Semantic Priming Effect on Relative Clause Attachment Ambiguity Resolution in L2

Introduction
Numerous  studies  have  focused  on  the
processing  of  structurally  ambiguous
sentences  in  order  to  shed  light  on  the
nature of the human sentence processing
mechanism.  Several studies have shown
that there are cross-linguistic differences
in  the  resolution  of  ambiguous  relative
clauses  (RCs)  like  Someone  shot  [the
servant]DP1 of [the actress]DP2 [who was
on  the  balcony]RC,  where  the  RC  can
modify  either  DP1or  DP2  in  the
preceding  complex  DP.    A  high  (DP1)
attachment  preference  has  been  reported
in  Dutch  (Brysbaert  &  Mitchell,  1996;

Desmet, Brysbaert & De Baecke, 2002),
French  (Baccino,  De  Vincenzi  &  Job,
2000;  Colonna,    Pynte,    &    Mitchell,  
2000;  Quinn,  Abdelghany  &  Fodor,
2000,  Zagar,  Pynte,  &  Rativeau,  1997),
German  (Hemforth,  Konieczny,
Scheepers  &  Strube,  1998;  Hemforth,
Konieczny  &  Scheepers,  2000),  Persian
(Arabmofrad  &  Marefat,  2008),  and
Spanish  (Cuetos  &  Mitchell,  1988),
whereas  a  low  attachment  (DP2)
preference  has  been  found  in  Brazilian
Portuguese  (Finger  &  Zimmer,  2000;
Miyamoto,  1998),  English  (Cuetos  &
Mitchell, 1988;  Frazier & Clifton, 1996;

Deevy, 2000), Norwegian, Swedish, and
Romanian  (Ehrlich,  Fernandez,  Fodor,
Stenshoel & Vinereanu, 1999). A number
of
attempts
have
been
made to account
for these cross-linguistic differences.
The dominant account in the literature for
RC  attachment  preferences  has  been
based  upon  Recency  (Gibson,
Pearlmutter,  Canseco-Gonzalez,  &
Hickok,  1996),  Late  Closure  principle
(Frazier  &  Fodor,  1978),  or  Right
Association  (Kimball,1973),  which
assume  that  all  human  languages  are
processed  in  the  same  way.  Frazier
(1978)  defined  Late  Closure  as  "When
possible,  attach  incoming  material  into
the  clause  or  phrase  currently  being
parsed"  (p.  49).  This  is  similar  to
Kimball's  (1973)  Right  Association
principle  "Terminal  symbol  is  optimally
associated  to  the  lowest  nonterminal
node"  (p.  24).  According
to  Recency
principle,  constituents  such  as  RC
modifiers  are  attached  to  the  most
recently  processed  (or  closest)  phrase,
regardless  of  the  language  being
processed  (Fernandez,  2003).  The  
principle    of  Late    Closure    has    been  
found      to    apply    in    a    number    of
languages,  with  a  wide    variety    of  
constructions.  
Cuetos and Mitchell (1988) were among
the  first  to  challenge  the  ‘universalist’
view; they showed that parsing does not
proceed in the same way in all languages.
In  their  experiment,  they  gave
constructions  comparable  to  Someone
shot  [the  servant]DP1  of  [the  actress]DP2
[who  was  on  the  balcony]RC  to  English
and Spanish native speakers. They found
that  English  subjects  tended  to  attach

RCs to DP2. On the other hand, speakers
of  Spanish  showed
an  overall  high
attachment  (DP1)  preference.  This  led
them  to  conclude  that  certain  parsing
strategies may not be universal, and that
there  exists  cross-linguistic  variation.
Following
Cuetos
and  Mitchell,  further
studies  were  conducted  in  order  to

examine  RC
attachment  preferences  in
other  languages.  The  results  have
provided additional support for the view
that the Late Closure principle might not
be  generalized  cross-linguistically,  and
there exist strategies that are more likely
to be operative in other languages.
The  data  obtained  from  Gibson  et  al.
(1996)  experiment  on  sentences  with
three  potential  antecedents  for  the
ambiguous  RCs  in  English  and  Spanish
provided  evidence  for  developing  the
Predicate Proximity principle, as another
factor  that  competes  with  the  universal
Recency  principle.  According  to
Predicate  Proximity,  incoming
constituents  are  favorably  attached  to  a
verb  argument.  Gibson  and  Pearlmutter
(1998)  argued  that  in  certain  languages
like
German,  Russian  and  Spanish,
Predicate  Proximity  outranks  the
Recency  principle  and  leads  to  a  high
(DP1)  attachment  preference.  By
contrast,  in  configurational  languages
such  as  English,  Norwegian,  and
Swedish,  ambiguous  modifiers  attach  to
the  most  recent  phrase  in  harmony  with
the locality principle of Recency (Felser,
Roberts, Marinis & Gross, 2003), leading
to low-attachment instead.  
 
Ambiguous RCs in Persian
According  to  Karimi  (2001,  p.  31),
“Persian  is  a  null-subject  verb  final
language  with  SOV  word  order  in
declarative  sentences  and  subordinate
clauses”. Similar to English, Persian RCs
are  post-nominal,  and  since  there  is  no
relative  pronoun  in  Persian,  RCs  are
always introduced by complementizer ke
in Persian (Taleghani, 2008, p. 84).  
Previous studies have shown that Persian
native  speakers  have  a  high  attachment

preference  (Arabmofrad  &  Marefat,
2008;  Marefat  &  Meraji,  2005;
Moghadassian, 2008). Forty five Persian-speaking monolinguals participated in an
on-line  study  conducted  by  Arabmofrad
and Marefat (2008). Fifteen sets of items
were  developed  and  normed;  each  set
contained  sentences  in  three  conditions:
1)  RC  was  semantically  related  to  DP1
and could be attached only to it and not to
DP2 (sentence 1 below); 2) RC could be
attached  only  to  DP2,  based  on  a
semantic relationship (sentence 2 below);
3) an  ambiguous condition in which RC
could  be  attached  to  both  DP1  and  DP2
(sentence 3 below).
(1)  an  mærd  pæræstar-e nozad [ke dašt
ghædæm mizæd] ra did
‘That  man  saw  the  nurse  of  the  infant
[who was walking].’
 
(2)  an  mærd  nozad-e  pæræstar [ke dašt
ghædæm mizæd] ra did
‘That  man  saw  the  infant  of  the  nurse
[who was walking].’
 
(3) an mærd dokhtær-e pæræstar ke dašt
ghædæm mizæd ra did
‘That man saw the daughter of the nurse
[who was walking].’
 
The rationale behind the study was that if
the  participants  prefer  DP1  attachment,
then  the  reaction  time  for  sentences  in
which  the  RC  is,  contrary  to  their
expectation, semantically  related only to
DP2  would  be  slower  than  that  for
sentences in which the RC is semantically
related  to  DP1.  And,  conversely,  if  they
have  a  tendency  to  attach  RC  to  DP2,
then, reaction time for sentences in which
the  RC  could  only  be  attached  to  DP1
would  take  longer.  The  participants,
tested  individually,  were  required  to
make  grammaticality  judgments  about
the  sentences  that  were  presented  in  a
non-cumulative way. Decisions as well as
decision  times  were  automatically
recorded.  The  results  for  accuracy
responses showed no difference between
the  three  conditions  which  implies  that
after  reading  the  sentences,  participants
accurately  made  grammaticality
judgments about the sentences. However,
analysis  of  the  reaction  times  for
grammaticality judgment of the sentences
in  the  three  conditions  showed  that  the
participants  took  shorter  reaction  times
for sentences in which due to a semantic
cue,  RC  had  to  be  attached  to  DP1,  but
longer  reaction  times  to  sentences  in
which the RC had to be attached to DP2.
Moreover,  there  was  no  difference
between  ambiguous  sentences  and  those
in  which  RC  referred  to  DP1.  These
results show a high attachment preference
by Persian native speakers.
 
Factors Affecting Attachment Preferences
within a Language
There  are  a  number  of  individual-level
factors,  such  as  proficiency  and  WMC,
which can cause intra-lingual differences
in attachment preferences.  
As  for  proficiency,  Miyao  and  Omaki
(2006)  used  an  off-line  and  an  on-line
self-paced  reading  task  to  examine  RC
attachment preferences of intermediate to
advanced Korean L2 learners of Japanese
and  Japanese  native  speakers.  Results
from  the  off-line  sentence  interpretation
task showed that both Korean L2 learners
and  Japanese  natives  preferred  high
attachment.  However,  results  from  the
on-line  self-paced  reading  task  showed
that  Japanese  natives  preferred  high
attachment  but  Korean  L2  learners
preferred low attachment. To account for
these  results,  Miyao  and  Omaki  stated
that there may be three different stages in
development  of  L2  processing:  L1
transfer  phase,  intermediate  phase,  and
target-like  phase.  They  defined  the  L1

transfer phase as the stage in which low-proficiency  L2 learners transfer their  L1
grammar,  including  their  L1  parsing
preference.  In  the  intermediate  phase,
medium-level L2 learners’ grammar and
parsing  preferences  are  still  developing
and  not  yet  efficient,  i.e.,  their  grammar
is  non-target  like  and  includes  traces  of
their  L1.  In  the  target-like  phase,  high-proficiency  L2  learners  have  target-like
parsing  preferences.  They  stated  that
participants  in  their  study  were  in  the
intermediate  phase,  and  as  a  result  they
resorted  to  a  parsing  strategy  that
minimized  their  processing  burden  (i.e.,
locality principle), which resulted in low
attachment.  
Although  some  L2  studies  on  RC
attachment  preferences  have  found  no
effect  of  WMC,  numerous  studies  have
shown  that  individuals  with  high  WMC
process  syntactically  ambiguous
sentences differently from those with low
WMC  (Kim  &  Christianson,  2013;
Mendelsohn & Pearlmutter, 1999; Swets,
Desmet,  Hambrick,  &  Ferreira,  2007;
Vos,  Gunter,  Schriefers,  &  Friederici,
2001).  Swets  et  al.  (2007)  examined  the
role  of  WMC  in  RC  attachment
preferences of English and Dutch native
speakers. Test sentences were ambiguous
structures such as: [The maid]DP1 of [the
princess]DP2  [who  scratched  herself  in
public]RC  was  terribly  embarrassed.
They  reported  a  negative  correlation
between  WMC  and  RC  attachment
preference. To account for these findings,
they proposed the ‘chunking’ hypothesis:
low  span  readers  do  not  have  enough
resources  to  parse  a  sentence  without
stopping  at  intermediate  places,  which
causes them to pause before the RC. This
pause  in  processing  makes  them  chunk
DP1  and  DP2  into  a  single  unit,  leading
them  to  attach  the  following  RC  to  the
head  of  the  complex  DP,  i.e.,  DP1  (the
maid).  
Fodor  (2002)  proposed  the  Implicit
Prosody  Hypothesis  (IPH)  to  justify  the
intra-lingual  variation  in  attachment
preferences.  According  to  IPH,  different
prosodic  groupings  of  a  sentence  can
result in different interpretations. In case
of  ambiguous  RCs,  a  prosodic  grouping
of  (DP1)  (DP2  RC)  reflects  a  low
attachment; while a prosodic grouping of
(DP1  DP2)  (RC)  demonstrates  a  high
attachment  interpretation.  Thus,  for
sentence  (4a),  where  the  prosodic
grouping  is  (DP1)  (DP2  RC)  and
prosodic boundary is put after DP1 (i.e.,
servant), the parser is likely to attach the
ambiguous RC to DP2 (i.e., the actress),
while,  in  sentence  (4b)  where  the
prosodic  grouping  is  (DP1  DP2)  (RC)
and prosodic boundary  comes after DP2
(i.e., the actress), the parser interprets the
RC as modifying DP1 (i.e., the servant).
(4)  a.  Someone  shot  the  servant  #of  the
actress who was on the balcony.
      b.  Someone  shot  the  servant  of  the
actress #who was on the balcony.
More support for IPH has been provided
by  Jun  (2003)  who  examined  native
speakers of English, French, Greek, Farsi
(Persian), French,  Japanese, Korean, and
showed  that  in  each  language  there  is  a
correlation  between  attachment
preferences  and  the  default  prosody
assigned  upon  reading  ambiguous
sentences containing RCs.
 
The Present Study
This  study  undertakes  to  see  if  the
different  findings  across  and  within
languages with regard to the resolution of
ambiguous RCs can be accounted for by
the  role  of  semantics.  Previous  research
has shown that the effect of semantics is
“strong  enough  to  over-ride  any  phrase
structure  based  locality  principle  that

might otherwise favor NP1 attachment”
(Felser, et al., 2003, P. 457). But the type
semantics Felser et al. refer to is limited
to  the  distinction  between  thematic
preposition  with  and  the  case  assigner
preposition  of.  Since  the  former
preposition  constructs  a  local  thematic
domain, the ambiguous RC is associated
with  the  DP  inside  this  domain.  Thus,
semantics  was  limited  to  the  lexical
semantic features of prepositions and the
results  from  many  studies  (Felser  et  al.,
2003;  Papadopoulou  &  Clahsen,  2003)
have shown that L2 learners are sensitive
to  the  bias  provided  by  the  preposition
with.  But  in  this  study,  semantics  has  a
broader  domain.  In  the  two  experiments
reported  in  this  study,  the  sentence
contexts are varied to establish a semantic
relationship  between  the  subject  of  the
main  clause  and  one  of  the  DPs  in  a
complex DP (Experiment 1) and the verb
of the main clause and one of the DPs in
the  complex  DP  (Experiment  2).  In
sentence  (5)  below,  the  subject  of  the
matrix sentence is related to DP1, and in
sentence (6) it is related to DP2.   
(5) The doctor saw [the nurse]DP1 of [the
pupil]DP2 [who was in the yard]RC.                                         
(6) The teacher saw [the nurse]DP1 of [the
pupil]DP2 [who was in the yard]RC.
When the parser starts reading a sentence,
the  first  piece  of  information  s/he
encounters is most often the subject of the
sentence. When the first  DP,  the doctor,
in  the  case  of  sentence  (5),  is  activated,
based on the Spreading Activation Model
(Collins  &  Loftus  1975;  Dell,  1986),
other words that are semantically related
to it also become activated, and when the
parser  gets  to  DP1  the  nurse,  the  two
semantically  associated  words  reinforce
the  activation  of  each  other  and  remain
more accessible in comparison to DP2 the
pupil. Encountering the  RC,  who was in
the yard, the parser is expected to attach
it to the more accessible DP which is the
nurse. Moreover, being the subject of the
sentence,  the  doctor  occupies  a
syntactically prominent position and thus
its  accessibility  is  enhanced  (Bock  &
Warren,  1985;  Brennan,  1995;  Brennan,
Friedman,  &  Pollard,  1987;  Prat-Sala  &
Branigan,  2000)  and  this  makes  its
semantically related item in the sentence
(i.e., the nurse) more accessible as well.
Another  kind  of  semantic  relationship
that may affect RC attachment preference
is  when  the  verb  of  the  main  clause  is
semantically related to one of the DPs in
the complex DP preceding RC. Altmann
and  Kamide  (1999)  reported  an  eye-tracking  study  in  which  participants
listened to sentences such as The boy will
eat  the  cake  and  The  boy  will  move  the
cake  while  they  viewed  a  scene
containing a boy  who was sitting on the
floor surrounded by various items such as
a toy train set, a ball, a toy car, a balloon,
and  a  birthday  cake.  When  participants
heard the verb eat, they tended to look at
the  cake  more  often  compared  to  when
they heard the verb move. This happened
because the selectional restrictions of the
verb eat prescribed that only one object in
the  visual  scene  could  be  relevant  (i.e.,
the  cake),  but  the  verb  move  could  refer
to all of the movable objects in the scene
(i.e., the toy train, the ball, the toy car, the
balloon,  and  the  cake).  This  finding
shows  that  by  using  the  selectional
restrictions  of  the  verb,  comprehenders
are  able  to  predict  an  upcoming  direct
object of the verb. In sentence (7) below,
encountering  the  verb  inject,  the  parser
“not  only  analyses  [it]  .  .  .  but  also
predicts  upcoming  unseen  elements”
(Arai & Keller, 2013, p. 525). Thus, when
the parser sees the nurse (DP1), a strong
association is developed between the verb
and this DP which is expected to make it
more  accessible  in  comparison  to  the

lawyer  DP2,  when  the  parser  is  at  the
stage of attaching the RC to a preceding
DP.  In  sentence  (8),  on  the  other  hand,
there  is  a  strong  semantic  relationship
between  the  verb  defend  and  DP2  the
lawyer compared to DP1 the nurse, and is
thus more accessible, and if priming plays
a  role,  the  RC  is  expected  to  attach  to
DP2.   
(7)  Someone  injected  [the  nurse]DP1  of
[the  lawyer]DP2  [who  was  on  the
balcony]RC.
(8)  Someone  defended  [the  nurse]DP1  of
[the  lawyer]DP2  [who  was  on  the
balcony]RC.  
Thus,  in  line  with  previous  studies
(Schafer,  Carter,  Clifton,  &  Frazier,
1996), we predicted that the primed word
would become more salient and therefore
would  attract  the  ambiguous  RC.  The
present  study  builds  on  the  above
findings and aims to delve into the role of
priming in RC attachment preferences of
Persian  L2  learners  of  English.  In
particular,  this  study  aims  to  provide
answers to the following question:
Does  priming  one  of  the  DPs  in  the
complex  DP  through  associating  it  with
the subject/verb of the sentence influence
the RC attachment preferences of Persian
L2  learners  of  English  with  different
proficiencies and WMCs?
 
Experiment 1
The aim of this experiment was to explore
the  impact  of  semantic  priming  created
by the association between the subject of
the  matrix  sentence  and  one  of  the  DPs
preceding  RC  on  participants’
preferences  with  different  proficiency
and WMCs.
 
Method
Participants
33  Persian-speaking  learners  of  English
(mean  age  20,  range18-22,  11  females)
majoring  in  English  Language  and
Literature at two universities participated
in  this  experiment  as  a  course
requirement.  They  were  B.A.  students
and had not received advanced linguistic
instruction and had no idea about the aim
of  the  study.  Persian  was  the  L1  of  all
participants  and  they  had  all  started
learning  English  at  high  school.  Two
participants  were  excluded  because  they
had  not  completed  the  proficiency  test
and 2 more ones were excluded because
they could not meet the criterion of 90%
comprehension  accuracy  of  fillers  (see
below for details). In this way, data from
29  participants  were  used  for  data
analysis.
Instruments
The Operation Span Task (OST)
The  vast  majority  of  previous  studies
have  used  the  reading  span  task  as  the
sole  index  of  WMC  while  there  exist
many  different  assessments,  which  tap
different  working  memory  mechanisms
(Conway,  Kane,  Bunting,  Hambrick,
Wilhelm  &  Engle,  2005).  The  reading
span  task  assesses  the  ability  to  sustain
and process information through reading
sentences while participants’ attention is
split. In this study, OST was used rather
than  the  reading  span  task.  In  OST,
instead  of  sentences,  mathematical
operations  are  used.  Both  tasks  (i.e.
reading  span  task  and  operation  task)
have  been  shown  to  predict  sentence
comprehension  performance  (Turner  &
Engle, 1989) but O’Rourke (2013) stated
that  only  operation  span  predicts
accuracy  for  syntactically  complex
sentences.
The  OST  was  administered  using
Microsoft PowerPoint 2010. Participants
were  presented  with  sets  of  simple
equations  ranging  from  two  to  five
equations per set. There were three trials
for  each  set  size,  resulting  in  a  total  of

orty-two  (3  ×  (2  +  3  +  4  +  5)  =  42)
equations for the entire test. A sample set
including three items is presented below:
a. 5 + (8 × 2) = 21, ?, Z
b. (2 + 9) – 4 = 5, ?, Y
c. 5 × (7 – 2) = 45, ?, B
???
Before starting the task, a 15-item warm-up  activity  was  administered.  Then  the
42-item  OST  was  presented  in  a  fixed
order  for  all  participants.  Each  item
appeared  on  the  screen  and  remained
there  for  5  seconds.  The  length  of  this
interval  was  based  on  the  findings  of  a
pilot study. Then a question mark (‘?’)
appeared  on  the  screen,  at  this  point
participants  were  instructed  to  judge  the
accuracy of the equations by saying ‘true’
or ‘false’. After the equation judgment,
participants  were  instructed  to  press  the
space bar. Then a capital letter appeared
on  the  screen  to  be  read  aloud.  After  3
seconds  this  letter  disappeared  and  the
participants  proceeded  to  the  next  item.
When  the  participant  reached  the  last
item in a set, three question marks (‘???’)
appeared.  The  participants  were
instructed to stop at this point and recall
the letters in the order in which they had
appeared  in  the  set.  The  experimenter
recorded  the  responses  on  an  answer
sheet.  
Two  scores  were  reported  based  on  this
test: one for the true/false answers to the
equations  (judgment  accuracy)  and  one
for  the  number  of  letters  recalled
accurately.  The  correlation  coefficient
between  these  two  scores  was  .80  (p  =
.000). Since there was the possibility that
the  participants  focus  on  the  letters  they
were  to  recall  (which  was  a  measure  of
their  WMC)  and  take  the  truth  value  of
the  equations  not  seriously,  their
performance  on  equations  was
considered as a criterion for selection.  
In  order  to  be  qualified  to  participate  in
the  study  participants  were  required  to
correctly  judge  at  least  38  equations  out
of 42 (i.e. 90 percent).
As for the OST, one point was given for
every  letter  correctly  recalled  in  the
correct order. The scores ranged between
25  and  42;  and  the  mean  and  standard
deviation were 32 and 5.31, respectively.
Reliability of this test, based on the KR-21 formula, turned out to be .63.
 
Proficiency Test
Prior  to  the main  test,  the participants’
proficiency  in  English  was  assessed
through  the  Oxford  Placement  Test
(OPT,  Allan,  2000).  The  scores  ranged
between  22  and  54  (out  of  a  maximum
possible  score  of  60);  and  the  mean  and
standard  deviation  were  39.59  and  7.25,
respectively. KR-21 reliability of this test
turned out to be .76.   
Descriptive  statistics  for  the  OST  as  a
measure of WMC and OPT as a measure
of proficiency are documented in Table 1
below.

The Main Test  
The  main  test  consisted  of  70  sentences
including  10  practice  items  (three  of
which  also  served  as  warm-ups  across
four  versions  of  the  main  test),  20  test
sentences, and 40 fillers
 
Test Sentences
The  test  sentences  were  all  structurally
ambiguous  sentences  containing  a  main
clause  and  an  RC,  which  could  refer  to
either  of  two  preceding  DPs  that  were
linked  together  by  genitive  of  and
functioned as the object of the sentence.
The RC, in all sentences, was introduced
by the relative pronoun who. The subjects

of  the  sentences  were  animate  and
represented different occupations such as
teacher, lawyer, doctor, etc.  
Based  on  the  relationship  between  the
subject  of  the  main  clause  and  either  of
the  two  DPs  in  a  complex  DP,  test
sentences  were  categorized  into  three
types:  DP1-biased  subject  in  which  the
subject of the main clause and DP1 were
semantically related; DP2-biased subject
in  which  the  subject  of  the  main  clause
and  DP2  were  related;  Unbiased  subject
with no specific relationship between the
subject  of  the  main  clause  and  either  of
DPs.  Examples  for  each  category  are
provided below:
 
DP1-biased subject  
(9) The doctor saw the nurse of the pupil
who was in the yard.
 
DP2-biased subject  
(10)  The  teacher  saw  the  nurse  of  the
pupil who was in the yard.
 
Unbiased subject  
(11) The lawyer saw the nurse of the pupil
who was in the yard.
 
Sentences  (9),  (10)  and  (11)  were
regarded as a set of test sentences.  
To  assess  the  relationship  set  by  the
researchers  between  the  subject  and
either  of  the  DPs,  a  norming  study  was
conducted.  Seven  applied  linguists  and
25 participants from the same pool as in
the  main  study,  none  of  whom
participated in the main experiment, took
part  in  the  norming  study.  They  were
asked  to  decide  whether  there  was  an
occupational  semantic  relationship
between  the  underlined  words.  All  the
items  were  likert-scaled,  ranging  from  0
to  4,  where  0  meant  there  was  not  any
semantic relationship between the words
and  4  meant  they  were  strongly  related.
Based  on  the  results  of  the  norming,  10
sets (those for which 90% of the subjects
had rated the semantic relationship as 4)
were  selected.  Each  set  had  3  test
sentences,  so  we  had  a  total  of  30
sentences to be used in the main test.
These 30 test sentences were divided into
two versions to make the test shorter and
to  discourage  the  participants  from
developing  any  strategies.  Each  version
included  10  subject-biased  sentences
(five  DP1  and  five  DP2  subject-biased
sentences).  For  each  DP  biased  item,
there was an unbiased item. So there were
10  unbiased-subject  sentences  in  each
version. If the DP1-biased-subject item of
a  set  appeared  in  Version  1,  it  was
replaced with DP2-biased-subject item in
Version  2.  The  unbiased-subject  item  of
each set was common in both versions.
 
Filler  Sentences  and  Comprehension
Questions
Forty  filler  sentences  were  developed.
Like test sentences, all the fillers included
RCs  which  were  introduced  with  a
relative  pronoun.  Thirty-five  out  of  40
filler sentences were not ambiguous. The
remaining  5  filler  sentences  were
ambiguous  and  looked  exactly  like  test
sentences,  but  it  was  the  unambiguous
part that was questioned.  
The filler items were used to ensure that
the participants attended to the content of
the  sentences  they  read  on  the  monitor.
So  participants  whose  score  on  filler
sentences  was  less  than  90%  were
excluded.  
All the sentences were followed by a fill-in-the-blank comprehension question; (1)
to find out which DP was recognized by
participants as the host of the RC; (2) to
check whether participants paid attention
to the content of the test. With regard to
fillers,  the  questions  only  served  the
second  purpose  because  their  answers

could  be  checked  for  accuracy.  The
answers  to  the  ambiguous  experimental
items  simply  indicated  a  participant's
preference and could not be checked for
their truth value. A sample item of a test
sentence  followed  by  the  corresponding
comprehension  question  is  provided
below.
(12)  The  doctor  saw  the  nurse  of  the
student who was in the yard.
………………was in the yard.  
 
To ensure that ordering had no effect, the
item  presentation  order  in  the  two
versions  of  the  main  experiment  was
reversed.  So  there  were  four  versions
including 63 items: 3 warm-up sentences,
20 test sentences, and 40 fillers. Warm-up
and filler sentences were the same across
the four versions.
 
Procedure
The  OST  and  the  main  test  were
administered  in  two  different  sessions
with  a  one  week  interval.  The  Rapid
Serial  Visual  Processing  paradigm  (in
which parts of a sentence are presented in
a  time-controlled  manner,  this  short
period of time between the presentations
of different parts of the sentence prohibit
development  of  any  strategy  by  the
participants)  was  adopted  and  each
stimulus was presented on the screen for
nine  seconds  in  black  letters  on  a  white
background.  This  interval  was  decided
upon  based  on  the  findings  of  a  pilot
study.  After  that  a  fill-in-the-blank
comprehension question appeared on the
screen  and  participants  were  required  to
provide  an  answer  to  the  question  by
typing in the specified space. Participants'
answers were automatically recorded.  
 
Scoring System
We  identified  whether  the  participants
referred  to  DP1  or  DP2  in  each  item.  In
case a participant completed the sentence
with the whole phrase “DP1 of DP2”, this
would count as a DP1 choice because it is
the head the whole phrase. Out of the 580
answers  (29  participants  ×  20  items)
provided,  378  were  DP1  and  178  were
DP2 and 24 (4.14%) were not included in
the analysis because they referred neither
to DP1 nor to DP2.  
In  this  study,  WMC  and  proficiency
scores  were  not  used  to  classify
participants  into  groups  because  such
classification leads to “loss of statistical
power”  (1983,  p.  260).  Moreover,  as
Conway et al. (2005) put it “information
and  power  are  lost,  because  less
variability is captured by categories than
a continuum” (p. 782).
 
Results
Before analyzing the data, answers to the
comprehension  questions  following  the
fillers  were  checked  to  ensure  the
participants  had  read  the  sentences
attentively.  Those  with  accuracy  scores
lower  than  90%  (2  participants)  were
excluded from analysis.
Table  2  shows  the  frequency  and
percentage  of  DP1  and  DP2  choices
across  the  three  conditions.  In  the  DP1
biased condition, 67.4% of the responses
referred to DP1 while only 32.6% of the
responses  referred  to  DP2.  Similarly,  in
the  DP2  biased  condition,  a  large
percentage of replies, i.e., 72.5% referred
to  DP1,  but  just  27.5%  of  the  replies
referred  to  DP2.  In  the  unbiased
condition, too, DP1 responses were twice
as  many  as  DP2  replies  (66.1%  vs.
33.9%,  respectively).  As  can  be  seen,
irrespective  of  the  semantic
manipulation,  the  participants  rarely
selected DP2 as the antecedent of the RC.

The  following  table  presents  the
correlation  coefficient  between  WMC
and  proficiency  and  DP1  attachment
preferences across different conditions.

As  the  table  displays,  WMC  correlates
with  DP1  choices  negatively  and
significantly  in  all  conditions,  meaning
that  as  WMC  increases,  preference  for
DP1  decreases.  As  for  proficiency,  the
correlations are negative in all cases but
do not reach significance.  
The  Pearson  correlation  between  total
DP1 choices (the total of DP1 choices in
the three conditions) and WMC indicates
that the effect size of the correlation was
medium (r = -.442, N = 29, R2 = .195). On
the  other  hand,  the  Pearson  correlation
between  total  DP1  choices  and
proficiency  was  not  statistical  and  the
effect size turned out to be small (r = -.225,
N = 29, R2 = .051).
The  following  figure  shows  the  scatter
plot  for  the  correlation  between  WMC
and  total  DP1  choices.  The  loess  curve,
having  smoothed  the  data,  shows  that
preference  for  DP1  across  different
conditions decreases as WMC increases.
As is evinced in the figure, there is a sharp
slope  indicating  that  when  WMC
increases,  preference  for  DP1  severely
drops,  while  those  with  lower  WMCs
prefer DP1.

We  used  a  Mixed  Effect  Model  to
evaluate  the  relation  between  different
variables with preference.  All statistical
analysis were carried out using R (R Core
Team  (2014).  R:  A  language  and
environment  for  statistical  computing.  R
Foundation  for  Statistical  Computing,
Vienna,  Austria.  URL).  Among  the  9
models including the simplest model with
only one variable to the most complicated
one  with  all  interactions,  we  chose  the
model with the best AIC. The inclusion of
interactions  did  not  improve  the  model,
so  they  were  not  included  in  the  final
model.  This  model  includes  the  cross
random  coefficient  of  condition  and
random intercept defined in the levels of
subject and item and with main effects of
condition,  standardized  proficiency  and
WMC.  The  results  are  demonstrated  in
Table  4.  These  results  revealed  no
significant  effect  for  condition  and
proficiency  (ps  >  .05),  but  WMC  had  a
significant effect (P < .05), indicating that
those  who  selected  DP1  had  a
significantly  lower  WMC.  In  this  way,
semantic manipulation was found to play
no role in the participants’ preferences.

Experiment 2
This  experiment  explored  whether
Persian-speaking  EFL  learners'  RC
attachment  preference  is  affected  by  the
semantic relationship between the verb of
the main clause and one of the DPs in a
complex DP across different proficiency
and WMCs.
 
Method
Participants
A  total  number  of  33  participants  (14
females)  were  selected  from  the  same
pool  as  those  in  Experiment  1.  None  of
them  had  participated  in  Experiment  1.
Their  ages  ranged  between  18-22  years.
Two participants were excluded because
they  did  not  show  up  for  the  main  test.
And 2 more were excluded since they did
not satisfy  the  requirement on the  fillers
(90% accuracy).
Descriptive  statistics  for  WMC  and
Proficiency are documented in Table 5.

Martials  
The  OST,  OPT,  and  the  practice  test,
warm-up sentences,  filler sentences,  and
fill-in-the-blank  comprehension
questions  were  similar  to  those  used  in
Experiment  1  in  terms  of  structure  and
number.  Test  sentences  that  were
different are elaborated below.
 
Test Sentences
The  structure  of  experimental  sentences
was similar to that used in Experiment 1,
but  in  Experiment  2  a  semantic
relationship was established between the
verb  of  the  main  clause  and  one  of  the
DPs  in  the  complex  DP.  Each  pair  of
words,  i.e.,  the  DP  and  the  verb,  was
chosen  on  the  basis  of  their  semantic
relationship.  The  following  website  was
consulted  to  determine  the  semantic
relationships, http://semantic-link.com.
Based  on  the  relationship  between  the
verb of the main clause and either of the
two  DPs  in  the  complex  DP,  test
sentences  were  categorized  into  three
types:  DP1-biased  verb  with  an
occupational  semantic  relationship
between the verb of the main clause and
DP1; DP2-biased verb with a relationship
between the verb of the main clause and
DP2; and Unbiased verb with no specific
relationship between the verb of the main
clause  and  either  of  the  DPs.  Examples
for each category are provided below:
 
DP1-biased verb  
(13)  Someone  cured  the  doctor  of  the
teacher who was preparing to go home.
 
DP2-biased verb
(14)  Someone  scored  the  doctor  of  the
teacher who was preparing to go home.
 
Unbiased verb
(15)  Someone  saw  the  doctor  of  the
teacher who was preparing to go home.
 
Sentences  (13),  (14),  and  (15)  are
regarded  as  a  set  of  experimental
sentences.  Similar  to  Experiment  1,  a
norming study was conducted to establish
the semantic relatedness of the verbs and
DPs.  
A sample experimental item and a sample
filler  item  followed  by  their
corresponding  comprehension  questions
are provided below.
 
(16)  Someone  cured  the  doctor  of  the
teacher who was preparing to go home.
………………was preparing to go home.
 
(17)  Someone  knew  the  police  officer  to
whom I gave my passport.
 I gave him my………………… .
 
As  in  Experiment  1,  there  were  four
versions of the main test and each version
included 63 items: 3 warm-up sentences,
20 test sentences, and 40 fillers.
 
Procedure
Procedure  was  identical  to  that  of
Experiment 1.
 
Scoring System
The scoring system was the same as that
of Experiment 1.  
 
Results
Table  6  presents  the  frequency  and
percentage  of  DP1  and  DP2  choices
across the three semantically manipulated
conditions.  As  in  the  previous
experiment,  the participants’ preference
for DP1 was stronger for DP1 than DP2
irrespective  of  the  condition.  In  each  of
the  three  conditions,  preference  for  DP1
was  twice  more  than  that  for  DP2.  This
finding  is  comparable  to  that  in
Experiment 1.

Table  7  displays  the  correlation
coefficient  between  WMC  and
proficiency  with  DP1  attachment
preferences  across  the  three  conditions.
In  the  same  vein  as  in  Experiment  1,
WMC correlated significantly negatively
with  DP1  choices  in  all  conditions  and
with  the  total  DP1  choices.  Proficiency,
too, had a negative correlation with DP1
preferences  but  failed  to  reach
significance.

The  Pearson  correlation  between  total
DP1 choices and WMC was -.508 and the
effect  size  was  medium  (R2  =  .258).  On
the  other  hand,  the  Pearson  correlation
between  total  DP1  choices  and
proficiency  was  not  statistical  and  the

effect size was small (r = -.252, N = 29,
R2 = .063).
The  figure  below  shows  the  scatter  plot
for  the  correlations  between  WMC  and

total  DP1  choices.  As  WMC  increases,
preference for DP1 decreases.
Figure 2. Scatter plot for the correlations
between WMC and total DP1 choices in
Experiment 2
 
To  analyze  the  data,  we  followed  the
same  procedure  as  in  Experiment  1,  by
evaluating different models including the
simplest one with only one variable to the
most  complicated  one,  with  all
interactions,  and  ending  up  in  a  model
with  the  lowest  AIC.  The  best  fitting
model  includes  the  crossed  random
coefficient  of  condition  and  random
intercept defined in the levels of subject
and  item  and  with  main  effects  of
condition,  standardized  proficiency  and
standardized  WMC.  Table  8  reports  a
summary of all coefficients for the choice
of DP. The main effect obtained belonged
to  WMC  (p  =  .008),  with  more  choices
for  DP1  by  those  having  lower  WMC.
Proficiency  and  condition  had  no  main
effects (ps > .05).

Discussion and Conclusion
This  study  investigated  whether
ambiguity resolution by Persian-speaking
learners  of  English  as  an  L2  is  sensitive
to priming one of the DPs in the complex
DP  by  creating  a  semantic  relationship
between  the  subject/  verb  of  the  main
clause and one of the DPs. The impact of
proficiency  and  WMC  as  individual
properties  of  the  participants  was  also
examined.  
The  findings  showed  no  priming  effect.
In DP1 related condition, as in The doctor
saw the nurse of the pupil who was in the
yard, the first DP, the nurse was expected
to  be  more  accessible,  following  the
Spreading  Activation  Model  (Collins  &
Loftus  1975;  Dell,  1986),  and  hence
selected as the host of the ambiguous RC,
but the second DP the pupil was not. As
far  as  L2  learners  with  low  WMC  and
proficiency are concerned, this prediction
was borne out, but this preference was not
due  to  DP1’s  being  primed  through  its
association  with  the  subject  of  the  main
clause,  because  even  in  the  DP2  related
condition, as in The teacher saw the nurse
of  the  pupil who  was  in the  yard,  where
DP2  the  pupil  was  expected  to  be  more
accessible, DP1 the nurse was selected as
 
the  antecedent  of  the  RC.  The  same
results  were  obtained  in  the  unbiased
condition as well. As for the verb related
data, the findings were exactly the same.
In  the  three  semantically  different
conditions,  the  low  WMC  L2  learners’
preference  was  determined  by  the
Predicate  Proximity  principle,  which
favors attachment to DP1. Earlier results
from  different  studies  (Arabmofrad  &
Marefat, 2008; Marefat & Meraji, 2005)
have  shown  that  in  Persian,  the  native
language of the participants of this study,
in which adjuncts can occur between the
verbs  and  their  complements,  Predicate
Proximity  is  operative  rather  than  Late
Closure.  Thus,  semantic  manipulation
doesn’t seem to influence the L2 learners’
preference  for  the  antecedent  of
ambiguous RCs. The theoretical account
for  these  findings  may  be  viewed  as
consistent  with  theories  in  which  a
structural  analysis  of  a  newly
encountered word is constructed (Frazier
& Fodor, 1978; Frazier & Rayner, 1982).
But  L2  learners  with  higher  WMCs
favored  DP2;  their  preference  was  like
that of the English native speakers.
English RC attachment preferences of the
L2 learners were not associated with their
proficiency.  This  result  is  inconsistent
with  Miyao  and  Omaki’s  (2006)
developmental stages of L2 processing.   
The  factor  shown  to  play  a  role  in  RC
attachment preferences of the L2 learners
in  this  study  was  WMC.  This  finding
provides  evidence  in  support  of  the
‘chunking’  hypothesis,  suggested  by
Swets  et  al.  (2007).  Based  on  this
hypothesis,  low  span  participants,  not
having  adequate  resources,  pause  at  the
boundary  between  the  complex  DP  and
the RC and, in this way, chunk DP1 and
DP2  into  a  single  unit,  producing,  as  a
result,  a  DP1  attachment.  In  this  study,
too,  participants  with  high  WMC  may
have taken in longer chunks, without any
break  at  the  boundary  between  the
complex  DP  and  the  RC  and  thus  have
attached  low.  Since  the  material  in  this
study was not presented chunk by chunk,
participants had the opportunity to chunk
it  themselves;  thus,  participants  with
different  WMCs  could  chunk  the
sentences differently leading to different
attachment preferences.  
This  finding  is  also  consistent  with  the
predictions  of  the  Implicit  Prosody
Hypothesis (Fodor, 2002) which predicts
that the prosodic grouping of the complex
DP  and  RC  in  a  way  that  a  pause  is
inserted  after  the  second  DP  -which  is
proposed  to  be  the  case  for  participants
with  low  WMC-  leads  to  a  DP1
attachment preference.