Introduction
This article looks at input, output, and
interaction as sources of language data for
second language learners to draw on. The
second part of the article looks at how
learners extract linguistic information from
this data and considers a crucial stage in the
input-to-acquisition process, namely intake.
The study of input
The role of language input is acknowledged
by researchers with different perspectives on
second language acquisition. All agree that
some form of input is needed for language
learning to occur (cf. Gass, 1997). How
input is related to learning is an area of
contention: ‘it is uncontroversial that a
learner needs input in order to acquire a
language […]. Unfortunately the consensus
stops about there. How much input is
necessary? What kind of input? Under what
conditions need it be provided?’ (Gregg,
2001, p.167).
What is input? Sharwood Smith defines
input as ‘the potentially processable
language data which are made available by
chance or by design, to the language learner’
(1993, p. 167) and thus emphasises the point
that input need not be used (processed), but
potentially can be. Also, the intent with
which the language data are presented or
sought out, does not determine what is and
what is not input. Carroll (2001) makes a
distinction between stimuli as ‘all […]
observable instantiations of the second language’
(p. 8) and input, which she
reserves for stimuli that have entered the
brain. To her, input is not physical
information (sounds, visual data), but a
mental representation, available for internal
processors to use.
Not all definitions of input are quite as
neutral as the above. Gregg’s definition
(2001) of input is ‘information that is fed
into an input-output device; the output is
grammar’, (p. 167). Gregg takes a mentalist
view of language learning that sees language
as innate and its development as set or
‘triggered’ by the language data available to
the learner. Input here is seen as ‘evidence’.
Several researchers see the role of input as
the main source of information from which
language develops. Krashen (1982, 1985)
makes a distinction between input in a
general sense and comprehensible input, i.e.
language data that are understandable to a
learner. Comprehensible input in Krashen’s
view is sufficient for acquisition to take
place. When talking about input, authors
such as Krashen (but see also Faerch,
Haastrup, & Phillipson, 1984; VanPatten
1996) thus often take it to mean a specific
kind of (qualitatively different) language
data.
Beebe (1985) argues that what constitutes
input is determined to a large extent by the
learner:
Studies of input in second language
acquisition must view non-native speakers
not simply as passive recipients of
comprehensible or incomprehensible input
from native speakers, but as active
participants in choosing the target language
This appears to differ somewhat from the
commonly held view in cognitive psychology that
‘stimuli’ are ‘anything in the environment we
respond to’.
models they prefer and thus acquiring 'the
right stuff' according to their values. In other
words, learners have 'input preferences' (or
'model preferences') in the sense that they
consciously or unconsciously choose to
attend to some target language models rather
than others. (p. 404)
Another way of saying this is that input is
what learners pay attention to in the
language. This appears similar to what Sato
& Jacobs (1992) write: ‘in the present
perspective, input is viewed as the object of
the learner's attention’ (p. 269). This is
clearly different from Sharwood Smith’s
‘language data that are potentially available
for processing’ in that it combines
information with the process of selecting it.
The different approaches to (research into)
input were described by Schachter (1986)
who identified 1) a data-oriented approach
which observes, records, and transcribes
actual learner - native speaker interaction, 2)
a language-model approach which attempts
to describe language itself (attempting to
answer questions such as ‘what does
knowledge of a language consist of?), and 3)
a processing-model approach which focuses
primarily on the processing learners engage
in when interacting with the second
language. Although these three approaches
are all valid ways of investigating second
language acquisition, most researchers
would probably agree with the basic
meaning of the word input as ‘linguistic
data’. This is also the working definition I
adopt here.
Positive and negative evidence
Research has shown that the input learners
receive does not provide all the information
they need to learn a language. This has been
referred to as the ‘logical problem’ of
language acquisition (Bley-Vroman 1989).
Chomsky (1959, 1965), most prominently,
suggested that humans make use of internal
building blocks (‘universal grammar’), and
that the input they receive serves as evidence
of what is and what is not possible in the
language that is learned. Language is thus to
a large extent innate and the input triggers
its development. In second language
acquisition research the most vocal
subscriber to this view is Krashen who has
argued for the importance of comprehensible
input as a necessary and sufficient condition
for acquisition to take place (1982, 1985).
Language develops on the basis of positive
evidence, i.e. examples of actual language
use.
Others, however, have argued that such
comprehensible input alone is not sufficient
for learning to take place; it can be used for
comprehension only (e.g. Faerch & Kasper,
1980) without affecting change in the
learner’s interlanguage. ‘Paradoxically,
comprehensible input may actually inhibit
learning on occasion, because it is often
possible to understand a message without
understanding all the structures and lexical
items in the language encoding it, and
without being aware of not understanding
them at all’ (Long, 1996, p. 425).
Chomskyan researchers have proposed that
in addition to universal rules, learners are
also endowed with a set of language
constraints, limiting the number of
grammatical possibilities. However, if
participants make use only of the input they
receive and the positive evidence contained
in it, then how is it possible that they make
mistakes they themselves have never
encountered before?
L. White found (1989) that learners’
language behaviour supports the transfer
hypothesis. This states that second language
learners primarily make decisions of
acceptability on the basis of their first
language. When the first language (e.g.
English) is more restrictive (for example in
requiring adverb adjacency) than the second
(e.g. French), learners will limit themselves
to the more restrictive use found in their first
language. To avoid this type of transfer
White has argued that negative evidence is
needed (e.g. L. White, 1989, 1991; see also
Bley-Vroman, 1986). Negative evidence is
defined as ‘the type of information that is
provided to learners concerning the
incorrectness of an utterance’ (Gass, 2003,
p. 225) and may help by drawing learners’
attention to language form. Trahey writes:
‘exposure to positive evidence can lead to
changes in linguistic competence when the
structure to be acquired is readily available
in the input. However when the problem is
one of 'unacquiring', as in the case of
SVAO, positive evidence appears not to be
sufficient’ (1996, p.134).
The role of negative evidence, however, is
disputed. Some have claimed that it is not
clear that it occurs at all outside instruction
(Pinker, 1989). However, several studies
have shown it to exist in both first (Farrar,
1992), and second language acquisition
(Oliver, 1995). Mackey, Oliver & Leeman
(2003) investigated 48 native speaker/non-native speaker and non-native speaker/non-native speaker dyads consisting of both
adults and children while completing an
information gap task. Although they found
differences between the various groups
(with native speakers notably providing
more feedback than non-native speakers), in
all cases at least 30% of errors resulted in
feedback and between 25% and 41% of
these resulted in modified output.
It has, however, been argued that even if
negative evidence occurs, it is not relevant
to learning. Schwartz (1993) writes:
If there were a translation algorithm that
could take the knowledge that results from
being told “This is not a grammatical
sentence in this language” and convert it
into information that is in a form the
processes in the language module could
make use of, then ND [negative data] could
be usable. It appears, however, that there
exists no such mechanism’ (p. 158).
Schwartz does not argue that negative
evidence cannot result in learning, but says
that it does not result in the type of
knowledge underlying second language
proficiency. However, evidence exists that
learners can in fact make use of negative
data. An often cited study in this respect is
one by Bohannon & Stanowicz (1988)
which showed that children do benefit from
negative evidence in learning their first
language
. Also for second language
learning there appears to be evidence
showing a facilitative effect on learning (see
below), but it has been pointed out that in
interpreting the results of such studies one
has to be careful to distinguish between
immediate effects (uptake) and delayed
effects (learning) (cf. Birdsong, 1989).
Negative evidence can be provided in
different ways. Long (1996) distinguishes
between explicit negative evidence (such as
error correction), and implicit negative
evidence (such as when one interlocutor
shows not to have understood the other, or
reformulates an utterance without it
interrupting the flow of the conversation (i.e.
recasts). Negative evidence can perform
different functions. Firstly, it can help
learners ‘notice the gap’ between the input
and their own output. R. Ellis (1995)
Although their study has been criticised for only
investigating one particular socio-economic stratum
of society, thus questioning its generalisability to
other groups of learners.
describes this as making a cognitive
comparison. As a result of realising this
gap, participants can then attempt to
reformulate their utterance or store
information about that aspect of the
language. It may also result in quite sudden
shifts in the learner’s interlanguage, for
example when it leads to a realisation that
certain forms cannot be used in the target
language at all. Secondly, negative evidence
can also increase learners’ awareness of the
target language in a broader sense. By
drawing attention to what is not possible in
the target language, negative feedback
necessarily contrasts different linguistic
forms and encourages learners to understand
the differences (Schmidt, 1990).
Research on various types of negative
evidence has shown facilitative effects on
learning. Carroll & Swain (1993) for
example, investigated instruction of English
dative alternation combined with one of four
types of feedback. Learners either 1)
received metalinguistic information about
their responses, 2) were told when their
answer was wrong, 3) received a recast with
a model answer, or 4) were asked if they
were sure about their answer. The results
showed that all four groups improved
performance, but that the metalinguistic
group did best.
Han (2002) investigated the effects of
recasts for tense consistency provided
during eight sessions over a period of two
months in written and oral narration tasks.
Feedback was given individually, and was
consistent; all errors were recasted. Using a
pretest-posttest-delayed posttest design she
Although this term appears broader in that it need
not involve a ‘gap’ in the sense of an inability to
express oneself, i.e. the comparison could be made
between two sources of input. Swain (1998) refers to
this as noticing a ‘hole’.
found recasts to lead to increased awareness
(and much greater awareness than in the no-recasts group), and improved tense
consistency both on the production tasks and
on an immediate and delayed posttest
Recasts were given in individual settings,
and Han suggests that this may have
contributed to the results.
A number of other studies have also found
benefits for negative feedback on second
language learning (e.g. Pica, 1994; Mackey
& Philp, 1998), however a small number did
not (e.g. Sanz & Morgan-Short, 2004).
Mackey, Oliver & Leeman (2003) point out
that a range of variables (types of dyads,
type of interaction, age, etc) affects the
provision, uptake and learning effects of
negative evidence and that more research
needs to be done to isolate and investigate
those variables.
Input and learnability
For some researchers the primary role of
input is to trigger the development of innate
knowledge of language (Krashen, 1981;
Schwartz, 1993; L. White, 1989).
Pienemann (1984, 1985, 1988, 1989, 1998),
sharing this viewpoint, argues that language
(including a learner’s second language)
develops in a predictable, and pre-determined way. His language processing
theory posits a number of determinants of
the relative difficulty and acquisition order
of various linguistic features in the input.
The first of these is psychological
complexity and refers to the extent to which
‘the language learner must re-order and re-arrange linguistic material in the process of
mapping underlying semantics onto surface
It has to be pointed out that the participants in the
study were at an upper-intermediate level and
reported high motivation for learning English.
forms’ (1988, p. 89). The second is saliency;
items that are more salient are easier and
will be acquired earlier. Items are more
salient if they come in sentence initial or
sentence final position. The third
determinant is the distance between an item
that triggers a transformation and the place
in the sentence where the transformation is
effected. The greater the distance, the harder
the item. The most important premise of
processing theory, however, is that learners
must go through various “levels” which they
cannot skip. At the first level learners are
unable to organise lexical material into word
classes or categories and cannot identify
where the information in the sentence is. At
the second level they are partially able to do
so but generally only on the basis of salient
sentence items (e.g. sentence first or
sentence final). At the third level learners
are able to organise some of the lexical
material into categories but transformation is
limited to either word or constituent initial
or final position. At the fourth level they are
able to recognise all elements in a sentence.
The fifth level, S-procedures, deals with an
exchange of information between internal
constituents. Pienemann sees these levels as
an ‘implicational hierarchy of processing
procedures’. Implicational here means that
each procedure necessarily builds on the
other; ‘the learner cannot acquire what
he/she cannot process’ (p. 87). Again,
language development is seen as
predetermined.
Others see a more active role for the learner
and more possible variation in the
acquisition of various features in the input.
Goldschneider & DeKeyser conducted a
metastudy (2001) of second language
research on the acquisition of morphemes.
Using multiple regression analysis they
identified five determinants explaining a
large portion of the variation in performance
found in the individual studies. These were:
1) perceptual saliency, or how easy it is to
hear or perceive a given structure
Semantic complexity or how many
meanings are expressed by a certain form. 3)
Morphophonological regularity as the
degree to which a grammatical feature is
affected by its phonological environment. 4)
syntactic category as the difference between
lexical and functional and free versus bound
items. 5) frequency. All of these factors will,
to some extent, impact on the order in which
they are acquired
N. Ellis (2002a, 2002b) specifically
investigated point 5) above and argued that
frequency largely determines acquisition.
Second language acquisition, in his view, is
to a large extent exemplar based and not
based on abstract rules or structures (2002a):
‘the acquisition of grammar is the piecemeal
learning of many thousands of constructions
and the frequency-biased abstractions of
regularities within them’ (p. 168). He sees
second language learning as a form of
implicit learning, dependent on input.
However, he points out that this may not
apply to initial registration of language
forms, which may require attention and
conscious noticing. What gets registered
initially gets grouped later, by ‘unconscious
processes of association to form larger units
Saliency is a term also used by other researchers
with sometimes different or additional meanings. For
example Bardovi-Harlig (1987) defines it ‘in terms of
availability of data’ (p. 401), (which appears to be
similar to frequency) and uses it to explain findings
of acquisition order (preposition stranding before
preposition pied piping) which appear to go against
the order predicted by universal grammar (unmarked
construction such as preposition stranding are
acquired first).
Goldschneider & DeKeyser point out that ‘we have
argued that these five factors are not a completely
heterogeneous set, but can all be seen as aspects of
salience in a broad sense of the word’ (p. 37).
that are henceforth used in pattern
recognition’ (p. 174).
Other aspects of the input have been
identified as affecting noticing and learning.
VanPatten (1985, 1990, 1996; VanPatten &
Cadierno, 1993) argues that learners assign
varying degrees of “communicative value”
to different aspects of the input, defined as
‘the meaning that a form contributes to
overall sentence meaning’ (2001, p. 759)
and ‘the relative contribution a form makes
to the referential meaning of an utterance
[…] based on the presence or absence of two
features: inherent semantic value and
redundancy within the sentence-utterance’
(1996, p. 24). The communicative value of a
form is greater if its semantic value is
greater and if it is not redundant. In his 1990
study, VanPatten asked subjects to mark all
the occasions they heard the Spanish article
la, the third-person plural verb morpheme -n, or the word inflación. His results showed
that attending to the verb morpheme or the
definite article resulted in lower
comprehension levels. He attributed that to
their lower communicative value and
suggested that learners first look for content
words in the input. If sufficient attentional
resources are available grammatical forms
may then be processed. VanPatten’s concept
of communicative value has come under
heavy criticism. DeKeyser, Salaberry,
Robinson and Harrington (2002) argue that
both form and meaning can be processed
simultaneously and that to expect the
internal parser to scan all content for
communicative value first while saving
certain parts of the input to be processed
later is not congruent with current thinking
and findings from research into sentence
processing.
Output and interaction
The roles of output and interaction on
acquisition have received considerable
attention since the publication of Hatch’s
early papers (1978a, 1978b) in which she
drew attention to their potential benefits to
learning. For some (e.g. Krashen, 1981) the
role of interaction is predominantly that it
provides learners with comprehensible input
(i.e. input that is attuned to their
developmental level). For others, interaction
and the output it generates directly
contribute to learning. Long is well-known
for his interaction hypothesis (1981, 1983,
1996) which emphasises the crucial role of
the process of negotiation on learning.
Negotiation, or the ‘modification and
restructuring of interaction that occurs when
learners and their interlocutors anticipate,
perceive, or experience difficulties in
message comprehensibility’ (Pica, 1994, p.
493), has a number of beneficial effects.
Firstly, it aids in increasing understanding,
and thus results in the learner receiving
more, and more comprehensible input,
necessary for learning to take place.
Negotiation exchanges are said to result in
‘denser’ than average speech, with more
repetitions, reformulations, expansions,
extra stress, and a range of other features, all
of which increase frequency and saliency of
aspects of the input. Learners are also more
likely to benefit from this enhanced input as
they have at least partial control over the
semantic content of the interaction and can
thus free up attentional resources to pay
attention to form in the input. They are
likely to be alert as they try to get their
meaning across and as a problem in the
communication occurs.
Secondly, interaction takes place in a
context that is meaningful to the
interlocutors. From this context learners
derive a degree of support which helps them
in their understanding as well as in getting
their meaning across. They also derive
support from their conversation partners
who may supply words, or restate
utterances, and in so doing provide
scaffolding, allowing learners to express
meaning they would otherwise be unable to.
Next, interaction can also lead to the
occurrence of negative feedback (one form
of negative evidence, see above), i.e.
information about what is and is not
understandable and/or correct in a speaker’s
output:
Negative feedback is generally facilitative of
L2 acquisition, and necessary for the
acquisition of specifiable L2 structures (such
as the English adverb-placement example
for French speakers) for which positive
evidence will be insufficient. A mechanism
is posited whereby, while correct-form
meaning associations are strengthened both
by positive evidence and that negative
feedback that contains positive evidence,
incorrect associates are weakened and in
some cases ultimately relinquished
altogether as a result of both negative
evidence and prolonged absence of support
in the input. (Long, 1996, p. 430).
Gass (1997; Gass & Varonis, 1994) has
argued that since such negative feedback is
situated in a communicative context and is
thus linked to actual communicative goals, it
is more likely to be usable to the learner.
However, it is not entirely clear how much
negotiation of meaning takes place, with
some claiming that it is substantial (Pica,
1994), and others that it is not (Skehan,
1989). It has also been pointed out that
although interaction can have beneficial
effects, conversational success in itself does
Towards a definition of intake
not necessarily result in learning (Faerch &
Kasper, 1980).
Swain has argued for the important role of
learner production in learning. Her “output
hypothesis” (1985) developed from
observations of Canadian immersion
students who, despite years of receiving
exposure to the second language, did not
fully develop in particular certain
grammatical aspects of the target language.
Swain found that the immersion classes
were characterised by a lack of opportunities
for output and afforded few opportunities for
“pushed output”, i.e. output that required
them to ‘stretch their interlanguage’. Many
students were able to get by using
communication strategies to get their
meaning across and were never challenged
to further develop their language. Swain
suggested that by requiring learners to
produce comprehensible output, they would
be pushed to be more accurate and to pay
attention to both form and meaning, and in
so doing move from semantic to syntactic
processing. In addition, Swain (1998)
suggested that output would 1) induce
noticing, 2) allow for hypothesis formation
and testing (see also R. Ellis & He, 1999;
Pica 1988), and 3) give opportunities for
meta-talk. The effect of output on 1)
noticing, was investigated in another article
(Swain & Lapkin, 1995) in the context of a
writing task with a think-aloud protocol. The
authors found that learners do become aware
of problems in their writing and engage in
strategic thought processes to solve those
problems. Swain (1985) has suggested that
output can also serve to help with
developing automaticity (referred to as the
fluency function of output) and this seems to
have been corroborated by research showing
that when producing the language,
connections in the brain are strengthened,
aiding the process of automatisation (cf. de
Bot, 1992, 1996). Izumi, in several studies,
investigated the effect of output on noticing
(2002, 2003; Izumi, Bigelow, Fujiwara, &
Fearnow, 1999). He suggested (2002) that
Swain & Lapkin’s study may have found an
effect for the noticing induced by the output
because their measurement was immediate
task performance. To investigate if there was
also a delayed effect he made use of a
written reconstruction task to measure
noticing of English relative clauses,
followed by a series of posttests. He found
that in comparison with a control group who
received a receptive (meaning-focused) task
only, output did have an effect on both
noticing and learning.
Benefits have been found for non-interactive
language production. In the general learning
domain, Baddely (1990) writes about the
effects of producing an item: ‘the act of
successfully recalling an item increases the
chance that that item will be remembered.
This is not simply because it acts as another
learning trial, since recalling the item leads
to better retention than presenting it again; it
appears that the retrieval route to that item is
in some way strengthened by being
successfully used’ (p. 156).
In the second language acquisition domain
N. Ellis & Sinclair (1996) found that
subjects encouraged to rehearse foreign
language utterances were better than both
silent controls and subjects prevented from
rehearsal by articulatory suppression at a)
comprehension and translation, b) explicit
metalinguistic knowledge of the detailed
content of grammatical regularities, c)
acquisition of the foreign language forms of
words and phrases, d) accuracy in
pronunciation, e) some aspects of
productive, but not receptive, grammatical
fluency and accuracy.
Although not rejecting a role for output and
interaction, VanPatten (1996, 2002;
VanPatten & Cadierno, 1993) has argued
that the role of input and input processing
are crucial for language development.
VanPatten & Cadierno (1993) compared
traditional form-focused instruction (rule
presentation followed by output practice) on
direct object pronouns with ‘Processing
Instruction’ which ‘involves explanation and
practice/experience processing input data,
taking learner strategies in input processing
as the starting point for determining what
explicit instruction should look like’ (p.
225). In the study, participants receiving
processing instruction were given an
explanation of the target structure as well as
‘explanations of important points to keep in
mind about the position of object pronouns
of Spanish’ (p. 231). This was followed by a
reading or listening exercise in which
participants had to demonstrate
understanding of the structure, and an
activity in which they had to respond to the
content of the input. VanPatten & Cadierno
found that input processing led to significant
gains for both comprehension and output
skills, compared with a significant
improvement for output skills only for the
traditional group. VanPatten argued that
second language instruction should include
an increased focus on improving the way
learners process the input, as opposed to
focusing primarily on output practice.
However, in a replication study, DeKeyser
& Sokalski (1996; see also DeKeyser et al.,
2002), found that the relative effectiveness
of production versus comprehension practice
depends on the complexity of the target
structure and on the delay between practice
and testing; there may an immediate, but no
lasting effect for comprehension practice.
Allen (2000) investigated acquisition of
French causatives and compared the effects
of processing instruction and production-based instruction. She did not find an
advantage for the processing instruction
group compared with the production group,
but both groups did improve compared with
a control group. Other studies have reported
similar findings (Erlam, 2003; Salaberry,
1997).
Despite VanPatten’s suggestions, there
appears to be evidence of a facilitative effect
for output and interaction
. In sum, output
and interaction can:
- provide additional input
- result in comprehensible input which
impacts on learning
- enhance fluency by allowing
participants to produce the target
language
- facilitate form-meaning connections
- result in negotiation of meaning
which in turn can raise awareness of
the target language
- provide opportunities for negative
feedback
- impact on learning directly as a
result of verbalisation
Definitions of intake
The term intake ‘has taken on a number of
different meanings, and it is not always clear
what a particular investigator means in using
it’ (McLaughlin, 1987, p.13). That was true
nearly two decades ago, and it still is true
today. The purpose of this section, then, is to
review these different meanings and identify
the commonalities and differences between
them.
VanPatten (2002) has recently clearly
acknowledged a number of important roles for
output, such as that it can function as a ‘focusing
device’, drawing learners to mismatches between the
input and their own output, and he has acknowledged
its role in the development of fluency.
Definitions of intake come into three broad
categories: those that see intake as a product,
those that see it as a process, and those that
see it as a combination of the two. Coming
into the first of these categories is Corder
(1967) who provides the earliest recorded
definition of the term as: ‘a mental
representation of a physical stimulus’ (p.
165). For Corder, intake is thus something
that has been detected but has not yet been
integrated into the learner’s developing
second language system as it is still linked to
the physical stimulus. Others also see intake
as a product, but give it quite a different
meaning. Krashen, for example, on the basis
of an investigation of caretaker speech
(1978) concludes: ‘ “intake” is, simply,
where language acquisition comes from, that
subset of linguistic input that helps the
acquirer learn the language‘ (1981, p. 101)
and: ‘intake is first of all input that is
understood’ (p. 102; emphasis in original).
Interestingly, Krashen talks of input in first
language acquisition containing ‘a high
proportion of intake’ (p. 102, 1981;
emphasis in original) by which he means
language input of which a great deal is
understood. This is interesting because it
shows that for Krashen the occurrence of
intake is something that is not so much
dependent on the learner as on the quality of
the input. It appears that in Krashen’s view
one cannot help but understand
appropriately used input, and thus be
provided with intake. It must be noted that,
perhaps as a result of this interpretation,
Krashen appears to use the terms input and
intake somewhat arbitrarily. For example,
on a different occasion, he talks about
language acquisition developing better when
the intake is communicative and understood.
Finally, Krashen claims that intake ‘builds’
acquisition, but how this happens is not
elaborated on.
Also Faerch & Kasper (1980) see intake as a
product but make a distinction between
intake for communication and intake for
learning. Intake for communication is
detected input that the learner has
comprehended (maybe partially on the basis
of non-linguistic aspects relating to the
communication that takes place), whereas on
the basis of intake for learning ‘the learner
forms her hypotheses about the L2 rules and
tests them out subsequently’ (p. 64). Intake
for learning is clearly processed more deeply
as it requires the learner to (consciously or
not) make a comparison between current
knowledge and new information, whereas
this is not the case with intake for
communication. Loschky & Bley-Vroman
(1993) see intake for communication as
depriving the learner of the potential for
feedback and thus a chance to notice a
difficulty with his/her performance:
Thus, it may be possible to (1) comprehend
native speaker input, or (2) make one's
interlanguage output comprehensible to a
native speaker without (3) focusing on or
using the target form of instruction. […] this
is certainly possible through use of strategic
competence. Second, as a consequence of
this, negative feedback which could
potentially destabilize one's target language
hypotheses may be either absent or non-salient. The learner may never 'notice a gap'.
(p. 131).
Sharwood Smith (1986) does not
specifically discuss intake but does make a
similar distinction as do Faerch & Kasper.
He talks about input having dual relevance,
for immediate communicative purposes but
the input may also ‘contribute to the
substantiation or reflection of some current
hypothesis about the target language system’
(p. 243). Input may not be relevant for
acquisition where the learner is not
developmentally ready or where the
demands of the communicative exchange are
heavy, even though the input can be
interpreted. Sharwood Smith later defined
intake as ‘that part of input which has
actually been processed by the learner and
turned into knowledge of some kind’ (1993,
p. 8). He specifies this by saying ‘input is, as
it were, the goods that are presented to the
customer, including the articles that the
customer picks up to look at. Intake is what
is actually bought and taken away from the
shop, i.e. what passes into the ownership of
the customer’ (pp. 8-9). This is ambiguous.
What is meant by ‘ownership’ here? Is it a
hire-purchase which may be returned at any
time, are ‘the goods’ consumed before the
buyer even arrives home, are they shelved
for future use, or are they used every day?
Carroll (2001) also sees intake as a product
but makes a very clear distinction between
comprehension and intake. She describes
comprehended speech as a ‘speech signal
which has been successfully parsed and
re-encoded in semantic terms’ (p. 9). Carroll
rejects the view that intake consists of
comprehended speech as it would mean that
all learning would involve concept learning.
To her, comprehending speech is ‘something
that happens as a consequence of a
successful parse of the speech signal’ (p .9;
emphasis in the original). She sees intake as
a subset of the input; stimuli that are
perceived by the learner. She defines it as
‘that which is taken in by the hearer’ (p. 10).
In addition, perceived stimuli are
characterised as “transduced stimuli”, or
stimuli neurally available for processors to
extract further information from. Carroll
emphasises that intake is not input to the
learning mechanisms, but input to speech
parsers. Her view of intake thus diverges
from that of the other authors cited here.
Gass’s (1997) model of second language
acquisition consists of a number of stages
starting from raw input. Several factors
(including time pressure, frequency, affect,
salience, associations and prior knowledge)
influence whether input gets noticed, or
apperceived. Apperception is conceptualised
as a priming device. It prepares the learner
for the possibility of subsequent analysis.
Some or all of the noticed input may be
comprehended, with comprehension relating
to a continuum of properties of the
apperceived input, from meaning-related
properties to deeper, grammatical features.
What gets comprehended may (depending
on a range of factors) become intake, which
Gass defines as the ‘process of assimilating
linguistic material’ (p. 5). Intake can thus be
conceptualised as apperceived input that has
been further processed. This further
processing can take the form of hypothesis
testing, rule strengthening, storage for later
use, or the intake may remain unused. It is
interesting to note that unlike Faerch &
Kasper (1980), Gass sees comprehension
(comprehended input) as a prerequisite for
intake to take place.
VanPatten’s (1996) definition of intake is
similar to Gass’s: ‘intake is the subset of
filtered input that serves as the data for
accommodation by the developing system. It
is the input that has been processed in some
way by the learner during the act of
comprehension. Intake […] are the data
made available for further processing (e.g.
internalization) once the input has been
processed’ (p. 10). More recently VanPatten
(2002) defined intake as ‘the linguistic data
actually processed from the input and held
in working memory for further processing’
(p. 757). For VanPatten, as for Gass, intake
follows or occurs simultaneously with the
process of comprehension. Leow holds a
similar view: ‘intake, […] is that part of the
input that has been attended to by second
language learners while processing the
input’ (1993, p. 334).
Others have approached intake as a process,
rather than as a product. Chaudron’s
definition of intake (1985), for example,
encompasses the processes Carroll, Gass and
others refer to: ‘intake is processing of target
language input’, or ‘the mediating process
between the target language available to
learners as input and the learners’
internalized set of L2 rules and strategies for
second language development’ (p. 1). And:
'in speaking of intake we are, in effect,
referring not to a single event or product, but
to a complex phenomenon of information
processing that involves several stages,
roughly characterized as (1) the initial stages
of perception of input, (2) the subsequent
stages of recoding and encoding of the
semantic (communicated) information into
long term memory, and (3) the series of
stages by which learners fully integrate and
incorporate the linguistic information in
input into their developing grammars’ (p. 2).
Chaudron refers to this process as a
continuum from preliminary to final intake,
although he concedes that the two ends of
the continuum constitute very different
categories of cognitive activity (1983, pp.
438-439).
Also, Boulouffe (1987) conceptualises
intake as a process and calls it ‘the
notoriously impenetrable interval between
input and output’ (p. 245). She describes
intake as ‘the locus of the learner’s active
search for inner consistency’ (p. 246) by
which she means a process of equilibration
through accommodation or assimilation of
new knowledge. She gives examples of
students receiving feedback on an incorrect
production of a target sentence. The number
of attempts represent the intake process
whereby the new knowledge (the correct
sentence structure) is assimilated. She
clearly sees intake as the process of learning,
hypothesis testing, transfer etcetera. It is also
a process that requires the learner’s active
participation as it is something that is
subject to control.
Similarly, Hatch (1983) writes: ‘if the
learner “casts a net” into the input, the result
is supposedly intake’ (p. 79) and ‘for me, all
input is intake if the learner does respond in
some way to it’ (p. 81).
We might say that input is what the learner
hears and attempts to process. That part that
learners process only partially is still input,
though traces of it may remain and help in
building the internal representation of the
language. The part the learner actually
successfully and completely processed is a
subset called intake. That part, then, is the
language that is already part of the internal
representation. (p. 81).
Yet others acknowledge that intake can be
seen as both a product and a process.
Kumaravadivelu (1994) provides an
overview of the preceding 25 years of
theorising into intake and suggests that a
focus on intake as either a product or a
process is flawed. He proposes a synthesis
of a range of intake factors (age, affective
factors, negotiation etc) and intake processes
(structuring, inferencing, transfer etc) that
dynamically interact and are co-requisites
for intake. Intake factors determine which
aspects of the input get engaged. These
receive attention and as a result of this,
consciously or unconsciously, a mismatch is
detected between that aspect of the input and
existing knowledge the learner has. As a
result of this mismatch a range of intake
processes execute that form the process of
learning. Like Chaudron then,
Kumaravadivelu includes within intake the
whole process from detection to final
acquisition. It is difficult to see how this
proposal of intake can be distinguished from
one of learning in general.
Finally, although not concerning itself so
much with attempts to define intake, some
recent neurobiological research has made
efforts to come to a more precise
understanding of what constitutes intake and
where in the brain the product or process is
localised. Sato & Jacobs (1992) identified
the nucleus reticularis thalami (NRT) as that
area of the brain that seems to facilitate
processing in other areas of the brain known
to be involved in (language) learning and
production, including the hippocampus, the
cerebellum, the basal ganglia, and the
cingulate gyrus. The authors propose that
the NRT functions as a “gating mechanism”
that allows or inhibits information flow to
these areas. As such the NRT can be said to
facilitate intake from input in a literal sense.
Their assertion, however, that ‘the key
assumption here is that information
ascending through the NRT to the cerebral
cortex constitutes intake (or at least potential
intake)’ (p. 287) casts some doubt on their
own claims as it shows that by adding the
afterthought between brackets, the authors
make a distinction, perhaps implicitly,
between what enters the system and aspects
of that information that may be used for
language learning. Although the authors
admit that ‘the NRT's posited role in
language acquisition is at present neither
directly observable nor testable’ (p. 287),
identifying the neurobiological correlates of
the intake process is a promising approach
to arriving at a more precise, and meaningful
interpretation of the processes that the
concept of intake is said to embrace.
Research into brain activity may help make
such processes ‘observable’. One relevant
example is the work of Buckner (2000) who,
quoting himself, writes:
Neuroimaging data suggest a pattern relating
localized brain activity and memory
encoding. Several neuroscientific
hypotheses have proposed that certain
regions within the frontal cortex participate
in the short term maintenance and
manipulation of information over brief
periods of time, as would be required during
many kinds of information processing tasks.
Deep processing tasks and intentional tasks
make use of such processing, while shallow
processing tasks do not (Buckner & Tulving
1995).’ (p. 285)
Buckner relates this to the formation of
memory traces and writes: ‘one speculation
would be that the critical cascade that drives
episodic human memory formation occurs
when frontal activity provides a source of
information (input) to medial temporal lobes
and functions to bind together the outcomes
of information processing from frontal and
other cortical regions to form lasting,
recollectable memory traces’ (p. 285). The
same could apply to language learning
where frontal processing provides input to
the rest of the developing system. This is an
interesting, but as of yet little explored
avenue for research into the topic of intake.
A working definition of intake
Above a wide range of existing definitions
have been discussed. Next I turn to what I
believe are the key elements in these
definitions and drawing on these I will
propose a working definition of intake.
First, I make a distinction between input and
stimuli, where stimuli are seen as the
Towards a definition of intake 01
language potentially available to a learner,
and input as those stimuli that have entered
the learner’s system. Secondly, and in line
with most of the authors mentioned above, I
see intake as resulting from that subset of
the input that is detected by the learner.
Following Tomlin & Villa (1994), I do not
equate detected input with noticed input. A
definition of intake or an operationalisation
of it, should probably leave open the
question of whether or not intake can only
be derived from noticed input, until greater
evidence for either position has been found
(cf. Schmidt, 2001). For some (e.g. Carroll,
2001; Sato & Jacobs, 19929) this detected
input equals intake. However, detected input
can be used for comprehension only and this
need not involve any attention to the formal
aspects of the input, nor does there have to
be any link with long-term memory. To
resolve this, Faerch & Kasper (1980)
proposed a distinction between intake for
comprehension and intake for learning.
However, this seems to create unnecessary
confusion, as these two terms refer to
different representations and subsequent
uses of input. It may be more accurate to
distinguish between detected input, the
influence of which does not reach beyond
the moment, and intake, the influence of
which is potentially lasting.
Tomlin & Villa (1994) define detection as
‘the process by which particular exemplars
are registered in memory and therefore
could be made accessible to whatever the
key processes are of learning, such as
hypothesis formations and testing.’ (pp. 192-193). Assuming that ‘memory’ in the above
definition refers to ‘working memory’, I see
intake as a subset of this detected input,
which is accessible (and not merely could be
Sato & Jacobs specify this by suggesting that only
detected input that enters certain areas of the brain
affect language learning.
made accessible) to ‘whatever the key
processes are of learning, such as hypothesis
forming and testing’. Which exemplars or
what subset of the detected input becomes
intake depends on a large range of factors,
including (but perhaps not necessarily) the
amount of attention given to that subset of
the detected input. This in turn depends on
aspects of the input, such as its saliency, on
the learner and the state of her
interlanguage, and on other, non-cognitive
factors such (e.g. motivation).
Intake is thus detected input that goes
beyond what is held in working memory for
immediate recognition and comprehension.
It establishes a link with long-term memory.
In case of a subsequent encounter with the
particular linguistic phenomenon these links
and/or connections are strengthened. This
would explain frequency effects (N. Ellis,
2002a, 2002b). The degree to which this
happens, again, depends on a large range of
factors including the amount of attention
paid to the input, the strength of existing
connections, as well as learning aptitude etc.
As a working definition then, I propose the
following:
Intake is a subset of the detected input
(comprehended or not), held in short-term
memory, from which connections with long-term memory are potentially created or
strengthened.
The above is quite similar to how Tomlin &
Villa (1994, p .196) describe what precedes
potential acquisition by a learner. They
write:
1. The learner must discern the
presence of some element of
grammatical form.
2. The learner must discern that there is
a new or unusual character to the
event representation witnessed.
Applied Research in English: 1(2) 29
3. The learner must discern that there is
a relationship holding between these
two levels of grammatical form and
mental representation.
4. The learner must send those
observations off for further
processing (hypothesis formation
and testing).
The result of stages 1-3 is intake. However,
in an attempt to accommodate connectionist
views on acquisition, the working definition
includes the possibility of the strengthening
of existing knowledge as opposed to the
learner discerning only a ‘new or unusual
character to the event representation’.
Operationalising intake
Intake has been operationalised and
measured in a number of different ways.
Several authors operationalise intake as a
change in performance. Zobl (1985) simply
sees changes in a learner's rule output after
an exposure session as evidence of intake.
Similarly, Rosa & O’Neill (1999)
recommend performance measures when
attempting to measure intake such as recall
protocols, cloze tests, grammaticality
judgements, and rule formation, all to be
administered soon after the treatment or
exposure to the target input. They accede
that intake tasks that introduce production as
a factor may be inadequate in that ‘there is
some potential for interference from
inappropriately automatized production
routines’ (p. 286). In their own study they
made use of a multiple-choice recognition
task. Interestingly Rosa & O’Neill write in a
footnote to their 1999 article: ‘in order to
minimize the possibility of learners
performing the posttest on the basis of
memorized material, all of the test sentences
containing the target structure were different
than the sentences included in the treatment
task’ (p. 549). This appears to measure
learning, not intake.
Leow (1993, 1995) also used multiple-choice recognition tasks and gave
participants very limited time to complete
their tasks, which were administered
immediately after exposure.
To measure learners’ intake of linguistic
items in the input, a multiple-choice
recognition assessment task was carefully
designed to address only the linguistic item
that had been attended to by the learners in
the input. The three factors crucial to this
assessment task were a) the administration
of the assessment task immediately after
exposure to the input, b) the limited amount
of time learners had to complete the task (cf.
Chaudron, 1985), and c) a single, final
answer. (1993, p. 337).
The fact that only items that had been
attended to by the learners in the input were
included in the recognition test may have
raised participants’ awareness of those
items. In a later study (2001a) Leow also
made use of think-aloud protocols. In that
study he aimed to investigate the effects of
awareness on acquisition and recorded
correct verbal production of the target form.
Chaudron (1985) warns that production
measures need to be used with caution as
they could cause interference from previous
knowledge.
Shook (1994) made use of both production
tests (cloze test, sentence completion) and a
recognition test (multiple-choice sentence
completion) all of which were administered
immediately following the exposure. Shook
claims that ‘it is most improbable that the
data collection procedures used could reflect
anything except the immediacy of Process I
[the input-to-intake stage], and thus this
Towards a definition of intake 02
study does not reflect any acquisition of the
grammatical input’ (p. 85).
The above measures of intake have in
common that they attempt to probe beyond
what is held in short-term memory (and as
such aim to measure intake as opposed to
detected input). Likewise, they attempt to
avoid measuring (performance based on)
previous knowledge. A measure of intake
should also avoid measuring acquisition;
any measure that requires retention of
knowledge for extended periods of time is
an indication of knowledge in long-term
memory, not intake. Intake tests, then, can
only be administered after, but reasonably
soon after exposure to the second language.
This does not preclude task performance as
a measure of intake, however, exposure to
the target language (i.e. a listening or
reading passage) needs to be separated in
time from the activities participants are
asked to perform on that input.
Forced recognition tests, grammaticality
judgment tests, as well as measures
containing a degree of production, including
fill-in-the-gap, jumbled sentences tests
etcetera, are all potentially valid measures,
despite their individual drawbacks (provided
they are administered not too long after
exposure to the input). However, more
subtle measures such as the forced
recognition tests, are more likely to be
sensitive to intake in the early stages of the
learning process. Measures requiring
production need to be used with care. Free
production, and measures such as fill-in-the-gap without multiple-choice options are
more likely to measure integrated
knowledge, and can sometimes be
awareness-raising. Verbal protocols can help
to provide an additional measure of intake
and can act as a comparison with
performance measures.
Conclusion
Although there are great differences in the
various positions researchers have taken in
relation to the roles of input, output,
interaction, and intake in second language
acquisition, there is certainly also
considerable agreement. First, there appears
to be a consensus that language learning
cannot take place without input (although, as
mentioned above, input can affect the
development of aspects of the language not
contained in the input itself). There appears
to be a reasonable degree of consensus that
certain types of input are more favourable to
learning than others, and that input at the
very least has to be comprehensible. A range
of characteristics of the input determine
what is acquired (within the constraints of a
predetermined developmental order) and a
manipulation of those can affect learning.
There is evidence, at least for second
language learning, that both output and
interaction to some extent facilitate learning,
either through the provision of more
comprehensible input, or by drawing
attention to certain aspects of the input, or
the learner’s own output.
Of all the information available to the
learner, only some remains. This is where
the road forks. Connectionists (cf. Plunkett
& Elman 1997). argue that information that
is not detected is discarded, but all
information that is detected, affects the
learner’s developing system. It is unclear
what the role of intake in such a system is, if
indeed there is any; it is perhaps telling that
I was unable to find a clear account in the
literature of how such a process would fit
into a connectionist approach (but see N.
Ellis 2002a, 2002b, for a discussion of the
role of noticing in essentially frequency
drive implicit learning). Many second
language researchers working from an
information-processing perspective view
this differently: information may be detected
but that does not mean that it has any lasting
effect. It can be used for communication
purposes or be lost before it is stored.
Although this group of researchers holds
very different views on the exact process
affecting what information is incorporated
and what is not, they do appear to agree that
there is a level of processing that takes place
on the input, that determines whether
aspects of it may potentially be learned or
not. Some have labelled one or more of the
various stages in this processing as ‘intake’
(one exception appears to be Carroll, for
whom intake equals something more akin to
input, as defined by most other second
language researchers), others (e.g. Sharwood
Smith, 1993) appear to be talking about this
stage without using the same term. Leow
points out that the concept is useful to have
and apply: ‘the distinction between input
and intake has theoretical value because it
proposes that there is at least one
intermediate stage of input processing
through which the input second language
learners receive must pass before any or all
of it can become part of learners’ developing
linguistic system’ (1993, p.334). Although
some refer to the actual processing as intake,
others to its product, and yet others to both,
investigating this processing or its product is
crucial in understanding how input becomes
part of a learner’s system.
