The relationship between working memory and L2 reading comprehension

Introduction
Language  learning  aptitude  is  one  of  the
sources of individual differences in second
language  learning  (Carroll,  1965;  Skehan,
1991).  Based  on  this  view,  individuals
may not have an undifferentiated talent for
learning  languages,  but  rather  a  multi-component  talent  from  which  each
component  may  vary  relatively
independently  from  the  others  (Skehan,
1991).  Recent  research  has  suggested  that
working  memory  (based  on  Baddeley  and
Hitch’s  1974  model)  may  form  a  central
component  of  second  language  learning
aptitude  (Miyake  &  Freidman,  1998;
Sawyer & Ranta, 2001; Skehan, 2002).
     
Working memory is defined as a cognitive
workspace (e.g., Baddeley & Hitch, 1974;
Baddeley,  2007)  with  a  limited  pool  of
attentional resources for temporary storage
and  processing  information  while
performing  higher  order  cognitive  tasks
such  as  comprehension,  learning  and
reasoning  (Baddeley  &  Logie,  1999).
Working  memory  is  comprised  of  four
components:  the  phonological  loop,  the
visuospatial  sketchpad,  the  central
executive,  and  the  episodic  buffer
(Baddeley,  2000a).  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  Ellis,  N.

C.  (2001),  “It  regulates  information  flow
within  working  memory,  activates  or
inhibits  the  whole  sequences  of  activities,
and  resolves  potential  conflicts  between
ongoing  schema-controlled activities” (p.,
33). The phonological loop is in charge of
temporary  storage  and  processing  of
verbal  information,  (Baddeley,  2007,
2000a;  N.  Ellis,  2001)  while  the
visuospatial  sketchpad  provides  an
interface  between  visual  and  spatial
information  received  either  through  the
senses  or  from  long-term  memory
(Baddeley & Hitch, 1974, p., 854). Finally,
the  episodic  buffer  acts  as  a  temporary
storage  space  where  information  from  the
other  components  are  integrated
(Baddeley, 2000a).
 
A  strong  body  of  research  implicates
working  memory  capacity  and  first
language  abilities  such  as  fluency  of
speech  (Daneman,  1991),  ability  to  learn
new words (Daneman & Green, 1986) and
reading  comprehension  (e.g.,  Daneman  &
Carpenter, 1980; Waters & Caplan, 1996).
Emerging  research  in  second  language
acquisition has linked working memory to
second  language  learning  in  areas
including  word  naming  and  vocabulary
learning  (Atkins  &  Baddeley,  1998),
online  parsing  performance  (Juffs,  2004),
interactional  feedback  (e.g.,  Ando,
Fukunaga,  Kurahachi,  Stuto,  Nakano,  &
Kage,  1992;  Mackey,  Philp,  Egi,  Fujii,  &
Tatsumi,  2002,  Mackey,  Adams,  Stafford,
&  Winke,  2010)  and  reading
comprehension  (e.g.,  Chun  &  Payne,
2004; Harrington & Sawyer, 1992; Lesser,
2007; Walter, 2004).  
 
Reading comprehension
Reading  is  “a  multifaceted,  complex
construct in that it consists of a number of
component  operations,  each  dependent  on
a  wide  range  of  competencies”  (Koda,
2005,  p.,  XV),  the  goal  of  which  is  to
construct  text  meaning  based  on  visually
encoded  information  (Koda,  2005,  p.,  1).
Successful  comprehension  occurs  when
extracted  text  information  interacts  with  a
reader’s  prior  knowledge  in  three  critical
operations:  decoding  the  linguistic
information  from  the  text,  integrating  the
extracted  information  into  phrases,
sentences  and  paragraphs,  and
synthesizing  text  information  with  prior
knowledge  (Koda,  2007,  p.,  4).  Because
reading is a complex cognitive function, it
is  likely  that  individual  learner  capacities,
like  working  memory,  may  influence
reading comprehension.
 
This  may  be  particularly  the  case  for
second  language  (L2)  reading,  because,
unlike  reading  in  the  first  language  (L1),
L2  reading  involves  dual-language
involvement  in  each  operation  (e.g.,
Beach, 1997; Cain & Oakhill, 2006; Grabe
&  Stoller,  2002;  Koda,  2005).  Compared
to most L1 readers who begin to read with
considerable tacit language knowledge, L2
readers  have  a  much  wider  range  of
language  proficiency  when  they  start
learning  to  read.  Moreover,  the  cognitive
and  linguistic  resources  accessible  to  L2
readers vary considerably more than those
available  to  L1  readers  (Grabe,  2009;
Koda,  2005).  Therefore,  one  of  the  major
areas  of  differences  between  L1  and  L2
reading  lies  in  the  linguistic  and  cognitive
processing  domain.  Readers  may  have
differing  amounts  of  lexical,  grammatical
and  discourse  knowledge  of  initial  stages
of  L1  and  L2  reading,  greater
metalinguistic  and  metacognitive
awareness  in  L2  settings,  differing
amounts  of  exposure  to  L2  reading,
varying  linguistic  differences  across  any
two languages, varying L2 proficiency as a
foundation  for  L2  reading,  and  varying
language  transfer  influences  and
interacting  influence  of  working  with  two
languages  (e.g.,  Beach,  1997;  Cain  &
Oakhill,  2006;  Grabe  &  Stoller,  2002;
Koda, 2005).
 
Role  of  working  memory  in  L2  reading
comprehension
The  research  on  the  development  of
reading  comprehension  skills  and  sources
of individual differences in comprehension
indicates a strong relationship between  L1
reading skills and cognitive variables such
as  working  memory  (Just  &  Carpenter,
2002) and inhibitory control (Gernsbacher,
Varner  &  Faust,  1990).  Since  working
memory is considered a mental workspace
where  the  processes  of  retrieving,
integrating,  updating  and  revising  of
information  is  performed,  it  plays  an
important  role  in  understanding  a  text.
First,  to  identify  the  words,  the  reader
needs  to  recode  written  symbols  into
phonological  codes  to  allow  for
computations  to  recognize  linguistic
structure.  Then,  they  develop  a  coherent
and  integrated  representation  of  the
concepts  through  making  links  between
successive  sentences.  This  requires  the
reader  to  maintain  the  recently  read
material  in  working  memory  to  make
inferences  (Schmalhofer,  McDaniel,  &
Keefe  2002),  while  simultaneously
processing  the  same  or  other  information
either  recoded  from  the  text  or  retrieved
from  the  long-term  memory.  Finally,
working  memory  plays  a  role  as  a  buffer
of  the  just  read  propositions  in  a  text,  and
the  information  retrieved  from  the  long-term  memory  to  both  establish  a  local
coherence between sentences and facilitate
its  integration  with  the  activated
background knowledge (e.g., Beech, 1997;
Graesser, Singer, & Trabasso, 1994).
 
L1 research suggests an important role for
working memory in first language reading
(e.g.,  Daneman  &  Green,  1986;  Waters  &
Caplan,  1996).  In  the  same  line,  research
on  the  role  of  working  memory  is
emerging as an area of concern for second
language  acquisition  (e.g.,  Miyake  &
Freidman,  1998;  Robinson,  1995,  2002,
2005).  However,  little  is  known  about  the
role of working memory in the processing
of  second  language  learning  in  general,
and  in  reading  comprehension  in
particular.  Prior  L2  studies  on  reading
present evidence of a relationship between
working  memory  and  reading
comprehension (Alptekin & Erçetin, 2009;
Chun  &  Payne,  2004;  Harrington  &
Sawyer,  1992;  Lesser,  2007;  Walter,
2004).  For  example,  Harrington  and
Sawyer  (1992)  investigated  the
relationship  between  L2  working  memory
capacity  and  L2  reading  among  Japanese
learners  of  English.  The  subjects  were
required to complete three memory tests in
L1  and  L2  as  well  as  L2  English  reading
comprehension.  The  memory  assessment
consisted  of  digit  span,  word  span,  and
reading  span  tests.  L2  English  reading
comprehension  consisted  of  the  grammar
and  reading  sections  of  the  TOEFL  and  a
cloze  passage.  Results  indicated  a
significant,  strong  correlation  between
working  memory  capacity  (L2  reading
span),  and  both  TOEFL  reading  (r=.54)
and  TOEFL  grammar  (r=.57).
Furthermore, there was a weak correlation
between  L2  reading  span  and  cloze
passage,  (r=.33).  However,  no  significant
correlations were found between digit span
and  word  span  measures  on  the  one  hand
and L2 English reading comprehension on
the other.
 
Walter  (2004)  examined  the  question  of
whether  the  transfer  of  reading
comprehension  skill  from  L1  to  L2  is
linked  to  the  development  of  verbal
working  memory  in  L2,  which  turned  out
to  take  place  at  a  much  lower  level  of  L2
proficiency  than  that  found  by  Harrington
and  Sawyer  (1992).  Two  groups  of  L1
French  ESL  learners  participated  in  her
study.  The  first  group  consisted  of  19
lower-intermediate ESL learners, while the
second  group  consisted  of  22  upper-intermediate ESL learners. Three measures
were  administered  by  Walter  (2004),  each
in both languages (French and English): 1)
a  baseline  comprehension  assessment
where  the  participants  were  required  to
complete  a  gapped  summary  of  the  story
 
they had just read, 2) a pro-form resolution
test  where  the  participants  were  told  to
read  a  story  and  stop  when  they
encountered  an  expression  printed  in  red,
then read the word aloud, tell the meaning
of the word, and identify its first mention,
3)  a  verbal  working  memory  measure
where  the  participants  were  asked  to  read
an  increasingly  longer  sets  of  sentences
and  judge  if  they  were  logical  or  illogical
and  then  recall  the  sentence-final  words
across the sets.  
 
The  results  indicated  a  significant
correlation  between  working  memory
scores and L2 summary completion scores.
However,  the  correlation  was  higher  for
lower-intermediate  group  (.79***,  P  <
.0001)  than  for  upper-intermediate  group
(.46**,  P  <  .01).  This  implies  that  the
lower-intermediate  group’s  success  in
summary  completion  tasks  relied
significantly  on  their  working  memory
capacity.  This  supports  the  idea  that  there
is  a  link  between  the  development  of
verbal working memory in L2 and success
in  L2  reading  comprehension.  This  study
also  revealed  that  success  of  upper-intermediate  group  in  L2  reading
comprehension  relied  more  on  reading
skills  (the  ability  to  build  well-structured
mental  representations  of  texts)  than  on
working  memory.  These  findings  suggest
that  the  role  of  working  memory  in  L2
reading  development  may  be  mediated  by
L2 proficiency.
 
Recent  research  by  Alptekin  and  Erçetin
(2009)  provides  additional  evidence  of  a
mediating  role  of  proficiency  in  the
relationship  between  working  memory
capacity  and  L2  reading  ability.  In  their
study,  30  L1  Turkish  undergraduate
students  with  advanced  L2  English
proficiency  (enrolled  in  English  language
teaching  courses)  were  required  to
complete  two  working  memory  measures
and  a  reading  comprehension  test.  Results
of  their  study  indicated  a  moderately
significant  correlation  (r=  .40*,  P  <.05)
between  scores  on  one  working  memory
measure  and  learner  ability  to  make
inferences  in  the  texts.  However,  no
further  relationships  were  found  among
working  memory  and  reading  measures.
These  findings  raise  questions  about  the
importance  of  working  memory  in  L2
reading at higher levels of proficiency.
 
As  well  as  proficiency,  Lesser’s  (2007)
research  suggests  that  the  role  of  working
memory  in  L2  reading  may  also  be
moderated  by  prior  content  knowledge.  In
his  study,  94  high  beginner  L2  Spanish
learners completed a computerized version
of  an  L1  RST  as  a  measure  of  working
memory  capacity, a  recall protocol task to
measure passage comprehension, and form
recognition and tense identification tests to
determine  processing  of  future  tense
morphology.  The  results  of  the  study
suggested  that  topic  familiarity  was  an
important  factor  in  L2  reading
comprehension  as  it  played  a  significant
role in beginning L2 readers’ recognition
of  target  forms  and  their  ability  to  make
form-meaning  connections.  Working
memory  also  played  a  significant  role  in
learners’ comprehension and processing of
grammatical form, depending on the extent
to which it interacted with learners’ prior
knowledge.  A  more  significant  role  of
working  memory  in  reading
comprehension  was  observed  as  the
participants’  prior  knowledge  about  text
topic increased.
 
It  should  be  noted  that  other  studies  have
not  uncovered  connections  between
working  memory  and  reading
comprehension.  For  example,  Chun  &
Payne  (2004)  examined  the  role  of
individual  differences  in  the  L2  German
reading  comprehension  and  vocabulary
acquisition  of  13  L1  English  students  in  a
second  year  German  language  course.  A
computer-delivered  version  of  Daneman
and Carpenter’s (1980) L1 RST as well as
a  non-word  repetition  task  were  used  to
measure  working  memory.  A  German
 
short  story,  including  four  sets  of
comprehension  exercises  followed  by  a
recall  protocol,  was  used  as  a  measure  of
reading  comprehension.  The  results
indicated  a  strong  relationship  between
phonological working memory capacity as
measured  by  word  recognition  based  on
non-word repetition and look-up behavior,
measured  as  the  number  of  annotations
which  had  been  looked  up  and  recorded
while  reading  an  L2  text.  Learners  with
low  phonological  short-term  memory
capacity  looked  for  an  average  of  three
times  more  words  than  participants  with
high  phonological  short-term  memory
capacity. However, they did not report any
significant  findings  for  working  memory
on  any  of  the  comprehension  or
vocabulary acquisition measures
 
Role  of  phonological  short-term
memory in L2 reading comprehension  
A  considerable  body  of  evidence  suggests
that  PSTM,  as  a  component  of  WM,  may
be  an  essential  cognitive  mechanism
underlying  successful  L2  reading  (e.g.,
Masoura  &  Gathercole,  1999,  2005;
Papagno,  Valentine  &  Baddeley,  1991;
Service  1992,  Service  &  Craik,  1993;
Service  &  Kohonen,  1995).  For  example,
in  a  longitudinal  study  that  lasted  for  four
years, Service (1992) examined the role of
PSTM  in  English  as  a  foreign  language
learning  of  44  L1  Finnish  primary  school
students.  PSTM  was  measured  through  a
pseudoword repetition task conducted each
year  of  the  study.  In  each  task,  the
participants  were  required  to  listen  to  two
lists of pseudowords, an English-sounding
list and a Finnish-sounding one, and repeat
aloud  the  pseudowords  they  heard  as
quickly  as  possible.  Service  (1992)  found
a  strong  relationship  between  PSTM,  as
measured  by  the  English-sounding
pseudoword lists, at the start of the English
instruction and the performance on tests of
listening,  reading  comprehension  and
writing 2.5  years later. She also suggested
that  PSTM  underlies  the  acquisition  of
new  vocabulary  items  in  a  foreign
language.
 
In  a  follow-up  longitudinal  study,  Service
and  Kohonen  (1995)  investigated  whether
the  relationship  between  PSTM  and
foreign language learning is accounted  for
by  vocabulary  acquisition.  They  recorded
42  (9-10  year-old)  Finnish  participants'
performance  on  pseudoword  repetition,  as
a measure of PSTM, over four consecutive
years.  They  also  recorded  the  participants'
performance  on  different  individual  L2
English tasks during the fourth year of the
longitudinal  study.  These  tasks  measured
the  participants'  L2  reading,  listening,
writing,  vocabulary  and  knowledge  of
grammatical  structures.  Their  regression
analyses on pseudoword repetition and L2
tasks  revealed  significant  correlations
between  pseudoword  repetition  and
foreign  language  learning,  even  after  a
measure  of  general  academic  achievement
had been partialed out. By varying second-step  factors  in  their  regression  analysis,
they were able to show that L2 vocabulary
performance  and  pseudoword  repetition
accounted  for  the  same  variance  in
performance  for  foreign  language
measures.  Service  and  Kohonen  (1995)
interpret  these  findings  as  an  indication
that PSTM influences vocabulary learning,
which  in  turn  influences  success  in  other
areas  of  L2  performance.  This  data
provides  evidence  of  a  specific
relationship  between  PSTM  (as  measured
by  pseudoword  repetition)  and  vocabulary
learning.
 
Furthermore,  Masoura  and  Gathercole
(2005)  found  an  important  role  for  PSTM
in  the  L2  English  vocabulary  learning  for
Greek  children.  They  investigated  the
contributions  of  PSTM  and  existing
foreign  vocabulary  knowledge  to  the
learning  of  new  English  words.  Their  L1
Greek  children  completed  a  paired-associate learning task as a measure of L2
vocabulary  learning,  two  non-word
repetition tasks as measures of PSTM, and
 
a  nonverbal  ability  task.  Masoura  and
Gathercole (2005) found that PSTM made
a  large  contribution  to  L2  vocabulary
learning  at  earlier  stages  of  L2  learning,
but  as  the  familiarity  with  L2  knowledge
increased,  the  existing  L2  knowledge
played  a  mediating  role  in  L2  vocabulary
learning.
 
However,  Kormos  &  Sáfár  (2008)  found
no  significant  correlation  between  PSTM
and  L2  proficiency.  They  investigated
whether  there  is  a  relationship  between
PSTM  and  WMC  and  performance  in  L2
language  skills,  as  measured  by  an  L2
proficiency  test.  They  asked  121
secondary  school  students  to  complete  a
non-word  repetition  test,  a  Cambridge
First  Certificate  Exam,  and  a  backward
digit  span  test  after  an  intensive  language
training  program.  Their  results  indicated
that  there  was  no  significant  correlation
between PSTM and L2 language skills, but
there was a significant correlation between
WMC,  as  measured  by  a  backward  digit
span  test  and  L2  language  skills  (reading,
listening,  and  speaking),  with  the
exception  of  writing.  Kormos  &  Sáfár
(2008)  suggested  that  PSTM  and  WM  are
distinct  constructs,  and  play  a  different
role  in  instructed  second  language
acquisition.
 
Purpose of the study
Although  the  studies  reviewed  above,
except  for  Chun  and  Payne  (2004),
provide  evidence  for  a  relationship
between  working  memory  and  L2  reading
comprehension,  there  has  been  little
attention  paid  to  the  role  of  working
memory  in  reading  comprehension  across
language  proficiency  levels.  The
proficiency  of  the  participants  in  these
studies  varies  from  the  advanced  level
(Alptekin & Erçetin, 2009; Harrington and
Sawyer,  1992)  to  the  intermediate  (Chun
&  Payne,  2004),  upper  and  lower
intermediate  (Walter,  2004)  and  high
beginning levels (Lesser, 2007).  
 
Similarly,  there  are  limitations  with  the
prior  studies  on  the  role  of  PSTM  in  L2
reading. These studies differed in terms of
the  findings  and  the  proficiency  level  of
the participants used. While some research
has  linked  PSTM  to  L2  reading  ability
(e.g.,  Masuora  &  Gathercole,  2005;
Service,  1992),  some  others  found  no
relationship  between  PSTM  and  L2
reading ability (e.g., Harrington & Sawyer,
1992;  Kormos  &  Sáfár,  2008).
Furthermore, none of the prior studies has
examined the role of PSTM in L2  reading
comprehension  across  language
proficiency levels.
 
 These  limitations  and  differences  in
research  findings  point  to  the  need  to
examine the relationship between WM and
PSTM  and  L2  reading  ability  across
proficiency  levels.  Thus,  the  current  study
was  designed  and  proposed  the  following
research  questions  to  investigate  whether
WM  and  PSTM  influence  L2  reading
ability at different levels of proficiency.
 
Research questions
This  study  has  been  designed  to  address
the following questions:
 
1-  What  is  the  relationship  between
learners’ working memory and L2
reading  comprehension  across
proficiency levels?
2-  What  is  the  relationship  between
phonological  short-term  memory
and  L2  reading  comprehension
across proficiency levels?
 
These  research  questions  were  developed
based  on  the  hypothesis  that  working
memory  capacity  and  L2  proficiency  play
important  roles  in  different  stages  of
second  language  reading  development.
Prior  L2  studies  suggest  that  individual
differences  in  working  memory  capacity
predict reading ability at lower proficiency
levels  (e.g.,  Harrington  &  Sawyer,  1992;
Lesser,  2007;  Walter,  2004).  They  all
suggest  that  learners  with  higher  working

 
memory  capacity  outperform  those  with
lower  working  memory  capacity  on  given
reading  tasks.  This  may  be  that  these
learners rely more on memory resources in
processing  and  arriving  at  the  text
meaning. However, as their L2 proficiency
develops,  their  automaticity  in  processing
the text meaning will increase, as well, and
consequently  their  reliance  on  memory
resources  will  decrease.  The  more  the
participants  develop  L2  proficiency,  the
more  they  draw  on  automaticity,  and  the
less  they  rely  on  memory  resources.  Then
this research was designed to examine this
hypothesis  and  determine  how  the  role  of
working  memory  in  L2  reading  ability  is
mediated  by  L2  proficiency  level  at
different stages of development.
 
Methodology
Subjects
Fifty  five  L1  Persian  EFL  learners  at
beginning,  intermediate,  and  advanced
level  participated  in  the  study.  As
indicated  in  Table  1,  they  included  both
males  and  females,  with  the  mean  age  of
19, studying English as a foreign language
in a private language school. Identification
of  the  proficiency  level  of  the  participants
was  based  on  the  Kanoon  language
institute  (KLI)  Placement  test,  a  test  used
in  the  language  school  where  the
participants  were  selected.  This  test  was
originally  developed  to  identify  the
optimal  level  for  students  entering
programs  where  KLI  books  and  materials
are taught. This test includes 120 multiple-choice  questions  measuring  the
participants’  L2  proficiency  in  listening
and  reading  skills  as  well  as  vocabulary
and grammar and is completed within two
hours.  One  point  is  allocated  to  each
correct  answer  with  the  total  of  120  for
this  test.  The  range  of  cut  off  scores  for
placing the participants into three levels of
proficiency is 0-40, 41-80, and 80-120 for
beginning,  intermediate  and  advanced
levels  respectively.  The  participants  had
taken  the  test  two  weeks  before  this  study
was  conducted.  As  indicated  in  Table  1
fewer  students  were  enrolled  in  the
advanced  level  course  at  this  school,  so
this group was the smallest.

Materials
The  participants  each  completed  a  battery
of  reading  and  memory  measures  over  a
total  of  an  hour  and  a  half.  Some  of  the
measures  were  administered  in  a  group
setting,  others  one-on-one  with  one  of  the
researchers.  The  reading  measure
consisted  of  two  reading  passages  at  each
level  of  proficiency.  The  memory
measures  consisted  of  a  RST  and  non-word  recognition  task.  All  participants
followed the same order in completing the
tests.  
 
Memory measures
Reading  Span  Test.  A  RST  was  first
introduced  by  Daneman  &  Carpenter
(1980).  It  was  used  to  measure  working
memory  capacity  and  give  an  index  of
processing and storage, the components of
working  memory.  In  this  test,  the
participants  are  asked  to  read  sets  of
sentences  and  report  on  semantic
acceptability  of  each  sentence  (processing
assessment)  and  then  recall  the  final  word
of  each  sentence  when  prompted  (storage
assessment).  As  a  measure  of  working
memory  capacity,  this  test  has  been  used
in  several  studies  (e.g.,  Chun  &  Payne,
2004;  Daneman  &  Carpenter,  1980;
Harrington & Sawyer, 1992; Lesser, 2007;
Light & Anderson, 1985; Osaka & Osaka,
1992;  Osaka,  Osaka,  &  Groner,  1993;
Swanson, 1994).
 
In  this  study,  a  Persian  RST  was  used  to
measure  working  memory  capacity.  This
was  based  on  the  prior  research  that
indicates  working  memory  is  language
 
independent  (Osaka  &  Osaka,  1992;
Osaka,  Osaka  &  Groner,  1993).
Furthermore,  measuring  working  memory
in  L1  could  help  to  avoid  conflating
working memory and L2 proficiency. This
test  was  developed  by  one  of  the
researchers,  and  its  problems  were
identified  and  removed  over  three  pilot
studies.  The  test  was  designed  with  64
items. For each item, the participants were
required  to  judge  whether  the  sentence
made  sense  or  was  not  and  also  to
remember  the  final  word.  After  sets  of  3,
4,  5,  or  6  sentences,  the  participants  were
asked  to  recall  the  final  words  and  write
them down in correct order in their answer
sheets.  The  test  was  administered  using  a
projector in for full-classes.
 
Non-word  recognition  test.  A  non-word
recognition  task  was  used  to  measure
phonological  short  term  memory
Phonological-short  term  memory  is  in
charge  of  temporary  storage  and
processing  of  verbal  information  (e.g.,
Baddeley, 2000a, 2007; Baddeley & Hitch,
1974;  N.  Ellis,  2001).  It  is  a  separate
construct  from  working  memory  (Juffs  &
Harrington,  2001).  Research  suggests  that
learning the sound structures of new words
in  L2  is  mediated  by  phonological  short-term  memory  (e.g.,  Gathercole  &
Baddeley,  1990a;  Miyake  &  Freidman,
1998;  Valler  &  Papango,  2002;  Skehan,
1989).  Measures  of  phonological  short-term  memory  are  commonly  included  in
studies  of  working  memory  and  second
language  learning  (e.g.,  Chun  &  Payne,
2004;  Harrington  &  Sawyer,  1992;
Mackey,  Philp,  Egi,  Fujii,  &  Tatsumi,
2002; Trofimovich, Ammar, & Gatbonton,
2007).
 
This  test  was  based  on  recommendations
by  Gathercole  and  her  colleagues  (e.g.,
Gathercole & Baddeley, 1989; Gathercole,
Frankish,  Pickering,  Peaker,  1999;
Gathercole,  Pickering,  Hall  &  Peaker,
2001;  Gathercole,  Willis,  Emslie  &
Baddeley,  1991)  for  the  most  valid
measure  of  phonological  short-term
memory.  This  test  consisted  of  sequences
of  English  non-words.  Non-words  were
used  to  minimize  the  influence  of
vocabulary  knowledge  on  phonological
short-term  memory  and  yield  a  relatively
accurate  estimate  of  it  (Gathercole  &
Pickering, 2001). Following the procedure
used  by  Trofimovich  et  al.,  (2007)  the
participants  heard  two  consecutive
sequences  of  non-words  and  judged
whether they were in the same or different
order.  The  length  of  each  sequence  was
gradually increased across the pairs within
the range of 4 to 7 non-words.  
 
As a result, the test was administered to 55
participants  at  three  proficiency  levels  in
this  study.  This  test  was  conducted  in  full
class, and the participants were required to
listen  to  each  pair  of  sequences  to
determine  if  the  order  of  non-words  in
both  sequences  was  the  same  or  different
by checking the boxes next to each choice
in their answer sheets.
 
Reading measure
At  each  level  of  proficiency,  two  reading
passages  were  selected  from  the  language
school’s syllabus material resources where
the  difficulty  level  of  the  passages  had
been  controlled  for  each  level  of
proficiency.  The  participants  reported  that
they  had  not  viewed  the  passages  prior  to
the  study.  The  reason  for  choosing  two
passages was to minimize the effect of the
participants’  background  knowledge  as
well  as  text  genres  on  comprehension
(Alderson,  2000;  Beach,  1997;  Koda,
2005;  Mitchell,  1982;  Nation,  2009).  In
post  study  de-briefing  questionnaires,
participants claimed that they had not been
familiar with the text content. All passages
included  social  and  science  topics
followed  by  5  multiple-choice
comprehension  questions  each.  The
multiple-choice  comprehension  questions
included  both  literal  and  inferential  types
of  comprehension  questions  to  assess  the
 
participants’  lower  and  higher  level
processing of information respectively.  
Scoring
The  participants’  raw  scores  on  each
memory  and  reading  measure  were
calculated.  To  score  RST,  one  mark  was
allocated  to  correct  judgement  and  one
mark to their correct recall with the total of
54  each.  To  control  the  recency  effect,  no
marks were given to the last target of each
set if it was recalled first (Turner & Engle,
1989;  Waters  &  Caplan,  1996).  This  has
proved  to  be  a  reliable  method  of  scoring
(Convey,  Kane,  Bunting,  Hambrick,
Wilhelm,  Engle,  2005).  (Recency  effect
influence  recall  of  information.
Essentially,  the  most  recently  presented
items  will  most  likely  be  recalled  best.
Thus, controlling this effect may give us a
more  accurate  estimate  of  working
memory capacity.) Furthermore, to control
any  trade-off  between  the  working
memory  components,  a  composite
working  memory  z-scores  was  computed
as  an  indicator  of  working  memory
capacity  (e.g.,  Alptekin  &  Erçetin,  2009;
Lesser,  2007;  Turner  &  Engle,  1989;
Waters & Caplan, 1996).  
 
To  score  non-word  recognition  test,  one
mark was allocated to each correct answer
with  the  total  of  22  for  this  measure.  The
scores  in  this  measure  were  placed  at  a
wide  range  within  each  group.  They  were
situated  in  the  ranges  of  7-21,  10-21,  and
9-21  for  the  beginning,  intermediate  and
advanced  groups  respectively.  To  score
reading measures, one mark was allocated
to  each  correct  answer.  Since  the  number
of  passages  and  questions  were  consistent
across  proficiency  levels,  each  participant
was  expected  to  obtain  a  reading  score
within  a  range  of  0-10.  Results  indicated
relatively  wide  ranges  in  the  reading
scores  within  the  groups,  particularly  at
the beginning and intermediate levels. The
scores  at  these  two  levels  ranged  from  3-10 and 4-10 respectively. To be consistent
throughout  the  study,  the  scores  of  non-word  recognition  test  and  reading
measures  were  converted  into  z-scores
before the inferential analysis is obtained.  
 
Results
Descriptive  statistics  across  proficiency
levels
Descriptive  statistics  for  memory  and
reading  z-scores  at  each  level  of
proficiency are indicated in Table 2 which
report  mean,  standard  deviation,  range,
minimum,  maximum  and  number  of
participants for each proficiency level.

As  indicated  in  Table  2,  the  means  and
standard deviations at the intermediate and
advanced  levels  are  nearly  the  same,  and
different from those at the beginning level.
Furthermore,  the  results  of  one-way
ANOVA indicated a significant difference
between  the  beginning  and  intermediate
groups’ performance on working memory
capacity, as measured by RST (F (2, 52) =
5.87,  P=.005;  P<.05).  However,  they
indicated an overall similarity between the
groups  on  PSTM  (F  (2,  52)  =  2.27,
P=.113, P<.05), suggesting that there were
no significant differences on participants’
performance on PSTM here.
Working memory and reading
In  order  to  weight  tests  equally,  z-scores
were  calculated  for  all  measures.  Then
correlations  between  and  within
explanatory and responsive variables were
obtained. To control any trade-off between
the  components  of  working  memory  and
provide  a  more  stable  measure  of  the
working  memory  capacity,  composites
were  created  from  unit-weighted  z-scores
of  storage  and  processing  measures
through  averaging  the  sum  of  storage  and
processing  z-scores  (e.g.,  Lesser,  2007;
Turner  &  Engle,  1989;  Waters  &  Caplan,
1996).  Table  3  displays  the  results  of  the
correlation  analysis.  Only  one  significant
correlation  was  found  in  these  data,
between  the  RST  scores  and  reading
comprehension at the beginning level. This
was a moderate, positive correlation. 

Regressions  were  run  where  there  was  a
statistically significant correlation between
explanatory  and  responsive  variable.  This
was  to  see  how  much  the  explanatory
variable  makes  contribution  to  the
prediction  of  the  responsive  variable.
Overall,  one  regression  was  run  at  the
beginning  level.  Summary  results  are
presented in Table 4.

To  determine  how  much  effect  can  be
attributed  to  the  influence  of  working
memory on the reading measure, the effect
size  for  the  model  was  obtained  from  the
R²  value  in  the  regression  model.    As
displayed  in  Table  4,  R²  value  indicated
that RST composite  accounted for 25% of
the variance for the reading measure. This
model had a β of .50, indicating that each
increase  of  1  point  in  the  RST  composite
scores  predicted  a  half  point  increase  in
the reading measure scores.
 
General discussion
Research question 1
The  first  research  question  addressed  the
relationship between working memory and
L2  reading  comprehension  and  also  if  the
relationship  differed  at  different
proficiency  levels.  The  results  provide
limited evidence of a relationship between
working  memory  capacity  and  L2  reading
comprehension.  There  was  a  significant
correlation  between  the  reading  measure
and  working  memory  composite  (r=.501*,
P<.05)  at  beginning  level.  Regression
results  also  indicated  that  RST  could  be  a
good  predictor  of  L2  learners’  reading
performance  at  beginning  level.  The  large
effect  size  found  here  suggests  that
working memory (as measured by RST) is
a  good  predictor  of  individual’s
performance in the reading comprehension
tests.  These  results  suggest  that  working
memory capacity, as measured by RST, is
a potential source of individual differences
in  explaining  L2  reading  ability  at  the
beginning  level.  Indeed,  the  analysis
suggested  that  a  quarter  of  the  variance
among  learners  at  this  level  may  be
explained  by  differences  in  working
memory  capacity.  This  signals  in  an
important  role  for  working  memory  in
beginning  level  L2  reading.  This  suggests
that,  at  lower  proficiency  levels  where
much of the language processing may still
be  controlled  and  effortful,  learners  with
higher  cognitive  resources  may  have  an
advantage.   
 
These  results  provide  further  evidence  for
that  working  memory,  as  measured  by
RSTs,  plays  a  role  in  reading  ability  both
in  L1  (e.g.,  Daneman  &  Carpenter,  1980;
 
Waters & Caplan, 1996; Just & Carpenter,
1992) and  L2 (e.g., Harrington & Sawyer,
1992;  Lesser,  2007;  Walter,  2004).  The
results  also  provide  evidence  for  claims
that  the  relationship  between  working
memory capacity and reading ability might
be  most  important  at  the  beginning  levels
of  language  ability  (e.g.,  Ellis  &  Sinclair,
1996;  Temple,  1997,  Williams  &  Lovett,
2003).  
 
However, as the proficiency increases, the
relationship between working memory and
L2  reading  disappears.  This  suggests  that
low  proficiency  L2  learners  rely  on
working  memory  more  than  high
proficiency  L2  learners  during  reading
tasks.  This  supports  Temple’s  (1997)
proposal  that  working  memory  plays  an
especially  important  role  in  early  L2
learning. It is also consistent with findings
of  previous  study  conducted  by  Lesser
(2007)  that  working  memory  plays  an
important  role  in  beginning  Spanish
learners’  comprehension  as  well  as
Walter’s  (2004)  findings  that  for  lower-proficiency  L1  French  ESL  learners  the
transfer  of  reading  comprehension  skill
from  L1  to  L2  relies  on  verbal  working
memory.  She  suggested  that  success  of
higher-proficiency  learners  relied  more  on
reading skills than on working memory. It
seems  that  for  higher  level  learners,  with
greater  language  knowledge  and  greater
automaticity  in  the  reading  process,  the
reading tasks presented less of a burden on
working  memory  than  for  lower  level
learners  who  relied  more  on  memory
capacity.  At  this  point  then,  differences  in
working  memory  capacity  no  longer  lead
to  differences  in  reading  comprehension.
Rather,  as  proficiency  develops,  language
knowledge  takes  the  major  role  in
extracting  the  text  information  (Alderson,
2000;  Cain  &  Oakhill,  2006;  Grabe  &
Stoller,  2002;  Koda,  2005;  Leslie  &
Caldwell,  2009),  perhaps  because  of
greater  automaticity  in  language
processing  at  higher  levels  of  proficiency.
This  suggests  a  dual  view  of  individual
reading  comprehension  in  a  second
language;  at  the  beginning  levels  of
proficiency, learners with greater cognitive
capacity  may  be  better  readers  while  at
higher  levels,  learners  with  greater
language  knowledge  may  be  better
readers.  The  more  fluent  the  learners  are,
the  more  automatic  their  processes,  and
the  less  memory  demanding  L2  reading
will be.  
 
In  summary,  RST,  as  a  working  memory
measure, was found to be a good predictor
of  reading  comprehension,  but  only  for
low  proficiency  learners.  In  other  words,
individual  differences  in  beginners’
working  memory  capacity  may  play  an
important  role  in  their  reading  ability.  As
the  results  of  this  study  and  prior  studies
(e.g.,  Harrington  &  Sawyer,  1992;  Lesser,
2007;  Walter,  2004)  suggest  beginning
level  learners  with  high  working  memory
capacity  outperformed  the  learners  with
low  working  memory  capacity  in  their
reading ability.   
 
Research question 2
The second research question looked at the
relationship  between  PSTM  and  L2
reading.  It  also sought to determine if this
relationship  differed  across  proficiency
groups.  The  results  of  the  study  indicated
no  significant  correlation  between  this
variable  and  reading  measures  at  each
level  of  proficiency.  PSTM  is  a  poor
predictor  of  the  participants’  reading
ability.  These  findings  may  reflect  earlier
evidence that PSTM plays a mediating role
in  reading  comprehension,  possibly
limited  to  vocabulary  acquisition  both  in
first  and  second  language  (e.g.,  Baddeley,
Papagno,  &  Vallar,  1998;  Gathercole  &
Baddeley,  1989;  Gathercole  &  Baddeley,
1990)  which  in  turn  impacts  on  reading
comprehension  (Cheung,  1996;
Gathercole,  Willis,  Emslie  &  Baddeley,
1991;  Masuora  &  Gathercole,  2005).  This
may be why no direct relationship between
PSTM  and  reading  measures  was  found.
These  findings  are  consistent  with  prior
 
research  in  which  PSTM  did  not  explain
individual  differences  in  L1  (e.g.,
Daneman  &  Carpenter,  1980)  and  L2
reading  comprehension  (Harrington  &
Sawyer, 1992; Kormos & Sáfár, 2008).
 
Limitations of the study and suggestions
for further research
There  was  a  relatively  small  sample  size,
particularly  at  advanced  level.  Further
research  among  a  large  group  of
participants  could  provide  a  more  reliable
view  of  the  role  of  proficiency  in  the
relationship between working memory and
L2 reading. Second, in this study, memory
measures  included  a  PSTM  test  and  a
working  memory  capacity  measure,  RST.
Both  of  these  tests  were  based  on  verbal
processing.  Further  research  may  include
non-verbal  measures  of  working  memory,
for  example,  a  math  span  test,  to  more
accurately  measure  working  memory
capacity.  Finally,  a  broader  testing  battery
of  reading  measures,  beyond  multiple-choice  testing,  may  better  illuminate  the
relationship between working memory and
L2 reading comprehension.
 
Conclusion
This  study  examined  whether  the
relationship  between  WM  and  L2  reading
ability  differs  across  proficiency  levels.
Similar to prior studies (e.g., Harrington &
Sawyer,  1992;  Lesser,  2007;  Walter,
2004),  the  present  study  indicated  that
there  is  a  relationship  between  WM  and
L2  reading  ability.  However,  this  study  is
distinguished from the prior studies in that
it  adds  a  unique  theoretical  implication  to
the  research  area  of  WM  and  L2  reading
ability.  
 
The  implication  of  this  study  is  that  the
relationship  between  WM  and  L2  reading
ability  differs  according  to  proficiency
level.  The  findings  here  suggest  that
working memory capacity can well predict
participants’  reading  ability  at  the
beginning  level.  At  higher  proficiency
levels,  other  factors  such  as  language
knowledge  may  play  an  important  role  in
predicting reading ability.
 
The  present  study  also  investigated
whether  the  relationship  between  PSTM,
as  a  component  of  WM,  and  L2  reading
ability is mediated by L2 proficiency. This
study  provides  further  support  for
Harrington  &  Sawyer’s  (1992)  and
Kormos  &  Sáfár’s  (2008)  studies,  which
suggest  that  PSTM  does  not  play  a  direct
role in L2 reading ability. This is likely to
be  because  simple  processing  in  PSTM
(articulatory  rehearsal)  may  not  be  a  good
predictor  for  multi-level  processing  in  L2
reading ability.
 
The  findings  of  this  study  should  be
considered  as  preliminary  steps  in
exploring  the  relationship  between
working  memory  capacity  and  L2  reading
comprehension,  providing  new  directions
for  further  studies  in  this  area.  These
studies  could  examine  the  relationship
between working memory capacity and the
lower  and  higher  level  reading  processes.
Further  studies  could  also  investigate
whether  there  is  a  relationship  between
PSTM  and  L2  reading  comprehension  by
the  mediating  role  of  L2  vocabulary
development.