Full text of DOI:10.3758/BF03194964

Memory & Cognition
2002, 30 (4), 629-636
Semantic aspects of morphological processing:
Transparency effects in Serbian
LAURIE BETH FELDMAN
State University of New York, Albany, New York
and Haskins Laboratories, New Haven, Connecticut
DRAGANA BARAC-CIKOJA
Gallaudet College, Washington, D.C.
and
´
ALEKSANDAR KOSTIC
University of Belgrade, Belgrade, Serbia
and Haskins Laboratories, New Haven, Connecticut
We examined the contribution of semanticsto morphological facilitationin the visuallexicaldecision
task at two stimulus onset asynchronies (SOAs) with Serbian materials. Primes appeared in Roman or
Cyrillic characters. Targets always were printed in Roman. When primes were presented at an SOA of
250 msec, decision latencies to verbal targets (e.g.,
ally (e.g.,
) showed greatest facilitation after inflection-
volim
) relatedprimes, significantlyless aftersemanticallytransparentderived primes (e.g.,
vole
za-
), and less again after semanticallyopaque derived primes (e.g.,
vole
). Latencies after seman-
prevole
tically transparent and opaque derived target words did not differ at an SOA of 48 msec. Both were
slower than afterinflectionallyrelatedprimes. Statedgenerally,effectsof semantictransparencyamong
derivationally related verb forms were evident at long SOAs, but not at short ones. Under alphabet-
alternatingconditions, magnitudes of facilitationwere greateroverall,but the patternwassimilar.The out-
come suggeststhat restrictedprocessingtime for the prime limits the contribution of semanticsto mor-
phological processing and callsinto question accounts that posit a task-invariantsemantic criterionfor
morphological decomposition within the lexicon.
lowance
, whose meaningscannotbe derivedfrom knowl-
For a variety of languages with an assortment of word
recognitiontasks, researchers have demonstrated that prior edge of the constituent morphemes, tend to be semanti-
presentationof a word formed from thesame base morpheme cally opaque.
as the target (morphologicallyrelated) speeds decision la-
A source of debateamong those who study morphology
tencies to the target (see Feldman, 2001, for a review). Re- is whethermorphologicalknowledgeaboutcomplexforms
liableevidenceof morphologicalfacilitationindicatesthat is representedexplicitlyin the mental lexicon and whether
a word’s morphologicalstructure influences the processes the same characterization applies for both opaque and
of word recognition.Because morphologicalrelativestend transparent derivations. For a decompositional account,
to be formed by adding an affix to a base morpheme,words morphologically complex words are accessed or repre-
that share a base morpheme tend to be similar in form and sented in terms of their constituentmorphemes (e.g., Taft
similar in meaning.However, there can be variabilityin the & Forster, 1975), and morphological facilitation arises
degreeofsimilarity. For example,themeaningofsome com- from multiple activations of a shared base morpheme
allowable
plex forms (
meanings of its components (
fore, morphological relatives such as
) is fully predictable from the (Marslen-Wilson,Tyler, Waksler, & Older, 1994).Fora full-
allow 1 able
), and there- form account, morphologicallycomplex words are repre-
allowable
and sented as full forms (e.g., Butterworth, 1983) that are in-
allow
other relatives (e.g.,
from its components (
are relatively close in meaning. The meaning of terconnected in accord with a morphological principle
allowance
) is not fully predictable (Lukatela, Gligorijevic´, Kostic´, & Turvey, 1980) so as to
allow 1 ance
al-
). Words like
produce facilitation by activation along a shared pathway.
When patterns of morphologicalfacilitation differ among
morphologically complex forms, depending on their se-
mantic transparency, a possible locus is the lexical archi-
tecture. Accordingly, semantically opaque derived forms
fail to produce facilitation because they are represented in
the lexicon as (separate) whole units, whereas semanti-
cally transparent derived forms produce facilitation be-
This study was supported by funds from National Institutes of Child
Health and Development Grant HD-01994 to Haskins Laboratories and
by funds from the Faculty of Philosophy at The University of Belgrade
in Serbia. We thank Matthew Pastizzo for analyzing the data. Please ad-
dress correspondence to L. B. Feldman, Haskins Laboratories, 270
Crown Street, New Haven, CT 06510 (e-mail: lf503@albany.edu).
629
Copyright 2002 Psychonomic Society, Inc.

´
FELDMAN, BARAC-CIKOJA, AND KOSTIC
630
cause they can be decomposed into morphological com- presentations. Conversely, however, when primes were
ponents and the base morpheme is activated by the prime forward masked and presentedin immediatesuccession at
and then by the target (e.g., Marslen-Wilson et al., 1994). very short prime durations in Hebrew, the magnitudes of
In essence, decomposition and, therefore, morphological facilitation after opaque and transparent primes formed
facilitation apply only to transparent forms in English. In from the same base morpheme as their target were com-
contrast to the accountdevelopedfor English,Frost and his parable (Frost et al., 1997).
associateshave asserted that because of its productivemor-
Even in unmaskedpriming contextsthat typicallyreveal
phology, transparent and opaque morphological relatives an effect of semantic transparency, stimulus onset asyn-
in Hebrew are represented in the same decomposed man- chrony(SOA) is critical.In English,morphologicalfacilita-
ner and produce comparable magnitudes of facilitation tion for visual targets preceded by a semanticallytranspar-
according accordance
) and by
(Bentin & Frost, 1995; Deutsch, Frost, & Forster, 1998; ent suffixed form (e.g.,
Frost, Forster, & Deutsch, 1997).
an opaque suffixed form (e.g.,
–
accordian accordance
–
)
An alternativeperspective posits no explicitrepresenta- differed significantly at an SOA of 250 msec but did not
tion of morphologyand no decomposedlexicalentries.In- differ at an SOA of 48 msec (Feldman & Pastizzo, 2002;
stead, it emphasizesgraded similarityof form and similar- Feldman & Soltano, 1999; Feldman et al., 2001). Collec-
ity of meaning as they contribute to distributed patterns tively, patterns of divergence between inflections and de-
that underlie morphologicalfacilitation (Plaut & Gonner- rivationsand betweensemanticallytransparentand opaque
man, 2000; Rueckl, Mikolinski, Raveh, Miner, & Mars, morphological relatives that vary with experimental ma-
1997). To the extent that outcomes vary across languages, nipulationssuch as prime durationchallengethe adequacy
differences reflect the overall systematicity in the map- of a decompositional account of morphological facilita-
ping between form and meaning (Plaut & Gonnerman, tion and call into question a semantic criterion for de-
2000). Universally within a language,however, inflections composition of lexical entries into their constituent mor-
allowed
parent derivations(e.g.,
(e.g.,
) tend to be more systematic than trans- phemes.
allowable
), which tend, in turn,
The implication of the foregoing is that although coor-
to be more systematic than (even partially) opaque de- dinated studies across languages are important to ascer-
allowance
rivations (e.g.,
tudes of facilitation are anticipated.
). Therefore, graded magni- tain the generality of a lexical architecture and, in partic-
ular, the extentto which semantic transparency determines
Time- and task-varying patterns of facilitation compli- the manner in which morphologicallycomplex words are
cate assessments of the role of semantic transparency in represented (e.g., Marslen-Wilson et al., 1994), proce-
morphologicalprocessing and dictatethat cross-task vari- dures are not interchangeable,and variations across tasks
ation must be differentiated from cross-language varia- and SOAs within a language cannot be ignored. Accord-
tion. A review of the literature reveals remarkably similar ingly, the present study extended recent work conducted
findings across languageswhen comparisons are within a by Feldman et al. (2001) to the morphologicallyrich lan-
task (Feldman, Soltano, Pastizzo, & Francis, 2001). For guage of Serbian, where extensive families of derivation-
example, results in the long-term repetition priming vari- ally related verbs can be created by affixing a derivational
ant of the lexical decision task, where an average of 10 prefix to a base morpheme (Bybee, 1985;Partridge,1964).
items intervened between the presentation of the prime We replicated with Serbian verbs differences between in-
and that of the target, failed to show an effect of semantic flectionsand derivations,as well as the interactionof SOA
transparencyamong derivationalrelativesin eitherHebrew and semantic transparency among derivations previously
(Bentin & Feldman, 1990) or English (Feldman & Stotko, documented in English. For example, the verb infinitives
zavoliti
privoliti
1990, cited in Feldman, 1992). That is, the magnitudes of
morphologicalfacilitationat long lags after derived primes formed from the same base morpheme (
(to fall in love) and
(to convince) are
vol voleti
), as is
creation
(e.g.,
(e.g.,
) that were semantically close to targets (to love). They necessarily differ word initially, however,
) and those that were semantically more re- because of the presence of a derivational prefix. There-
create
creature
) did not differ. Similarly, inflections fore, inflected forms of the verbs
zavoliti
privoliti
and
voleti
by de-
mote (e.g.,
(e.g.,
folded
folder
) tended to dif- are related to inflected forms of the verb
) and derivations(e.g.,
fer in the degree of similarity they shared with the base rivation. To elaborate, the inflected (third person plural
fold
zavole privole
morpheme (e.g.,
), but differences in long-term tar- present tense) forms
and
are derivations
volim
get facilitation failed to emerge in English (Raveh & of the first person singular present tense form
Rueckl, 2000).
. It is
useful to point out that, in Serbian, there are multiple pre-
za pri u iz
, , ) that can combine with verbs and
With immediate unmasked visual (Raveh, 1999) and fixes (i.e.,
,
cross-modal (Pastizzo & Feldman, 2001b) variants of the that a particular prefix can enter into a relatively semanti-
primed lexical decision task in English, facilitation was cally opaquederivationwhen combined with one verb, but
greater after inflectionsthan after derivations.Similarly in into a relatively transparent derivation when combined
Hebrew, effects of semantic transparency among deriva- with another verb. In sum, the repertoire of derivational
tionallyrelated primes and targetsarose with visual(Bentin prefixes is not confounded with degree of transparency.
& Feldman, 1990), as well as with cross-modal (Frost,
In Experiment 1A, visual primes and visual targets
Deutsch, Gilboa, Tannenbaum,& Marslen-Wilson,2000), were presented in the immediate priming variant of the

SEMANTIC TRANSPARENCY IN SERBIAN
631
lexical decision task at an SOA of 250 msec. On the basis Feldman & Barac-Cikoja, 1996). A secondary but related
of previous work at SOAs of 250 msec or greater (Bentin goal was to replicate differences in morphological facili-
& Feldman, 1990;Feldman & Raveh, in press; Feldman & tation after inflectional and derivational primes. Concur-
Soltano, 1999; Raveh & Feldman, 1998), effects of se- rently, we varied processingtimefor the prime so as to alter
mantic transparency among verb forms created from a the potentialsemantic contributionto morphologicalpro-
common base morpheme were anticipated. The goal of cessing.
Experiment 1B was to contrast the effect of transparent
and opaque morphologicalrelatives on decision latencies
to visual targets when primes appeared at a 48-msec SOA.
On the basis of the absence of semantic transparency ef-
fects at very short SOAs with English derivations (Feld-
man & Soltano, 1999) and the absence of a difference be-
tweenEnglishinflectionsandderivationsthatwere matched
on orthographicoverlapto the target (Raveh, 1999; Raveh
& Rueckl, 2000) in Serbian, we did not expect to observe
differences between opaque and transparent forms under
those experimentalconditionsthat tend to be relativelyin-
sensitive to semantic dimensions of relatedness. Because
they retained greater orthographicsimilarity,as well as se-
mantic similarity, to the target, we anticipated that even at
short SOAs, inflectionalprimes would produce greater fa-
cilitation than would derivational primes. This finding
would be consistent with the claim that at short SOAs,
form also influences the magnitude of morphological fa-
cilitation (Pastizzo & Feldman, 2001b).
METHOD
Participants
Forty-eight students participated in each experiment defined by
SOA (250 and 48 msec) and by alphabet of the prime (Roman and
Cyrillic). All the participants were recruited from psychology courses
at the University of Belgrade. All were native speakers of Serbian,
with normal or corrected-to-normal vision and no known reading
disorders. No one participated individually in more than one exper-
iment.
Materials
Morphologically complex words were selected as primes and as
targets. Forty words were selected as critical targets. Primes were paired
with each target to create four conditions: inflected forms of the
same verb (e.g., vole–volim), derived forms that were transparent
with respect to base morpheme (e.g., zavole–volim), derived
forms that were opaque with respect to base morpheme (e.g.,
privole–volim), and unrelated but morphologically complex forms
(e.g., Ïstampaju–volim). Accordingly, the repeated base morpheme
appeared word medially in prefixed derivational primes (e.g., za-
vole, privole), but word initially in targets and in inflectional
The goal of Experiments1C and 1D was to contrast the
effect of transparent and opaque morphological deriva- primes (e.g., volim, vole).
All related primes shared the base morpheme (e.g., vol) with the
tions on decision latencies when primes and targets were
presented in different alphabets. Alphabet alternation in
Serbian provideda techniqueto essentiallyeliminatefrom
the pattern of facilitation the effect of orthographic (but
not phonological) similarity between prime and target
(e.g.,Lukatela,Lukatela,Carello,& Turvey, 1993;Lukatela,
Turvey, & Todorovic´, 1991) because, with the exception
of about five phonemes, graphemes differ in the Roman
and Cyrillic scripts. In Experiment 1C, visual primes in
Cyrillic appeared for 250 msec. In Experiment 1D, visual
primes in Cyrillic appeared for 48 msec.
To summarize, we exploitedthe productivesystemof de-
rivationbased on verb prefixationinSerbian and compared
the magnitudesof morphologicalfacilitationafter opaque
and transparent derivational primes at long and at short
SOAs in the immediate priming task. Our intent was to
use the pattern of morphologicalfacilitation to determine
whether a decomposed lexical representation of a word’s
target. Targets were first person singular (e.g., 1 im, 1am) and
primes were third person plural (e.g., 1 e, aju) affixed forms. In the
transparent morphological condition, meaning was also shared be-
tween the full forms of the prime and the target. In the opaque mor-
phological condition, prime and target shared the base morpheme, but
the full forms were not similar in meaning. An informal rating study
by native speakers (n 5 10) was conducted to assess which of the de-
rivationally related primes was more closely related to the target, and
all items had at least 70% agreement. Unrelated primes were
matched for person, number, and tense to the related primes and did
not overlap orthographically with the target.
In addition, 40 morphologically complex pseudoword targets
were created and presented. They were formed by changing one or
two letters in a verbal base morpheme other than those of the word
targets. The proportions of primes that were paired with pseudo-
words mimicked that of word primes. To reduce the proportion of re-
lated trials, filler trials in which the prime and the target did not
match on form or meaning were introduced. The inclusion of 40
filler word–word trials and 40 filler word–pseudoword trials reduced
the relatedness proportion for word–word trials to about 38%. Filler
morphological structure was restricted to semantically primes did not share phonology, orthography, or meaning with their
targets.
transparent forms. Contrasting magnitudes of facilitation
for opaque and transparent forms would be compatible
Procedure and Design
with the outcome reported in English with cross-modal
We constructed four experimental lists, each containing 160 trials.
Both prime and target letter strings appeared in the Roman alphabet
and with long-durationvisual presentationsand would at-
test to the contributionof semanticsto morphologicalpro-
cessing. Comparable magnitudes of facilitation would be
compatible with the outcome reported in Hebrew with a
forward masked procedure and would imply an attenuated
role of semantics. To this end, we also made use of the bi-
alphabetic fluency of skilled readers in Serbia to reduce
the orthographically based form similarity between tar-
gets and their morphologically related primes (see, e.g.,
in Experiments 1A and 1B. Primes were presented in Cyrillic, and
targets were presented in Roman, in Experiments 1C and 1D. Primes
were counterbalanced across the four lists so that each list included
all types of primes. If a particular target appeared with a particular
prime on one experimental list, its paired prime was different on the
other three lists. No target was repeated within an experimental list.
In all experiments, a trial consisted of a fixation point (1) pre-
sented in the center of the computer monitor for 250 msec. Then a
prime appeared for 250 msec (Experiments 1A and 1C) or for 48 msec

´
FELDMAN, BARAC-CIKOJA, AND KOSTIC
632
MS 5
3.68,
F
5
(Experiments 1B and 1D) in the same location, after which the tar-
get appeared alone one line below where the prime had been. The tar-
get remained on the screen until the participant made a lexical deci-
sion response or until 1,500 msec had elapsed. A 1,000-msec inter-
trial interval followed each trial. Responses were registered on a
generic keyboard. The participants pressed the L key for word re-
sponses and the A key for nonword responses, and latencies were mea-
sured from the onset of the target. Each session began with 16 prac-
tice trials to familiarize the participant with the task.
2,367] and the accuracy measure [ 1(3,141)
MS 5 MS 5 p
0.003, < .07].
e
F
5
0.005; 2(3,105) 2.46,
e
e
Planned comparisonsrevealed that decisionlatenciesto
targets that followed inflected primes (620 msec) and
transparent primes (649 msec) were significantly faster
than those that followed unrelated primes [674 msec;
F
5
MS 5
F
5
MS 5
1(1,141) 35.96,
1,924; 2(1,105) 30.19,
MS 5
e
e
F
5
F
5
2,367; and 1(1,141) 7.33,
1,924; 2(1,105)
2,367,respectively].Moreover, latenciesafter
opaque primes and unrelated primes (3 msec) did not dif-
e
MS 5
5.62,
e
RESULTS AND DISCUSSION
F
fer ( s < 1). Most important, latencies to targets that fol-
Incorrect keypresses and outliers(defined as a standard
deviation of three or more from each participant’s mean)
were classified as errors. Outliers constitutedabout 1% of
the responses. No data were deleted because of high item
error rates or high participant error rates. A total of four
prime–target pairs were eliminated from the critical item
set because they were incorrectly spelled in one or more
conditions. By eliminating the 4 items from all four lists
across experiments and SOAs, we standardized the set of
lowed transparent derived primes (649 msec) were signif-
icantlyreducedrelativetothoseafter opaquederivedprimes
[671 msec; 1(1,141) 6.02,
4.80,
flected and transparentderived primes differed [ 1(1,141)
F
5
MS 5
F
5
1,924; 2(1,105)
2,367]. Finally, decision latencies after in-
e
MS 5
e
F
5
MS 5
F
5
MS 5
10.82,
1,924; 2(1,105) 9.75, 2,367]. Fa-
cilitation (unrelated minus related) after teransparent and
e
opaque primes is plotted in Figure 1, panel A.
targets. Mean response times, accuracies, and difference Experiment 1B
scores based on the remaining 36 items are reported in
Table 1.
The overall one-way ANOVA on the latency data at the
48-msec SOA when the prime and the target appeared in
In general, target latencies were faster after a morpho- Roman characters also revealed an effect of prime type
F
5
MS 5
F
5
logicalrelativethan after an unrelatedprime, and the mag-
nitude of facilitation varied with type of relatedness. Un-
[
1(3,141) 11.75,
1,976; 2(3,105) 11.96,
1,769]. The effect of prime type was significant
MS 5
e
MS 5
e
F
5
less otherwise indicated,all effects were significant at the with the accuracy measure as well [ 1(3,141) 2.93,
e
p
F
5
MS 5
p
< .05 level. We present the results of analyses on each 0.003; 2(3,105) 2.58,
0.002, < .06].
e
combination of SOA, along with planned comparisons.
Planned comparisonsrevealed that decisionlatenciesto
Although alphabet and SOA were manipulated between targets that followed inflected primes (677 msec) were
participants, items did repeat. Therefore, we also include significantlyreduced relative to those that followed unre-
F
5
MS 5
lated primes [728 msec; 1(1,141) 31.86, 1,976;
e
a combined analysis across SOA and alphabet.
F
5
MS 5
2(1,105) 32.52, 1,769]. The difference between
Experiment 1A
transparent and unrelateed primes just missed significance
F
5
MS 5
MS 5
p
p
The overall one-way analysis of variance (ANOVA) on [713 msec; 1(1,141)
2.87,
1,976, < .09;
1,769, < .10]. In addition, la-
get appeared in Roman characters revealed that the effect tency differencesafter opaqueprimes and unrelatedprimes
e
F
5
the data at the 250-msec SOA when the prime and the tar-
2(1,105) 2.76,
e
F 5
F 5
p
2.14; <
of prime type was significant with both the latency mea- were not significant (13 msec;
1.93;
sure [ 1(3,141) 15.53,
1,924; 2(3,105) 13.11, .20). In contrast to the 250-msec1SOA outco2me, latencies
F
5
MS 5
F
5
e
Table 1
Mean Decision Latencies (RTs, With SDs) and Percentages Correct
250-msec SOA
48-msec SOA
RT
RT
Prime Type
M
SD % Correct Facilitation
M
SD
% Correct
Faciliation
Roman–Roman
Unrelated
Inflection
Derivation transparent
Derivation opaque
674
620
649
671
91
80
91
91
94
94
97
94
728 117
677 112
713 117
715 106
97
98
95
96
54*
25*
3
51*
15
13
Cyrillic–Roman
93
Unrelated
Inflection
Derivation transparent
Derivation opaque
660
603
621
639
91
80
90
83
710
655
676
679
93
91
78
73
91
99
97
97
97
97
96
57*
39*
21*
55*
34*
31*
Note—The prime–target pairs for the prime types were Ïstampaju–volim for the unrelated, vole–volim for
the inflection, privole–volim for the derivation transparent, and zavole–volim for the derivation opaque.
*Significantly different from the unrelated condition by subject and by items.

SEMANTIC TRANSPARENCY IN SERBIAN
633
Figure 1. Facilitation as a function of prime type and stimulus onset asynchrony (SOA). Panel A shows the data for Roman
primes and Roman targets; panel B shows the data for Cyrillic primes and Roman targets. Error bars represent the 95% con-
fidence interval.
F
5
MS 5
e
to targets that followed transparent primes (713 msec) cant with the accuracy measure [ 1(3,141) 12.77,
F
F
5
MS 5
and opaque primes (716 msec) did not differ ( s < 1). Fi- 0.005; 2(3,105) 9.77,
0.002].
e
nally, decision latenciesafter inflected and transparentde-
Planned comparisonsrevealed that decisionlatenciesto
1,976; targetsthat followed unrelatedprimes (710 msec) were sig-
nificantly slower than those that followed either inflected
F
5
1,769].
MS 5
rived primes differed [ 1(1,141) 15.60,
MS 5
e
F
5
2(1,105) 19.33,
e
F
5
MS 5
1,531;
e
primes [655 msec; 1(1,141)
47.50,
2,605] or transparent primes
MS 5
F
5
F
MS 5
Experiment 1C
The overall one-way ANOVA on the Cyrillic prime la- [676 msec; 1(1,141) 18.56,
MS 5
2(1,105) 30.21,
e
5
F
5
1,531; 2(1,105)
2,605]. Latencies after opaque primes and
e
tency data at an SOA of 250 msec revealed a significant 10.56,
MS 5
e
F
5
F
5
MS 5
effect of prime type [ 1(3,141) 20.43,
1,423; unrelated primes also differed [ 1(1,141) 15.70,
e
e
F
5
MS 5
5
F
5
MS 5
2,605]. Finally, laten-
e
2(3,105) 9.96,
2,208]. Prime type was also sig- 1,531; 2(1,105) 10.66,
MS 5
e
F
F
5
nificant[ 1(3,141) 4.27,
MS 5
0.004; 2(3,105) 2.95, cies to targets after transparent primes (676 msec) and
0.004] with the accuracy measure. opaque derived primes (679 msec) did not differ at the
Planned comparisonsrevealedthat decisionlatenciesto short SOA ( < 1). However, target latencies after in-
e
e
F
s
F
5
targets that followed inflected primes and transparent flected and transparent derivations did [ 1(1,141) 6.68,
MS 5
F
5
MS 5
1,531; 2(1,105) 5.04, 2,605].
e
primes (621 msec) were significantly faster than those
to targets that followed unrelated primes [660 msec;
e
F
5
MS 5
F
F
5
1(1,141) 55.57,
1,423; 2(1,105) 26.01, Combined Analyses
MS 5
Because decisionlatenciesafter opaqueand transparent
e
MS 5
5
2,208; and 1(1,141)
5
26.38,
2,208,respectively].Moreover, primes were our focus and because inflected prime facil-
in Experiment 1C, latencies after opaque primes were itationdid not vary over SOA, difference scores (unrelated
1,423;
e
e
F
MS 5
2(1,105) 14.10,
e
F
5
faster than those after unrelated primes [ 1(1,141) 7.63, minus related) for target latencies after opaque and trans-
MS 5
F
5
MS 5
p
2,208, < .07]. parent primes in Experiments1 A–1D were computed and
1,423; 2(1,105) 3.29,
Moest important, latencies to targets ethat followed trans- combined into one analysis. The main effect of alphabet
F
5
MS 5
F
5
parent derived primes (621 msec) were reduced relative to was significant[ 1(1,188) 5.38,
5,835; 2(1,35)
5,305], as was the main effect of trans-
MS 5
e
MS 5
those that followed opaque derived primes [639 msec; 5.41,
MS 5 MS 5
e
F
F
5
F
5
5
F
5
1(1,141) 5.63,
2,208]. Finally, target latencies after inflected and trans- 4.53,
parent derivationsdiffered only in the analysis by partici- conditionproduced greater magnitudesof facilitationthan
1,423; 2(1,105) 3.77,
parency [ 1(1,188) 10.43,
1,205; 2(1,35)
e
e
e
MS 5
2,332]. Accordingly, the alphabet-alternating
e
F
5
MS 5
F
5
pants [ 1(1,141) 5.38,
MS 5
1,423; 2(1,105) 1.81, did the pure Roman condition, and the magnitude of fa-
cilitation was reduced after opaque, relative to transpar-
e
p
2,208, < .20].
e
ent, morphologicalrelatives.Most important,transparency
F
Experiment 1D
did not interactsignificantlywith alphabet( s < 1) but did
F
5
5.99,
MS 5
The overall one-way ANOVA on the latency data at an interact with SOA [ 1(1,188)
1,205;
e
F
5
MS 5
SOA of 48 msec revealed an effect of (Cyrillic)prime type
2(1,35) 8.38, 1,079]. Stated succinctly,consis-
[
1(3,141) 16.19,
1,531; 2(3,105) 10.24, tent with the plannedecomparisonsindividuallyreported in
F
5
MS 5
F
5
e
MS 5
2,605]. The effect of prime type also was signifi- Experiments 1A–1D, whether opaque and transparent
e

´
FELDMAN, BARAC-CIKOJA, AND KOSTIC
634
primes differentiallyaffected target decision latencies de- earlier findings in English, we observed that in the mor-
pended on SOA. Finally, because it did not interact with phologically rich language of Serbian, the effect of se-
F
alphabet( s < 1), the interactionof transparencyand SOA mantic transparency among derivationsformed by prefix-
was present in the Roman–Roman alphabet condition, as ation was evident at the longer, but not at the shorter,
well as in the Cyrillic–Roman alphabet condition.
In summary, whether alphabet alternated so as to re-
prime duration.
We introducedan alphabetmanipulationbetween prime
duce the orthographic similarity between primes and tar- and target to reduce their orthographic similarity. Interest-
gets presented at an SOA of 250 msec or whether it ingly, thepatternoffacilitationacross experimentsrevealed
remained consistent, target decision latencies after se- an effect of alphabet such that the overall magnitude of
mantically transparent and opaque morphological rela- morphological facilitation was greater by about 30 msec
tives differed significantly.Latenciesafter transparentand when the prime and the target were presented in different
opaque primes did not differ at the shorter SOA of 48 alphabets than when the alphabet was consistent. To our
msec, however. Finally, facilitation after inflected primes knowledge, this is the first report that in immediate un-
was significantat both SOAs, and its magnituderemained masked priming, the magnitude of facilitation in Serbian
virtually unchanged.
increased with alphabetalternationand the consequentre-
Although the transparent and the opaque derivational ductionin similarity between prime and target. A possible
primes differed with respect to semantic transparency and interpretation is that orthographic inhibition attenuates
similarity of meaning, repetition of the base morpheme morphological facilitation when prime and target look
vol
e
(e.g.,
variation of only the prefix (e.g.,
the primes were matched for orthographic overlap with a naming task at SOAs of 550 msec or greater, same-
volim
) and the inflectional suffix (e.g., ) along with similar, as in the pure alphabet condition. For phonolog-
pri za
,
) guaranteed that ically ambiguous Serbian targets with masked primes in
the target (e.g.,
). Therefore, differences between alphabet contexts tended to facilitate naming more than
transparent and opaque derivational primes in the present different-alphabet contexts did, presumably because al-
outcome are consistent with claims that effects of ortho- phabet consistency helped to resolve phonological ambi-
graphic similarity are inadequate to account for patterns guity in the target (e.g., Lukatela et al., 1993; see also
of morphological facilitation (Feldman & Moskovljevic´, Lukatela et al., 1991). Perhaps more relevant, for phono-
1987). Nevertheless, it is possible that abstract form sim- logically unambiguous Serbian targets presented at an
ilarity may interact with semantic similarity under partic- SOA of 700 msec, decision latencies in same-alphabet
ular temporal configurations of prime and target. In fact, contextswere about 20 msec faster than those in different-
because form facilitationarises at short SOAs (e.g., Drews alphabetcontexts (Lukatela, Feldman, Turvey, Carello, &
& Zwitserlood, 1995;Feldman, 2001), we suspect that the Katz, 1989). Of course, the presence of phonologically
SOA-invariant facilitation after inflected primes reflects ambiguous words and longer SOAs may limit the applic-
the contributions of (abstract) form in conjunction with ability of those findings. In the present study, by contrast,
high semantic similarity.
alphabet alternation between prime and target benefited
decision latencies. Because we manipulated alphabet
between participants and because the effect of alphabet
did not interact with prime type, SOA, or the interaction
GENERAL DISCUSSION
In the present study, we compared decision latencies of prime type and SOA, we do not interpret the outcome
after inflectionally and derivationally related primes and further.
differences among derivations that varied along a dimen-
The results of our experiments were strikingly similar
sion of semantic transparency. Inflections and derivations despitethe manipulationof alphabetfor the prime. In both
differed in the degree of form, as well as semantic similar- the consistent-alphabet(Experiments 1A and 1B) and the
ity, that theyshared withthetarget. Transparentand opaque alternating-alphabet (Experiments 1C and 1D) contexts,
derived primes differed only along a semantic dimension. the magnitudeof facilitationfollowingsemanticallytrans-
Our study in Serbian complementspreviousprimed visual parent primes was reduced relative to inflectional primes
lexicaldecision studies in English and in Hebrew, because and the relative difference did not change over SOA. In-
derivations were formed by the addition of a derivational flectionalprimes differed from the target only with respect
e
im
prefix to the base and because primes and targets were al- to an inflectional suffix (i.e., vs. ). Therefore, these
ways verbal formations. By tracking the magnitude of words were similarto the target both in form and in mean-
morphological facilitation after transparent and opaque ing. Greater overall (phonological and orthographic, as
primes across SOAs within the unmasked lexicaldecision well as semantic) similarity is typical for inflections rela-
task, we were able to explore the time course over which tive to even transparent derivations. We suggest that the
effects of semantic transparency emerged in morphologi- greater combinedsimilarityof form and meaningcouldac-
cal processing. In our experiments, visual primes pre- count for augmented facilitation after inflectionally re-
ceded visual targets at SOAs of 48 or 250 msec. From the lated, relative to derivationallyrelated, primes in immedi-
work in English, we knew that effects of semantic trans- ate priming.
parency were variable over this range of SOAs (Feldman
The interactionof SOA andsemantictransparencyfor de-
& Soltano, 1999; Feldman et al., 2001). Consistent with rivations likewise dominated the outcomes of the experi-

SEMANTIC TRANSPARENCY IN SERBIAN
635
ments. Planned comparisons indicated that the difference
For a decompositionalaccount,it is oftenargued thatlex-
between opaque and transparent derivational primes was icalrepresentationsfor morphologicallycomplexwords are
significant at the 250-msec SOA but was absent at the decomposed into their constituent morphemes and that
shorter SOA. The pattern in Serbian replicated that re- morphologicalfacilitation reflects the repeated activation
ported by Feldman et al. (2001) in English. Like associa- of a morpheme in prime and target. Consequently, when
tive facilitation that increases with SOA (Lorch, 1982; transparent and opaque primes reduce decision latencies
Raveh, 1999), Feldman and her colleagues claimed that for a morphologicallyrelated target in a similar way, it is
the semantic contribution to processing among morpho- sometimes argued that decomposition is not sensitive to
logical relatives progresses over time, so that differences semantic similarity among relatives. In particular, on the
between opaque and transparent forms become greater as basis of results with a forward masked priming task,
SOA increases. In the present study, significant facilita- Deutsch et al. (1998) have proposed a model of the (He-
tion after inflectionally related primes accompanied the brew) lexicon in which “all words derived from the same
3
SOA transparencyinteraction,therebyrenderingimplau- root are clustered via a shared representation of the root
sible an interpretationbased on lack of power to detectdif- morpheme.” Moreover, “this organization is independent
ferences at the 48-msec SOA.
of semantic factors” (p. 1250). Conversely, when transpar-
There are two methods by which to assess morpholog- ent, but not opaque,derivationsproducemorphologicalfa-
ical effects in the literature, and it is interestingto note that, cilitation,some have argued that all morphologicallycom-
in the present study, reliance on one in isolationleads to an plex words may not be represented in the same manner
incomplete characterization of the role of semantic trans- within the lexicon. For example, on the basis of the out-
parency. The traditional method focuses on the difference come of a cross-modal priming task, Marslen-Wilson et al.
between target latenciesafter related and unrelated primes (1994) have claimed that, in English, only those deriva-
and is subject to potential problems associated with the tions that are semantically transparent are decomposed
construction of the unrelated baseline. For example, Frost with respect to their constituents.
et al. (1997) relied on an orthographically similar prime,
The interactionof SOA and transparency in the present
whereas Feldman et al. (2001) and Marslen-Wilson et al. study challenges the adequacy of a decompositional ac-
(1994) preferred a dissimilar prime. Crucially, however, countof morphologicalfacilitation.On the basisofpatterns
there is evidence that when morphological relatives were of facilitation across alphabets at a long SOA, one would
fell fall
), claim that, as in English,onlyinflectionsand semantically
similar in form and matched in length (e.g.,
–
evaluationsof facilitationrelative to orthographicand un- transparent derivations are reliably decomposed. By con-
related baselinesdiffered (Feldman & Pastizzo, 2002; Pas- trast, on the basis of patterns of facilitation across alpha-
tizzo & Feldman, 2001a). Assessed against an unrelated bets at a short SOA, one would claim that, as in Hebrew,
prime, when both the prime and the target were presented not only inflections, but also all derivations are decom-
in Roman at a 250-msec SOA, in the present study we ob- posed. Stated generally, time-varying effects of semantic
served that opaque primes had no effect (3 msec) on target transparency highlight the shortcomings of simply inter-
decision latencies but that transparent primes produced preting significant morphological facilitation at a single
significant morphological facilitation (25 msec). When SOA as evidence of a static lexical architecture. To elab-
primes were in Cyrillic and targets were in Roman, we ob- orate, that degree of semantic similarity can influence the
served that opaqueprimes producedfacilitation(21 msec) pattern of morphological facilitation only at a long SOA
on target decision, as did transparent primes (39 msec), is difficult to reconcile with a semantic criterion for de-
and that both were significant.
composition of lexical entries into their constituent mor-
In a study whose focus is semantic transparency, essen- phemes.
tial information derives from a direct comparison of tar-
An account that emphasizes similarity of form and of
get decisionlatenciesafter opaqueand transparentprimes. meaningas they contributeto word recognitionand,by way
One advantage of this approach is that potential con- of distributedpatterns, of morphologicalactivation(Plaut
founds related to the nature of the unrelated controlcan be & Gonnerman, 2000; Rueckl et al., 1997) may be more
avoided. When we compared target latencies after opaque compatiblewith graded magnitudesof facilitation.Accord-
and transparent primes directly, results replicated across ingly, in the present study, because (inflected and) trans-
alphabetenvironments. That is, the effect of transparency parent derived forms are related in form and in meaning,
did not interact with alphabetof the prime, althoughit did they produce facilitation, but because opaque forms are
interactwith SOA. Evidently, a focus restricted to whether similar in form but less reliable semantically, their effect
a morphologicallyrelated prime produced facilitationthat on a target depends on the relative sensitivity of the par-
meets the criterion for statistical significancefails to cap- ticularexperimentalprocedureto meaning.In essence, the
ture the overall pattern. Across alphabet contexts in Ser- critical interaction of transparency and SOA reflects the
bian, the effect of semantic transparency, as assessed by a differential contributionsof semantic similarity to perfor-
significant difference in target decision latencies after mance in lexicaldecision with changesin processing time
opaque and transparent derived primes, was present at the for the prime. The present findings complement another
long SOA but absent at the short SOA.
recent finding as well: Effects of semantic transparency

´
FELDMAN, BARAC-CIKOJA, AND KOSTIC
636
among morphological relatives appear to be systemati- Feldman, L. B., Soltano, E. G., Pastizzo, M., & Francis, S. (2001).
When does semantic transparency influence morphological process-
ing? Manuscript submitted for publication.
Frost, R., Deutsch, A., Gilboa, O., Tannenbaum, M., & Marslen-
Wilson, W. (2000). Morphologicalpriming:Dissociation of phonolog-
cally graded across targets, dependingon the productivity
or family size of their base morphemes (Feldman & Pas-
tizzo, 2002; Feldman et al., 2001).
In the present study, we demonstratedthat the influence
of semantic transparency on morphological facilitation
varies over SOA in Serbian, a languagewith a highly pro-
ductivemorphology. Moreover,thefindingis not restricted
to visual primes and targets presented in the same alpha-
ical, semantic, and morphological factors. Memory & Cognition, 28,
1277-1288.
Frost, R., Forster, K., & Deutsch, A. (1997). What can we learn from
the morphologyof Hebrew? A masked-priming investigation of mor-
phological representation. Journal of Experimental Psychology:
Learning, Memory, & Cognition, 23, 829-856.
bet. In essence, by manipulatingSOA, we have replicated Lorch, R. (1982). Priming and search processes in semantic memory:
A test of three models of spreading activation. Journal of Verbal
Learning & Verbal Behavior, 21, 468-492.
Lukatela,G., Feldman,L. B., Turvey,M. T., Carello,C., & Katz, L.
(1989). Context effects in bi-alphabetical word perception. Journal of
the contrasting patterns of transparency reported with
masked primes in Hebrew and with cross-modal presen-
tations in English. In conclusion, time- and task-varying
patterns within a language call into question claims that
languages differ with respect to a semantic criterion for
morphologicaldecomposition.
Memory & Language, 28, 214-236.
Lukatela, G., Gligorijevic´, B., Kostic´, A., & Turvey, M. T. (1980).
Representation of inflected nouns in the internal lexicon. Memory &
Cognition, 8, 415-423.
Lukatela,G.,Lukatela,K., Carello,C., & Turvey,M. T. (1993).Re-
duction in alphabet priming with delay and degradation. Journalof Ex-
perimental Psychology: Learning, Memory, & Cognition, 19, 1094-
1100.
REFERENCES
Bentin, S., & Feldman, L. B. (1990). The contributionof morpholog-
ical and semantic relatedness to repetition priming at short and long
lags: Evidence from Hebrew. Quarterly Journal of Experimental Psy-
chology, 42A, 693-711.
Bentin, S., & Frost, R. (1995). Morphological factors in visual word
identification in Hebrew. In L. B. Feldman (Ed.), Morphological as-
pects of language processing. (pp. 271-292). Hillsdale, NJ: Erlbaum.
Butterworth, B. (1983). Lexical representation. In B. Butterworth
(Ed.), Language production (Vol. 2, pp. 257-294). San Diego: Acad-
emic Press.
Bybee, J. H. (1985). Morphology. Amsterdam: John Benjamins.
Deutsch, A., Frost, R., & Forster, K. (1998).Verbs and nounsare or-
ganized and accessed differently in the mental lexicon: Evidence from
Hebrew. Journal of Experimental Psychology: Learning, Memory, &
Cognition, 24, 1238-1255.
Drews, E., & Zwitserlood,P. (1995). Morphologicaland orthographic
similarity in visual word recognition.Journal of Experimental Psychol-
ogy: Human Perception & Performance, 21, 1098-1116.
Feldman, L. B. (1992). Morphological relationships revealed through
the repetition priming task. In M. Noonan, P. Downing, & S. Lima
(Eds.), Linguisticsand literacy (pp. 239-254).Amsterdam: John Ben-
jamins.
Feldman,L. B. (2001).Are morphologicaleffects distinguishablefrom
the effects of shared meaning and shared form? Journal of Experimen-
tal Psychology: Learning, Memory, & Cognition, 26, 1431-1444.
Feldman, L. B., & Barac-Cikoja,D. (1996). Serbo-Croatian: A biscrip-
tal language. In P. T. Daniels & W. Bright (Eds.), The world’s writing
systems (pp. 769-772). New York: Oxford University Press.
Feldman, L. B., & Moskovljevic´ , J. (1987). Repetition priming is not
purely episodicin origin. Journalof Experimental Psychology:Learn-
ing, Memory, & Cognition, 13, 573-581.
Feldman, L. B., & Pastizzo, M. J. (2002). Graded semantic composi-
tionality alters morphological facilitation. Manuscript submitted for
publication.
Feldman, L. B., & Raveh, M. (in press). Semantic aspects of morpho-
logical processing. In J. Shimron (Ed.), Languageprocessing and lan-
guageacquisitionin languageswith root-based morphology.Amster-
dam: John Benjamins.
´
Lukatela, G., Turvey, M. T., & Todorovic, D. (1991). Is alphabet bi-
asing in bialphabeticword perception automatic and prelexical? Jour-
nal of Experimental Psychology: Learning, Memory, & Cognition, 17,
653-663.
Marslen-Wilson, W., Tyler, L., Waksler, R., & Older, L. (1994).
Morphology and meaning in the English mental lexicon. Psychologi-
cal Review, 101, 3-33.
Partridge, M. (1964). Serbo-Croatian practical grammar and reader.
Belgrade: University of Belgrade Press.
Pastizzo, M. J., & Feldman, L. B. (2001a). Discrepancies between or-
thographic and unrelated baselines in masked priming undermine a
decompositionalaccount of morphologicalfacilitation. Journal of Ex-
perimental Psychology: Learning, Memory, & Cognition, 28, 244-
249.
Pastizzo, M. J., & Feldman, L. B. (2001b). Does prime modality in-
fluence morphological processing? Brain & Language, 81, 28-41.
Plaut, D. C., & Gonnerman, L. M. (2000). Are non-semantic morpho-
logical effects incompatible with a distributedconnectionist approach
to lexical processing? Language& CognitiveProcesses, 15, 445-486.
Rav eh, M. (1999). The contributionof frequency and semantic similar-
ity to morphological processing. Unpublished doctoral dissertation,
University of Connecticut.
Raveh, M., & Feldman, L. B. (1998, September). The contribution of
morphologicalrelatives and affix position to morphologicalprocess-
ing. Paper presented at the X Congress of the European Society for
Cognitive Psychology, Jerusalem.
Raveh, M., & Rueckl, J. (2000).Equivalenteffects of inflected and de-
rived primes: Long-term morphologicalpriming in fragment comple-
tion and lexical decision. Journal of Memory & Language, 42, 103-
119.
Rueckl, J. G., Mikolinski, M., Raveh, M., Miner, C. S., & Mars, F.
(1997). Morphologicalpriming, connectionist networks, and masked
fragment completion. Journal of Memory & Language, 36, 382-405.
Taft, M., & Forster, K. I. (1975).Lexical storageand retrieval of prefixed
words. Journal of Verbal Learning & Verbal Behavior, 14, 638-647.
Feldman,L. B., & Soltano, E. G. (1999). Morphologicalpriming:The
role of prime duration, semantic transparency and affix position.
Brain & Language, 60, 33-39.
(Manuscript received June 23, 2000;
revision accepted for publication February 15, 2002.)