Theoretical Research and Second Language Acquisition Theory

Stephen Krashen

I will attempt to summarize the theoretical research that has gone on in the last
decade by stating five hypotheses about second language acquisition. Following
this, we will look at the implications of these five hypotheses for other theoretical
work in second language acquisition, and then see what this theory has to say, or
predict, about second and foreign language teaching. There will be discussion of
the gaps in the research literature, places where more support or testing of
hypotheses is called for.

Five Hypotheses About Second Language Acquisition

I. The Acquisition/Learning hypothesis

This hypothesis states that we have two different and independent ways of
developing ability in second languages. We can acquire and we can learn.
Acquisition is defined as the process children use to acquire first language. It is
subconscious in two different ways. First, the process is subconscious; we are
usually not aware that we are acquiring while we are acquiring. Rather, we have
the impression that we are doing something else, such as having an interesting
conversation or reading an interesting book. We are also not always aware that
we have acquired something; the knowledge itself is subconscious. This is
illustrated by the fact that native speakers do not always "know" (consciously) the
rules of their language. It is the purpose of current formal linguistics
(transformational-generative grammar) to describe these subconscious intuitions
about language, the native speaker's "tacit" knowledge. Learning is conscious, or
explicit knowledge about language. Learning is developed, it is thought, by
explicit, or formal instruction, and is thought to be aided by the practice of error
correction. Error correction, supposedly, helps the learner come to the correct
mental representation of a rule. In everyday language, acquisition is "picking up"
a language, while learning is "grammar, " or "rules." It is difficult to test the
acquisition/learning hypothesis directly, but it plays an important role in all the
other hypotheses, so evidence in favor of them serves as evidence supporting
the acquisition/learning hypothesis. The issue of error correction has been
investigated directly, and we will return to the possibilities in a later section, once
the relationship between acquisition and learning is stated.

II. The Natural Order hypothesis

This hypothesis states that we acquire (not learn) grammatical structures in a

predictable order, that is, certain structures tend to be acquired early, and others
tend to be acquired late. This order can be altered by first language influence
(see below), but cannot be altered by the effects of instruction. Much of the
research supporting the Natural Order Hypothesis comes from English. Brown,
for example, reported that children acquiring English as a first language tend to
acquire certain morphemes early (plural /s/ and progressive ing) and others late
(third person singular ending /s/ and the possessive `s) (Brown, 4). Similar claims
were made for child second language acquisition initially by Dulay and Burt and
subsequently by others (Dulay and Burt, 7), and for adult second language
acquisition, beginning with Bailey, Madden, and Krashen (14). The second
language order is not identical to the first language order, but there are some
similarities (Krashen, 12). Related to the Natural Order hypothesis is the
phenomenon of transitional forms. It appears to be the case that second
language learners pass through predictable stages on their way to acquiring the
correct form for many, if not all, structures. For example, in the acquisition of
English negation, both first and second language acquirers often pass through a
stage in which the negative marker is placed outside the sentence, as in 1) No
like it now (Ravem, 21). A second stage consists or placing tile negative marker
between the subject and verb, as in 2) This no have calendar (Schumann, 23).
Finally in the third stage, which not all acquirers reach, the correct form is
acquired. There are many areas of research and unanswered questions
regarding the natural order hypothesis. A major one, of course, is why an order
exists. What determines the order? Why are some structures early and others
late? There has been considerable research and speculation on this topic. See,
for example, Clark and Clark, 6; Hatch and Wagner-Gough, 10; Dulay and Burt,
8; and Larsen-Freeman, 16. While this is a theoretically interesting question, it
may not be crucial to second language (Caching practice. We need to know that
an order exists to understand why students make the errors they do and to alter
our expectations accordingly, but it is not yet clear that we need to know the
determinants of the order. Another possible priority is to expand our knowledge
of what the order is for as many languages and as many structures as possible.
We have enough evidence to state the hypothesis, but our data is limited to a few
well studied languages and a handful of structures. Such additional confirming
evidence is desirable in order to support the hypothesis, or to disconfirm it, but it
may not be crucial for language teachers. As I will point out later, in discussing
the Input hypothesis (hypothesis IV), the major implication of the Natural Order
hypothesis is not that we teach along tile Natural Order, beginning with those
structures shown to be acquired early. In fact, the available evidence strongly
suggests that we should not use a grammatical syllabus at all, no matter what it
is based on! A very important question is when the order appears and when it
does not, that is, what conditions are necessary to show that a Natural Order is
indeed present. This is discussed as part of the next hypothesis.

III. The Monitor hypothesis

The acquisition/learning hypothesis stated that two separate processes exist for
developing ability in a second language. The Monitor hypothesis describes their
interrelationship and how each is utilized by the second language performer. The
Monitor hypothesis claims that acquisition, not learning, is responsible for our
fluency in second language performance, for our ability to use second languages
easily and comfortably. Conscious learning does not contribute to fluency, but
has only one function: it can be used as an editor, or Monitor. We use conscious
learning to make corrections, to change the form of the output of the acquired
system before we write or speak, or sometimes after (self-correction). This
hypothesis regulates learning to a less than starring role in second language
performance. Research over the last few years suggests strongly that the use of
conscious learning is very limited. Not only is it restricted to the Monitor function,
but it is not easy to use the Monitor effectively. I have posited that three
conditions need to be met in order to use the Monitor, conditions that are
necessary, but not sufficient:

1) Time: In normal conversation, there is rarely enough time to consult and utilize
conscious rules, although some performers claim to be able to do this very well.
2) Focus on form: just having time is not enough. Even when acquirers have
plenty of lime, they do not always think about grammatical correctness. Dulay
and Burt have pointed out that a second condition is necessary: the acquirer
must be focused on form, or correctness. 3) Know the rule. This is a formidable
condition, considering how incomplete our knowledge of formal grammar is.
Linguists concede that they have described only fragments of natural languages,
and teachers and students have access only to a portion of these descriptions
Research support for this hypothesis comes originally from studies of the natural
order phenomenon. It has been found that second language performers show
"natural orders" (in this case, a difficulty order similar to the longitudinal order of
acquisition) in what can be considered "Monitor-free" situations, in which use of
the conscious grammar is precluded, when there is little time or focusing on form.
When the conditions for Monitor use are met, the natural order does not appear,
which represents the intrusion of the conscious grammar. It is interesting to note
just how the natural order is disturbed by the Monitor. What we see is a rise in
rank of late acquired items that are "easy" to learn. In English as a second
language, for example, accuracy on items such as the third person singular
morpheme will rise in rank and accuracy when students are given time and are
focused on form. I would expect that similar increases would occur on similar
items in other languages, items that are late acquired but are "learnable" and
taught early. The simple de + le = du rule in French is a good example, as are
case endings in German. Our students are often able to perform well on these
items on pencil and paper grammar tests, but are much less accurate in free
conversation. Another very interesting point is our current hypothesis that for
most students anything short of a discrete-point grammar test will not invoke(' the
conscious grammar to any great degree. We see only mild, if any, Monitor use in
composition, and in other situations in which students are ask d to "be careful."
Only a grammar test seems to meet all three conditions. A great many questions
can be asked with respect to this hypothesis. Most obviously, since most of the
data in support of this hypothesis comes from English as a second language, we
need to confirm its validity for foreign language situations. This work has been
begun by some students at the University of Southern California. We also need
to gather more data on when the Monitor is used and when it is not: what
situations bring out the grammar? Is it true that extensive grammar use is only
possible on grammar tests? The question of individual variation is another
important one. I have suggested that we find basically three types of performers:
Monitor over-users, who Monitor all the time, to the detriment of their fluency;
Monitor under-users, those who fail to consult the conscious grammar at all; and
optional users, performers who use the grammar when they can, but only if such
use does not interfere with communication (Krashen, 9). Optimal users typically
are able to use grammar rules in writing; for example, but may not always use
them in speaking. This individual variation schema was developed from a
consideration of case histories, and has found some support in other studies (So,
24; Stafford and Coritt, 25). It needs not only further confirmation, but
investigation of what sorts of programs might be optimal for different types of
learners.

IV. The Input hypothesis

This hypothesis deals with the important question of how we acquire. It consists
of three interrelated parts:

1) We acquire by understanding input containing structures that. are a bit beyond

our current competence. In terms of the Natural Order hypothesis we move from
our current level i to the next level i + 1 by understanding input containing i + 1.
We acquire, the hypothesis states, by going for meaning, by focusing on what is
said rather than how it is said. We are aided in this process by extralinguistic
context, and our knowledge of the world. We do not acquire by first learning
about the structure of the language. We try to understand the message, and
structure is thereby acquired. 2) Speaking "emerges." We do not teach speaking
but give acquirers comprehensible input. Speech will come on its own, when the
acquirer feels ready. Early speech is not grammatically accurate, but accuracy
develops as the acquirer obtains more comprehensible input. 3) The best input is
not grammatically sequenced. Rather, if the acquirer understands the input
presented, and enough of it is made available, i + 1, the structures the acquirer
needs for further development, will be automatically provided. Thus, the best
input is not grammatically sequenced. Not only is it not necessary, but it may be
harmful, when the goal is acquisition (this is not the case when the goal is
conscious learning). The acquirers will receive comprehensible input containing
structures just beyond them if they are in situations involving genuine
communication, and these structures will be constantly provided and
automatically reviewed. They need not worry about missing a class and thereby
missing the past tense forever (or at least until next year). With natural,
comprehensible input the hypothesis predicts that they will hear the past tense
again and again. In other words, part (3) of the Input hypothesis claims that input
for acquisition need not focus only on i + 1, it just needs to contain it. f + I will be
supplied, and naturally reviewed, when the acquirers obtain enough
comprehensive input. Evidence for the Input hypothesis is given in some detail in
other publications (Krashen, 12, 13), but it is useful to briefly mention two.
phenomena in second language acquisition that relate to and are consistent with
the Input hypothesis. The first is the presence of the silent period, a period of
time before the acquirer actually starts to speak. The silent period is very
noticeable in child second language acquisition; young children in a new country
facing a new language may say nothing (except for some memorized sentences
and phrases) for several months. According to the Input hypothesis, this is a time
during which they are building up competence via input, by listening and
understanding. When they are ready, they start to talk. We generally do not allow
our students to have a silent period. We insist on production, and accurate
production, right away. When adults have to talk too early, before they have had
a chance to acquire much of the second language, they have only one choice,
and that is to fall back on their first language, an idea first proposed by Newmark
(18). They "think" in the first language, supply lexical items from the second
language, and use the conscious grammar as best they can to make repairs.
According to this view, first language interference is not interference at all, but is
the result of using old knowledge. Its cure is acquisition, or more comprehensible
input. A great deal of research related to this central hypothesis needs to be
done. Most important, we can ask whether acquirers who have had a. chance to
get more comprehensible input actually do better than those who do not. Also,
does comprehensible input with imperfections, that is, the speech of other
acquirers, help or hurt? The "first language interference" hypothesis is also quite
testable (Krashen , 12). The input hypothesis is perhaps the most crucial of all,
since if it is correct, it will revolutionize our methodology in second language
teaching (see Asher, 1; Nord, 19; Postovsky, 20; Terrell, 29; and Winitz and
Reed, 31).

V. The Affective Filter hypothesis

This hypothesis deals with the role of affective variables. Briefly, the research
literature in second language acquisition tells us that the following affective
variables are related to success in second language acquisition):

I) Anxiety: The lower the level of anxiety, the better the language acquisition. In
Stevick's terms, the student should be "off the defensive" (26). 2) Motivation:
Certainly, higher motivation predicts better second language acquisition. Certain
types of motivation are more effective in certain situations, "integrative"
motivation helping most in long-term, and "luxury" second language acquisition
and "instrumental" helping for short-term acquisition where there is a practical
need for the language (Gardner and Lambert, 9). 3) Self-confidence: The
acquirer with more self-esteem and self-confidence tends to do better in second
language acquisition I have hypothesized that these affective factors relate more
directly to subconscious language acquisition than to conscious learning) since
we see stronger relationships between these affective variables and attainment
in second languages when communicative-type tests are used, and when we test
students who have had a chance to acquire the language (and not just learn it).
Dulay and Burt (8) have made this relationship more explicit and clear by positing
the presence of an "Affective Filter." According to the Affective Filter hypothesis,
acquirers in a less than optimal affective state will have a filter, or mental block,
preventing them from utilizing input fully for further language acquisition. If they
are anxious, on the defensive, or not motivated, they may understand the input,
but the input will not enter the "language acquisition device." According to this
hypothesis, given performers with identical comprehensible input, there may still
be variation in rate of acquisition and ultimate attainment in acquisition. The one
with the "lower filter" will go faster and farther. As with the other hypotheses,
there is much work left to do. The generalizations concerning affective variables
and subconscious acquisition were made post hoc, on the basis of previously
reported studies. This relationship needs 10 be experimentally verified. (I have
similarly claimed that what we call aptitude in second language acquisition is
closely tied to conscious learning, also on the basis of a post hoc analysis of
previously done experiments and reports.) Also, the Filter hypothesis itself is
indirectly testable. Will we find that acquirers with similar input progress at
different rates, and are these differences traceable to affective factors? What
other affective factors contribute to the filter, and will we find different factors
prevalent in different situations? (Krashen, 12; Schumann, 22)

The Causative Variable in Second Language Acquisition

We can summarize the five hypotheses with a single claim: People acquire
second languages when they obtain comprehensible input, and when their
Affective Filters are low enough to allow the input "in." Thus, comprehensible
input is the true and only causative variable in second language acquisition. This
predicts that other variables posited to be related to success in second language
acquisition are actually intervening variables for comprehensive input. This is
quite testable, arid, of course, quite exciting. The first testable component of this
hypothesis is that instruction itself is an intervening variable, that it helps second
language acquisition only when it provides comprehensible input. The little data
that is available supports this, but it is truly shocking how little data is available. In
short, the research done to date suggests that instruction helps only when it is
the main source of comprehensible input, when the acquirer has no other source.
It helps, for example, for beginners, even those in second language situations
(who are not yet competent enough to understand the language outside the
class), but does not seem to help intermediate students who have another rich
source of comprehensible input. This makes sense. It suggests that the goal of
instruction is not to produce advanced native-like speakers but to bring students
to the point where they can begin to take advantage of the natural input available
to them outside of class. Similarly, studies probing the effect of exposure and
reported use of second languages come to similar conclusions. They predict
attainment in second language when they genuinely reflect reause of the
language, interaction, comprehensible input. Again only a handful of studies
speak to this question, and my conclusions are again post hoc. This is thus a
wide-open area for research. It can even be hypothesized that age is an
intervening variable. The literature tells us that older acquirers are faster
acquirers, for early stages. Younger acquirers, however, are superior in terms of
ultimate attainment (Krashen, Long, and Scarcella, 14). The older acquirers'
advantage in rate may be due to their ability to obtain comprehensible input
(Scarcella and Higa, 32), while the younger acquirers' superiority in eventual
attainment has been hypothesized to be due to the strengthening of the affective
filter. As Scarcella and Higa have demonstrated, these are very testable
hypotheses.

Implications for Teaching

The theory makes definite predictions for second and foreign language teaching,
all of which are testable, and some of which have been tested to some extent in
applied research experiments. Briefly, the theory predicts that successful second
language teaching programs will have these characteristics:

1) They will supply a great deal of comprehensible input that is interesting and
relevant to the students. The goal of this input will not be to provide practice on
specific points of grammar, but will be to transmit messages of interest. 2) They
will not force students to speak before they are ready and will be tolerant of
errors in early speech. 3) They will put grammar in its proper place. Some adults
(and very few children) are able to use conscious rules to increase the
grammatical accuracy of their output, and even for these people, very strict
conditions need to be met before the conscious grammar can be applied. Several
methods come close to meeting these requirements, including Asher's Total
Physical Response method, Terrell's Natural Approach, Lozanov's
Suggestopedia and recent materials developed by Harris Winitz. In addition,
several non methods meet these requirements nicely. Successful conversation
with a speaker of the language you are trying to acquire may be the best lesson
of all, as long as the speaker succeeds in making the speech comprehensible to
you. According to the theory, acquirers profit not from what they say, but from
what the native speaker says. Output thus makes an indirect but powerful
contribution to acquisition by inviting comprehensible input! Also, pleasure
reading has the potential for supplying comprehensible input helpful for
acquisition. Of course, the claim that a method is successful is quite testable, as
we shall see in the next section. In addition, the hypothesis that
conversationalists and pleasure readers acquire (but not necessarily learn) more
is also testable, but has not, to my knowledge, been tested directly in theoretical
research. A prediction that the theory makes that has been extensively tested is
that subject matter teaching can help language acquisition. The research is quite
extensive for children, coming from the very well studied immersion programs
(Lambert and Tucker, 15), but the possibility that adults may also benefit remains
a theoretical prediction.

Applied Research
We will focus on just one aspect of applied research here, that of method
comparison studies, the attempt to determine which of two methods is better by
direct comparison. While this sort of research looks to be straight-forward, it is
loaded with difficulties and confounds. Nevertheless, a substantial number of
studies have been done, and, since they involve fairly large numbers of students
in many different settings, and they give remarkably consistent results, they are
worth examining. The studies can be divided historically into two time periods.
First came what can be termed "traditional" method comparison studies, studies
comparing the more common methods, such as audiolingual, cognitive-code, and
grammar-translation. We can summarize the results of these studies (Chastain,
5; Mueller, 17; van Elek and Oskarsson, 30):

1) There is little difference between audiolingual-type teaching and grammar-

based approaches (grammar-translation and cognitive-code style) for
adolescents. 2) For adults, cognitive type methods are a little better. Cognitive
students outperform audiolingual students, but the differences are quite small.
This result had two different kinds of reactions in the field. Stevick noted the
implicit contradiction, asking how methods based on totally different theories of
language acquisition could produce such similar results (Stevick, 26). Many
methodologists and teachers, however, simply assumed that the answer was to
be eclectic, to choose parts of each system in the belief that the answer must be
somewhere in the middle. As fair-minded as this sounds, it often resulted in
teachers choosing the worst from each, the parts least likely to cause acquisition:
drill from audiolingual and extensive grammar explanation from grammar-
translation and cognitive code!

There is an explanation. Neither approach provided much in the way of

comprehensible input, and neither met the requirements outlined above for
successful language teaching methods. This predicts that methods that do not
meet these requirements should do much better, and there is some indication
that they do. Asher has provided the profession with an extensive series of
studies demonstrating the efficacy of Total Physical Response teaching, showing
that TPR is not just a little better, but is much better than audiolingual teaching. In
one study, in fact, his TPR students acquired five times faster than controls
(Asher, 2). Swaffar and Woodruff also confirm that methods focusing on
comprehensible input and that keep grammar in its place do much better than the
older alternatives (Swaffar and Woodruff 27). These kinds of results are very
encouraging. They support the theory, even though they were done without this
particular theory in mind, and point to exciting new directions. What remains to
be done? A great deal. We need, despite their flaws, more comparisons, more
program development research, using the model of Swaffar and Woodruff. This
includes more direct tests of newer methods. Also useful would be case histories
of students in these methods, a means of getting at their reactions and problems.
Experimentation I would be particularly interested in involves a new direction for
the language laboratory. Up to now, the lab has been a place where students can
come to exercise their output and have it corrected. A far easier and
technologically simpler use for the lab is as a supplementary source of
comprehensible input. Here are some possibilities: taped Stories, with pictures to
aid comprehension and add to the enjoyment (as currently being developed by
Harris Winitz), radio programs, commercials on tape, a "cheap" library, for casual
pleasure reading, filled with books that the lab is not afraid of losing. And perhaps
most important, a native speaker, willing to chat with whomever comes in! The
theory predicts that such a lab would do great things for language acquisition,
and this prediction is certainly testable. The question of materials can also be
considered an (applied) research priority. The predictions made for methods also
can be applied to materials. Simply, the best materials will be those that supply
comprehensible input, that do not force overuse of grammar, and that keep the
student off the defensive. This predicts that specialized readers, readers on
topics that truly interest students, aids to comprehension in the classroom (e.g.,
visuals), and materials that aid Monitor use without creating over-users will
succeed. This is quite testable, and calls for a new tradition of field testing. In the
past, for example, we assumed our readers were acceptable if they contained
certain structures and avoided others. The theory predicts now that successful
readers are simply those that arc comprehensible and interesting. If these
requirements are met, the structural requirements (that i+1 be present) will
automatically be met.

Conclusion

We have touched on a few places where more research needs to be done. In all
cases, the gaps I have identified are related to current hypotheses. Of course,
new hypotheses will be developed, but this is unpredictable. What is predictable
is that if we continue our thinking and research, research designed to lest old
hypotheses, new ideas will come, deeper generalizations to account for
anomalies. What is most important is the realization that both theoretical and
applied research can contribute to progress in foreign language education, and
that accepting research as a determinant of practice does not necessarily mean
rejecting teachers' experiences and intuitions. But language teaching should not
be based entirely on fashion, and matters of methodology should not be settled
by committee and vote. Our top priority should be to form a real partnership
between theoretical researcher, applied researcher, and language teacher, so we
can work together toward a common goal.

Krashen, Stephen D. (1983). Practical applications of research. Psycholinguistic

Research ACTFL Yearbook. Lincolnwood, Illinois: National Textbook. P. 54-65.