Written word recognition by the elementary and advanced level Persian-English bilinguals

Introduction
As  about  half  of  the  world’s  population
know  more  than  one  language  (Grosjean,
1982),  research  on  bilingualism  has  been
on  the  rise  (Costa,  Santesteban,  &  Cano,
2005;  Kroll,  Bobb,  &Wodniecka,  2006;
Kroll  &  Curley,  1988;  Kroll  &  Stewart,
1994; Ju& Luce, 2004; Lagrou, Hartsuiker,
&Duyck,  2011;  Marian  &  Spivey,  2003;
Potter, So, Von Eckardt, & Feldman, 1984;
Schulpen,  Dijkstra,  Schriefers,  &Hasper,
2003;  Spivey  &  Marian,  1999;  Weber  &
Cutler,  2004;  Weinreich,  1953).  Two  main
models explore the mental representation of
languages in a bilingual mind. According to
traditional  models,  two  separate  lexicons
exist,  one  for  L1  and  the  other  for  L2.
When reading in one language, only words
from  the  relevant  lexicon  are  activated.  In
this  sense,  the  bilingual  lexical  memory  is
activated  selectively;  the  language  used
determines  the  words  to  be  retrieved.  As  a
result,  only  orthographic  and  phonological
representations  of  words  from  the  same
language  are  activated  and  no  activation
spreads  to  the  orthographic  and
phonological  representation  of  the  other
language. This view works for a number of
models  such  as  the  Word  Association
Model (Kroll & Curley, 1988), the Concept
Mediation  Model  (Kroll  &  Curley,  1988;
Potter  et  al.,  1984),  and  the  Revised
Hierarchical  Model  (Kroll  &  Stewart,
1994).  
 
Nonetheless,  many  studies  have  provided
evidence  for  an  alternative  viewpoint  that
incorporates  an  integrated  lexicon  (Dijkstra,
 
2005).  These  studies  suggest  that  lexical
representations  of  L1are  accessed  when  the
bilingual  is  reading  in  L2  (Brysbaert,  van
Dyck,  &  van  de  Poel,  1999;  Costa,
Caramazza&  Sebastian-Galles,  2000;
Dijkstra,  Grainger,  &  Van  Heuven,  1999;
Dijkstra,  Timmermans,  &Schriefers,  2000;
Duyck, 2005; Duyck, Diependaele, Drieghe,
&Brysbaert,  2004;  Haigh&  Jared,  2007;
Jared  &  Kroll,  2001;  Lemhöfer&Dijkstra,
2004;  Schwartz,  Kroll,  &  Diaz,  2007)  and
vice  versa  (Duyck,  2005;  Van  Assche,
Duyck,  Hartsuiker,  &Diependaele,  2009;
Van  Hell  &Dijkstra,  2002).  For  example,
Distributed  Lexical/Conceptual  Feature
Model  (Kroll  &  De  Groot,  1997)  proposes
that  lexical  and  conceptual  features  are
shared  between  languages  and  are  stored  in
a  distributed  mode.  There  are  language-specific  lemmas  including  syntactic
information  mediating  these  mental
representations. Such processing leads to the
assumption  of  nonselectivity  of  language  in
reading.  The  bilingual  interactive  activation
model+  (BIA+)  is  based  on  the  same
assumption  (Dijkstra&  Van  Heuven,  2002).
Additionally,  BIA+  postulates  one  unified
lexicon for both languages.
 
There  has  been  continuous  debate  over  the
selectivity vs. nonselectivity assumptions. A
number  of  researchers  have  argued  that
cross-language  orthographical  differences
may  restrict  language  nonselectivity  when  
the  two  scripts  differ  (Nakayama,  Sears,
Hino,  &Lupker,  2012).The  reason  seems  to
be  that  orthographical  differences  guide
incoming  sensory  information  toward  the
appropriate  lexical  system  such  that
nontarget language representations are never
contacted. If this is the case, Persian-English
bilinguals  are  supposed  to  show  language
selectivity  when  reading  in  English  due  to
cross-script differences between Persian and
English.  Therefore,  we  base  the  theoretical
foundation  of  this  study  on  the  models  that
support the language selective view.
 
Literature review
RHM  is  a  dominant  model  in
psycholinguistics  research  (Kroll  &
Steward,  1994;  Kroll  &Tokowicz,  2001).
This  model  acknowledges  two  levels  of
word  representation:  the  conceptual  and  the
lexical.  The  model  proposes  that  both
languages  of  a  bilingual  share  the  same
conceptual  store;  however,  each  language
has  its  own  separate  store  at  the  lexical
level.  There  are  connections  between  the
two languages of a bilingual speaker at both
lexical  and  conceptual  levels.  The
conceptual  store  is  connected  to  L1  and  L2
lexical  stores  via  routes  called  the
conceptual connections. There are also some
links  connecting  the  L1  lexical  store  to  the
L2  lexical  store  called  the  lexical  links.
Lexical  processing  may  occur  at  the  lexical
level  through  lexical  routes  or  at  the
conceptual  level  via  conceptual  connections
(Figure 1).

A number of studies have provided evidence
for such mental structure. Kroll and Stewart
(1990),  for  example,  made  Dutch-English
learners  translate  two  groups  of  words  in
both  forward  and  backward  directions.  One
of  the  lists  included  semantically
categorized  while  the  other  contained
randomly  organized  words.  It  was  assumed
that  forward  translation  would  be  affected

 
by  the  semantic  manipulation  of  words
because it occurs at the conceptual level but
backward  translation  would  not  be  affected
by  the  semantic  categorization  of
wordsbecause  it  occurs  at  the  lexical  level.
The  findings  of  the  study  confirmed  these
two assumptions.
 
Other  studies  concluded  that  for  both  early
and  advanced  L2  learners,  forward
translation  takes  more  time  and  is  more
sensitive  to  semantic  manipulation  than
backward  translation  (de  Groot,
Dannenburg&  Van  Hell,  1994;  Sanchez-Casas,  Davis  &  Garcia-Albea,  1992;  Sholl,
Sankaranararyanan& Kroll, 1995).
 
One basic assumption made by RHM (Kroll
&  Stewart,  1994)  is  that  at  early  stages  of
language  acquisition,  L2  learners  rely
mainly  on  lexical  links  between  L2  words
and  that  their  L1  translation  equivalence  is
established  once  a  language  is  learned.
Hence,  the  lexical  links  from  L2  to  L1
translation  equivalents  are  stronger  than  the
reverse  links.  The  existence  of  strong  L2  to
L1  lexical  links  means  strong  priming  from
L2  to  L1.  Several  cross-language  studies
have  failed  to  find  translation  priming
effects  from  L2  to  L1  for  noncognates
although  strong  priming  effects  have  been
found  from  L1  to  L2(Basnight-Brown
&Altarriba,  2007;  Gollan,  Forster,  &  Frost,
1997;  Jiang,  1999,  Jiang  &  Forster,  2001;
Kim & Davis, 2003; Voga& Grainger, 2007;
Williams,  1994).  Generally,  the  magnitude
of forward priming is greater than backward
priming,  although  the  evidence  for  L2-  L1
translation priming is less consistent.
 
What  these  aforementioned  studies  have  in
common  is  that  they  have  failed  to  report
significant  masked  translation  priming
effects  when  a  lexical  decision  task  was
used.  Bradley  (1991)  observed  L2-L1
priming when he tested unrelated word pairs
in  a  speed  recognition  memory  task.  The
task  was  to  decide  as  rapidly  as  possible
whether  the  presented  word  was  one  of  the
words  already  learned.  The  words  were
preceded by a masked version of an already
learned or a completely new word. What he
found was strong L2-L1 priming.
 
In  order  to  see  if  the  task  presented  would
influence  priming  effects  in  forward  or
backward  directions,  Jiang  and  Forster
(2001)  gave  episodic  and  lexical  decision
tasks  to  a  number  of  Chinese–English
bilingual  speakers  using  masked  priming
paradigm.  The  results  of  the  study
demonstrated  significant  masked  translation
priming  effects  in  the  backward  direction
when  an  episodic  task  was  used  and
significant  priming  effects  in  the  forward
direction  when  a  lexical  decision  task  was
used. To interpret the findings, these authors
put  forward  a  separate  memory  system
model.  According  to  this  model,  lexical
memory  and  episodic  memory  constitute
separate memory modules (Figure 2). 

Jiang  and  Forster  (2001)  argued  that  in  the
case  of  L2-L1  priming,  both  the  prime  (L2
word) and the episodic memory trace of the
L1  target  are  represented  within  episodic
memory.  As  overt  response  to  the  target  is
controlled  by  information  coming  from  the
 
 
84            Written word recognition by the elementary
same  memory  (episodic),  L2-L1  priming
occurs.  Moreover,  the  L1-L2  priming  was
effective in the lexical decision task because
the  L1  prime  activates  the  shared  semantic
features  through  strong  L1  conceptual
connections.  Nonetheless,  in  the  backward
direction,  as  the  conceptual  connection
between  the  L2  prime  and  the  conceptual
store  is  weak,  the  target  is  not  preactivated
(Kroll & Stewart, 1994).  In fact, the results
of  the  study  suggest  that  different  tasks
involve different links in a bilingual memory
and  the  presence  of  both  lexical  and
conceptual  connections  cannot  be
demonstrated  in  one  particular  task.  Lexical
links  show  priming  effects  in  episodic
recognition tasks but conceptual connections
show priming in lexical decision task.
The present study
RHM  predicts  strong  L2-L1  lexical
connections  at  the  early  stages  of  language
learning,  which  do  not  disappear  with
increasing  proficiency,  although  the  nature
of  the  links  might  change.    There  is  little
evidence  whether  or  not  initial  dependence
on  L1  would  play  a  continuing  role  during
L2  processing  even  at  higher  levels  of
proficiency. If L1 simply provides a way for
L2 to find its way into the cognitive system,
the  same  sort  of  activity  might  be  absent  at
higher  stages  as  L2  learners  become  more
proficient.  Very  little  information  from  the
existing  literature  tells  us  how  and  through
what process the nature of these connections
change  with  increasing  proficiency.  The
main  purpose  of  this  study  is  to  further
investigate the issue. According to Jiang and
Forster  (2001),  testing  L2  learners  in  an
episodic  recognition  task  can  explain  the
existence  and  strength  of  lexical  links.
Hence,  to  serve  the  main  goal  of  the  study,
four  experiments  were  designed  to  test  two
groups  of  elementary  and  two  groups  of
high  proficiency  L2  learners  with  an
episodic  task  in  forward  and  backward
directions  as  summarized  in  the  following
questions:
 
  If according to RHM there are strong
L2-L1  lexical  links  at  low  levels  of
language proficiency, will significant
priming  be  obtained  in  episodic
recognition  tasks  for  elementary  L2
learners in the backward direction?
  If  according  to  RHM  L2-L1  lexical
links  do  not  disappear  even  at  high
stages  of  language  proficiency,  will
significant  priming  be  obtained  in
episodic  recognition  tasks  for  highly
proficient  L2 learners in  the forward
direction?
 
Method
Four experiments were conducted to explore
two  main  predictions  of  RHM.  In
experiments  1  and  2,  two  groups  of
elementary and highly proficient L2 learners
were  tested  in  an  episodic  task  in  the
backward  direction.  The  same  method  was
followed  in  experiments  3  and  4  to  test  the
two  other  groups  of  elementary  and  highly
proficient learners in an episodic task in  the
forward direction.  
 
Participants
Twenty  four  Persian  learners  of  English
were selected out of 60.  All the participants
were undergraduate students of TEFL at the
Islamic Azad University, Najafabad Branch.
They  were  Persian  native  speakers;
however,  they  had  received  formal
instruction  in  English  in  high  school,  at  the
university,  and  in  language  institutes.  They
had  no  exposure  to  English  in  natural
settings.
 
The  grammar  part  of  the  Oxford  Placement
Test  (OPT,  Allan,  2004),  including  100
grammatical  multiple  choice  questions,  was
administered  to  homogenize  the  learners
based  on  their  general  knowledge  of
 
 
English.  Those  who  scored  between  52  and
59 were identified as elementary participants
and  were  selected.  The  reliability  index  of
the test was estimated through Chronbach’s
alpha (α = .78).
 
The  participants  were  randomly  assigned  to
two  groups  for  the  first  and  third
experiments.  Group  1  consisted  of  L2
participants  who  took  part  in  an  episodic
task in the backward direction while group 2
involved  L2  participants  who  anperformed
episodic task in the forward direction.
 
Another sample of 24 was selected from the
same  pool  delineated  above;  this  time,
however,  these  learners  were  identified  as
highly  proficient  participants  based  on  the
OPT  manual.  The  participants  were
randomly  assigned  to  two  groups  for  the
second  and  fourth  experiments.  Group  1
consisted  of  L2  participants,  who  took  part
in an episodic task in the backward direction
and  group  2  involved  L2  participants  who
performed  an  episodic  task  in  the  forward
direction.
 
Stimuli and design
The  stimuli  used  in  experiments  1  and  2
included 60 English-Persian translation pairs
and 60 Persian nonwords (Appendix A).The
Persian  target  words  were  divided  into  two
sets  (A  &  B)  of  30  in  the  study  phase
(Appendix  B)  and  two  sets  of  the  same
words in the test phase (Appendix C).In the
study  phase,  each  set  was  shown  to  half  of
the  participants  in  each  group.  This  set,
therefore,  was  considered  as  old  and  the
other as new. The same was followed for the
second  group.  In  the  test  phase,  two
presentation  lists  were  constructed,  each
with 30 old items and 30 new items. Half of
the  Persian  targets  (both  old  and  new)  on
each  presentation  list  were  paired  with
English  translation  primes,  and  the  other
half were paired with English control primes
that were unrelated to the target. In order to
make  the  stimuli  homogenous,  the  control
primes  were  matched  with  the  translation
primes  on  length,  frequency,  and
concreteness.The  translation  and  its  control
prime were similar to each other in terms of
length,  frequency  and  concreteness,  yet
different  from  each  other  in  terms  of
semantic  relation  to  the  target;  i.e.,    related  
(translation  prime)  vs.  unrelated  (control
prime).  Thirty  English  control  primes  were
generated  by  the  MRC  Psycholinguistic
Database (Cullings, 1988).  
The  same  Persian-English  translation  pairs
were used in experiments 3 and 4(Appendix
E).The  English  target  words  were  divided
into  two  sets  (A  &  B)  of  30(Appendix
F).The  same  procedure  was  adopted  to
create  two  presentation  lists  (Appendix
G).Half of the English targets (both old and
new)  on  each  presentation  list  were  paired
with  Persian  translation  primes,  and  the
other  half  were  paired  with  Persian  control
primes.  Bijankhan  corpus
(Amiri&AleAhmad,  n.d.)  was  used  for  this
purpose.  Moreover,  60  nonwords  were
generated  by  the  ARC  nonword  database
(Rastle,  Harrington,  &Coltheart,  2002).All
the  nonwords  were  preceded  by  unrelated
primes. Ten additional translation pairs were
selected to be used as practice items.
Procedure
The procedure was divided into study phase
and  test  phase.  In  the  study  phase,  the
participants  were  given  a  list  of  30  Persian
target  words  as  well  as  10  practice  Persian
words  to  study  and  memorize  so  that  they
would  be  tested  on  a  memory  test(test
phase)  later  on.  Each  group  of  participants
was  divided  into  two  groups  of  6.  The  first
half received the Persian target words in set
A  and  the  other  half  received  Persian  target
words  in  set  B.  In  other  words,  each  group
received  only  one  set  of  Persian  words
considered  as  old  in  the  test  phase.  They
 
were  given  as  much  time  as  they  needed  to
memorize  the  words  on  the  list.  Then  they
received an initial recognition task on paper,
in which they were asked to circle the words
they  had  studied  on  the  study  list.  In  cases
when  the  performance  was  90%  or  more
accurate, they received the computer version
of  the  recognition  task  in  which  the
participants  were  to  decide  as  quickly  as
possible  whether  the  word  presented  on  the
screen  was  one  of  the  words  they  had
studied in the study phase.
 
Following  Forster  and  Davis  (1984),
presentation of items in experiments 1 and 2
included  the  following  masked  priming
sequence:  First,  the  participants  were
presented  with  a  row  of  10  hash  marks  for
500  ms  which  served  to  mask  the
subsequently  presented  prime.  Second,  the
prime  word  immediately  appeared  for  50
ms.  Next,  a  blank  interval,  consisting  of  a
row  of  hash  marks,  was  presented  for  150
ms.  The  target  immediately  followed  the
backward mask. The target remained on the
screen until the participants made a response
(Appendix  D).  The  inclusion  of  the  blank
space  and  the  backward  mask  was  for  the
purpose  of  increasing  the  amount  of  prime
processing time (see Jiang 1999, Experiment
4). Normally when the prime is in the L2, its
processing  is  slower  than  when  it  is  in  the
L1; as a result, there would be no chance for
the  L2  prime  to  have  any  effect  on  the  L1
target (see Jiang 1999, Experiment 4). After
each  trial  was  completed,  the  participants
received feedback on the speed and accuracy
of their performance.
 
In  experiments  3  and  4,  each  trial  consisted
of  the  following  sequence:  First,  a  forward
mask  of  10  hash  marks  appeared  for  500
ms.This  forward  mask  was  followed  by  the
prime  which  was  presented  for  50  ms.
Finally,  the  target  word  immediately
followed  the  prime  and  remained  on  the
screen until the participants made a response
(Appendix  H).  The  participants  were  asked
to  decide  as  rapidly  as  possible  whether  the
word presented on the screen was one of the
words on the study list.
Apparatus
The  DMDX  package  developed  at  the
University  of  Arizona  by  J.C.  Forster
(Forster  &  Forster,  2003)  was  used  to
presentthe stimuli.
 
Results
Elementary L2 learners (forwardddirection)
RTs  longer  than  1400  ms  and  incorrect
responses,  which  included  25.5  %  of  the
data,  were  excluded  from  the  analysis
(Gollan  et  al.,  1997;  Keatley,  Spinks,  &  de
Gelder,  1994).  The  descriptive  statistics  of
the  RTs  in  the  forward  direction  are
provided  in  Table  1  (tables  appear  after
References).
The  means  for  response  times  were  13.63
ms faster for the translation items in the old  
group and 0.7 ms faster for the control items
in the new group. To compare the means of
the  noncognate  translation  and  the
noncognate control items in the old and new
groups, two paired samples t-tests were run,
the  results  of  which  showed  that  the
noncognate  translation  and  the  noncognate
control  items  were  processed  similarly  by
both groups (see Table 2).
 
 
Elementary L2 learners (backwarddirection)
As  in  the  previous  analysis,  7.22  %  of  the
data,  which  included  the  scores  over  1400
ms  and  incorrect  responses,  were  excluded
from  the  analysis  (Gollan  et  al.,  1997;
Keatley  et  al.,  1994).  The  descriptive
statistics  of  the  RTs  in  the  backward
direction are provided in Table 3.
 
The mean RTs were 83.81 ms faster for the
translation  items  in  the  old  and  25.83  ms

 
faster  for  the  translation  items  in  the  new
group.
 
In  order  to  compare  the  means  of  the
noncognate  translation  and  the  noncognate
control  items  in  old  and  new  groups,  two
paired  samples  t-tests  were  applied.  The
results  show  that  noncognate  translation  the
noncognate  control  items  were  processed
similarly  in  the  forward  direction;  however,
the  noncognate  translation  items  were
reacted  to  significantly  faster  than  the
noncognate  control  items  in  the  backward
direction (see Table 4).
 
Highly  proficient  L2  learners  (forward
direction)
The  incorrect  responses  and  the  RTs  longer
than 1400 ms, which included 17.5 % of the
data,  were  excluded  from  the  analysis
(Gollanet  al.,  1997;  Keatley  et  al.,  1994).
The  descriptive  statistics  of  the  RTs  for  the
noncognates  in  the  forward  direction  are
provided in Table 5.
 
The mean RTs were 37.55 ms faster for the
control items in the old and 24.64 ms faster
for the control items in the new group.
 
Two  paired  samples  t-test  were  run  to
compare  the  means  of  noncognate
translation and the noncognate control items
in old and new groups. The results show that
noncognate  translation  and  noncognate
control  items  were  processed  similarly  (see
Table 6).
 
Highly  proficient  L2  learners  (backward
direction)
Response  times  longer  than1400  ms  and
incorrect responses, which included 21.52%
of the data, were excluded from the analysis
(Gollan  et  al.,  1997;  Keatley  et  al.,  1994).
The  descriptive  statistics  of  the  RTs  for  the
noncognates  in  the  forward  direction  are
provided in Table 7.
The  mean  response  times  were  41.51  ms
faster for the translation items in the old and
6.98  ms  faster  for  the  control  items  in  the
new group.
 
Two  paired  samples  t-test  run  to  compare
the means of the noncognate translation with
the noncognate control items in old and new
groups show that the noncognate translation
and  the  noncognate  control  items  were
processed  similarly  by  both  groups  (see
Table 8).
 
Discussion and conclusion
The  main  purpose  of  the  experiments
conducted  in  this  study  was  to  evaluate
predictions  made  by  RHM  indicating
whether or not the strong L2-L1 lexical links
formed  at  the  beginning  stages  of  language
learning remain unchanged  at higher stages.
Two  groups  of  elementary  and  two  groups
of  high  proficiency  L2  learners  were  tested
on  noncognate  stimuli  with  an  episodic
recognition  task  in  both  forward  and
backward directions.
 
The  results  obtained  for  the  elementary
learners  showed  no  significant  priming  in
the  forward  but  significant  priming  in  the
backward  direction.  This  pattern  is
consistent with the basic prediction made by
RHM  regarding  elementary  stages  of
language  learning;  however,  no  such  links
seem to exist in the forward direction.  
 
Two  other  lines  of  research  confirm  this
pattern.  In  a  series  of  experiments  done  by
Kroll  and  Curley  (1988),  beginner  English-German  L2  learners  were  tested  on  picture
naming  and  translation  tasks  in  the
backward  direction.  Kroll  and  Curley  came
to  the  conclusion  that  for  beginners,
backward  processing  occurs  at  the  lexical
level.  
 
In  another  experiment  done  by  Kroll  and
Stewart  (1994),  a  number  of  L2  learners
were  tested  on  picture  naming  and
translation  tasks  in  either  semantically
categorized  or  randomized  context.  The
findings of the study showed slower RTs for
the  picture  naming  and  forward  translation
in  the  semantically  categorized  context.
Moreover,  semantic  manipulation  of  the
context  did  not  affect  backward  translation.
The  authors  suggested  that  forward
translation and picture naming tasks proceed
along  the  conceptual  routes;  however,
backward  translation  exploited  lexical  links
between L1 and L2.
 
Shorter RTs for backward translation proves
the  existence  of  the  strong  lexical
connections  in  the  backward  direction.
Moreover, the fact that backward translation
is  not  sensitive  to  semantic  factors  when
compared  with  forward  translation  can  be
taken  as  evidence  suggesting  that  backward
translation  occurs  through  L2-L1  lexical
routes. The present study shed more light on
this  issue  by  showing  consistent  priming
effects in the backward direction.
 
The observed pattern also supports the Word
Association Model (Figure 3). According to
this  model,  there  is  a  common  conceptual
system  for  the  lexical  systems  of  the  two
languages.  L2  learners  retrieve  the  meaning
of  L2  lexical  items  via  links  to  their
translation  equivalents  in  the  first  language
(L2-L1 lexical links).

Another  finding  of  the  study  was  that  no
significant  priming  was  obtained  in  either
direction  for  highly  proficient  L2  learners.
This  pattern  shows  that  at  higher  levels  of
proficiency, L2-L1 lexical links are lost and
L1  has  almost  no  role  in  retrieving  the
semantic  content  of  L2  words.    This  is  not
consistent  with  the  prediction  of  RHM,  as
this  model  predicted  the  existence  of  these
routes  even  at  higher  stages  of  language
learning. The observed pattern shows that as
L2  learners  become  more  proficient,  they
begin to make direct conceptual connections
from the L2 lexicon. In fact, at higher stages
of  L2  acquisition,  the  role  of  L2-L1  lexical
access  decreases  and  is  replaced  by  L2
conceptual  connections.  As  L2  learners
increasingly rely on these connections, such
connections  gradually  become  stronger
(Figure  4).  This  finding  is  consistent  with
the Concept Mediation Model. 

A  study  conducted  by  Kroll  and  Curley
(1988)  confirms  this  pattern.  In  this  study,
advanced  and  elementary  English-German
learners  performed  translation  and  picture
naming tasks. It was concluded that early L2
learners  performed  picture  naming  task
through  conceptual  connections  and
backward  translation  through  lexical  routes.
However,  proficient  L2  learners  relied  on
the  conceptual  connection  whendoing  both
tasks.
 
RHM  predicts  that  lexical  links  would
remain  unchanged  even  at  higher  stages  of
language  development.  However,  the
 
absence of lexical links for highly proficient
learners in the present study shows that they
are  lost  at  this  stage.  As  proficiency  in  L2
increases,  learners  begin  to  use  the
conceptual connections instead of the lexical
routes.  The  presence  of  the  lexical  links  in
early  learners  and  the  absence  of  such  links
in highly proficient learners provide enough
support  for  a  shift  from  word  association  to
concept mediation.  
 
Based  on  the  discussion,  the  following
implications  seem  to  be  in  order  for  L2
vocabulary  teaching  and  learning.  The  use
of L1 has several advantages. It provides the
core meaning of words,which is the first step
in  associating  the  form  with  meaning  and
reinforcing  the  connection.  As  Grabe  and
Stoller  (1997,  p.114)  put  it,  "perhaps,  for
adults,  there  are  times  when  it  is  important
to  know  that  a  word  is  understood
accurately." Furthermore, using L1 may link
an  L2  word  to  its  firm  semantic  and
linguistic  structure  which  can  serve  as  the
steadiest  “cognitive  hook  to  hang  the  new
item  on"(Fraster,  1999,  p.238).  This  way,
learners  canretain  the  words  in  long  term
memory more efficiently.