acquisition.
Language Learning, 31, 409-438.
Schachter, J. (1986). Three approaches to the study of input. Language Learning, 36, 211-225.
Schmidt, R. (1990). The role of consciousness in second language learning. Applied Linguistics, 11, 129-158.
Sharwood Smith, M. (1991). Speaking to many minds: On the relevance of different types of language
information for the L2 learner. Second Language Research, 7, 118-132.
Tomlin, R., & Villa, V. (1994). Attention in cognitive science and second language acquisition. Studies in
Second Language Acquisition, 16,183-203.
Varonis, E., & Gass, S. (1985). Non-native/non-native conversations: A model for negotiation of meaning.
Applied Linguistics, 6, 71-90.
Wagner-Gough, K., & Hatch, E. (1975). The importance of input in second language acquisition studies.
Language Learning, 25, 297-308.
White, L. (1987). Against comprehensible input: The input hypothesis and the development of second-
language competence. Applied Linguistics, 8, 95-100.
Woken, M., & Swales, J. (1989). Expertise and authority in native-non-native conversations: The need for a
variable account. In S. Gass, C. Madden, D. Preston, & L. Selinker (Eds.), Variation in second language
acquisition: Discourse and pragmatics (pp. 211-227). Clevedon, UK: Multilingual Matters.
Zuengler, J. (1989). Performance variation in NS-NNS interactions: Ethnolinguistic difference or discourse
domain? In S. Gass, C. Madden, D. Preston, & L. Selinker (Eds.), Variation in second language acquisition-
Discourse and pragmatics (pp. 228-244). Clevedon, UK: Multilingual Matters.