What is working memory?
Working memory (WM) can be defined as a
cognitive workspace with a limited capacity
pool of attentional resources for the
temporary storage of information while
performing higher order cognitive tasks such
as reasoning, learning and comprehension
(Baddeley & Hitch, 1974; Baddeley &
Logie, 1999). Baddeley and his colleagues
view WM as that which simultaneously
maintains and processes the input it receives
through different channels of
communications (e.g., touch, long-term
memory, sight, and hearing) (Baddeley,
1986, 1996, 2003, 2007; Baddeley & Hitch,
1974; Baddeley & Logie, 1999; Gathercole
& Baddeley, 1993). A three-component
model of WM was proposed by Baddeley
and Hitch (1974). This model consists of a
central executive and two “slave”
components, the phonological loop and the
visuo-spatial sketchpad. This model was in
use until 2000, when Baddeley added a new
component to it, the episodic buffer, to
account for the studies on densely amnesiac
patients with long-term memory deficits.
This model, as shown in Figure 1, specifies
a functional role of memory as well as an
economical and coherent account of
information on each memory component.
Baddeley’s (2000) model of WM, revised to
incorporate links with long-term memory
(LTM) by way of both the subsystems and
the newly proposed episodic buffer.
The most important component in this
model is the central executive or supervisory
attentional system, which is a limited
capacity pool of general resources.
According to N. Ellis, (2001), “It regulates
information flow within WM, activates or
inhibits the whole sequences of activities,
and resolves potential conflicts between on-going schema-controlled activities” (p., 33).
The reading or listening span tests are
usually used to measure central executive
and give an index for WM.
The phonological loop is in charge of the
temporary storage and processing of verbal
information. It plays a role as a phonological
store by holding phonological
representations of auditory information for a
brief period of time, and as an articulatory
rehearsal system by enabling the reader to
use inner speech to refresh the decaying
representations in the phonological store
(Baddeley, 2000, 2007; N. Ellis, 2001).
Phonological loop is often measured by
presenting spoken lists of words (word
span), digits (digit span) or non-words (non-word span), and asking participants to recall
the lists of words and/or digits in the order
in which they are presented. The maximum
number of items that the individual can
correctly recall is considered to be their
phonological memory score.
The visuo-spatial sketchpad is an interface
between visual and spatial information
received either through the senses or from
long-term memory (Baddeley & Hitch,
1974, p., 854). It is also involved in
generating visual images, temporarily
maintaining them, and manipulating
information with visual or spatial
dimensions. Furthermore, it can be activated
by spoken words by using long-term
knowledge to convert the auditory presented
words into visuo-spatial code (Baddeley,
2007; N. Ellis, 2001). To measure visual
memory, Della Sala, Gray, Baddeley,
Allamano & Wilson’s (1999) pattern span
test is usually used by researchers. In this
test, the individual is presented with 2 x 2
matrixes, with two of the cells filled. Then
after 3 seconds, the individual is asked to
indicate which cells were filled in the
stimulus matrix, using an empty 2 x 2
matrix. The size of the matrix is increased
by two cells every three trials, with half of
the cells of each matrix being randomly
filled. The individual’s pattern span is
determined by the maximum number of the
cells that the participant is able to recall
correctly.
The Corsi Block task is typically used to
measure spatial memory (Milner, 1971). In
this test, the subject is presented with an
array of nine cubes arranged at random
locations on a board placed between the
tester and the participant. The test starts with
the tester initially tapping two of the blocks
one after the other and then asking the
subject to imitate the sequence. The
sequence of taps gradually increases to a
point at which performance breaks down.
The episodic buffer (Baddeley, 2000) is a
limited capacity temporary storage system.
According to Baddeley (2007), “It combines
information from the loop, the sketchpad,
long-term memory, or indeed from
perceptual input into a coherent episode” (p.,
148). Moreover, it plays a role in interfacing
between WM and long-term memory
through the central executive, interacting
phonological loop and sketchpad. It is also
proposed that retrieval from the episodic
buffer is through conscious awareness.
However, no method of measurement has
been proposed yet to assess the episodic
buffer (Baddeley, 2007).
Rationale of the study
Since an important role for working memory
has been found in the first language
acquisition (e.g., Daneman, 1991; Daneman
& Green, 1986; Waters & Caplan, 1996),
research on the role of working memory is
emerging as an area of concern for second
language acquisition (e.g., Atkins &
Baddeley, 1998; Miyake & Freidman, 1998;
Robinson, 2002, 2005). Working memory is
typically measured by a reading span test
(RST) or listening span test in L1 or L2.
The Reading span tests were first introduced
by Daneman & Carpenter (1980). They were
used to measure and give an index for
working memory capacity (WMC). In a
reading span test (RST), participants are
asked to read sets of sentences, report on the
semantic acceptability of each sentence
(processing assessment), and then recall the
final word of each sentence when prompted
(storage assessment). Since the introduction
of the RST by Daneman and Carpenter
(1980), many researchers have used either
Daneman and Carpenter’s original RST or
the modified versions of that in their studies
(e.g., Alptekin & Erçetin, 2009; Chun &
Payne, 2004; Daneman & Carpenter, 1980;
Harrington & Sawyer, 1992; Lesser, 2007;
Osaka & Osaka, 1992; Swanson, 1993;
Walter, 2004).These studies measured WM
either through an L1 RST (Chun & Payne,
2004; Lesser, 2007), an L2 RST (Alptekin &
Erçetin, 2009), or both L1 and L2 RSTs
(Harrington &Sawyer, 1992; Walter, 2004).
As prior research indicated that WM is
language independent (e.g., Miyake &
Freidman, 1998; Osaka & Osaka, 1992;
Osaka, Osaka & Groner, 1993), measuring
WM in L1 was then became popular in
cognitive psychology and studies in second
language learning. This could also help to
avoid conflating WM and L2 proficiency.
However, while there may be considerable
number of L1 RSTs for some languages;
there are few L1 RSTs in some others. In
Persian, there may be few reliable versions
of RST, and if any, none of them has been
published or accessible for the use in other
L2 studies. This issue points to the need for
the development of a RST in this language
for the research with L1 Persian EFL
learners. The present study focused on the
process of development and validation of an
L1 Persian RST for the use in second
language learning studies. More specifically,
this study describes the stages at which RST
items were developed, piloted, revised, and
finally employed in the research with L1
Persian participants.
Methodology
Subjects
74 L1 Persian EFL learners at three
proficiency levels participated in three pilot
studies. Then the newly developed test was
administered to 140 L1 Persian EFL learners
in an experimental study. They included
both males and females, 16-35 years old,
studying English as a foreign language in a
private language school in Iran.
Material
A corpus of Persian sentences was
constructed by an expert in the Persian
language. The sentences contained general
information, and lacked of any technical and
scientific content. 64 sentences were
selected from this corpus to form the RST.
This test included 10 practice session
sentences and 54 test sentences, all of which
were in an active and affirmative form
within a range of 13-16 words. Half of the
sentences were constructed as ‘nonsense’
sentences. This was done by rearranging a
few words in such a way that sentences were
semantically anomalous (Chun & Payne,
2004; Harrington & Sawyer, Lesser, 2007,
Turner & Engle, 1989; Waters & Caplan,
1996). This was to make sure that the
participants processed sentences for
meaning without focusing only on the
retention of recall items. This test was
administered individually using a computer-based format. Because Persian sentences
follow SOV syntax (the sentences initiate
with a subject followed by an object and a
verb respectively), each sentence ends in a
verb, similar to the reading span tests in
Japanese (Osaka & Osaka, 1992) and
German (Osaka et al., 1993; Roehr &
Ganem-Gutierrez, 2008). Each verb
appeared only once in the test. Therefore,
the final words in this test were 64 different
verbs. The verbs in each set were not
semantically related. The sentences in the
test were arranged in three sets of 3, 4, 5,
and 6 sentences. Half of the sentences in
each set were nonsense.
Test procedure
After the initial form of the RST was
developed, three pilot studies were
administered to three groups of L1 Persian
EFL learners. This was to identify the
potential problems with the test. Then the
newly developed test was used in an
experimental study for the measurement of
working memory capacity.
The test was in a PowerPoint format and
was taken individually. It assessed two WM
components, processing and storage (e.g.,
Chun & Payne, 2004; Daneman &
Carpenter, 1980; Harrington & Sawyer,
1992; Lesser, 2007; Waters & Caplan,
1996). The participants had to read each
sentence, judge whether or not it made sense
and say their judgment aloud while their
answer was recorded. This was the measure
of WM processing. They also had to
remember the last word of each sentence up
to the end of the set until a visual prompt
(three hash keys) along with a two-second
auditory prompt appeared on the computer
screen. The pilot study results suggested that
these two simultaneous prompts could well
put a clear boundary between the sets and
help the participants not to miss the recall
time. At this time, the participants had to
recall the sentence-final words and say them
out loud while their answers were recorded
by the researcher. This was the measure of
the WM storage component. To control the
recency effect, the participants were
required to recall the words in the order in
which they appeared (Baddeley & Hitch,
1993; Waters & Caplan, 1996).
A test instruction guide followed by an oral
explanation which included an example set
of three sentences was given to the
participants prior to the test. Then they were
given a practice session consisting of 10
sentences in two sets of three and a set of
four sentences. Then the test began with a
set of 3 sentences, and as the test progressed,
the number of sentences presented on each
trial increased successively from three to
six, with three trials being presented at each
series length. The prompt slide transitions
increased accordingly from 12 to 18 seconds
based on the length of each set.
Pilot studies
To identify the potential problems with the
RST, three pilot studies were administered
to three different groups of L1 Persian EFL
learners. In the first pilot, a group of 12 L2
participants completed the RST, followed by
a retrospective report. In their retrospective
report, they all reported that the transition
time, 6 seconds, for each slide was too short
to read through the sentence. They also
wrote that a few sentences were too vague
for them to determine whether they made
sense or not. The results of an item analysis
indicated that there were some poor test
items in the test. They were identified as
being too difficult. These results indicated
that the participants had performed poorly
on both the processing and recall
components. The sentences which the
students had identified as too vague were
located among the ones which had been
identified as too difficult by the item
analysis. In consultation with the Persian
language expert, these sentences were either
revised or replaced with new sentences.
Then the transition time for each slide was
increased to 8 seconds as well.
In the second pilot study, similar to the
procedure in the first pilot study, a group of
18 L1 Persian EFL learners completed the
revised RST followed by a retrospective
report. In their retrospective report, they
wrote that they had had sufficient time to
read through the sentence on each slide and
even rehearse the sentence final words
(target). They also reported a case where
two sentence final words were semantically
related, and they had been able to make an
association between them for better recall.
The results of this study supported the
participants’ claims. Their performance on
the RST was better than the prior group’s.
Most of them were also able to obtain the
scores for the two semantically related
targets. Since the participants’ rehearsing
could have inflated the recall scores, the
transition time for each slide was decreased
to 7 seconds. Furthermore, one of the
sentences including a semantically related
word was replaced with a new sentence
including a different target word. The new
sentence was developed and proposed by the
same Persian language expert.
In the third pilot study, the revised reading
span test was administered to a group of 44
participants. They reported that the
transition time for each slide was just
enough to read the sentence through and
decide whether it made sense or not. No one
reported any opportunity for rehearsing the
targets. Moreover, they believed that all
sentences throughout the test had been
neither too easy nor too difficult for them.
The results of the item analysis also
indicated that each item made a good
contribution to the test. The internal
reliability for this test, as indicated by
Cronbach’s Alpha, was .834 & .737 for the
RST recall and processing respectively.
Application of the newly developed
reading span test in L2 research
The final test was used in an experimental
study conducted by the researcher. This
study investigated the relationship between
WM and L2 reading ability on 140 L1
Persian EFL learners at three proficiency
levels. The sentences in the test were
arranged in three sets of 3, 4, 5, and 6
sentences. Half of the sentences in each set
were nonsense. Each sentence appeared on
screen for 7 seconds, when the computer
transitioned to the next slide. After each set,
a slide with 3 hash keys and a two-second
auditory prompt appeared. This was to
signal to the participants to recall the final
word of each sentence in the set.
To score the test, one mark was allocated to
the participants’ correct judgment and one
mark to their correct recall of the test session
items, with the total of 54 each. Thus, since
there were 54 sentences across all the trial
sets, the range of the participants’
processing and recall scores was between 0
and 54 for each participant. No marks were
given to the practice session items. This was
consistent with the scoring method in recent
studies (e.g., Alptekin & Erçetin, 2009).
Then a composite WM score was used as an
indicator of the participants’ WMC (e.g.,
Lesser, 2007; Waters & Caplan, 1996). The
composite WM was obtained by adding the
processing and recall z-scores. This is a
more reliable scoring method of WMC
compared to the traditional span scores that
quantify the highest set size completed or
the number of words in correct sets
(Freidman & Miyake, 2005). An item
analysis was conducted on this measure. The
internal reliability for this measure, as
indicated by Cronbach’s Alpha, was .844
and .790 for the RST processing and recall
respectively. This suggests that the newly
developed RST is reliable enough and could
be used for the measurement of WM in
future studies.
Conclusion
This study described developing an L1
Persian reading span test for the
measurement of L1 Persian EFL learners’
working memory capacity. The Persian
reading span test was developed, piloted and
successfully used in a study with 140
participants. As the internal reliability of this
measure was quite high, the test can be used
to measure working memory capacity in
future second language learning studies. The
same procedure could also be used to
develop a reading span test for speakers of
other languages.
