SSLA, 19, 195–221. Printed in the United States of America.

BEYOND EXPLICIT
RULE LEARNING

Automatizing Second
Language Morphosyntax

Robert M. DeKeyser
University of Pittsburgh

This study is a fine-grained analysis of extensive empirical data on the

automatization of explicitly learned rules of morphosyntax in a second
language. Sixty-one subjects were taught four morphosyntactic rules
and 32 vocabulary items in an artificial language. After they had
reached criterion on a set of metalinguistic tests of grammar and vocab-
ulary, they engaged in systematic, computer-controlled comprehen-
sion and production practice for 8 weeks. Comprehension practice
consisted of choosing between pictures displayed on the computer
screen to match a sentence; production practice consisted of typing
the correct sentence corresponding to a picture. All subjects were
taught the same rules and then practiced them, and all subjects had
the same amount of comprehension and production practice, but
which rules were practiced in comprehension and which in production
varied between groups. Results show that the learning of morphosyn-

This study was supported by U.S. Department of Education grant P017A50064. I thank Chris Connors and
Jim Rankin for their expert programming and Ben and Philippa Benson-Xu, Jeanine Carlock, Lorien Clemens,
Jing-Fu Fan, Kiduk Kim, Jannine Markizon, Jeri and Scott Misler, Yong-Ping Mou, David Novinksi, John Smith,
David Steinitz, and Zander Teller for their superb acting performances. Thanks are also due to the experi-
´
menters Keiko Iijima, Jannine Markizon, Andre Mather, Don Peckham, Leonore Rodrigues, Michelle Sadlier,
Clay Taylor, Paul Toth, Eugenia Wan, and Bill Williams. I gratefully acknowledge the advice and encourage-
ment from Carol Baker, Alan Juffs, Donald McBurney, Daniel Everett, Robert Henderson, David Malicki, Chris-
tina Paulston, Charles Perfetti, and Richmond and Sarah Thomason and the financial support of the Univer-
sity of Pittsburgh Central Research Development Fund for a pilot study as well as that of the Linguistics
Department for various expenses. Kathleen Bardovi-Harlig, Nick Ellis, Jan Hulstijn, Donald Peckham, Peter
Robinson, and Lynne Yang provided helpful comments on an earlier draft of this paper. Finally, I wish to
express my utmost gratitude to Jim Rankin for computational feats far beyond the call of duty.
Address correspondence to Robert DeKeyser, Department of Linguistics, University of Pittsburgh, Pitts-
burgh, PA 15260; e-mail: rdk1@vms.cis.pitt.edu.

 1997 Cambridge University Press 0272-2631/97 $7.50 + .10 195

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 196 Robert M. DeKeyser

tactic rules is highly skill-specific and that these skills develop very
gradually over time, following the same power function learning curve
as the acquisition of other cognitive skills. These results are consistent
with current skill acquisition theory.

In recent years a number of empirical studies have been forthcoming that show the
superiority of explicit over implicit processes in second language learning, at least
for certain rules and certain learners (e.g., Alanen, 1995; de Graaff, this issue; De-
Keyser, 1995; N. Ellis, 1993; House, 1996; Master, 1994; Robinson, 1996). A weakness
of these studies is that they stop short of demonstrating that the explicitly learned
knowledge can be used automatically, that is, fast, with a low error rate, and with
little interference from other cognitive tasks.
Whereas a substantial literature on automaticity and automatization exists in
cognitive psychology, and the automatization of rules has long been recognized as
a central problem in applied linguistics by those who believe rules are useful at all,
hardly any fine-grained experimental literature is available on the process of rule
automatization in the course of second language learning. Robinson and Ha (1993)
provided evidence for a certain degree of automaticity in grammaticality judgments
for sentences in a pseudo-artificial language, but they did not investigate the develop-
ment of automaticity over time. The present study aims to contribute to filling the
gap in empirical research on automatization by presenting a fine-grained longitudinal
account of the acquisition of the rules of an artificial language (a miniature linguistic
system) under different conditions of practice.

AUTOMATICITY
Automaticity has received considerable attention in the cognitive literature of the
last 20 years. Despite continuing disagreements about the exact nature of capacity
limitations and the role of attention (cf., e.g., Pashler, 1994) about the relationship
between automaticity and implicit knowledge (Seger, 1994), and about how automa-
ticity is brought about (with competing models as a result, such as Anderson’s
[1993] ACT-R (Adaptive Control of Thought—Rational) model; Schneider and Detweil-
er’s [1988] connectionist-control model, and Logan’s instance learning model [Logan,
1988, 1992; Logan & Etherton, 1994; Logan, Taylor, & Etherton, 1996]), substantial
progress has been made. It is now clear that there is a continuum of automaticity
rather than an automatic-controlled dichotomy. Researchers widely agree that phe-
nomena such as gradual drop-offs in reaction time and error rates, and diminished
interference from and with simultaneous tasks (cf., e.g., Logan & Etherton, 1994),
are reliable criteria of automaticity. Some of the experiments showing these effects
have become classics (e.g., Schneider & Shiffrin, 1977).
The most widely accepted theory on how automaticity is brought about is Ander-
son’s ACT1 model of the human cognitive architecture (cf., esp. Anderson, 1987,
1992, 1993). According to this model, knowledge typically starts out as explicit

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 197

(declarative) information, “knowledge that,” which is turned into specialized proce-

dural rules, and “knowledge how,” for very specific behaviors through analogy with
a series of examples and with the help of very general behavioral rules. It is then
fine-tuned over time as a function of cost-effectiveness (probability of being correct
and cost in terms of mental resources). The result of this last process is a gradual
drop-off in reaction time and error rate. This drop-off is generally held to take the
form of a power function (Newell & Rosenbloom, 1981; Anderson, 1993), that is, a
function of the form y = x n, reflecting that the amount of improvement (the change
in y, be it reaction time or error rate) decreases as a function of the amount of
practice x, the power n being negative.
The ACT model has been tested very extensively and successfully in empirical
studies on the learning of algebra, geometry, and computer programming. Anderson’s
(1983) application of ACT* to first language learning, however, has remained contro-
versial, because of its insistence (subsequently relaxed; see Anderson & Fincham,
1994) that all knowledge starts out in declarative form.
In what follows, I will draw on rule-based theories of automaticity in general and
on Anderson’s ACT-R model in particular. For an empirical study drawing on memory-
based theories of automaticity in general, and Logan’s instance theory in particular,
see Robinson (this issue).

AUTOMATIZATION OF SECOND LANGUAGE RULES

Automatization is the process that leads to automaticity. In the field of applied
linguistics, the notion of automatization was used loosely by proponents of audiolin-
gualism (e.g., Rivers, 1964) and cognitive code (e.g., Chastain, 1971) and elaborated
didactically by early communicative methodologists (Paulston & Bruder, 1976).
These authors claimed that automaticity was to be achieved by overlearning of
stimulus–reaction chains (Rivers), grammar teaching followed by extensive drill
practice (Chastain), and the careful sequencing of mechanical, meaningful, and
communicative drills (Paulston & Bruder). At that time, the computational tools for
providing strictly controlled treatments and conducting fine-grained data collection
and analysis were not available to applied linguists, and no empirical studies were
carried out on the exact nature of automatization.
Since the late 1970s, the notion of automatization in second language learning
has been questioned by numerous authors. Some, most notably Krashen (e.g., 1982,
1985, 1994), question the usefulness of any rule learning or focus on form and argue
that the spontaneous use of a rule after it was taught and practiced by no means
results from automatizing that rule but is the result of a completely independent,
implicit process of acquisition, resulting from exposure to large amounts of compre-
hensible, meaningful input. Others, whereas recognizing that focus on form is useful,
still see it as a way of improving the implicit intake of structures from the input and
therefore consider systematic and form-focused practice in production to be of no
use to acquisition (R. Ellis, 1993; VanPatten & Cadierno, 1993a, 1993b).
A variety of recent studies, in both the classroom (e.g., Scott, 1989; Spada &
Lightbown, 1993; White, Spada, Lightbown, & Ranta, 1991) and the laboratory (e.g.,

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 198 Robert M. DeKeyser

Alanen, 1995; Doughty, 1991; N. Ellis, 1993), have provided evidence that some sort
of focus on form is useful, and some theoreticians have endorsed the notion of
automatizing explicitly learned second language rules through production practice,
such as Gatbonton and Segalowitz (1988), Johnson (1996), O’Malley and Chamot
(1990), and Schmidt (1992). Johnson (1996), in particular, discusses the conditions
for automatization at length, defining “the skill of automatization” (automaticity in
our terminology) as “the ability to get things right when no attention is available
for getting them right” (p. 137). He argues that learners should be given a series of
tasks at the level of ra − 1 (required attention minus one), that is, where the amount
of attention that can be paid to the forms in question is slightly less than comfortable
for the learner at each stage. This has traditionally been accomplished by increasing
focus on meaning, but Johnson (1996) insists that other kinds of form defocus, such
as time pressure and competing tasks are equally useful distraction techniques, and
hold promise in the computer age (p. 176).
Virtually no empirical evidence for the usefulness of practice in production exists
yet, and certainly no precise documentation of the process of automatization of
rules through such practice. The few studies in the domain of second language
learning that have built on Schneider and Shiffrin’s (1977) methodology have been
limited to vocabulary learning (e.g., Favreau & Segalowitz, 1983), reading (e.g.,
McLeod & McLaughlin, 1986), or word recognition in reading (e.g., Segalowitz &
Segalowitz, 1993); see McLaughlin (1987) for a review. On the other hand, the litera-
ture on temporal variables (e.g., Möhle, 1984; Raupach, 1987; Towell, Hawkins, &
Bazergui, 1996; for a critical review, cf. Griffiths, 1991) provides detailed evidence
of speed-up and chunking in advanced learners, but its failure to trace the learning
of specific rules prevents it from being convincing evidence for automatization
processes. However, several studies, in particular those by Towell et al. (1996) and
Raupach (1987), are interesting because of their wealth of fine-grained performance
data from different stages of the learning process, even if their database is small,
and the interpretation in terms of skill acquisition models is highly speculative.
In summary, the issue of automatization of grammar rules through practice in
production remains largely uninvestigated because the applied linguistics research
methodology of the 1960s and 1970s was insufficiently developed to address the
complexities of automatization in language learning and because second language
researchers in the 1980s were reluctant to conduct empirical studies on it, due to
the prevailing doubts about the usefulness of learning and practicing rules. But the
time is now ripe to carry out such research, given the renewed confidence in teaching
and learning at least some kinds of rules (see de Graaff, this issue; DeKeyser, in
press; and Hulstijn & de Graaff, 1994, for more discussion of the explicit learning of
different kinds of rules) and the experimental and analytical tools now at our disposal
for the study of automatization.

THE SPECIFICITY OF PRODUCTION RULES

Related to the controversy over whether automatization exists in second language
learning, or whether second languages are learned in a completely different way,
due to the autonomous nature of the linguistic module of the mind, is the issue of

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 199

the specificity of knowledge. Those who claim that linguistic competence in a second
language is acquired implicitly also argue that the same knowledge, once acquired,
is drawn on for both production and comprehension, making practice in production
unnecessary (Krashen, 1985; VanPatten & Cadierno, 1993a, 1993b). Those who argue
that all cognitive skills are acquired through the same mechanisms of proceduraliza-
tion of declarative knowledge see proceduralized knowledge (production rules)
as highly specific, so that comprehension and production, for instance, draw on
completely different sets of rules and therefore require separate practice (see, e.g.,
Anderson, 1993, pp. 37–38, where this is illustrated in detail for reading and writing
computer code).
Until recently, the issue of skill specificity had received virtually no attention in
empirical second language acquisition research. Recent experiments conducted by
VanPatten and his associates suggest that practice in production does not contribute
anything to acquisition beyond form-focused teaching followed by comprehension
practice (Cadierno, 1995; VanPatten & Cadierno, 1993a, 1993b), or at least that
comprehension practice alone is enough to bring about significant improvement in
both comprehension and production skills (VanPatten & Sanz, 1995). These experi-
ments dealt with the form and position of object clitics (VanPatten & Cadierno,
1993a, 1993b; VanPatten & Sanz, 1995) and the morphology of the past tense (Cadie-
rno, 1995) in Spanish. DeKeyser and Sokalski (1996), however, have challenged the
generalizability of these findings to other elements of morphosyntax (such as the
conditional) and have argued that the effect of input and output practice is largely
skill-specific.
In all these studies, the period of training was short, and the outcome measures
were largely paper-and-pencil tests; even in the case where oral testing was con-
ducted (VanPatten & Sanz, 1995), it allowed for a certain amount of reflection. No
research so far has investigated the longitudinal effect of systematic practice on
real-time performance in comprehension and production. Therefore, it was decided
to conduct a longitudinal study on the differential effect of comprehension practice
and production practice in the highly controlled environment of the computer-
assisted learning of a miniature linguistic system.

METHOD

Hypotheses
1. Once second language grammar rules have been explicitly assimilated, practice will
lead to gradual automatization, as measured by reduced reaction time, reduced error
rate, and decreased interference from simultaneous tasks (cf., e.g., Johnson, 1996;
Schmidt, 1992).
2. The reduction in reaction time and error rate will follow a power function for individual
grammar rules (cf., e.g., Newell & Rosenbloom, 1981).
3a. The practice effect will be skill-specific in the sense that students with only comprehen-
sion practice will improve more in comprehension skills than students with only
production practice and vice versa.
3b. Students with an equal amount of practice in both skills will perform at an intermediate
level in both skills (cf., e.g., Anderson, 1993).

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 200 Robert M. DeKeyser

The Language

As will become clear later, it was important for this study to have rules that were
completely independent of each other, that is, to avoid rules that cooccur, such as
plural verb forms cooccurring with plural subject nouns, for instance. It was also
necessary to have grammatical categories that correspond to characteristics of
people and activities that can easily be presented visually. Furthermore, rules needed
to apply without exception and present learners with binary options. Therefore, it
was necessary to use an especially created miniature linguistic system (but with
the characteristics of a natural language, unlike the artificial grammars generating
letter strings used by cognitive psychologists). Such a miniature linguistic system
offers the additional advantages that no students have any previous knowledge of the
language or related languages and that no differences in integrative or instrumental
motivation are to be expected. (See Hulstijn [this issue] for other studies of second
language acquisition that have used (semi-)artificial languages.)
Autopractan, especially created for this study, is an agglutinative language with
flexible word order. The noun receives number and case marking; the verb is
marked for gender and instrumentality. The singular is marked with the suffix -li,
the plural with -to. Nominative case is marked with the suffix -fu, accusative with
-ka. Masculine and feminine are expressed by the suffixes -ne and -pi, respectively.
The verb carries the suffix -ra when the action referred to is carried out by means
of some instrument or tool, and the suffix -so when no tool is involved. Instrumental
case is expressed by the suffix -te. Subject noun phrases can be dropped; Autoprac-
tan is a null-subject language. These rules comprise the entire morphosyntax of
Autopractan.
The Autopractan lexicon contains 32 words, 16 nouns and 16 verbs. None of the
lexemes resemble English words or words in the most commonly taught languages.
See Appendix A for a list of the vocabulary items and their English equivalent; see
Appendix B for some example sentences.

Subjects

The learners in this experiment were 61 paid volunteers, divided in three groups of
about 20 each. Thirty-two participants were female and 29 were male. Most were
undergraduates; a few were graduate students or staff members. The amount of
payment depended on performance: Subjects were told at the beginning of the study
that, at the end of the study and based on their performance throughout, they would
receive $8 per hour if they scored in the top third of the group, $7 in the middle
third, and $6 in the bottom third. As the rules to be taught involved gender marking
and case marking, experience with Romance languages, German, Russian, Latin,
or other commonly taught languages could have affected the results; therefore,
participants in the three groups were matched on their experience with these lan-
guages.

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 201

Design
The study had mostly a within-subjects design. Hypotheses 1 and 2 were tested by
analyzing data of all the subjects over time (separately for comprehension and
production, for reaction time and error rate, and for a single-task condition and a
dual-task condition, the latter involving mental arithmetic during the language task).
Hypothesis 3a was tested by comparing the same subjects on different (sets of) tasks.
Hypothesis 3b, however, was tested by means of a between-subjects comparison.

Materials
A set of 400 color pictures were taken, corresponding to sentences in Autopractan;
16 were used for the initial vocabulary presentation, 256 for the practice and testing
sessions, and 128 for the final testing session. All the pictures were digitized for use
on a Macintosh computer.
A text about 10 computer screens long was written to present the Autopractan
grammar rules listed in the section “The Language” to the subjects. See Appendix
C for an example screen. The last part of the grammar presentation contained picture
and sentence combinations.

Instruments
Three multiple-choice metalinguistic tests were developed to practice and test sub-
jects’ explicit rule knowledge; each test consisted of 10 items.
Each practice session consisted of both comprehension and production exercises
and tests. For comprehension, sets of four pictures appeared on the screen with a
sentence corresponding to one of the four pictures; the other three pictures would
have required one or two different morphemes in the sentence. For production, one
picture appeared with one or more blank fields to type in sentences or word endings.
For both production and comprehension, sentences were scored as wrong as soon
as there was any mistake. All sentences were either two or three words (six or nine
morphemes) long.
The Modern Language Aptitude Test (MLAT) Form A (Carroll & Sapon, 1959) was
administered to all subjects before the experiment in order to make sure that the
different practice groups had the same level of language learning aptitude.

Procedure
The entire experiment, from the initial presentation of grammar and vocabulary to
the final testing, took place on Power Macintosh 7200/75 computers; the software
for all the components was written in HyperCard. Each individual participated in
22 sessions of an hour or less, spread over an 11-week period.
All students were taught the four grammar rules and 32 vocabulary items of
Autopractan by means of a traditional presentation of grammar rules, along with a

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 202 Robert M. DeKeyser

presentation of the vocabulary accompanied with pictures to represent the people

(nouns) and actions (verbs) involved. To make sure the grammar was thoroughly
understood, a number of sentence and picture combinations were then presented
with sets of questions about the inflectional morphemes and their relationship with
the meaning represented by the pictures.
After going through this presentation twice (Sessions 1 and 2), the students were
tested on their explicit knowledge of vocabulary and grammar (Sessions 3–6): Recall
of all vocabulary items was tested in randomized order, and a fill-in-the-blanks test
about explicit grammar rule knowledge was administered. In Sessions 3–5, a different
10-item test was used in each session. The participants received explicit feedback
on every response, explaining in detail for each error why the response was wrong.
In Session 6, a combined test of all 30 previously used items was administered. At
that time, all students demonstrated complete knowledge of both vocabulary and
grammar, in the sense that they made no mistakes (in most cases) or just one or
two (most likely due to distraction, as was often clear when subjects said “oops”
immediately after choosing an answer).
Afterward, the subjects were assigned to one of three practice conditions. Group
A practiced the rules for number of the noun and instrumentality of the verb in
comprehension and the rules for case marking on the noun and gender marking on
the verb in production. For Group B it was the other way around: Participants in
Group B practiced the rules for number of the noun and instrumentality of the verb
in production and the rules for case marking on the noun and gender marking on
the verb in comprehension. Group C practiced all four rules in both comprehension
and production, but with half as many items each as Groups A and B, so that the
total number of items was the same for all three groups. The subjects practiced
comprehension by selecting one of a set of four pictures that matched a sentence
displayed in the middle of the screen; the four pictures were always chosen so that
the two grammar features being practiced were critical to choosing the correct
picture (number marking on the noun and instrumentality marking on the verb for
all the sets of pictures shown to Group A; case marking on the noun and gender
marking on the verb for all the sets of four pictures shown to Group B; and one of
these two combinations for each set of four pictures shown to Group C, depending on
the set). Production practice took the form of typing in the sentence corresponding to
a picture displayed on the screen. Again, two grammar rules were critical to giving
a correct answer (case marking on the noun and gender marking on the verb for
all the pictures shown to Group A; number marking on the noun and instrumentality
marking on the verb for all the pictures shown to Group B; one of these two combina-
tions for each picture shown to Group C, depending on the picture). Having a null-
subject language made it possible never to present sentences with subject–object
contrasts in comprehension for Group A and in production for Group B (by presenting
sentences without subject for comprehension and prompting for sentences without
subject in production). The other contrasts were simply avoided for the group in
question by making sure all pictures or picture sets represented only masculine
subjects, or only singular nouns, or only actions carried out without a tool. Having
complete control over what grammatical contrasts appeared in a given picture was

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 203

necessary to make sure that the participants would only practice two out of four
grammatical contrasts in production and the other two in comprehension.
In each of the 15 practice sessions (Sessions 7–21), subjects worked through
eight blocks of items. Four blocks for practice were always followed by four blocks
for testing. Within each group of four, there were two blocks of production items
and two blocks of comprehension items; for both comprehension and production,
one of the two blocks was single-task and one was dual-task. In the dual-task condi-
tion, subjects saw a number between 100 and 1,000 in the middle of the screen
before the picture(s) appeared. While they were carrying out the comprehension
or production task, they heard beeps at irregular intervals.2 The secondary task
consisted of memorizing the three-digit number, counting the beeps, and subtracting
the number of beeps from the original number immediately after giving a response
to the comprehension or production item.3 The outcome of the subject’s computation
was recorded along with the answer to the main task and the reaction time.4 The
reason for the dual-task condition was to allow for an assessment of how automatized
the language skills were: The more a skill is automatized, the less it should interfere
with a secondary task or show interference from it; that is, the more it is automatized,
the less difference is expected between the single-task and dual-task conditions.
The four testing blocks of picture and sentence combinations each consisted of
16 items in all 15 practice sessions; the four practice blocks always consisted of 8
items each. The order in which pictures or sets of pictures were presented was
randomized for each block; so was the choice of the combination of morphemes
out of the possible set of four combinations, as well as the position on the screen
for the one correct and three incorrect pictures in the comprehension blocks. In
the first session, the practice items were in English instead of Autopractan. In this
way, the subjects had the opportunity to get used to the testing formats before ever
having practiced a single Autopractan sentence in comprehension or production.
This set-up allowed for the collection, in the second half of the first practice session,
of ideal baseline data for Autopractan comprehension and production skills: The
learners had not had any previous practice of the comprehension and production
skills themselves but had practiced the ancillary skills (e.g., dealing with the keyboard
and screen layout) and the relevant declarative knowledge (Autopractan vocabulary
and grammar as taught in Sessions 1 and 2 and practiced in declarative form in
Sessions 3–6). During the practice blocks, feedback was provided; each error trig-
gered an automatic explicit explanation of what grammatical categories in the answer
were wrong and why, for both the production and comprehension tasks.5 No feedback
was given during the testing blocks.
For the production tasks, there was a slight difference in format between the
practice and the testing blocks. In the four practice blocks in each session, subjects
had to type in the complete sentence corresponding to the picture on the screen.
In the testing blocks, the lexemes were provided and the subjects only needed
to type in the grammatical morphemes. In this way, the practice was maximally
representative of language production, but the reliability and validity of the test items
was improved by avoiding all problems of vocabulary knowledge and minimizing the
likelihood of typographical errors.

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 204 Robert M. DeKeyser

The four practice blocks always came before the four testing blocks, but the
order of comprehension versus production blocks and single-task versus dual-task
blocks was counterbalanced over sessions (even vs. uneven session numbers) and
subjects (even vs. uneven subject codes). Thus, when subjects with an uneven code
started with a certain order of blocks in Session 1, then subjects with an even code
would have the opposite order for both comprehension versus production and
single-task versus dual-task in that session, and the subjects with an uneven code
number would have the opposite order in Session 2, and so on.
To summarize, all subjects were taught the same rules explicitly and then received
the same amount of practice and exposure for each rule. All subjects also had the
same amount of practice in both skills and therefore also the same amount of
experience with both practice and testing formats. The only difference between the
three groups was the amount of practice of specific rules in specific skills and
formats. Thus, the design allows for testing the specificity of procedural skill (com-
prehension vs. production) acquired as a result of systematic practice of that skill
after all subjects had achieved the same quantity and quality of explicit knowledge.
After the 15 sessions of practicing and testing specific combinations of rules and
skills, a final session (Session 22) consisted of two parts. In one half of the final
session, the subjects were tested, with a new set of sentences and pictures, on the
usual combinations of rules and skills (without warm-up this time) (“same” condi-
tion). In the other half, they were tested on the opposite combination: The rules
that were previously practiced and tested in comprehension were now tested in
production, and vice versa (“reverse” condition). Whether subjects were first admin-
istered the items in the “same” and then the “reverse” condition or the other way
around was counterbalanced over subjects (low vs. high numbers in subject codes).
The order of comprehension versus production blocks and single-task versus dual-
task blocks was also counterbalanced over subjects (even vs. uneven subject codes).
It should be stressed here that the subjects did not engage in any new tasks at this
stage and were not tested on any new rules, only on task and rule combinations
they had not encountered before.

RESULTS

The Learning Curve

Figures 1–4 show the mean performance of all participants as a function of practice
for both comprehension and production in terms of reaction times and error rates.
Reaction times for comprehension are shown in Figure 1. Performance under
single-task and dual-task conditions is very similar, except for the first session, in
which it is slightly better for the single-task than for the dual-task condition.
Reaction times for production are presented in Figure 2. Reaction times in both
single-task and dual-task conditions show a very slow decline. Performance for
the dual-task conditions is virtually indistinguishable from that of the single-task
condition.
Figure 3 shows the number of errors for comprehension. The curves are somewhat

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 205

Figure 1. Reaction time (ms) in the comprehension task

as a function of practice (solid line = single-task condition;
dashed line = dual-task condition).

Figure 2. Reaction time (ms) in the production task as

a function of practice (solid line = single-task condition;
dashed line = dual-task condition).

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 206 Robert M. DeKeyser

Figure 3. Number of errors in the comprehension

task as a function of practice (solid line = single-task
condition; dashed line = dual-task condition).

bumpy for both conditions. Subjects in the single-task condition did better than
those in the dual-task condition for the first three sessions. After that, there appears
to be no pattern to the relative performance of single-task and dual-task conditions.
Figure 4 shows the number of errors for production. The single-task and dual-
task conditions yielded largely similar results, except for the first session. Here
subjects did better in the single-task condition, too.

Transfer Across Skills

Table 1 presents the means and standard deviations for the various measures of
performance in the “same” condition of the final test (in which the participants used
the rules in comprehension that they had always practiced in comprehension and
used the rules in production that they had always practiced in production) and in
the “reverse” condition (in which they had to use each rule in the opposite skill
compared to what they were used to), along with significance statistics for the
difference between the two conditions. For both production tasks and for the single-
task condition of the comprehension task, the difference between same and reverse,
as measured by a paired-samples t test, is always significant at p < .05, and usually
at p < .01 level, for both reaction times and error data. For the comprehension and
dual-task combination, however, only reaction time shows a significant difference
for the two conditions.
Table 1 also presents data from the mixed-practice group (C) for comparison,

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 207

Figure 4. Number of errors in the production task

as a function of practice (solid line = single-task con-
dition; dashed line = dual-task condition).

along with significance data for the comparisons between the performance of this
group on the final test and that of the specialized-practice groups (A and B) under
the “same” and “reverse” conditions. After determining that no significant difference
existed between Groups A and B on the one hand and Group C on the other in terms
of MLAT scores, and therefore that an analysis of covariance was not necessary, a
set of independent-sample t tests was carried out for the various performance
measures. As the last four columns in Table 1 show, the differences between Group
C and the same condition are never significant, except for the dual-task condition
of the comprehension task. The differences between Group C and the reverse condi-
tion are always significant, with the exception again of the comprehension and dual-
task combination.

DISCUSSION

Hypothesis 1
The hypothesis of gradual automatization as a function of practice was largely
confirmed. For both reaction time and error rates, for both production and compre-
hension, and for both single- and dual-task conditions, a clear and gradual drop-off
takes place, which slows down over time, from a dramatic decrease between Sessions
1 and 2 to an almost imperceptible decline for the last few sessions. These results
confirm the findings of a pilot study with the single-task condition only, with fewer

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 208
Table 1. Performance measures for the “same” and “reverse” conditions of the final test
Group C
Same Reverse (Same/ Difference
(Groups A/B) (Groups A/B) Reverse) Same vs. Difference Difference
(n = 41) (n = 41) (n = 20) Reverse C vs. Same C vs. Reverse

M SD M SD M SD t p t p t p

Comprehension
Single task
Reaction time (ms) 3,641 1,923 5,767 4,039 3,406 1,199 4.01 .000 0.58 .562 3.45 .001
Errors (out of 16) 0.63 0.77 1.42 1.25 0.75 0.72 3.11 .003 −0.58 .566 2.64 .011
Dual task
Reaction time (ms) 3,917 1,939 4,800 1,837 3,301 962 2.27 .029 1.66 .103 4.18 .000
Errors (out of 16) 1.24 1.58 1.07 1.17 0.65 0.61 −0.52 .605 2.11 .039 1.86 .068
Production
Single task

http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
Reaction time (ms) 4,964 2,126 8,325 4,014 5,182 1,319 3.48 .000 −0.39 .697 3.15 .000
Errors (out of 16) 1.00 1.84 2.68 3.75 0.60 0.50 2.09 .043 1.29 .202 3.50 .001
Dual task
Reaction time (ms) 5,506 2,531 7,591 2,845 4,997 876 3.09 .004 1.15 .255 4.68 .000
Errors (out of 16) 0.93 2.03 2.93 3.76 0.63 0.63 2.69 .010 .87 .388 3.82 .000

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
 Automatizing L2 Morphosyntax 209

subjects and fewer sessions (DeKeyser, 1996). The curves are very smooth for
reaction time and bumpier for the error data. It is well known that error data are
less reliable (especially with short tests) and therefore harder to fit to a curve than
reaction time data (cf., e.g., Anderson, 1995). The difference between the single-task
and dual-task conditions did not decline as slowly and gradually as expected. The
difference is small in most cases, except for the error rate in the first two or three
sessions. A two-way repeated-measures analysis of variance (first vs. last session ×
single-task vs. dual-task condition) yielded no significant interaction effect, only a
main effect for session. It appears that the arithmetic task was too easy to interfere
much with the language task once subjects had a modest amount of practice. As a
result, the interaction is in the expected direction, but it is not significant.

Hypothesis 2
The second hypothesis is more specific. Longitudinal studies of skill acquisition in
various domains, including perceptual-motor as well as cognitive skills, have yielded
data that follow a power law (Newell & Rosenbloom, 1981). That means that when
practice (expressed in terms of time elapsed or number of learning sessions) is x
and reaction time or error rate is y, then y = x n. When empirical data that fit a power
function are plotted on a logarithmic scale (log(y) vs. log(x)), they come close to
yielding a straight line; the correlation between log(y) and log(x) is close to 1. Other
functions exist that are characterized by a gradual decline, such as an exponential
function (y = e x). If a function is exponential, then a plot of log(y) versus x is a
straight line and the correlation of log(y) with x is 1, but Newell and Rosenbloom
(1981) show that power functions give a consistently better fit than exponential
functions for data that reflect the effect of practice, in the sense that the correlation
between log(y) and log(x) is very large (typically above 0.95) and higher than the
correlation between log(y) and x.
Our data for reaction time show a very close fit to a power function: The correla-
tions between log(reaction time) and log(session number) vary from −.983 for pro-
duction in the single-task condition to −.991 for production in the dual-task condition
(see Table 2). The corresponding coefficients for an exponential function (rlog(y)x) are
−.872 and −.891. Visual inspection of the corresponding graphs showed that, even
with correlation coefficients as high as −.991, there was still a slight curvature in
the spread of the data points around the regression line, mainly due to the fact that
the data for the first session have an unduly strong influence on the position of the
regression line. Therefore, following Newell and Rosenbloom (1981), the data were
analyzed again with the first session removed. Although this had only a slight
influence on the correlation coefficients (now ranging from −0.990 to −.996), the data
points are now symmetrically distributed around the regression line. In Figures 5–8
the regression line drawn is the closest fit for the data points from Sessions 2–15.
The data point for Session 1 has been plotted, too, however, in order to show how
it deviates from the pattern for the subsequent sessions. These graphs show even
more convincingly than the correlation coefficients by themselves that the data are
a perfect fit for a power function in the sense that the data points are virtually on

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 210 Robert M. DeKeyser

Table 2. Correlations between number of practice sessions

and performance measures as measured by rlog(y)log(x) and rlog(y)x
All Sessions Sessions 2–15

rlog(y)log(x) rlog(y)x rlog(y)log(x) rlog(y)x

Reaction time
Comprehension
Single task −.987 −.878 −.992 −.932
Dual task −.987 −.881 −.994 −.941
Production
Single task −.983 −.872 −.990 −.934
Dual task −.991 −.891 −.996 −.942
Error rate
Comprehension
Single task .783 .570 .744 .636
Dual task .849 .647 .847 .726
Production
Single task .807 .604 .825 .741
Dual task .809 .607 .872 .788

Figure 5. Reaction time as a function of number of

practice sessions (Session 1 removed from analysis
but not from graph): comprehension task/single-task
condition.

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 211

Figure 6. Reaction time as a function of number of

practice sessions (Session 1 removed from analysis
but not from graph): comprehension task/dual-task
condition.

the regression line and do not show a consistent residual pattern. The graphs for
the corresponding exponential function (not included in this article) consistently
show a nonsymmetrical distribution of the data points, above the regression line
for the first and last few sessions, below in the middle. In other words, the exponential
function is a less good fit, not just because of scatter around the regression line,
but because it fails to capture the exact shape of the learning curve as well as the
power function does.
It is important to note that the data from Session 1 are not just being excluded
in order to provide a better fit for the theoretical functions. In fact, it follows from
the model of skill acquisition outlined in the literature review that initial practice
of a task has very different effects (proceduralization of declarative knowledge)
compared to subsequent practice (automatization). Anderson and Fincham (1994),
in particular, argue for the qualitatively different nature of the initial practice and
the speed with which proceduralization takes place.6
The data for error rates show a less good fit, as is to be expected from a cursory
visual inspection of the graphs in Figures 1–4, but even there the power function
provides a consistently better fit than the exponential function (see Table 2).7
In summary, our data show that practice in the use of second language grammar
rules, whether through comprehension or production tasks, has exactly the same
effect on learning as it does in the acquisition of other cognitive skills. This strongly
suggests that the same mechanisms are involved and that the ability to comprehend

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 212 Robert M. DeKeyser

Figure 7. Reaction time as a function of number of

practice sessions (Session 1 removed from analysis but
not from graph): production task/single-task condition.

Figure 8. Reaction time as a function of number of

practice sessions (Session 1 removed from analysis but
not from graph): production task/dual-task condition.

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 213

or produce sentences in a second language is not necessarily acquired through the

implicit mechanisms of a separate mental module (as is generally accepted for first
language acquisition).

Hypothesis 3
If the hypothesis that the effect of practice is skill-specific is correct, then students
with comprehension practice only should improve more in comprehension skills
than students with production practice only and vice versa, whereas students with
an equal amount of practice in both skills should perform at an intermediate level
in both skills.
This hypothesis was largely confirmed, too, by the data from the final test (with
all new pictures). The “same” and “reverse” condition columns in Table 1 show that
the subjects almost always performed worse in the “reverse” condition than in the
“same” condition in terms of both reaction times and error rates, as the participants
in the pilot study (DeKeyser, 1996) did. For seven out of eight combinations of task,
task condition, and criterion measure, the difference is rather large (especially for
production, in which reaction times and error rates are roughly 1.5–3 times as high
in the “reverse” condition compared to the “same” condition), and it is significant
at p < .05 or better in a pairwise t test. For error rate in the comprehension task
under dual-task conditions, the difference is not significant, and it is actually in the
opposite direction from what was expected. This one failure to reach significance
is probably due to the low reliability of the comprehension task (16 items, four-way
multiple-choice).
It was further predicted that the results for Group C (mixed-practice group)
would always be in-between the results for the “same” and “reverse” conditions for
the specialized-practice groups (A and B). Table 1 shows that this was not the case:
Six out of the eight measures even show a slight (nonsignificant) advantage for the
mixed-practice group over the “same” condition of the specialized-practice group.
As predicted, however, the mixed-practice group always shows a distinct advantage
over the reverse condition for the single-practice group. This difference is significant
at p < .01 for all measures except the two comprehension error rate measures (single-
task and dual-task conditions), in which it narrowly fails to reach significance (at
p < .01 and p < .05, respectively). Here, again, the failure to reach significance for
this one particular pair of measures is probably due to the low reliability of the
multiple-choice task.
Overall, our data lend strong support to the hypothesis that practice has a skill-
specific effect in the sense that performance in comprehension or production is
severely reduced if only the opposite skill was practiced. This is contrary to the idea
of linguistic competence acquired through comprehension being equally available for
production and comprehension. At the same time, however, the data for the mixed-
practice group show that, once a certain level of practice in a skill has been reached,
additional practice in that same skill (in the sense of more items over the same time
period) does not yield better results. It appears that, within a given session, the
learning effect tends to asymptote after a certain numbers of items. This may be

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 214 Robert M. DeKeyser

another illustration of the same power law that was shown in the change from
session to session.8
It should be stressed that these findings go beyond those of VanPatten and
associates (Cadierno, 1995; VanPatten & Cadierno, 1993a, 1993b; VanPatten & Sanz
1995) on the one hand and DeKeyser and Sokalski (1996) on the other. Whereas the
disagreement between those studies concerns the short-term effect of practice be-
fore any automatization has taken place, the present findings address the long-
term effect of systematic practice, once the automatization process has reached
asymptote or come close, that is, once learning has bottomed out in terms of both
reaction time and error rate.

CONCLUSIONS AND IMPLICATIONS

This study has shown that the learning of second language grammar rules can
proceed very much in the same way that learning in other cognitive domains, from
geometry to computer programming, has been shown to take place. The learning
curves observed in this study follow the same power law as for those other skills.
As has been found in several studies in other domains, the first practice session
presented an exception to the pattern followed by the other sessions. This evidence
supports the model of skill acquisition that posits that during initial practice declara-
tive knowledge is turned into qualitatively different procedural knowledge and that
subsequently a much slower process of gradual automatization takes place, which
requires little or no change in task components, only a quantitative change within
the same components.
Evidence has also been presented to the effect that practice leads to highly
specific skills. Just as was found previously in studies of computer programming,
this experiment has shown that large amounts of practice in comprehension and
production of second language grammar rules lead to spectacular improvements in
the skill practiced but have far less impact on the reverse skill. When subjects had
to use rules in comprehension that they had previously practiced in production
only, and vice versa, the reaction time and error rate often went up by a factor
between 1.5 and 3, especially for production.
The way the study was designed makes it clear that the two main findings
presented here, the gradual improvement following the power law and the skill
specificity of the learning process, are language learning phenomena and not just
task habituation effects. Not only did subjects get practice in the mechanics of the
various task formats before the first test data were collected but, more importantly,
the students’ performance between the “same” and “reverse” conditions of the final
test shows a big difference between rules practiced in the skill tested and rules only
practiced in the other skill. As there was absolutely no difference between these two
conditions for the group as a whole in terms of task format, pictures, or sentences, the
differences in performance are clearly due to the degree to which grammar rules
had been practiced.
In terms of teaching methodology, findings such as these suggest that there is a
role for systematic practice of specific rules for specific skills in the second language

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 215

curriculum. This does not imply that practice should be as focused and repetitive
as it was in this study, not even that a structural syllabus should be used. It does
imply, however, that the sequence of explicit rule learning, followed by a short
period of activities focused on using explicit knowledge during performance of the
target skills, and finally by a long period of repeated opportunities to use that
knowledge, is likely to yield knowledge that is highly automatized.
Three caveats are in order, though. First of all, proceduralization and automatiza-
tion may not be equally successful for all learners and all rules. Second, no claim
is being made here that the automatized knowledge documented in this study is
equivalent to the implicit knowledge typically acquired in the native language (as
an artificial language was used, no native speaker comparisons could be made). It
remains to be seen whether more prolonged practice or practice that encourages
less focus on form could induce such fundamental change. Third, no specific recom-
mendations for teaching methodology should be made before the findings of this
study have been put to the test in the more ecologically valid context of real second
language classrooms.
In terms of research methodology, this study has shown that research in a
laboratory context allows forms of intensive, fine-grained, and highly controlled data
collection that, in turn, allow for more precise testing of more specific hypotheses
than would be possible in a more naturalistic context. Findings from such laboratory
research should not be overgeneralized, just as findings from biochemical studies
should not be used to recommend large-scale use of new drugs without careful
clinical studies. By the same token, however, it can be ill-advised to test complex
hypotheses in highly variable and poorly understood classroom contexts, just as it
would be counterindicated to test out new drugs on patients before their biochemical
properties are at least known in principle. Encouraged by the results of the present
study, which shows that second language grammar rules can be learned and automa-
tized like rules in other domains, at least under laboratory conditions, we are now
planning a number of classroom experiments on the long-term effect of practice
activities designed according to Anderson’s model of skill acquisition.
Finally, the use of an artificial language has been shown to be useful for various
reasons. It not only has the obvious advantages of controlling previous knowledge,
outside exposure, or integrative motivation, but perhaps, more importantly, it allows
the researcher to remove a number of linguistic constraints on an already complex
experimental design, which otherwise may become completely unmanageable.

NOTES

1. I am using ACT here as a general term for the original ACT model (Anderson, 1976), the ACT* (pronounced
ACT–star) version introduced in Anderson (1983), and the ACT-R (ACT–Rational) version introduced in
Anderson (1993).
2. During practice, for comprehension, the first beep occurred after 3 seconds; subsequent beeps followed
after intervals varying between 5 and 6 seconds in length. For production, the first beep occurred after 3
seconds, and subsequent beeps after intervals of 5.5–6 seconds. During testing, for comprehension, the first
beep occurred after 3.5 seconds; the intervals for subsequent beeps varied between 5 and 6 seconds in length.
For production, the first beep occurred after 3 seconds; subsequent beeps occurred after 2–3-second intervals.
These intervals were determined on a trial-and-error basis during the piloting.
3. Arithmetic tasks are commonly used as secondary tasks in experimental psychology to limit the amount

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 216 Robert M. DeKeyser

of cognitive resources that participants can allocate to the primary task when both are executed simultaneously
(see, e.g., Pashler, 1994). Counting backward in steps of seven from a three-digit number, for instance, is a
commonly used procedure. During the initial stages of piloting the Autopractan experiment, I found that
counting in steps of more than one was unduly difficult in this case and that simply asking subjects to subtract
each time they heard a beep was likely to lead to a situation where some subjects did this literally, and
others would count the number of beeps, and then subtract that number from the original number. The
procedure adopted here was meant to avoid these problems.
4. Reaction time for the comprehension tasks was measured as the time elapsed between presentation
of the stimulus sentences and pictures and hitting the button corresponding to one of the four picture choices.
For the production tasks, reaction time was the time elapsed between presentation of the picture stimulus
and hitting the return button after finishing typing in the sentence. Reaction time for the arithmetic task was
not measured, nor was the correctness of the arithmetic task taken into account for the data analysis. Subjects
were motivated to do well on this task, however, because the correctness of the arithmetic task was recorded
and taken into account to determine their payment.
5. In the comprehension format, errors were detected by a comparison of a series of grammar codes for
the sentence in question with the grammar codes for the picture chosen. In case of mismatch for one of the
codes, participants were told that there was a problem with the corresponding category (i.e., gender, number,
instrumentality, or case marking). Otherwise, they were told the picture was correct. In the production format,
the program scanned the input for all the required words in any order. When an error was detected, the
student was told what the expected word was, along with a description of its grammatical features. In the
metalinguistic format (Sessions 3–6) ad hoc feedback was given to each specific choice made in the multiple-
choice format for the grammar items. In case of errors in the vocabulary test, the right word was presented
to the student. In this vocabulary test, the ERRATA program developed by Robert Hart (University of Illinois
at Urbana–Champaign) was used for comparing input words to expected answers.
6. To determine how fast proceduralization takes place, a separate analysis was performed on the data
from the subjects for whom the dual-task condition preceded the single-task condition in the first session
(i.e., half of the subjects, as the order of conditions was counterbalanced). For these subjects, the results
for the dual-task condition show an even more dramatic deviation for the first session (the first data point
is almost twice as far removed from the regression line as for the group as a whole), whereas for the single-
task condition the first data point falls exactly on the regression line. This shows that the proceduralization
process is complete not only by the end of the first session but already by the end of the first 16-item block
for each skill.
7. The coefficients in the bottom half of Table 2 refer to the correlation with number correct out of 16
instead of number of errors out of 16, contrary to the previous tables and graphs, which present number of
errors. This was done to avoid having to take the logarithm of 0 in each case where there were no errors,
which would have led to many missing values.
8. I thank Nick Ellis for pointing this out to me.

REFERENCES

Alanen, R. A. (1995). Input enhancement and rule presentation in second language acquisition. In R. Schmidt
(Ed.), Attention and awareness in foreign language learning (pp. 259–302). Honolulu: University of Hawaii
Press.
Anderson, J. R. (1976). Language, memory, and thought. Hillsdale, NJ: Erlbaum.
Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard University Press.
Anderson, J. R. (1987). Skill acquisition: Compilation of weak-method problem solutions. Psychological Review,
94, 192–210.
Anderson, J. R. (1992). Automaticity and the ACT* theory. American Journal of Psychology, 105, 165–180.
Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Erlbaum.
Anderson, J. R. (1995). Learning and memory. An integrated approach. New York: Wiley.
Anderson, J. R., & Fincham, J. M. (1994). Acquisition of procedural skills from examples. Journal of Experimental
Psychology: Learning, Memory and Cognition, 20, 1322–1340.
Cadierno, T. (1995). Formal instruction from a processing perspective: An investigation into the Spanish past
tense. The Modern Language Journal, 79, 179–193.
Carroll, J. B., & Sapon, S. (1959). Modern Language Aptitude Test. Form A. New York: The Psychological
Corporation.
Chastain, K. (1971). The development of modern language skills: Theory to practice. Philadelphia: Center for
Curriculum Development.
de Graaff, R. (1997). The eXperanto experiment: Effects of explicit instruction on second language acquisition.
Studies in Second Language Acquisition, 19, 249–276.

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 217

DeKeyser, R. (1995). Learning second language grammar rules: An experiment with a miniature linguistic
system. Studies in Second Language Acquisition, 17, 379–410.
DeKeyser, R. (1996). Exploring automatization processes. TESOL Quarterly, 30, 349–357.
DeKeyser, R. (in press). Beyond focus on form: Cognitive perspectives on learning and practicing second
language grammar. In C. Doughty & J. Williams (Eds.), Focus on form in classroom language acquisition.
New York: Cambridge University Press.
DeKeyser, R., & Sokalski, K. (1996). The differential role of comprehension and production practice. Language
Learning, 46, 613–641.
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.
Ellis, N. (1993). Rules and instances in foreign language learning: Interactions of explicit and implicit knowledge.
European Journal of Cognitive Psychology, 5, 289–318.
Ellis, R. (1993). The structural syllabus and second language acquisition. TESOL Quarterly, 27, 91–113.
Favreau, M., & Segalowitz, N. (1983). Automatic and controlled processes in the first- and second-language
reading of fluent bilinguals. Memory and Cognition, 11, 565–574.
Gatbonton, E., & Segalowitz, N. (1988). Creative automatization: Principles for promoting fluency within a
communicative framework. TESOL Quarterly, 22, 473–492.
Griffiths, R. (1991). Pausological research in an L2 context: A rationale and review of selected studies. Applied
Linguistics, 12, 345–364.
Hart, R. (1994). ERRATA: Response analysis and error diagnosis tools (Tech. Rep. No. LLL-T-23-94). Language
Learning Laboratory, University of Illinois at Urbana–Champaign.
House, J. (1996). Developing pragmatic fluency in English as a foreign language. Studies in Second Language
Acquisition, 18, 225–252.
Hulstijn, J. H. (1997). Introduction. Second language acquisition research in the research lab: Possibilities
and limitations. Studies in Second Language Acquisition, 19, 131–144.
Hulstijn, J., & de Graaff, R. (1994). Under what conditions does explicit knowledge of a second language
facilitate the acquisition of implicit knowledge? A research proposal. AILA Review, 11, 97–112.
Johnson, K. (1996). Language teaching and skill learning. Oxford: Blackwell.
Krashen, S. D. (1982). Principles and practice in second language acquisition. Englewood Cliffs, NJ: Prentice-
Hall.
Krashen, S. D. (1985). The input hypothesis. New York: Longman.
Krashen, S. D. (1994). The input hypothesis and its rivals. In N. Ellis (Ed.), Implicit and explicit learning of
languages (pp. 45–77). London: Academic Press.
Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review, 95, 492–527.
Logan, G. D. (1992). Shapes of reaction-time distributions and shapes of learning curves: A test of the instance
theory of automaticity. Journal of Experimental Psychology: Learning, Memory and Cognition, 18,
883–914.
Logan, G. D., & Etherton, J. L. (1994). What is learned during automatization? The role of attention in
constructing an instance. Journal of Experimental Psychology: Learning, Memory and Cognition, 20,
1022–1050.
Logan, G. D., Taylor, S., & Etherton, J. (1996). Attention in the acquisition and expression of automaticity.
Journal of Experimental Psychology: Learning, Memory and Cognition, 22, 620–638.
Master, P. (1994). The effect of systematic instruction on learning the English article system. In T. Odlin (Ed.),
Perspectives on pedagogical grammar (pp. 229–252). New York: Cambridge University Press.
McLaughlin, B. (1987). Theories of second-language learning. London: Edward Arnold.
McLeod, B., & McLaughlin, B. (1986). Restructuring or automaticity? Reading in a second language. Language
Learning, 36, 109–123.
Möhle, D. (1984). A comparison of the second language speech production of different native speakers. In H.
W. Dechert, D. Möhle, & M. Raupach (Eds.), Second language productions (pp. 28–49). Tübingen, Germany:
Gunter Narr.
Newell, A., & Rosenbloom, P. S. (1981). Mechanisms of skill acquisition and the law of practice. In J. R.
Anderson (Ed.), Cognitive skills and their acquisition (pp. 1–55). Hillsdale, NJ: Erlbaum.
O’Malley, J. M., & Chamot, A. U. (1990). Learning strategies in second language acquisition. Cambridge: Cambridge
University Press.
Pashler, H. (1994). Dual-task interference in simple tasks: Data and theory. Psychological Bulletin, 116,
220–244.
Paulston, C. B., & Bruder, M. N. (1976). Teaching English as a second language: Techniques and procedures.
Cambridge, MA: Winthrop.
Raupach, M. (1987). Procedural learning in advanced learners of a foreign language. (Linguistic Agency University
of Duisburg Paper B167). Duisburg, Germany: Universität Duisburg Gesamthochschule.

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 218 Robert M. DeKeyser

Rivers, W. M. (1964). The psychologist and the foreign language teacher. Chicago: University of Chicago Press.
Robinson, P. (1996). Learning simple and complex second language rules under implicit, incidental, rule-
search and instructed conditions. Studies in Second Language Acquisition, 18, 27–67.
Robinson, P. (1997). Generalizability and automaticity of second language learning under implicit, incidental,
enhanced, and instructed conditions. Studies in Second Language Acquisition, 19, 223–248.
Robinson, P., & Ha, M. A. (1993). Instance theory and second language rule learning under explicit conditions.
Studies in Second Language Acquisition, 15, 413–438.
Schmidt, R. (1992). Psychological mechanisms underlying second language fluency. Studies in Second Language
Acquisition, 14, 357–385.
Schneider, W., & Detweiler, M. (1988). The role of practice in dual-task performance: Toward workload modeling
in a connectionist/control architecture. Human Factors, 30, 539–566.
Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: Part I.
Detection, search, and attention. Psychological Review, 84, 1–66.
Scott, V. M. (1989). An empirical study of explicit and implicit teaching strategies in French. The Modern
Language Journal, 73, 14–22.
Segalowitz, N. S., & Segalowitz, S. J. (1993). Skilled performance, practice, and the differentiation of speed-up
from automatization effects: Evidence from second language word recognition. Applied Psycholinguistics,
14, 369–385.
Seger, C. A. (1994). Implicit learning. Psychological Bulletin, 115, 163–196.
Spada, N., & Lightbown, P. M. (1993). Instruction and the development of questions in L2 classrooms. Studies
in Second Language Acquisition, 15, 205–224.
Towell, R., Hawkins, R., & Bazergui, N. (1996). The development of fluency in advanced learners of French.
Applied Linguistics, 17, 84–119.
VanPatten, B., & Cadierno, T. (1993a). Explicit instruction and input processing. Studies in Second Language
Acquisition, 15, 225–243.
VanPatten, B., & Cadierno, T. (1993b). Input processing and second language acquisition: A role for instruction.
The Modern Language Journal, 77, 45–57.
VanPatten, B., & Sanz, C. (1995). From input to output: Processing instruction and communicative tasks. In
F. Eckman et al. (Eds.), Second language acquisition theory and pedagogy (pp. 169–185). Mahwah, NJ:
Erlbaum.
White, L., Spada, N., Lightbown, P., & Ranta, L. (1991). Input enhancement and L2 question formation. Applied
Linguistics, 12, 416–432.

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 219

APPENDIX A

AUTOPRACTAN–ENGLISH DICTIONARY

balit, V “hit,” “punch,” “beat” kupif, N “sports fan”

barat, N “soldier” ladet, N “waiter”
barip, V “tickle” limot, V “push”
bimet, N “scout” mator, V “watch,” “look at”
biruf, N “sailor” nabok, N “worker”
bomap, N “rugbyplayer” nemok, V “strangle”
bulot, V “point at” patek, V “pull,” “drag”
butak, N “motorcyclist” pemek, N “clown”
fatis, V “feed” potam, N “doctor,” “physician”
fatuk, V “aim at,” “shoot at” pumaf, V “spank”
fitop, N “king,” “queen,” “monarch” samek, V “examine”
fokap, N “student” selek, N “cowboy”
fumas, V “block,” “stop,” “halt” sipuk, V “measure”
kabom, N “cook” sotot, N “viking”
kapit, V “scare” titom, N “backpacker”
kotif, V “poke” topim, V “trip”

Note: N = noun; V = verb.

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 220 Robert M. DeKeyser

APPENDIX B

SOME SAMPLE SENTENCES IN AUTOPRACTAN

(1) selek-li-fu fatuk-ne-ra fitop-li-ka

cowboy-SG-SUBJ shoot-MASC+INST king-SG-OBJ
or
fitop-li-ka fatuk-ne-ra selek-li-fu
king-SG-OBJ shoot-MASC+INST cowboy-SG-SUBJ
“The cowboy is shooting at the king.” (with a gun, for instance)

(2) potam-li-fu samek-pi-so titom-li-ka

doctor-SG-SUBJ examine-FEM-noINST backpacker-SG-OBJ
or
titom-li-ka samek-pi-so potam-li-fu
backpacker-SG-OBJ examine-FEM-noINST doctor-SG-SUBJ
“The doctor [fem.] is examining the backpacker.” (without an instrument)

(3) barat-to-fu mator-ne-so kabom-to-ka

soldier-PL-SUBJ look-MASC-noINST cook-PL-OBJ
or
kabom-to-ka mator-ne-so barat-to-fu
cook-PL-OBJ look-MASC-noINST soldier-PL-SUBJ
“The soldiers are looking at the cooks.” (without an instrument)

(4) mator-pi-ra bimet-to-ka

look-FEM-INST scout-PL-OBJ
or
bimet-to-ka mator-pi-ra
scout-PL-OBJ look-FEM-INST
“They [fem.] are looking at the girlscouts.” (with binoculars, for instance)

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf
 Automatizing L2 Morphosyntax 221

APPENDIX C

SAMPLE SCREEN FROM THE GRAMMAR PRESENTATION

Nouns in Autopractan differ depending on whether they are subject or object of the sentence—in
other words, depending on whether they express who does the action or who undergoes it.
The ending -fu on a noun marks a grammatical subject, the ending -ka marks a grammatical
object. Compare the following:

pemek-li- fu mator-ne-so fokap-li-ka

the (male) clown (SUBJ) is watching the student (OBJ)

fokap-li- fu mator-ne-so pemek-li-ka

the (male) student (SUBJ) is watching the clown (OBJ)

As can be seen, a noun normally occurs with two endings in Autopractan, one for singular (li)
versus plural (to) and one for subject ( fu) versus object (ka).

Downloaded from http:/www.cambridge.org/core. Ateneo de Manila University, on 10 Oct 2016 at 18:56:21, subject to the Cambridge Core terms of use, available at http:/www.cambridge.org/core/terms.
http://content-service:5050/content/id/urn:cambridge.org:id:article:S0272263197002040/resource/name/S0272263197002040a.pdf