Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
An investigation of the pseudohomophone effect: where and when it occurs
(USC Thesis Other)
An investigation of the pseudohomophone effect: where and when it occurs
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
AN INVESTIGATION OF THE PSEUDO HOM OPH ONE EFFECT:
W HERE AN D W HEN IT OCCURS
ty
A lan Scott Petersen
A Thesis Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY O F SOUTHERN CALIFORNIA
In Partial Fulfillm ent of the
R equirem ents for the D egree
MASTER O F ARTS
(Psychology)
A ugust 1995
C opyright 1995 A lan Scott Petersen
UNIVERSITY O F S O U T H E R N C A L IFO R N IA
THE GRA DUATE SC H O O L
U N IV ER SITY PARK
LO S ANQKLCS. C A L IFO R N IA SOOOT
This thesis, written by
Alan S c o tt P ete rsen
under the direction of h.±B Thesis Committee,
and approved by all its members, has been pre
sented to and accepted by the Dean of The
Graduate School, in partial fulfillment of the
requirements for the degree of
M aster o f A rts (P sych ology)
D**m
c./SL
Date July...1 2 . 1995
THESIS COMMITTEE
ACKNOW LEDGEM ENTS
I w o u ld like to th an k m y frien d a n d office m ate, Kim
D augherty, for all his help and advice over the years. I w ould especially
like to thank Beatrice D upuy, w ho w as alw ays su p p o rted m e in every
en d eav o r.
This research w as su pported in part by funds aw arded to M ark S.
Seidenberg by the N ational Science Foundation (grant DBS-9120415).
TABLE OF CONTENTS
ACKNOW LEDGEM ENTS ................................................................................. ii
LIST OF FIGURES .............................................................................................. v
ABSTRACT .......................................................................................................... vi
CHAPTER 1: INTRODUCTION ...................................................................... 1
1.1 The P seudohom ophone Effect .................................................... 2
1.2 Issues C oncerning P seudohom ophone Stim uli .................... 4
CHAPTER 2: REPLICATING THE PSEUDOHOM OPHONE EFFECTS 10
2.1 M ethod ................................................................................................ 10
2.1.1 Subjects .............................................................................. 10
2.1.2 M aterials and Design .................................................... 10
2.1.3 Procedure .......................................................................... 11
2.2 Results ................................................................................................ 12
2.2.1 N am ing Task .................................................................... 12
2.2.2 Lexical Decision Task .................................................... 14
2.2.3 A nalyses of subject speed ............................................ 15
2.3 D iscussion ......................................................................................... 18
CHAPTER 3: A DELAYED N A M IN G EXPERIMENT .......................... 24
3.1 M ethod ................................................................................................ 25
3.1.1 Subjects .............................................................................. 25
3.1.2 M aterials and Design .................................................... 25
3.2 R esults ................................................................................................ 27
3.3 Discussion ...................................................................................... 29
CHAPTER 4: GENERAL D ISC U SSIO N ....................................................... 33
4.1 Are there lexical nodes in lexical m e m o ry ? ............................ 35
4.2 C ould there be sem antic involvem ent in nonw ord nam ing
u n d er other c o n d itio n s? ................................................................ 37
REFERENCES .................................................................................................... 40
i v
LIST OF FIGURES
Figure 1. N am ing and lexical decision latencies as a function of subject
speed ................................................................................................................ 17
Figure 2. Pronunciation m odel including units and connections
relevant to generating phonological and articulatory o u tp u t .... 22
Figure 3. M ean nam ing latencies in Experim ent 2..........................................28
v
ABSTRACT
T his thesis exam ined the processing of p seu d o h o m o p h o n es
(nonw ords w hich sound like w ords, like BRANE or JOAK) an d non
pseudohom ophones (nonw ords w hich do not sound like w ords, like
BRONE and JOAP). First, pseudohom ophones yielded faster nam ing
la te n c ie s a n d s lo w e r lex ica l d e c is io n tim e s th a n n o n
pseudohom ophones, replicating previous results (M cCann & Besner,
1987; M cCann, Besner, & D avelaar, 1988). Post-hoc analyses revealed
that the effect w as related to subject speed in lexical decision but not
nam ing. Next, the non w ords w ere presented in im m ediate and delayed
nam ing conditions. A significant pseudohom ophone effect w as seen
that d id n o t change across delay conditions. This in d icates th at
pseu d ohom ophone effects in the lexical decision and nam ing tasks
h a v e d iffe re n t b a ses. In lexical d e c isio n , th ey re fle c t the
p se u d o h o m o p h o n e 's a c tiv a tio n of p h o n o lo g ic al an d se m an tic
inform ation associated w ith w ords. In nam ing, they reflect differences
in the ease of articulating pronunciations. Im plications of these results
concerning m odels of w ord recognition are discussed.
CHAPTER 1: INTRODUCTION
R esearch on visual w ord recognition focuses on the types of
kn o w led g e rep re se n ta tio n s an d processing m echanism s involved in
reading an d pronouncing w ords. It is felt that an u n d e rstan d in g of the
u n d erly in g principles that are involved in the recognition of w ords in
isolation will give insight into general read in g processes, ju st as an
aw aren ess o f g eo m etry is n eed ed for a com plete u n d e rsta n d in g of
calculus.
The extent to w hich the various factors in w ord recognition can
be assessed d ep en d s on how well other potentially confounding factors
are controlled. For exam ple, observing the effects of fam iliarity or age
of acq u isitio n on w o rd recognition d e p e n d s u p o n co ntrolling o th er
factors such as w ord frequency. The in terco rrelated n a tu re of these
factors, how ever, m akes this task difficult. A lso, som e elem ents are
m ore objectively m easured than others; factors such as w ord length,
n u m b er of syllables, o r initial phonem e. This is in contrast to factors
su ch as frequency, w hich are m ore statistical in n a tu re a n d can
th e re fo re be an a d d itio n a l so u rce o f e rro r (G ern sb ach er, 1984).
1
T herefore, it is very im p o rtan t to ensure th at effects sh o w n in stu d ies
are replicated using additional m aterials.
1.1 The Pseudohomophone Effect
The concerns expressed above are also p resen t w hen dealing
w ith p ro n o u n ceab le n o n w o rd s (stim uli th at can be p ro n o u n ced b u t
w hich are not actual w ords). M any stu d ies have exam ined w hat has
becom e know n as the pseudohomophone effect. P seu d o h o m o p h o n e s
are n o n w o rd s th at so u n d like actual w ords (e.g. BRANE, FOW ND). In
e x p e rim e n ts , th e se ty p e s o f stim u li a re c o m p a re d to n o n
p seu d o h o m o p h o n es, n o n w o rd s w hich are p ro n o u n ceab le b u t d o not
so u n d like real w ords (e.g. FRANE, YOW ND). By com paring these tw o
types of n o n w o rd stim uli, the effects of phonological inform ation in
w ord recognition can be assessed: if people access phonological codes
w hen read in g , pseu d o h o m o p h o n e stim uli w ould be expected to yield
different results than their non-p seu d o h o m o p h o n e counterparts.
Tw o articles by M cCann an d Besner (1987) and M cCann, Besner
an d D avelaar (1988) have p ro v id ed a g reat deal of d ata reg ard in g
pseudoh o m o p h o n es in lexical decision an d nam ing tasks. M cCann and
B e sn e r fo u n d s ig n if ic a n tly f a s te r n a m in g la te n c ie s fo r
p seu d o h o m o p h o n es com pared to n o n -pseudohom ophones, su ggesting
that the act of pronouncing a p seu d ohom ophone involves accessing its
real-w ord co u n terp art. In this w ay, the p ro n u n ciatio n of FO W N D is
2
assisted by FO UND, but YOW ND receives no such benefit.
M cC ann et al. (1988) used the m aterials from M cC ann an d
B e sn e r's s tu d y in a lexical d e c isio n task . T h ey fo u n d th a t
p se u d o h o m o p h o n es p ro d u ce d longer lexical decision latencies than
n o n -p seu d o h o m o p h o n es, a result o p p o site to th at fo u n d in nam ing.
These results w ere seen as su p p o rtin g the hypothesis th at processing a
p se u d o h o m o p h o n e in v o lv es accessin g its base w o rd in lexical
m em ory. In nam ing, the base w ord facilitates the pronunciation of the
p se u d o h o m o p h o n e , w h ereas in lexical d ecisio n , it in terferes w ith
resp o n d in g th at the pseudohom ophone is not a w ord.
In b o th stu d ie s, c o rre latio n calcu latio n s rev e ale d th a t the
response latencies for p seu d o h o m o p h o n es w ere u n related to the base
w o rd 's frequency. This w as taken as an indication that access to the base
w o rd w h en p ro cessin g the p se u d o h o m o p h o n e is n o t g o v e rn e d by
lexical frequency. The resu lts of these tw o stu d ies have im p o rta n t
im plications for theories of lexical representation and processing. First,
they su g g est th at the m ental lexicon m ust in clu d e rep resen tatio n s of
in d iv id u al w ords. This calls into question w hether m odels th at d o not
in c lu d e in d iv id u a l w o rd r e p r e s e n ta tio n s can a c c o u n t fo r
p seu d o h o m o p h o n e effects. In d istrib u te d rep resen tatio n m odels such
as S eidenberg an d M cC lelland's (1989), it is difficult to see how the
w ord BRAIN can influence the processing of the p seu d o w o rd BRANE,
3
as BRAIN does not have its ow n representation in the lexical netw ork.
Secondly, the results of these studies call into question the assum ption
th at frequency influences lexical access. This notion is hard to reconcile
w ith any existing m odel of w ord recognition, as a w o rd 's frequency is
alm ost universally assu m ed to influence its access from the lexicon.
M cCann and Besner (1987) and M cCann et al. (1988) proposed a m odel
of w ord recognition en d o w ed w ith the properties d em an d ed by their
em pirical results.
1.2 Issues Concerning Pseudohomophone Stimuli
The a rg u m e n ts m ad e in these stu d ie s are d e p e n d e n t on the
n o tio n th a t th e p se u d o h o m o p h o n e a n d n o n -p s e u d o h o m o p h o n e
stim uli differ only in the fact that the p seu d o h o m o p h o n es so u n d like
real w ords. The validity of this assum ption needs to be exam ined, as
th e r e e x is ts a s im p le a lte r n a tiv e e x p la n a tio n fo r th e
p se u d o h o m o p h o n e effects they have seen (S eidenberg, 1992). For
exam ple, the p se u d o h o m o p h o n e stim uli m ay m ore closely resem ble
real w o rd s th an their n o n -p seu d o h o m o p h o n e co u n terp arts. C reating
n o n w o rd s th at so u n d like real w o rd s m ay c o n stitu te u sin g m ore
com m on com binations of letters an d p honem es than w h en creating
n o n w o rd s th at do not so u n d like real w ords. The FO W N D-YO W ND
pair illustrates this: m ore w ords in English begin w ith F than Y, m ore
begin w ith FO than YO, an d m ore begin w ith /fa e w / th an /y a e w /.
4
W hen one w ants to create a non-pseu d o h o m o p h o n e co u n terp art to
FOW ND, one m ust choose a relatively uncom m on onset, such as Y, as
o th e r m o re c o m m o n o n s e ts w o u ld th e m s e lv e s c re a te
pseudohom ophones w hen com bined with OW ND (e.g. B, S, W, H, M,
R). If this is a com m on characteristic of the stim ulus pairs, it could
explain all of M cCann and colleagues' findings: n o nw ords that are
m ore sim ilar to actual w ords will be easier to pronounce than less
w ordlike non w ords, as in Seidenberg and M cC lelland's (1989) m odel,
and they will be harder to distinguish from w ords in a lexical decision
task.
M cCann and Besner were careful to assess w hether their results
could be due to the use of different onsets in the pseudohom ophone
and non-pseudohom ophone item s. First, they calculated m ean bigram
frequencies, using the M ayzner and T resselt (1965) norm s. The
pseudohom ophones had a slightly higher m ean sum m ed positional
bigram frequency (147) com pared to the non-pseudohom ophones (125),
b u t this difference w as not significant by a t-test. A problem w ith
accepting this negative finding is that it is questionable w hether the
M ayzner and Tresselt norm s, which date from 1965 and w ere based on
a c o rp u s of 20,000 w o rd s, p ro v id e a ro b u st estim ate of bigram
frequencies. W e reco m p u ted the bigram frequencies using the 1
m illion w ord Kucera and Francis corpus (the Solso & Juel, 1980, norm s
5
are also based on this corpus). The algorithm exam ined each bigram in
a letter string, searched the Kucera and Francis corpus for the w ords of
the sam e length that had the sam e bigram in the sam e position, and
sum m ed their frequencies. The m ean sum m ed positional bigram
frequency for the non-pseudohom ophone item s is 3711, w hereas the
m ean for the pseudohom ophone item s is 4397, and this difference is
statistically reliable, t(79) = 2.979, p < .01.
This orthographic difference also appears if the calculation is
based instead on the frequencies of the onsets alone. For this analysis,
w e counted the num ber of w ords in the Kucera and Francis corpus
beginning w ith the each onset (e.g., PR in PREET; P in PAIT). The
m ean num ber of w ords containing the non-pseudohom ophone onsets
is 972, the m ean for the pseudohom ophone onsets is 1223, and the
difference is statistically reliable, t(79) = -2.98, p < .01. Sim ilar results are
obtained if instead of counting the num ber of types containing the
on sets, w e sum the frequencies of th e tokens: For th e n o n
pseudohom ophone stim uli the m ean sum m ed frequency is 20096, for
the p se u d o h o m o p h o n e stim u li it is 24332, a n d the d ifference
approaches significance, t(79) = -1.74, p < .09. In sum m ary, M cCann and
Besner's null effect of bigram frequency appears to have derived from a
lack of robustness in the estim ates provided by M ayzer and Tresselt
(1965). E stim ates based on larger sam ples indicate a system atic
6
difference betw een the stim uli.
M cCann an d Besner also assessed w h eth er the stim uli differ in
term s of an o th e r m easu re of o rth o g rap h ic stru c tu re , C o lth e art's N
(C oltheart, D avelaar, Jonasson & Besner, 1977), w hich is the n u m b er of
w o rd s th at can be d eriv ed from a letter strin g by m aking one-letter
su b stitu tio n s. M cCann an d Besner rep o rt that th eir stim uli differed
significantly on the N m easure; the p seu d o h o m o p h o n e item s have
h ig h er N values th an the n o n -p seu d o h o m o p h o n e item s. H ow ever,
they also p erform ed p o st hoc analyses in d icatin g th at w hile N w as
s ig n ific a n tly c o rre la te d w ith la te n c ie s a n d e rro rs o n n o n
p se u d o h o m o p h o n e item s, it w as not for p se u d o h o m o p h o n e s (see
M cCann & Besner, 1987, p. 18, Figures 1-2). G iven the lack of an effect of
N on the pseudohom op hone item s they concluded that it could not be
the source of the overall p seu d o h o m o p h o n e-n o n -p seu d o h o m o p h o n e
difference. T he problem w ith the regression p lo tted in th e figures,
h o w ev er, is th at it co m pares C o lth eart N a n d the mean resp o n se
latency for all of the item s w ith a given N. The n u m b er of item s
c o n trib u tin g to each m ean th erefo re varied. W e recalcu lated the
correlations betw een C oltheart N and the actual item m eans for these
stim uli. The C oltheart N s for this analysis w ere generated using the
a lg o rith m for this p u rp o se in the MRC P sycholinguistics D atabase
(W ilson, 1988) an d differ slightly from those reported by M cCann and
7
Besner b u t yield sim ilar results. The correlation betw een C oltheart N
an d latency for the n o n -p seu d o h o m o p h o n e item s, -.27, is significant,
w hereas the correlation for the p seu d o h o m o p h o n e item s, -.11, is not,
replicating M cCann an d Besner's findings. H ow ever, the effect for the
n o n -p seu d o h o m o p h o n e item s is d u e to the item s w ith no neighbors
(i.e., C o lth e art N values of 0), of w hich th ere w ere 16 in the n o n
p s e u d o h o m o p h o n e list a n d 7 a m o n g th e p se u d o h o m o p h o n e s .
E xcluding these zero-neighbor item s from the n o n -p seu d o h o m o p h o n e
list, the correlation betw een C oltheart N and latency becom es -.075, also
non-significant. T herefore, it ap p ears that the differences betw een the
p se u d o h o m o p h o n e an d n o n -p se u d o h o m o p h o n e stim u li are largely
d u e to differences in the num ber of zero-neighbor item s.
F in a lly , M cC an n a n d B esner in c lu d e d tw o a d d itio n a l
experim ental conditions in o rd er to assess the effects of the confounds
b etw een the stim uli. The first w as the d elay ed n am in g condition,
w hich yielded a significant residual p seu d o h o m o p h o n e effect th at w as
sm aller in m ag n itu d e th at in im m ed iate nam ing. T hese resu lts are
d iscu ssed in the th ird ch ap ter of this thesis. Second, M cC ann and
Besner conducted a control experim ent because, as they n o ted (p. 17),
" A lth o u g h th e a d v a n ta g e [for p se u d o h o m o p h o n e n o n w o rd s in
E x p erim en t 1] w as sig n ifican tly red u c ed [in th e d elay ed n a m in g
condition], it m ight be arg u ed that the additional em phasis on speed of
8
resp o n d in g in the delayed condition w ould reduce the m ag n itu d e of
any stim ulus effect com pared w ith online responding." The additional
c o n tro ls in E x p e rim e n t 2 in v o lv e d d e riv in g n e w n o n
p s e u d o h o m o p h o n e s fro m th e p s e u d o h o m o p h o n e a n d n o n
p seu d o h o m o p h o n e stim uli by changing the vow els but retain in g the
onsets. T hus, FOW ND becam e FOIND and YOW ND becam e YOIND.
Latencies for these stim uli d id not differ. This in d icated that the
pseu d o h o m o p h o n e ad vantage in the original experim ent w as not due
to the differences betw een the onsets used in the tw o lists of stim uli.
These control stim uli introduce a new problem , how ever: w hile they
retain the onsets from the originals they contain different vow els. The
effect of this change w as to elim inate the difference in C o lth eart N
b etw een the tw o lists. The m ean for the n o n -p seu d o h o m o p h o n e-
deriv ed stim uli is 3.2, the m ean for the p seu d o h o m o p h o n e-d eriv ed
stim uli is 3.5, and this difference is not significant, t(79) = 1.31, p > .15.
T hus, the p seu d o hom ophone effect in the original M cCann and
Besner experim ent appears to be due to the fact that the stim uli differed
in term s of o rth o g rap h ic p ro p erties indexed by m easu res su ch as
C o lth e a rt N a n d b ig ra m fre q u e n c y , n o t th e fact th a t th e
p se u d o h o m o p h o n e s so u n d e d like w ords. T his conclusion is also
consistent w ith the results of the experim ents described in this thesis.
9
CHAPTER 2: REPLICATING THE PSEUDOHOM OPHONE EFFECTS
T his first e x p e rim e n t a tte m p te d to re p lic a te the basic
pseudohom ophone effects in nam ing and lexical decision.
2.1 M ethod
2.1.1 Subjects. F orty-eight subjects from the U niversity of
Southern California com m unity participated in the experim ent, half in
the nam ing condition, an d the o th er half in the lexical decision
condition. All w ere native speakers of English. For their participation,
subjects either w ere paid or received course credit.
2.1.2 M aterials an d D esign. T hirty-tw o sets of m onosyllabic
n o n w o rd stim uli w ere constructed. Each set w as com posed of four
nonw ords form ed by crossing tw o onsets w ith tw o rim es to form tw o
p seu d o h o m o p h o n es and tw o non-p seu d o h o m o p h o n es (e.g. HOAP,
JOAK, H O A K , JO A P). The stim u li w ere d iv id e d in to tw o
counterbalanced lists: for half the quadruples, the HOAK and JOAP pair
appeared in List A and the HOAP and JOAK pair in List B. This m ethod
of assignm ent w as reversed for the other half of the quadruples. This
d iv isio n en su red th at onset+ nucleus a n d rim e p a tte rn s w ere not
10
repeated w ithin subjects.
Sixty-four m onosyllabic w ords w ere used as filler item s for the
lexical decision task. W ords th at are h o m o p h o n es of the n o n w o rd
item s w ere excluded. The average Kucera an d Francis frequency of the
w ords w as 24.2 (sd 17.3, range 7-72).
2.1.3 P ro c e d u re . In both n am ing an d lexical decision tasks,
stim uli w ere presented in d iv id u ally in the center of the screen of an
IBM M odel 70 PC ru n n in g the M icro Experim ental L aboratory (MEL)
softw are package. Subjects w ere seated in front of the co m p u ter at a
com fortable distance in a dim ly lit, quiet room . For the nam in g task,
subjects w ere instructed to nam e each letter string that appeared on the
screen quickly and accurately. Subjects spoke into a table m icrophone
co n n ected to the co m p u ter via a b u tto n box c o n tain in g a voice-
activated relay. For the lexical decision task, subjects w ere instructed to
use keyboard keys that w ere labeled "Y" and "N " to indicate w hether or
not the letter string that appeared w as an English w ord.
The sequence of events on each trial w as sim ilar for both tasks.
A fixation cross ap p ea re d at the center o f the screen for 200 m s,
follow ed by a 50 m s blank interval, follow ed by the target item in the
cen ter of the screen. The ta rg e t rem ained on the screen un til the
subject's nam ing o r lexical decision response. For the nam ing task, the
ex p erim en ter p u sh e d a b u tto n on an external b u tto n box to record
11
w h e th e r th e su b ject's p ro n u n c ia tio n w as co rrect o r in co rrect, or
w hether there had been a spoiled trial (e.g., early triggering of the relay
from an extraneous sound); this keypress initiated the next trial. For
the lexical decision task, there w as a 2 sec IT1 before the reappearance of
the fixation cross.
Subjects w ere presented w ith short lists of practice item s in o rder
to fam iliarize them w ith each task. Subjects in the lexical decision task
w ere then presented w ith either List A or List B, w hich w as m erged
w ith the list of w ord fillers. The random ly interm ixed stim uli w ere
presented in tw o blocks. Stim uli occurred in a different ran d o m o rd er
for each subject. The first block began w ith 4 w arm u p trials. There w as
no break betw een blocks. Stim uli w ere presented in the sam e m anner
for the nam ing subjects but w ith the w ord stim uli excluded. For both
tasks, half of the subjects w ere presented w ith each list.
2.2 R esults
2.2.1 N a m in g T ask. O verall results are su m m arized in Table 1.
L atency analyses w ere based on correct responses. A nalyses w ere
perform ed on both trim m ed and u n trim m ed d ata sets, w hich yielded
very sim ilar results. The trim m ing procedure treated as outliers scores
th at w ere greater than 3 stan d ard deviations above o r below a subject's
m ean latency. T his p ro ced u re affected 18 scores (1.2%), 8 from the
p s e u d o h o m o p h o n e c o n d itio n a n d 10 fro m th e n o n
12
pseudohom ophones. G iven that the scores w ere evenly d istrib u ted
across conditions, they w ere considered to be tru e outliers; th u s
analyses from th e trim m ed d ataset w ill be rep o rted . A nalyses of
variance w ere conducted using both subject (Fi) and item (F 2) m eans.
The factors w ere List (A or B), Block (first or second), an d Type
(pseud o h o m o p h o n e or non-pseudohom ophone). The List and Block
factors yielded n either significant m ain effects nor interactions w ith
other factors an d will not be considered further. P seudohom phones
w ere nam ed 15 m s faster than n o n -p seu d o h o m o p h o n es, a sm all
difference that w as reliable in both subjects and item s analyses, Fi(l,22)
* 12.180, p < .005, and F2(l,63) = 5.161, p < .05, respectively. Subjects
m ade essentially no errors in nam ing; accuracy w as greater than 99%
for both types of nonw ords. We exam ined the correlations betw een
p se u d o h o m o p h o n e n am in g latencies a n d tw o e stim a te s of the
frequencies of their base w ords: from the m illion-w ord Kucera and
Francis (1967) corpus and a corpus of 40 m illion w ords from the Wall
Street jo u rn al (M arcus, Santorini, & M arcinkiew icz, 1993). As in the
M cCann and Besner (1987) study, these correlations did not approach
significance (both 0 < r < .10).
13
TABLE 1
Mean Naming and Lexical Decision Latencies (msec)
TASK
N O N W O R D TYPE N am in g Lexical Decision
latency accuracy latency accuracy
P seu d o h o m o p h o n es 582 (19.4) 99.8 % 715(18.7) 90.2 %
N o n -p seu d o h o m o p h o n es 597 (19.8) 99.7 % 684 04.7) 94.1 %
Effect size -15 +31
Note: S tandard errors in parentheses.
2.2.2 Lexical D ecision T ask. Latency data w ere based on correct
responses in the nonw ord conditions. Analyses w ere again perform ed
on b o th trim m ed an d u n trim m ed d ata sets. T he 3sd trim m in g
procedure had a som ew hat different effect than in the nam ing task,
how ever: it affected 35 scores (2.4%), and m ore pseudohom ophones
scores (22) than non-pseudohom ophones (12). Because scores w ere not
evenly d istrib u te d across conditions, the analyses based on the
untrim m ed data are reported. The analyses of variance involved the
sam e factors as in the nam ing task. Latencies for pseudohom ophones
w ere 31 m s longer th an for n o n -p se u d o h o m o p h o n es, an d this
difference w as reliable in both subject and item analyses, Fi(1.22) =
110.79, p < .005, and F2(l,63) = 6.83, p < .05, respectively. Subjects were
m ore accurate on non-pseudohom ophones (94.1% correct) com pared to
14
p seu d o h o m o p h o n es (90.5% correct). This difference w as significant by
subjects, Fi(1.22) = 9.059, p < .01, and m arginally by item s, F2(l,63) =
3.419, p < .07. The correlations betw een the tw o estim ates of base w ord
frequencies a n d lexical decision latencies w ere again low an d n o t
statistically significant (both r = -.14).
2.2.3 A n aly ses o f su b ject speed. A dditional analyses exam ined
how the m a g n itu d e s of the p se u d o h o m o p h o n e effects re la te d to
su b jects' sp e e d of resp o n d in g . S eidenberg (1985a) fo u n d th at the
m ag n itu d e of effects of orthographic-phonological reg u larity in w ord
n a m in g d e p e n d e d on subject sp eed . It has been su g g e ste d th at
p h o n o lo g ical effects on lexical d ecisio n w ill be larg e r for low er
frequency w o rd s an d slow er subjects (e.g. Seidenberg, 1985b). These
o b serv atio n s led us to exam ine w h eth er p se u d o h o m o p h o n e effects
w ould vary in sim ilar ways.
For both tasks, three g ro u p s of 8 subjects w ere each form ed on
the b asis o f su b jects' m ean n o n w o rd n am in g latencies. F igure 1
p ro v id e s su m m ary d a ta for the n am in g (u p p e r panel) a n d lexical
decision (lo w er panel) tasks. T he effects of subject sp eed o n the
m ag n itu d e of the pseudohom ophone effect w ere quite different for the
tw o tasks. In nam ing, the pseudohom ophone effect w as essentially flat
across subject groups: num erically, the effects w ere -15, -14, and -16 m s
for th e slow est, m edium , an d fastest subjects, respectively. This w as
15
borne out by m ain effects of subject group, Fi(2,21) = 14.88, p < .001, and
stim ulus type, F i(l,21) = 11.45, p < .005, but no interaction betw een the
factors (F < 1).
For the lexical decision task, in contrast, the m agnitude of the
effect depended on subject speed. There w ere significant m ain effects of
group, Fi(2,21) = 52.84, p < .001, and stim ulus type, Fi(l,21) = 15.10, p <
.001, and a significant interaction betw een them , F](2,21) = 4.85, p < .02.
The effect w as 63 m s for slow er subjects, 35 m s for m edium subjects,
and 1 m s for the fastest subjects. Accuracy did not differ reliably across
groups (94, 91, and 93 percent correct for the fast, m edium , and slow
subjects, respectively).
16
1000
JS 900
u
c
a >
3
E
(0
o
5
Pseudohomophone
□ Non-pseudohomophone
* : ri-.'i
s j i j i i i j
| |
<0 10 0 0
>»
u
900
2i BOO
700
600
O
M
O
•
a
8
£ 500
x
o
400
Slowest Medium Fastest
Subject Speed
Pseudohomophone
E 3 Non-pseudohomophone
1 ! 1 } !
iiji
! ) [! I !
! K M
Slowest Medium Fastest
Subject Speed
Figure 1. Naming and lexical decision latencies as a function of
subject speed.
17
2.3 Discussion
P seu d o h o m o p h o n e s w ere n am ed m ore ra p id ly th an n o n
p se u d o h o m o p h o n es, b u t yielded longer lexical decision latencies,
replicating the overall p attern s found by M cCann and Besner (1987)
and M cCann et al. (1988), using m ore closely controlled stim uli. As in
their studies, pseudohom ophone latencies d id not reliably correlate
w ith frequencies of the w ords from w hich they w ere derived. The
lexical decision results are roughly sim ilar in m agnitude to those in
M cCann et al.'s (1988) Experim ent 1 (20 ms in M cCann et al.; 31 ms in
ou r experim ent). O u r stu d y adds the additional inform ation that the
size of the effect in lexical decision decreases as a function of subject
speed w ith this task. M oreover, there is w hat seem s to be a floor effect
for the fastest subjects, w ho show no pseudohom ophone effect at all.
For the nam ing task, the basic p attern is as in M cCann and Besner
(1987); how ever, the m agnitude of the effect (15 ms) is closer to that in
their delayed nam ing condition (11 ms) than their im m ediate nam ing
condition (35 ms). In this task, the size of the effect did not vary w ith
subject speed.
The differing effects of subject speed provide an im portant hint
that the pseudohom ophone effects derive from different sources in the
tw o tasks. T he lexical decision d ata are c o n sisten t w ith earlier
suggestions that there are greater phonological effects in tasks such as
18
lexical decision for slow er subjects and low er frequency w ords (e.g.
M cCusker, H illinger, & Bias, 1981; Jared & Seidenberg, 1991). If, as in
the recurrent netw orks described by Plaut and Shaltice (1993) and Plaut,
M cClelland, Seidenberg and Patterson (1994), phonological inform ation
is activated over tim e, slow er subjects will show larger phonological
effects sim ply because there is m ore tim e for this inform ation to
becom e available. Slow er subjects m ay also rely m ore on using this
in fo rm atio n in m aking th eir responses. W hen the stim u lu s is a
pseudohom ophone, the fact that its phonological code is that of a w ord
interferes w ith m aking a nonw ord response. This could occur either
because the phonological code itself is associated w ith a w ord, or
because it activates sem antic inform ation associated w ith the w ord
(Van O rden, Johnston, & Hale, 1988).
T h e fa s te s t su b je c ts sh o w n o in te rfe re n c e in th e
pseudohom ophone condition, w hich suggests tw o possibilities. O ne is
th a t th ey w ere able to m ake th eir resp o n se s w ith o u t u sin g
phonological inform ation. For exam ple, they m ight decide w hether
the stim ulus is a w ord or not directly from the orthography. A second
possibility is that the fastest subjects also activated phonological
inform ation in perform ing the task, b u t it d id not interfere w ith
m aking the response. For exam ple, the subjects m ight activate the
p h o n o lo g ic a l a n d s e m a n tic c o d e s a s s o c ia te d w ith th e
19
pseu d o h o m o p h o n e, as in the Van O rden et al. (1988) studies, and
rap id ly p erform a spelling check that allow s them to identify the
s tim u lu s as a n o n w o rd , p ro d u c in g n o d iffe re n c e b e tw e e n
pseudohom ophone and non-pseudohom ophone conditions (Lesch &
Pollatsek, 1993, develop this account). These interpretations are not
distinguished by the current data. This is largely because there are no
observable behavioral effects associated with the hypothesized spelling
check; it is assum ed to w ork equally well and take equal am ounts of
tim e for b o th p se u d o h o m o p h o n e a n d n o n -p se u d o h o m o p h o n e
stim uli. Thus there is no direct evidence as to w hether it has occurred
or not. A lthough further research is required to determ ine the role of
sp e llin g check m ec h an ism s in p ro c e ssin g h o m o p h o n e s a n d
p seu d o h o m o p h o n es, the im p o rtan t p o in t relevant to the p resen t
resu lts, h o w ev er, is th at b o th in te rp re ta tio n s su g g e st th at the
p seu d o h o m o p h o n e effect in the lexical decision task resu lts from
decoding processes that produce the activation of phonological or
sem antic codes by the pseudohom ophones.
T he n a m in g re su lts su g g e st a d iffe re n t lo cu s for the
p se u d o h o m o p h o n e effect seen in this task. S ubjects d iffered
significantly in overall nam ing latencies b u t not the m agnitude of the
pseudohom ophone effect. This im plies that the effect is due to factors
un related to decoding skill. O ne possibility is that the results reflect
20
processes involved in generating articulatory o u tp u t. T he nam ing
process is usually assum ed to involve decoding letters, generating a
phonological code, and producing articulatory m otor o u tp u t (Balota &
C hum bley, 1985; McRae, Jared, & Seidenberg, 1990; M onsell, Doyle, &
H aggard, 1989). This view is illustrated in Figure 2. The orthographic to
phonological com putation in the figure reflects processes involved in
identifying letters and generating the phonological codes associated
w ith them . Parts of these processes have been addressed in m odels
such as M cC lelland an d R u m elh art (1981) an d S eidenberg and
M cClelland (1989). The phonological to articulatory com putation in
the figure reflects the processes involved in converting an internal
phonological code into an explicit articulatory response. This process
can be envisioned in term s of a recurrent netw ork that produces time-
varying o u tp u t over u n its rep resen tin g articu lato ry features (e.g.
Jordan, 1986). Differences in subject nam ing speed reflect differences in
their ability to decode the in p u t and generate a phonological code.
Because the p seu d ohom ophone and non-p seu d o h o m o p h o n e stim uli
w ere closely m atched in term s of o rth o g rap h ic and phonological
properties, they do not differ w ith respect to these com ponents of the
nam ing process. By hypothesis, the stim uli do differ in term s of ease of
articulation, how ever. The pronunciations of pseudohom ophones are
fam iliar, overlearned articulatory p attern s associated w ith com m on
21
Context
M eaning
O rthography Phonology
MAKE
A rticulation
/ m / / A / /k/
time ------------------ ►
Figure 2: Pronunciation model including units and connections
relevant to generating phonological and articulatory output.
m onosyllabic w ords such as HOPE. The p ro n u n ciatio n s of n on
p seudohom ophones are novel articulatory p a ttern s such as /jO p /.
Thus, even if the stim uli do not differ in term s of factors relevant to
generating the phonological code, they m ay differ w ith respect to ease
of producing articulatory ou tp u t. In short, the process of generating
22
articulatory o u tp u t is affected by w hether the phonological p attern to be
p ro d u c e d is fam iliar o r n ot, b u t not by w h e th e r the o rth o g ra p h ic
stim u lu s th at g en erated this p attern is a w ord, p seu d o h o m o p h o n e, or
n o n -p se u d o h o m o p h o n e .
T hese o b serv atio n s m o tiv ated E xperim ent 2, w hich replicated
th e n a m in g stu d y u sin g b o th im m e d ia te a n d d e la y e d n a m in g
conditions. The m ethodology w as m odeled on earlier stu d ies using
the delay p a ra d ig m by Balota an d C h um bley (1987), M cC ann and
Besner (1987), M onsell et al., (1989), and McRae et al., (1990). In one of
their experim ents, for exam ple, M cRae et al. (1990) exam ined high and
low frequency h o m o p h o n es such as M AIN an d M ANE. T he low er
frequency w o rd M ANE p ro d u ced longer im m ediate nam in g latencies
than the higher frequency w ord M AIN, even though both involve the
sam e articulatory m otor ou tp u t. This resu lt suggested that frequency
affects the ease of c o m p u tin g a w o rd ’s p h o n o lo g ical code. T he
h o m o p h o n e s w e re n a m e d e q u a lly fast a fte r a d e la y , h o w e v er,
indicating no residual effects of frequency on g en eratin g articulatory
ou tp u t. The prediction tested in the next experim ent w as that if the 15
m s p se u d o h o m o p h o n e effect in n a m in g for th e stim u li u se d in
E x p erim en t 1 w as solely d u e to p ro cesses in v o lv ed in g en eratin g
a rtic u la to ry o u tp u t, it w o u ld be th e sam e m a g n itu d e in b o th
im m ediate an d delayed nam ing conditions.
23
CHAPTER 3: A DELAYED NAMING EXPERIMENT
In th is e x p e rim e n t, su b jects n a m e d a stim u lu s n o n w o rd
im m ediately after it a p p eared on the screen (as in E xperim ent 1) or
after a sh o rt o r long delay. Because subjects vary considerably in
n am in g sp eed s, delays w ere calibrated to the su b ject's o w n nam ing
sp eed (M cRae et al. 1990). The sh o rt delay w as set at 2sd over the
subject's baseline n am ing latency for a set of practice n o n w o rd s, and
the long delay w as 4sd over this baseline. Subjects w ere signalled to
resp o n d by the appearance of brackets around the visual target, as in
Balota an d C h um bley (1985) an d McRae et al. (1990); M cCann and
Besner (1987) used a tone for this purpose. In im m ediate nam ing, the
stim ulus ap p eared on the screen w ith brackets aro u n d it. In the delay
conditions, the stim ulus w as presented w ithout brackets, an d then the
brackets appeared after the short or long delay. Delays varied random ly
from trial to trial so that subjects could not anticipate w hen they w ould
have to begin responding. This procedure encourages subjects to begin
com puting the phonological code as soon as the nonw ord is presented,
rath er than w aiting until the response signal ap p ears before initiating
24
this com ponent of the nam ing process (see M onsell et al., 1989; McRae
et al., 1990).
3.1 M ethod
3.1.1 S u b jects. T h irty -six su b je cts from th e U n iv e rsity of
S outhern C alifornia com m unity w ere p aid for their participation. All
w ere native speakers of English.
3.1.2 M a terials an d D esign. The p seu d o h o m o p h o n e an d n o n
p se u d o h o m o p h o n e stim u li from E xperim ent 1 w ere used again as
s tim u li. T h irty a d d itio n a l p s e u d o h o m o p h o n e s a n d n o n
p seu d ohom ophones w ere used in the practice session to establish each
su b je c t's b a selin e n a m in g latency. S tim u li w ere p re s e n te d on a
M acintosh S E /30 co m p u ter usin g the PsyScope ex p erim en t control
softw are. Real-tim e m easurem ent w as controlled by an external tim ing
board interfaced to the com puter. A m icrophone connected to a voice-
activated relay w as placed ab o u t 10 cm from the subject's m o u th to
record spoken responses.
T he e x p e rim e n t in v o lv e d tw o stag es. S ubjects w ere first
presented w ith the random ized list of 30 pseudohom ophones and non
p se u d o h o m o p h o n e s in an im m ed iate n am in g p ro c e d u re th at w as
id en tica l to th a t u sed in E x p erim en t 1. T he m ean an d sta n d a rd
deviation of the subject's nam ing latency across correct trials w ere then
calculated an d used to establish the delay intervals used in the second
25
stag e of the ex p erim en t. The th ree delay co n d itio n s used w ere
Im m ediate N am ing (no delay), Short (2 sd over baseline), and Long (4
sd over baseline). Across subjects, the average baseline nam ing latency
w as 583 m s and the short and long delays averaged 977 and 1371 ms,
respectively.
In the experim ent proper, the stim ulus quadruples w ere divided
into 4 counterbalanced lists, with one m em ber of each q u ad ru p le and
equal num bers of pseudohom ophones and n o n -p seudohom ophones
in each list. Each subject saw all 4 lists, w ith the o rd er of these lists
counterbalanced across subjects. Each subject was presented with a third
of the stim uli at each delay, and assignm ent of the delay conditions to
stim ulus item s w as counterbalanced across subjects. The subjects were
in stru cted to pronounce the non w ord that ap p eared on the screen
w hen the pair of brackets appeared around it. They w ere inform ed that
the signal to respond w ould occur at different delays, and w ere given
practice trials to adjust to the procedure.
Each trial proceeded as follows: after a 200 m s fixation cross and a
50 m s blank screen, the nonw ord appeared in the center of the screen.
After the appropriate delay, the brackets appeared around the nonw ord;
in the im m ed iate n am in g co n d itio n , the b rack ets an d n o n w o rd
a p p eared sim ultaneously. N am ing latencies w ere m easured from the
presentation of the brackets to the initiation of the subject's response,
26
w hich caused the non w ord to be rem oved from the screen. The
experim enter then pressed a key on the keyboard to record w hether the
trial w as correct, incorrect, or spoiled (typically due to the m icrophone
not registering the subject's speech). The softw are w as program m ed to
not register in p u t from the voice key un til the brackets appeared
around the target. Therefore, if the subject responded prem aturely, the
screen w ould not clear and the latency w ould not be recorded.
3.2 R esults
T rials on w hich the subject resp o n d ed before the brackets
appeared or w hen the subject's response failed to trigger the voice key
(1.6% of trials) w ere excluded from latency analyses, as w ere trials w hen
the subject nam ed the nonw ord incorrectly (2.4% of trials). Both
trim m ed and untrim m ed data were analyzed, as in Experim ent 1, and
yielded very sim ilar results. Using the 3sd cutoff for outliers resulted in
the exclusion of 19 scores, 11 from the pseudohom ophones and 8 from
the non-pseudohom ophones. A nalyses using the trim m ed data set are
rep o rted . R esults are su m m arized in Figure 3. The factors in the
analysis of variance w ere Delay (im m ediate, short, long), List (1-4),
Block (1-4) and Type (pseudohom ophone or non-pseudohom ophone).
The list an d block factors d id not yield significant m ain effects or
interactions w ith o th er factors and will not be considered further.
There w as a m ain effect of Delay Fi(2,64) = 189.8, p < .001, F2(2,126) =
27
1825, p < .001; m eans in the im m ediate, short, an d long conditions
w ere 631, 437, an d 417 m s respectively. T here w as a m ain effect of
stim u lu s type, F](l,32) = 7.61, p < .01, F2<2,126) = 1.45, p < .001, because
p s e u d o h o m o p h o n e s w e re n a m e d f a s t e r th a n n o n
p se u d o h o m o p h o n es; the effects in the im m ed iate, sh o rt, an d long
d elay conditions w ere 16, 9, and 9 m s, respectively. The interaction
betw een delay and stim ulus type did not approach significance (Fs < 1).
700
■ PMudohomophon*
□ Non-pMUdohomophona
5 5 0 -
5 0 0 -
Shorl Long
Figure 3: Mean naming latencies in Experiment 2.
Subjects again m ade few errors. Perform ance w as slightly better
o n p s e u d o h o m o p h o n e s (98.7% c o rre c t) th a n o n n o n
28
p se u d o h o m o p h o n es (98.2%), b u t the difference w as not significant.
A ccuracy w as also very sim ilar across delays: 97.9, 98.5, and 98.9 percent
correct for the im m ed iate, sh o rt, an d long con d itio n s, respectively.
C o rrela tio n s b etw een n am in g laten cies an d the tw o m ea su re s of
frequency w ere again sm all an d non-significant: the correlations w ith
im m ediate nam ing w ere -.01 and +.11 for the K ucera and Francis and
W all Street Journal corpora, respectively.
3.3 Discussion
The ex perim ent yielded a significant p seu d o h o m o p h o n e effect
that w as essentially flat across delay conditions, as it w as across subject
g ro u p s in E xperim ent 1. W hile the effects w ere -15, -16, an d -14 m s in
the p rev io u s nam ing experim ent, they w ere -16, -9 an d -9 m s in the
p re se n t ex p erim en t. H ence these stim u li yield very sim ilar effects
regardless of subject speed or nam ing condition. The m ean effect across
the 6 conditions, 13 m s, is close to the 11 m s effect that M cCann and
Besner (1987) observed in their E xperim ent 1 delay condition. M cCann
an d Besner obtained a m uch larger 35 m s effect in im m ediate nam ing,
how ever. The sources of these effects can be understood in term s of the
fram ew ork illustrated in Figure 2. The stim uli in both the M cCann and
B esner (1987) a n d the p re s e n t s tu d ie s co n sist of item s w h o se
p ro n u n ciatio n s are fam iliar, ov erlearn ed articulatory m otor program s
(p s e u d o h o m o p h o n e s ) a n d stim u li w h o se p ro n u n c ia tio n s a re
29
u n fam iliar (n o n -p seudohom ophones). T here are sm all differences
betw een fam iliar and unfam iliar m onosyllables w ith regard to ease of
pronunciation, on the o rd er of 10-15 ms across studies. These effects
will be present w henever an overt pronunciation m ust be generated;
th u s they a re fo u n d in b o th im m ed iate an d d elay ed n am in g
conditions. These persisten t differences in ease of articulation are
apparently not related to differences in decoding ability (i.e. the ability
to reco g n ize letters and perform o rth o g ra p h ic to phonological
conversion); the effects w ere the sam e size for subjects w hose nam ing
latencies differed by literally hundreds of m illiseconds in Experim ent 1.
W ithin the Figure 2 fram ew ork, these effects arise in the com putation
from p h onological code to a rticu la to ry o u tp u t. In a sense, this
difference betw een fam iliar pronunciations (associated with w ords and
pseudohom ophones), and unfam iliar pronunciations (associated w ith
n o n -p se u d o h o m o p h o n e s) re p re se n ts a c o arse -g rain ed effect of
frequency on the generation of articulatory output. A fam iliar pattern
is one that has occurred m ore frequently than an unfam iliar p attern
and will have a bigger im pact on the settings of the w eights m ediating
the production of articulatory output. Like M cCann and Besner (1987),
h o w ev er, w e d id not observe a sig n ifican t co rrelatio n betw een
pseudohom ophone nam ing latencies and base w ord frequencies. The
difference betw een fam iliar pronunciations and novel pronunciations
30
w as only about 13 m s, suggesting w hy it m ight be difficult to detect
even sm aller effects d u e to differences in the relative frequencies of
fam iliar p ro n u n ciatio n s.
Stim uli can also differ in term s of factors that influence the ease
of generating the phonological code that is the in p u t to the articulatory
co m p o n en t. Such factors include the frequency an d consistency of
spelling-sound correspondences, length, and o rth o g rap h ic redundancy.
H ere M cC ann an d B esner's stim u li d iffered from o u r ow n. T heir
p s e u d o h o m o p h o n e s a re m o re w o rd lik e th a n th e ir n o n
p se u d o h o m o p h o n es, m ak in g it easier to g e n erate the p honolo gical
codes for the p seu d o h o m o p h o n es. The 35 m s effect in M cCann and
B esn er's E x p erim en t 1 im m ed iate n a m in g c o n d itio n th u s reflects
d ifferen ces in b o th ease of c o m p u tin g p h o n o lo g ical co d es from
o rth o g ra p h y an d ease of g en eratin g articu la to ry o u tp u t. W hen the
fo rm er differences are elim in ated by usin g m ore closely-controlled
stim u li o r by req u irin g a d elay ed resp o n se, only th e latter effect
rem a in s.
The McRae et al. (1990) stu d ies shed further light on the bases of
th e se effects. M cRae et al. e x a m in e d m a tc h e d h o m o p h o n e -
p seu d o h o m o p h o n e p airs such as PROVE-PRUVE an d PEARL-PIRL.
T he stim uli in these conditions differ in term s of ease of com puting
th e p h o n o lo g ic a l co d e, b e c a u se th e h o m o p h o n e s a re fa m ilia r
31
o rth o g rap h ic p a ttern s an d the p seu d o h o m o p h o n es are not; how ever,
they involve the sam e articulatory m otor program s. The hom ophones
y ield ed sig n ifican tly faster im m e d ia te n am in g laten cies th a n the
pseudoh o m o p h o n es (35 m s in one experim ent, 52 m s in another), but
this effect w as elim inated in delayed nam ing. In contrast, the HOAP-
HOAK stim uli used in the present experim ents are sim ilar in term s of
ease of com puting the phonological code, but differ w ith respect to ease
of g e n e ra tin g a rtic u la to ry o u tp u t (because half the stim u li have
unfam iliar pronunciations). Therefore, they yield the sam e size effect
in both im m ediate a n d delay ed nam ing. Finally, the M cC ann an d
B esner p se u d o h o m o p h o n e s an d n o n -p se u d o h o m o p h o n e s differ in
term s of both ease of com puting phonology and generating articulatory
o u tp u t. They therefore yield a larger difference in im m ediate nam ing
than in delayed nam ing.
32
CHAPTER 4: GENERAL D ISCU SSIO N
The experim ents w e have described replicate m ajor aspects of
the M cCann and Besner (1987) and M cCann et al. (1988) stu d ies, but
p ro v id e a d d itio n a l d a ta a n d an aly ses b e a rin g on th e locus of
p seu d o h o m o p h o n e effects in nam in g an d lexical decision. M cCann
and colleagues interpreted their data as indicating that (a) processing a
p seu d o h o m o p h o n e involves accessing the base w o rd th at it so u n d s
like, and (b) lexical access is not frequency sensitive. T heir conclusions
a b o u t m odels of w ord recognition follow ed from these m ain results.
B esner, T w illey, M cC ann, a n d S eergobin (1990) later a rg u e d th at
pseudo h o m o p h o n e effects are incom patible w ith the absence of lexical
n o d es in th e Seidenberg and M cClelland (1989) m odel. H ow ever, the
p re s e n t s tu d ie s su g g e st a so m e w h a t d iffe re n t in te rp re ta tio n of
p se u d o h o m o p h o n e effects. N o n w o rd p ro n u n cia tio n m akes use of
k n ow ledge stru c tu re s and processes th at are acq u ired in learning to
rea d a n d are n o rm ally used in read in g w ords. T he d ifficu lty of
p ro n o u n cin g a non w ord therefore d ep en d s on the extent to w hich it
resem bles (i.e., shares stru ctu re w ith) know n w ords. W e have isolated
33
tw o co m p o n en ts of the nam ing process th at are influenced by the
d eg ree to w hich non w o rd stim uli are "w o rd ish ": the co m p u tatio n s
from o rth o g ra p h y to phonology an d from p h o nology to articulatory
o u tp u t. P s e u d o h o m o p h o n e s n e c e s s a rily d iffe r fro m n o n
p s e u d o h o m o p h o n e s in term s of facto rs th a t affect th e seco n d
c o m p o n e n t b e c a u se p s e u d o h o m o p h o n e s h a v e m o re fa m ilia r
phonological and articulatory representations. In contrast, these tw o
types of n o n w o rd s d o not necessarily differ in term s of factors that
influence the m ap p in g from o rth o g rap h y to phonology. M cCann and
B esner's stim uli h ap p en ed to differ w ith reg ard to such factors; o u r
stim uli d o not.
The lexical decision results are different, insofar as the m ere fact
that a no n w o rd w hich sounds like a w ord has an im pact on processing,
o th er aspects of lexical stru ctu re aside. T hus, the fact th at BRANE
so u n d s like a know n w ord w ith a kn o w n m eaning in terferes w ith
decid in g that it is a nonw ord. Sim ilar effects have been observed in
stu d ies by Van O rden et al. (1988) an d others using sem antic decision
tasks. T hese resu lts reflect the fact th at o rth o g rap h ic codes rap id ly
activate phonological inform ation w hich can, w hen the stim u lu s is a
p seu d o h o m o p h o n e, resu lt in the activation of sem antic inform ation.
H ow ever, the o th er m ajor finding from the lexical decision experim ent
is th at there is no p seu d ohom ophone effect for faster subjects. These
34
m ore skilled read ers can identify a p seu d o h o m o p h o n e as a non w o rd
w ith o u t interference from know ledge of the hom ophone, possibly by
m apping directly from o rth o g rap h y to sem antics.
W ith this account of the basic phenom ena in h an d , w e can now
reconsider som e of the theoretical im plications th at have been attached
to p seu d ohom ophone effects.
4.1 A re there lexical n o d es in lexical m em ory?
The existence of p seu d o h o m o p h o n e effects is som etim es taken
as evidence that the lexicon m ust include w ord-specific representations
such as lexical entries or logogens and as a problem for m odels that do
n o t include such rep resen tatio n s (e.g., Seidenberg & M cC lelland, 1989;
see also Plaut et al., 1994). This conclusion needs to be reconsidered in
light of the p resen t results. The d ata for the nam ing task indicate no
effect of p seudohom oph ony on the com putation of phonological codes
from o rth o g ra p h ic in p u t, a re s u lt th a t d o e s n o t d e m a n d the
in tro d u ctio n of w ord specific representations. There are differences in
the ease of p ro d u cin g fam iliar an d u n fam iliar p ro n u n cia tio n s, and
in tro d u c in g w o rd -sp e c ific re p re s e n ta tio n s th a t a re accessed in
g enerating articulatory o u tp u t w ould be one w ay to accom m odate these
effects. H o w ev e r, e x istin g c o n n e c tio n ist a rc h ite c tu re s like th e
Seidenberg and M cClelland m odel already account for these results. It
w as once believed th at effects of w o rd frequency on o rth o g rap h ic-
35
phonological conversion also required w ord-specific lexical entries (e.g.
M orton, 1969; F o rster, 1976), b u t n e u ra l n e tw o rk s e m p lo y in g
distrib u ted representations show that this intuition is incorrect. The
Seidenberg and M cClelland (1989) m odel, for exam ple, produces robust
w ord frequency effects w ithout using w ord-specific representations.
W hat is im portant to producing the effect is that w ord frequency have
an im pact on the know ledge represented in the netw ork (in virtue of
its effects on the w eights d u rin g training), not that the know ledge
representations them selves be word-specific.
If the principles that govern such m odels are extended to the
production of articulatory o u tp u t, it should be possible to provide a
sim ilar account of frequency effects at this stage in processing. O ne
such extension is provided by recurrent neural netw orks that produce
sequential o u tp u t (e.g., Jordan, 1986; C leerem ans & M cClelland, 1991;
Elm an, 1990, 1991; P earlm utter, 1989). Instead of learning a pattern
associated w ith a pronunciation, such netw orks learn to produce a
trajectory of o u tp u ts, such as an integrated sequence of articulatory
m ovem ents. Such trajectories are influenced by the sam e factors that
give rise to frequency effects in feedforw ard netw orks (C leerem ans &
M cClelland, 1991). Thus, frequency effects involving different types of
inform ation can all be seen as deriving from facts about how frequency
affects the settings on w eights in certain types of connectionist m odels.
36
In su m m ary , it is im p o rta n t to u n d e rsta n d the d ifferen t k in d s of
frequency effects, but it does n o t follow from su ch ev id en ce th at
m o d e ls of th is k n o w le d g e n e c e ssa rily re q u ire w o rd -s p e c ific
rep resen tatio n s.
4.2 Could there be semantic involvement in nonword naming under
other conditions?
O u r resu lts su g g est th at the effects of p se u d o h o m o p h o n y on
overt nam in g are quite lim ited u n d er the conditions w e have studied:
they are specific to generating articulatory o u tp u t, not the com putation
of phonological codes. T aking into account differences betw een the
stim uli, the resu lts are consistent w ith those of M cC ann an d Besner
(1987). These results d o not im plicate the use of sem antic inform ation
associated w ith the base w o rd s in n am ing p seu d o h o m o p h o n es. In
contrast, the lexical decision resu lts sug g est th at p seu d o h o m o p h o n es
som etim es activate the m eanings associated w ith th eir base w ords,
interfering w ith the lexical decision.
The basic m ethodology in both o u r nam ing studies and M cCann
a n d B e s n e r's in v o lv e d p r e s e n tin g ra n d o m ly in te rm ix e d
p seu d o h o m o p h o n es an d non-p seu d o h o m o p h o n es an d asking subjects
to nam e them aloud using their know ledge of English spelling-sound
correspondences. T here is n o th in g ab o u t o u r account of n am ing that
precludes changing the conditions of an experim ent in w ays that m ight
37
p ro d u ce sem antic effects on no n w o rd nam ing, how ever. Subjects can
be induced to change their strategies for perform ing the n am ing task
eith er th ro u g h explicit instruction (e.g. M onsell, P atterson, T allon, &
H ill, 1989, in stru cted subjects to p ro n o u n ce exception w o rd s such as
PIN T as if th ey w ere ru le g o v e rn e d ) o r im p licitly by m ean s of
m an ip u latio n s of the types of stim uli included in an experim ent (e.g.
M onsell, P a tte rso n , G rah am , H u g h es, & M ilroy, 1992, o b tain ed
d ifferent effects d ep en d in g on w h eth er non w o rd and exception w ord
stim uli w ere blocked or interm ixed; see also Baluch & Besner, 1991).
O ther m anipulations of subject strategies could produce larger effects of
s e m a n tic fa c to rs o n p s e u d o h o m o p h o n e n a m in g . For e x am p le ,
p resen tin g both p seu d ohom ophones and non-p seu d o h o m o p h o n es but
in stru ctin g subjects to only nam e alo u d the form er w o u ld probably
resu lt in g rea ter effects of p ro p erties of the base w o rd s on nam ing
latencies. A pseudo h o m o p h o n e derived from a higher frequency w ord
or a w ord w ith a m ore concrete m eaning m ig h t then be n am ed m ore
rapidly than one derived from a low er frequency, abstract w ord.
Such strategic effects on pronunciation can be explained in term s
of their effects on the allocation of attention to different p arts of the
stim u lu s, how this affects co m p u tatio n s w ith in the lexical netw ork,
an d on how the com puted inform ation is used in m aking a response.
Exactly the sam e analysis applies to the lexical decision task, for w hich
38
it is clear th at subjects m odify their response strategies in response to
ch aracteristics of the stim uli an d o th e r ex p erim en t-sp ecific factors
(Jam es, 1975; W aters & Seidenberg, 1985). In sum m ary, the conditions
w e h av e stu d ie d p ro v id e in fo rm atio n rele v an t to id en tify in g how
certain types of know ledge are represented in lexical m em ory and used
in p erfo rm in g sim ple tasks. The studies provide evidence concerning
th e locus of the p se u d o h o m o p h o n e effects o b ta in e d in d ifferen t
experim ents and conditions. The conditions w e have stu d ied d o not
exhaust the range of possibilities afforded by such tasks, how ever.
39
REFERENCES
Balota, D. A. & Chum bley, J. I. (1985). The locus of w ord frequency effects in
the pronunciation task: Lexical access a n d /o r production? fournal of
M emory and Language. 24. 89-106.
Baluch, B. & Besner, D. (1991). Visual w ord recognition: Evidence for strategic
control of lexical and nonlexical routines in oral reading, journal of
Experim ental Psychology: Learning. M em ory, and C ognition. 17. 644-
652.
Besner, D., Twilley, L., McCann, R., & Seergobin, K. (1990). O n the connection
betw een connectionism and data: Are a few w ords necessary?
Psychological Review. 97. 432-446.
Cleerem ans, A. & M cClelland, J. L. (1991). Learning the structure of event
sequences. Toumal of Experim ental Psychology: General. 120. 235-253.
Coltheart, M., Davelaar, E., Jonasson, J., & Besner, D. (1977). Access to the
internal lexicon. In S. Dornic (Ed.), A ttention & Perform ance IV.
H illsdale, Nj: Erlbaum.
Elman, J. L. (1990). Finding structure in time. Cognitive Science. 14. 179-211.
Elman, J. L. (1991). D istributed representations, sim ple recurrent netw orks,
and gram m atical structure. M achine Learning. 7, 195-225.
Forster, K. I. (1976). Accessing the m ental lexicon. In E. W alker & R. J . Wales
(Eds.), N ew approaches to language m echanism s. A m sterdam : N orth-
H olland.
G em sbacher, M. A. (1984). Resolving 20 years of inconsistent interactions
betw een lexical fam iliarity and orthography, concreteness, and
polysem y. Toumal of Experimental Psychology: General. 113. 256-281.
40
Jam es, C. T. (1975). The role of sem antic inform ation in lexical decisions.
Journal of Experim ental Psychology: H um an Perception and
Perform ance. 104. 130-136.
Jared, D. & Seidenberg, M. S. (1991). Does w ord recognition proceed from
spelling to so u n d to m eaning? Journal of E xperim ental Psychology:
G eneral. 120, 358-394.
Jordan, M. 1. (1986). A ttractor dynam ics and parallelism in a connectionist
sequential m achine. Proceedings of the Eighth A nnual M eeting of the
C ognitive Science Society. H illsdale, NJ: Erlbaum .
Kucera, H. & Francis, W. N. (1967). C om putational analysis of present-day
A m erican E nglish. Providence, RI: Brown U niversity Press.
Lesch, M. F. & Pollatsek, A. (1993). A utom atic access of sem antic inform ation
by phonological codes in visual w ord recognition. Journal of
E xperim ental Psychology: Learning. M em ory, and C ognition. 19. 285-
294.
M arcus, M. P., Santorini, B. & M arcinkiew icz, M. A. (1993). B uilding a large
an n o tated corpus of English: The Penn Treebank. C o m p u ta tio n a l
Linguistics. 19. 313-330.
M ayzner, M. S. & Tresselt, M. E. (1965). Tables of single-letter and digram
frequency counts for various w ord length and letter position
com binations. Psychonom ic Science M onograph S upplem ent. 1. 13-32.
M cCann, R., & Besner, D. (1987). R eading pseudohom ophones: Im plications
for m odels of pronunciation assem bly an d the locus of w ord-frequency
effects in nam ing. lo u m al of Experim ental Psychology: H um an
Perception an d Perform ance. 13. 14-24.
M cCann, R., Besner, D., & D avelaar, E. (1988). W ord recognition and
identification: D o w ord frequency effects reflect lexical access? Journal
of Experim ental Psychology: H um an Perception an d Perform ance. 13.
693-706.
M cC lelland, J. L. & R um elhart, D. E. (1981). A n interactive activation m odel
of context effects in letter perception: Part 1. An account of basic
findings. Psychological Review. 88. 375-407.
41
M cCusker, L. X., Hillinger, M. L., & Bias, R. G. (1981). Phonological recoding
and reading. Psychological Bulletin. 89. 217-245.
McRae, K., Jared, D., & Seidenberg, M. S. (1990). O n the roles of frequency and
lexical access in w ord nam ing. Journal of M em ory and Language, 29.
43-65.
M onsell, S., Patterson, K. E., Tallon, J., & Hill, J. (1989, January). Voluntary
surface dyslexia: A new argum ent for two processes in oral reading?
Paper presented at the m eeting of the Experim ental Psychology Society,
L ondon.
Monsell, S., Doyle, M. C., & H aggard, P. N. (1989). Effects of frequency on
visual w ord recognition tasks: W here are they? Journal of
Experim ental Psychology: General. 118. 43-71.
M onsell, S., Patterson, K. E., G raham , A., Hughes, C. H., & Milroy, R. (1992).
Lexical and sublexical translation of spelling to sound: Strategic
anticipation of lexical status. Journal of Experim ental Psychology:
Learning. M emory, and Cognition. 18. 452-467.
M orton, J. (1969). Interaction of inform ation in w ord recognition.
Psychological Review. 76. 165-178.
Pearlm utter, B. A. (1989). Learning state space trajectories in recurrent neural
netw orks. N eural C om putation. 1. 263-269.
Plaut, D. C. & Shallice, T. (1993). Deep dyslexia: A case study of connectionist
neuropsychology. Cognitive N europsychology. 10. 377-500.
Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. E. (1994).
U nderstanding norm al and im paired w ord reading: C om putational
principles in quasi-regular dom ains. Technical Report PDP.CNS.94-5,
D epartm ent of Psychology, Carnegie M ellon University.
Seidenberg, M. S. (1985a). The time course of phonological code activation in
tw o w riting system s. Cognition. 19. 1-30.
Seidenberg, M. S. (1985b). The time course of inform ation activation and
utilization in visual w ord recognition. In D. Besner, T. G. W aller, & E.
M. M acKinnon (Eds.), Reading research: A dvances in theory and
practice (Vol. 5, pp. 199-252). New York: Academ ic Press.
42
Seidenberg, M. S. (1992). Dyslexia in a com putational m odel of w ord
recognition in reading. In P. G ough, L. Ehri, & R. Treim an (Eds.),
R eading A cquisition. Hillsdale, NJ: Erlbaum.
Seidenberg, M. S. & McClelland, J. L. (1989) A distributed, developm ental
m odel of w ord recognition and nam ing. Psychological Review. 96. 523-
568.
Solso, R. L., & Juel, C. L. (1980). Positional frequency and versatility of bigram s
for two- through nine-letter English w ords. Behavior Research
M ethods & Instrum entation. 12. 297-343.
Van O rden, G. C., Johnston, J. C., & Hale, B. L. (1988). W ord identification in
reading proceeds from the spelling to sound to m eaning, journal of
Experim ental Psychology: Learning. M emory, and Cognition. 14. 371-
386.
W aters, G. S. & Seidenberg, M. S. (1985). Spelling-sound effects in reading:
tim e course and decision criteria. M em ory and C ognition. 13. 557-572.
W ilson, M. (1988). MRC Psycholinguistic Database: M achine-usable
dictionary, Version 2.00. Behavior Research M ethods, Instrum ents, &
C om puters, 20. 6-10.
43
INFORMATION TO USERS
This manuscript has been reproduced from the m icrofilm master. UM I
films the text directly from the original or copy submitted. Thus, some
thesis and dissertation copies are in typewriter face, while others may
be from any type of computer printer.
H ie quality of this reproduction is dependent upon the quality of the
copy submitted. Broken or indistinct print, colored or poor quality
illustrations and photographs, print bleedthrough, substandard margins,
and improper alignment can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete
manuscript and there are missing pages, these will be noted. Also, if
unauthorized copyright material had to be removed, a note will indicate
the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and
continuing from left to right in equal sections with small overlaps. Each
original is also photographed in one exposure and is included in
reduced form at the back of the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6" x 9” black and white
photographic prints are available for any photographs or illustrations
appearing in this copy for an additional charge. Contact UMI directly
to order.
A Bell & Howell information Company
300 Nortfi Z eeb Road. Ann Arbor M l 48106-1346 USA
313/761-4700 800/521-0600
UMI Number: 1376499
UMI Microfora 1376499
Copyright 1995, by UMI Coapany. All rights reserved.
This aicrofora edition is protected against unauthorized
copying under Title 17, United States Code.
UMI
300 North Zeeb Road
Ann Arbor, MI 48103
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Syntactic Priming As Applied To Syntactic Ambiguity Resolution
PDF
The relationship between alcoholism and crime: autonomic and neurpsychological factors
PDF
Violent environments and their effects on children
PDF
Japanese pronunciation by English speaking students and English-Korean speaking students: contrastive and error analysis
PDF
Meta-analysis on the misattribution of arousal
PDF
The Müller-Lyer illusion: a new variant, some old and new results
PDF
Predictors of pretend play in Korean-American and Anglo-American preschool children
PDF
Rationalizing risk: sexual behavior of gay male couples
PDF
An analysis of private and social gains from plastics recycling
PDF
The Cloisters Cross: a re-examination of date and style
PDF
Distributional cues and subject identification in the production of subject -verb agreement
PDF
The Hall Canyon pluton: implications for pluton emplacement and for the Mesozoic history of the west-central Panamint Mountains
PDF
The application of Sarbin's theory of emotions as narrative emplotments to stories of two men diagnosed with cancer
PDF
Children's immediate reactions to interparental conflict
PDF
The Japanese demonstratives ko, so and a
PDF
Fine motor skills of two- to three-year-old drug exposed children
PDF
Work for the masters degree
PDF
On the limits of not being scripted: video-making and discursive prositioning in coastal southeast Sulawesi
PDF
The dream becomes a reality (?): nation building and the continued struggle of the women of the Eritrian People's Liberation Front
PDF
Locatives in Chinese and Japanese: distribution, case assignment and c-command
Asset Metadata
Creator
Petersen, Alan Scott
(author)
Core Title
An investigation of the pseudohomophone effect: where and when it occurs
School
Graduate School
Degree
Master of Arts
Degree Program
Psychology
Degree Conferral Date
1995-08
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
language, linguistics,OAI-PMH Harvest,psychology, experimental
Language
English
Contributor
Digitized by ProQuest
(provenance)
Advisor
Seidenberg, Mark S. (
committee chair
), Biederman, Irving (
committee member
), MacDonald, Maryellen C. (
committee member
)
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c18-3593
Unique identifier
UC11357886
Identifier
1376499.pdf (filename),usctheses-c18-3593 (legacy record id)
Legacy Identifier
1376499-0.pdf
Dmrecord
3593
Document Type
Thesis
Rights
Petersen, Alan Scott
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA
Tags
language, linguistics
psychology, experimental