JoURNAL oF VERBAL LEARNING AND VERBAL BEHAVIoR 15, &7-620
(1976)
Lexical Storage and Retrieval of Polymorphem ic
and Polysyllabic Words
M,C.RCus
T.trr AND KENNETHI. Fonsrpn
Monash Uniuersitv
Five experimentsare describedwhich examine how polysyllabic words (e.g., DAYDREAM, ATHLETE) are stored and retrieved from lexical memory. The first four experiments look at interferenceeffectscausedby the accessingof inappropriate Iexical entries.
It is found that compoundnonwordswhosefirst constituentis a word (e.g.,DUSTWORTH,
FOOTMILGE) take longer to classifyas nonwords than compound nonwords whosefirst
constituentis not a word (e.g.,TROWBREAK, MOWDFLISK). Moreover, the presence
of a word in the secondconstituent position appearsto be irrelevant. Theseeffectshold even
when the boundary betweenconstituentsis unclearon an orthographic basis(e.g.,TRUCERIN). It is also arguedthat first syllables,as opposedto last syllables,have independent
status in the lexicon since nonword first syllables (e.g., ATH) show interference effects,
while last syllables(e.g., CULE) do not. The fifth experiment revealsthat the frequency of
the first constituent of a compound word influencesclassification times. The results point
to the conclusion that polysyllabic words, regardlessof whether they are polymorphemic or
monomorphemic,are accessed
via their first syllable.
It is commonly acceptedthat a written word
is recognized and comprehended when some
coded version of the sensory representation
of that word is found to match with a memory
representation in the reader's long-term
word store, or lexicon (Norman, 1970;
Rubenstein, Garfield, & Millikan, 1970;
Forster & Chambers, 1973).However, it has
been shown that it is not necessaryfor this
matching procedure to be basedon the sensory
and lexical representations of the complete
word. Taft and Forster (1975)have produced
evidenceto suggestthat recognition of a prefixed word (e.9., REJUVENATE) entails the
decompositionof that word into its constituent
morphemes, namely its prefix (RE) and its
stem (JUVENATE), since the stem, not the
word as a whole, is apparently the target for
Iexical search. There is also evidence to indicate that inflected words, such as CATS or
FALLING are recognized by accessing the
lexical representations of their stems, that is,
CAT and FALL (Gibson & Guinet, l97l;
Kintsch, 1972; Snodgrass & Jarvella, 1972;
Murrell & Morton, 1974; Jarvella & Snodgrass, 1974).
If it is true that a polymorphemic word is
accessedon the basis of its stem, then what
happens in a case where the polymorphemic
word is effectively composed of two stems,
as in the case of compound words like DAYDREAM and MILESTONE ? In such a case,
is a lexical search carried out for both constituent morphemes, or is there a search for
just one of the morphemes ? If the latter is true,
which of the morphemes is used-the first or
the last? Another possible alternative is that
no morphological decomposition takes place
at all, so that a compound word is actually
stored and retrieved from the lexicon as a
whole word. Experiment I was designed to
investigate these questions.
Theauthorsaregratefulto DonaldM. Thomsonfor
his helpfulcommentsregardingthis paper.Requests
ExpmJMsNr I
for reprints should be addressedto Marcus Taft,
It has been demonstrated by Taft and
Department of Psycholory, Monash University,
Clayton,Victoria3I 68,Australia.
Forster (1975) that a nonword which ha$ the
lopyright @ tgT6byAcademictress, Inc.
AU rights ofreproduction in any form rescrved.
Pilotcd in Great Britain
e7
608
TAFT AND FORSTER
morphological structure of a prefix plus
a stem takes longer to classify as a nonword
if the stem has lexical status(e.g.,DEJUVENATE) compared with items where the stem
does not have lexical status (e.g., DEPERTOIRE). In both cases,the prefix is assumed
to be stripped off and a searchfor the stem is
undertaken; however, in the former case,the
searchis interrupted by the finding of a lexical
entry (JUVENATE, which is required for the
recognition of the word REJUVENATE)
and this slows up the classificationresponse.
The present experiment is designed on
similar lines: it is assumedthat classification
times for nonwords will be lengthened
if the lexical search is interrupted by the
discovery of an entry that ultimately turns
out to be inappropriate. If compound nonwords (e.g.,DUSTWORTH, MOWDFLISK)
are classifiedas nonwords on the basisof their
constituent units rather than on the basis of
the item as a whole. then those nonwords
whose constituent units have lexical status
(DUSTWORTH) would be associatedwith
longer classification times than those nonwords whoseconstituent units have no lexical
status(MOWDFLISK).
If classification time is based on lexical
searchfor only the first constituent unit, then
the lexical status of the second constituent
would be irrelevant. Thus, honwords like
FOOTMILGE would take longer to classify
than nonwords like TROWBREAK. Conversely,if classificationtime is basedon lexical
search for only the second constituent unit,
then TROWBREAK would take longer to
classify than FOOTMILGE. If both constituent units were important, then there
would be no differences between FOOTMILGE and TROWBREAK.
Method
Materials. Four different nonword conditions were set up with 10 items in each:
nonwords constructed from two unrelated
words (WW), that is, DUSTWORTH,
BRIEFTAX; nonwords constructed from a
word followed by a nonword (WN), that is,
FOOTMILGE,
CLEANMIP;
nonwords
constructed from a nonword followed by a
word (NW), thatis, TROWBREAK, THERNLOW; and nonwords constructed from two
nonwords (NN), that is, MOWDFLISK,
SPILKWUT. Constituent units were matched
acrossthe four item conditions for length and,
where they were words, for frequency of
usageaccording to the Kudera-Francis word
count (Kudera & Francis,1967').
The 40 nonword items (presentedin the
Appendix) were randomly interspersedwith
40 distractor word items which were constructed in the same way as the nonwords,
that is, from two unrelated words (e.g.,
BLACKMAIL, NICKNAME); from a word
followed by a nonword (e.g., SANDWICH,
NEIGHBOn); from a nonword followed by
a word (e.9., HENCHMAN, STALWART);
and from two nonwords (e.g., SHRAPNEL,
wALRUS).
To eliminate any doubts as to what the
constituent units were for any particular
item, the constituent boundary of every
compound nonword (and compound word)
was indicated by a letter pair that was of
zero, or at least very low, probability of
occurrence within a syllable, for example,
the TM of FOOTMILGE. Thus, FOOTMILGE has to be analyzed as a word plus a
nonword, that is, FOOT + MILGE, since
both FOO + TMILGE
and FOOTM +
ILGE involve orthographic "illegalities."
Procedure.The procedure was the same for
all experiments reported in this paper. Items
were typed on cards and presented by means
of a two-field tachistoscope for 500 msec.
The subject held a "Yes" response button in
one hand and a "No" responsebutton in the
other and was instructed to press the former
if the stimulus item was a valid English word
and the latter if the stimulus item was not a
word. SubjectSwere told to respond as quickly
but as accurately as possible. The intertrial
interval was approximately 5 sec, with the
experimenter saying "ready" prior to the
POLYMORPHEMIC AND POLYSYLLABIC WORDS
lSt
ds
a
\VO
K,
3d
d,
of
fd
e
.r
v
)
f
f
ls
presentationof each stimulus item. The items
were presented in a different order for each
subject. Twelve practice items were also
presented.Fifteen paid subjectswere used in
each experiment.
In this and in all subsequentexperiments,the
effects of isolated trials with exceptionally
long or short latencies were minimized by
establishinga cutoffpoint two standarddeviation units away from the meansfor each subject and by setting any outlying valuesequal
to the cutoff. Trials on which an error was
made were omitted.
no difference between WW and WN nonwords, min F'< l, with neither the subject
anal ysi s, Fr(\, 14):0.23, p> .05, nor t he
item analysis,^Fr(l, 9): 0.41, p > .05, being
significant. Finally, NW and NN nonwords
did not differ, min F'< l, with neither the
subject nor item analyses being significant,
f't(l, 14):0.18, p > .05 and Fr(l, 9):0.07,
p > .05.
Although more errors were made in the
WW condition than in any other, an analysis
of the errors across the four conditions revealed no overall significant difference, min
F ' , ( 3 , 2 0 ) : 1 . 2 7 , p> . 0 5 .
Results and Discussion
It is evidentthat somesort of morphological
analysis
must be involved in the processing
Mean reaction times and percentageerror
of
compound
items, sincethe lexical status of
rates over all subjectsfor each of the nonword
the
constituent
units of compound nonwords
conditionsare given in Table l.
is able to influence classification times. In
TABLE I
point of fact, it is clear that it is only the lexical
MraN Cl.assrncanoN Tlurs (MrlllsecoNos), SrlNostatus of the first constituent that is important
lnoEnRoRs, AND PrRcrNrece Ennon Rlrrs ronWW,
in this regard; whether or not the second
WN, NW, lNo NN NoNwonos (ExpEnrueNr I)o
constituent unit is a word appearsto be irrelevant to the speedof the classificationresponse.
Percentage
Condition
Example
RTD S.E'
error
This conclusion is basedon two results.First,
the only conditions showing longer response
WW
DUSTWORTH
758 3 7
9
times than the NN control condition were
WN
FOOTMILGE
765 45
2
those
in which the first constituent was a word
NW
TROWBREAK
682 32
5
NN
MOWDFLISK
(i.e., WN and WW). Second, no extra inter677 28
4
ference was produced when the second cono Standard errors are based
on between-su.bjects
stituent was a word (i.e., WW was no slower
variability.
D Reaction time.
than WN).
On the basis of these results, the following
A one-wayanalysisof variancewith repeated model of compound word recognition can be
measures revealed an overall difference in formulated. The item is first decomposed
reaction time across the four conditions, into its constituent units (Experiment IV will
m i n F ' (3 , 1 8 ): 6 .4 8 , p < .01. Looki ng at examine how this is achieved) and a lexical
individual comparisons, it was found that search is undertaken for the first constituent.
both WW nonwords (DUSTWORTH) and When the appropriate lexical entry is found,
WN nonwords (FOOTMILGE) were associ- information within the entry stipulateswhether
ated with longer reaction times than NW
or not the second consitituent can go together
nonwords (TROWBREAK), min F'(1, l5):
with the first to form a valid English word.
9 . 5 9 ,p < . 0 1 a n d m i n F ' ( 1 , 1 7 ) : 7 . 5 7 , p < Only if the second constituent is listed in this
.02, respectively, and also longer reaction way will the item be recognized as a word.
times than NN nonwords (MOWDFLISK),
In compound words, suchasDAYDREAM,
min F'(1, 19): 12.65,p < .01 and min F'(1, the first constituent (DAY) is both a Ulor23):8.31, p <.01, respectively.There was pheme and a syllable, and thus the results
610
ofthis experimenf
qro ^..r
ambiguous
storage
i, i" Jli'are
AND F'R'TER
-
asrowhether
svrra-b.rer:
,ffiT:?.'f,_il,"f.:iiigi=,
1'olil.ilj:JT[iiT:li::::_o,
,T:Tl
oe rnade of
v
* ;sH
;ii:;ft-nr;*J.
"+T;:"",'rTrtr'lgl***;::ll'.#
i' in,...,
wourd
ouuiiu,rv
,r,.*":;;|s'
stored
f i.".
aswhor"
as
"";r;;;
a
ifiiit.i:SJ;,1io.n
tecause
r,,.ll,n?sified nonwolo
Ho;'J;H,
nrst
of rnorphemes
:'fftr',df:
sylraul.
',1.'ltities.
i'po.;fl;,'lilf
iT: ;:11g.
"' ;I;':i'€rt
larseror the[:,'-' "r
wasaccesseo
:tro,,
couta
noi
Aru*her
u"
;;
"""'*'orpheme
;:liq
i*ii,*:i:'":"i"tiltl$i
"i*ln1'ff
!Xfit""it!""1
*Jl",l,il*."tifi::"":*ouro'
^:::,'1o
srarus
in
asimironwords'
I*'J1';,#
:rJ:i*:
4"tl1?
x#ftii,;ic'
+#n#l
;".:.,'.'fi
.*u'in,i;'' ;; ?J;.".'t
andFEcT.a^ojlf 'uvrNarl,*"aa",
ilffi,**"
wordentrvrotnraicrl.
,, l^1n.igruro,r""
ii'.[.:?:ild?''r:T#*Tf
& Forster,i
:#
L?''
:"Fr*tt' lfiff}H*tu
fi:',itl{
;l#,u,,"inlil,
::::, Forsrer
sytlabic*o.ol' ::^""" beargued,rn"r,-oor,r]
svllableffi:
Te.rePresented
'::*;' ;il t,:;Ji'J:rt:'J I
PLAr,
ArH., rA D shourd
i
crassiry
; ;;;:1
to t* *i.l'ffi1:Y*.*"s
ARN,
ofwords
ro.Sj
a setoipr.oi.rions
ot'J,.i,,'ij,ro,non_
words
";;;;:'-y:r,IF*; yliili'ffiIl'1i
which
are
arso
'fhe
'i
anacef'l:,.n:*;;rlt'lonser
which
,[::vt"ur"'"s'
;*;,xiif'l':i:":!tTiiT*.i"ni,iil|T
sfi
iff
#,",.,:
:?,::-*tr*:t
fitll*1*5[1ff$
;ii."'",Jif
#lT,[ffi:n:;
:-::'o;
ITI*:::;*{ j:"_?:::,T:
f::,:fr*i ;;1;
"#ffi
l;H''r:,i;,;
tF ifli;g1}[,4-1t"
tr#
asworo'
*ru.llffiX
thar
..lT ?t'.1;:',1*tli":,':
nonwords
rike:1,.,:j:tii;fftil'rr
dition
rl 'r,",iJ;;?J;T$Jtl;
6),
FAWN).
,rlT y:Tl.,ti?",rhen,
isthatDo'wn-r^
r:r
crassiry'-,;ul"tll
1..":1i:"
i*i, ffi-;$-',,.}
.".;;;-r,"L,1tr[t'j:
]"lolu""il
donor r
ii",i,t'"tilf1.ii1,i;gg#:
asBRor Merhod
rwen
*!:'""'"
t:""'utli iltv ConditionI nonwords
x*r"T:::#i*i)
,':,iil'$fi:!fiH:,j
entrv;;llY'ld 1rt" [il
Indicates
lnrormation
*,'
***i;',,*r,,,i*if
fttffilllf#fififfi
:ffi.ltr
courd
o"ii"iniiTlH;T.Tlb:TfiJ,Hl
'##;",.0ro.
fij:ffi'-T,-,?:.::h:.r
.r
in tl
tharir lj:y'
"""*".il;.:T:,fi"#
;ftffi:;
: " ' . ' 1 ; - ' ' , 1 : . ; ' . ' 3: ,;
611
POLYMORPHEMIC AND POLYSYLLABIC WORDS
TABLE 2
2, however, the removal of letters was not at a
from MrnN
syllable boundary (e.g., BROT
(MnusrcoNos),
CrAssrptcrrtoN TIurs
BROTHER). Condition 3 nonwords were not SraNo,cRo ERnons, lNo PnncTNTAGEEnRon Rarrs
the first parts of any English words (e.g., roR NoNwono ConoITIoNs l, 2, ANo 3 ltlo Wono
CoNotnous 4, 5, aNo 6 (Exrrnuvmvr II)
PREN). Length was matched across conditions.
Percentage
Each of the word conditions contained l5
error
Example RT ,SE
Condition
items. There were three word conditions
constructed in the same way as the three Nonwords
nonword conditions exceptthat the items were
(l) First syllable PLAT
7ll 35
1I
words and were matched across conditions
of word
l2
(2) Not first
672 30
BROT
for their frequency of occurrence. In both
syllableof
Conditions 4 and 5, the words from which
word
the experimental items were constructed
(3) Unrelated to PREN
654 30
were always more frequent than the experiwords
mental items themselves,that is, NEIGH from
Words
NEIGHBOR (Condition 4), SHREW from
(4) First syllable NEIGH
l8
650 29
SHREWD (Condition 5). Condition 6 conof another
sisted of words which were not part of any
word
16
(5) Not part of SHREW 602 24
other word (e.9., SCOFF). All items are
another
presentedin the Appendix.
Resultsand Discussion
Mean reaction times and percentageerrors
over all subjectsare presentedin Table 2.
Looking at the nonwords first, a one-way
analysis of variance with repeated measures
on the one factor showed that there was an
overall significant difference across the three
conditions, min F'(2, 33): 4.99, p < .05.
Further analysis revealed that, as predicted,
Condition 1 (PLAT) was associated with
longer reaction times than both Condition
2 (BROT), min F'(1,33) : 4.28,P < .05, and
Condition 3 (PREN), min F'(1, 33):9.02,
p < .0L In addition, Condition 2 did not differ
from Condition 3, min F'(1, 32):1.06, P >
.05, with neither Ft(\, 14):3.14 nor Fr(|,
19): 1.61 being significant. An analysis of
the errors revealed no difference across the
three conditions,rnin F'(2,33) < l.
Analysis of the reaction-time data for the
word conditions again revealed a significant
differenceacrossthe three conditions,min F'(2,
28): 3.82, p < .05. More detailed analysis
showed that, as predicted, Condition 4
(NEIGH) was associatedwith longer reaction
word
(6) Unrelated to SCOFF
another
word
603 26
l0
times than both Condition 5 (SHREW),
min F'(1, 25):4.70, p < .05, and Condition
6 (SCOFF), min F'(1, 28) : 7.39, P < .02.
In addition, Condition 5 did not differ from
Condition 6, min F'(1,28) < l, with neither
Fr(|, 14): g.g1 nor F2(l , l4):0.01 being
significant. An analysis of the errors revealed
no significant differences across the three
conditions,min F'(2,23): 1.12,P > .05.
The results support the hypothesis that the
first syllable of a polysyllabic monomorph has
lexical status. Presumably, such a word would
therefore be recognized by finding the lexical
representation of its first syllable and then
ascertaining from information stored in that
entry whether or not the rest of its syllables
can go together with the first syllable to form
a word.
It appears, then, that polysyllabic single
NEIGHmorphemes (e.9., PLATFORM,
BOR, ATHLETE) are recognired by the salrrc
6r2
TAFT AND FORSTER
procedure as are polysyllabic words con- word (e.g., PROBE). These items comprised
taining two morphemes(e.g., DAYDREAM,
Condition 4.
TEXTBOOK). Thus, as far as lexical storage
and retrieval are concerned,one may consider Results and Discussion
a nonmorphemic syllable (PLAT, ATH)
Table 3 presents mean reaction times andand a morphemic syllable (DAY, TEXT) as error rates
across subjects.
being functionally equivalent.
As predicted, there were no differences
betweeneither the nonword conditions or the
word conditions, min F' < | in both cases,
ExprRlurNr III
w i th Ft(I, 14):0.01, p > .05 and Fr(| ,
The pattern of interferenceeffectsobserved 14):0.02, p > .05 for the nonwords and
in ExperimentI suggestedthat the last syllable Ft(l , 14):0.02, p > .05 and Fr(| , 14):
of a polysyllabic word is not employed in the 0.003, p > .05 for the words. There were nor
initial search process. If this interpretation differences in error rates either, min F' < ll
:
is correct, then there would be no need to in each case.
The results support the hypothesis that
store the last syllable as a lexical entry.
Therefore, the effects obtained in Experiment the last syllable of a word is not independentlyt
II should not hold if last syllables,rather than accessedfor the recognition of that word, i
first syllables,are used as items. Thus, classi- unlike the first syllable. This conclusion is
consistent with the results obtained in studies
fication of CULE (from MOLECULE)
as a nonword should show no interference in word restoration (Bruner ,& O'Dowd, r
1958; Chambers, 1975) and redintegrativel
effect, and, similarly, recognition of LEDGE
memory
(Horowitz, White, & Atwood, ,
as a word should not be affected by the ex1968;
Horowitz,
Chilian, & Dunnigan, 1969;
istenceof the word KNOWLEDGE.
Method
TABLE 3
Turrrs (MnrrsrcoNos),
Materials. Condition I consisted of 15 MraN CusslnclrloN
nonwords which were derived by removing SuNoanp EnnoRs, .cNn PEncTNTAGE Ennon Rlrrs
ron NoNwonp
C-oNurroNs I .lNp 2 lNo Wonp
all but the last syllable of polysyllabic English
CoNprrroNs 3 aNo 4 (ExrrnrurNr
III)
words, that is, CULE (from MOLECULE).
These were matched in length and in consonPercentage
ant-vowel arrangement with 15 nonwords
Condition
Example RT SE
error
which were not parts of any English words,
that is, SUNE (Condition 2). SinceConditions Nonwords
(1) Last syllable CULE
2 and 3 in Experiment II did not differ, it was
ffi31
t5
of word
felt that there was no need in this experiment
(2) Not last
SUNE
639 31
t0
to include a condition where the nonwords
syllable of
were last letters of words but not last syllables
word
(like CALE from SCALE).
Words
Condition 3 consisted of 15 words, each of
(3) Last syllable LEDGE 553 20
which was the final syllable of a word of higher
ofanother
frequency, that is, LEDGE (where KNOWword
(4) Not last
LEDGE hasa higher frequencythan LEDGE).
PROBE 552 M
syllable of
Each Condition 3 word was matched in
arrother
frequency with a word that was not part of
word
any other more frequently occurring English
-
: . ; ..
:::r 1 _':j;11ri" r.i.ljr
\:"!,!-&11&f4ff.?-
.. . rl,#i-
Dolinsk
suggeste
word, o
ment, c
what thr
So far
syllable
m ar ked
let t er c(
would, i
than at i
effects I
possible
For exa
consider
TRUCE
syllable
In such
effects o
could be
weakene
The a
effects w
ments a)
syllabific
then Wl
boundari
way as V
(like FOt
with clea
MUN) i
syllableb
take lonl
words wt
(e'g., GL
have clea
MrP).
Method
Materi,
boundari
signedin
the:WN ,
these':wa
'nl
$isttil
613
POLYMORPHEMIC AND POLYSYLLABIC WORDS
Dolinsky, 1973). These studies have all
suggestedthat the beginning fragment of a
word, compared to a medial or a final fragment, contains the most information about
what the whole word is.
ExprRturxr lV
So far, we have dealt with caseswhere the
syllable or constituent boundary is clearly
marked by an extremely low probability
letter combination or by a sequencethat
would, in fact, be illegal at any other position
than at a syllable boundary. Would the same
effects be expected if there were several
possible syllabifications of the test items?
For example, there are no orthographic
considerations that force an analysis of
TRUCERIN as TRUCE + RIN. The first
syllable could be either TRU or TRUCE.
In such cases,it might be expectedthat the
effects of a word embedded in a nonword
could be either nonexistent or considerably
weakened.
The alternative is that the interference
effects we have observed in previous experiments are in no way dependent on prior
syllabification of the test item. If this were so,
then WN nonwords with unclear syllable
boundaries should be classifiedin the same
way as WN nonwords with clear boundaries
(like FOOTMILGE). So, both WN nonwords
with clear syllable boundaries (e.9., LATCHMUN) and WN nonwords with unclear
syllableboundaries(e.g.,TRUCERIN) should
take longer to classify than both NN nonwords which have unclear syllable boundaries
(e.g., GLICERAX) and NN nonwords which
have clear syllableboundaries(e.g.,BONCH-
MrP).
Method
Materials. The WN nonwords with clear
were deboundaries (e.g., LATCHMUN)
signed in the same way as the nonwords of
the WN condition in Experiment I. Each of
these was matched with a WN nonword
with an unclear boundary (e.9.,TRUCERIN)
for both length and frequency of occurrence
of the first constituent. Each of these WN
nonword pairs was matched in length to a
NN nonword with a clear boundary (e.9.,
BONCHMIP) and a NN nonword with an
unclear boundary (e.g., GLICERAX). There
were 16 such matching quadruplets and these
are presentedin the Appendix.
The distractor items were 48 polysyllabic
words which were constructed in the same
way as the nonword items; that is, words
which consisted of a word followed by a
nonword with a clear syllable boundary (e.9.,
SANDWICH), words consisting of a word
followed by a nonword with an unclear
syllable boundary (e.9., MINESTRONE),
words consistingof two nonwords with a clear
syllable boundary (e.g. TECHNIQUE), and
words consisting of two nonwords with an
unclear syllableboundary (e.g.,PEDESTAL).
Resultsand Discussion
Reaction time and error rate means are
presentedin Table 4.
A 2 x 2 analysis of variance with repeated
measureson each factor was carried out, the
factors beingclearnessof boundary and lexical
status of the first syllable. A significant main
effect of the lexical status of the first syllable
.
TABLE 4
(MllllsecoNps)'
TIurs
CrassrrrclrroN
MelN
SrnNnlnp ERnons, nNo PrncrNTAcE Ennon Rlrrs
wrrn CLs,cR AND Uxclrln
ron WN NoNwonns
BouNonnrrs aNo NN NoNwonos wtrH CrelR AND
UNcr-Enn BouNolnres (ExeEnIunNr IV)
Condition
Example
RT
SE
Percentage
error
Clear boundary
LATCHMUN
WN
BONCHMIP
NN
&034
594 26
5
4
Unclear boundary
TRUCERIN
WN
GLICERAX
NN
@3 38
603 28
5
2
6t4
TAFT AND FORSTER
w a s o b t a i n e d ,m i n F ' ( 1 , 2 4 ) : 9 . 6 9 , p < . 0 1 , effects when the first syllable of a nonword
with WN nonwords taking longer than NN
item was not a word, even when the second
nonwords. However, there was no effect of syllablewas a word.
c lear nesso f b o u n d a ry , m i n F ' (1 , 2 9)< 1,
In many cases,of course,a searchfrom left
w i t h F , ( 1 , 1 4 ) : 0 . 3 4 a n d F r ( | , 1 4 ): 0 . 2 6 , to right will involve the accessingof seueral
and there was no interaction betweenthe two lexical entries, only one of which can be
f ac t or s ,mi n F ' (1 , 2 5 ) < l , w i th F r(I, 14):
correct. For example, a search for the word
0.50 and Fr(|, 14):0.20. A similar 2x2
HENCHMAN
will encounter not only
analysisof varianceon the error data revealed HENCH, but also HE and HEN. These
no significanteffects,min F' < | in each case. latter two would be found to be inappropriIt is apparent from the results that the ate when it is ascertained that NCHMAN
clearness of the constituent boundary is does not combine to form a word with a
irrelevant: WN nonwords take longer to first syllable HE and CHMAN does not
classifythan NN itemsregardlessof how clear- combine with a first syllable HEN. Of course,
ly the word constituentcan be separatedfrom it is alsopossiblethat HE and HEN are rapidly
the rest of the nonword. This implies that the discardedas potential first syllableswhen it is
access procedure responsible for the inter- discovered that the remaining constituents
ferenceeffectsdoes not attempt to determine would contain orthographically illegal sethe syllable boundaries prior to access.But quences.
this conclusion creates problems. If the
It is of some interestto note that the results
accesssystem usesless than the whole word, of this experiment are very damaging to any
then what is it that determineswhich subset theory that requires that access can only
of letters to use, if not the putative syllable occur after the stimulus item has been reboundary ?
codedphonologically(e.g.,Hansen& Rodgers
In attempting to solve this problem, it is 1968; Spoehr & Smith, 1973). For example,
necessaryto postulate a segmentation pro- it seems most unlikely that TRUCERIN
cedurethat alwaysselectsas an initial segment would be recoded with a phonological reprethe sequenceof letters that forms a word, if sentation that preservesthe pronunciation of
such a segment exists. Otherwise, the inter- TRUCE. In fact, according to the rules of
ferenceeffectswould not have been observed. Hansen and Rodgers, this item would be
The only solution we can find is to suggestthat pronounced as TRU + CE + RIN. Thus,
there is a procedurewhereby a seriesof lexical there would be no lexical entry for its first
searchesis undertaken for successiveletter syllable, and, consequently, no interference
combinations in the test item that begin with effectswould be observed.
the first letter of that item. That is, for the item
So far, most of the evidencefor the importTRUCERIN, there is a search for T, TR, ance of the first syllable in lexical accesshas
TRU, TRUC, TRUCE, and so otr, until been derived from experiments using nonone of these letter combinations is found to words as test items. This is primarily due to
correspond to a lexical entry (in this case, the fact that the interference effect in the
TRUCE). If no entry is found, then the item classification of nonwords, caused by the
can be classified as a nonword at that point. locating of a lexical entry, is such a powerful
But if an entry is found, then the contents of vehicle for word-recognition research. Howthe entry would be examined to determine ever, it is important to ensure that the firstwhether the remaining letters in the test item syllable strategy is not somehow restricted to
could be combined with the first syllable to nonword items. Experiment V is addressed
form a valid English word. Such a strategy to this issue and examines whether the first
would not, of course, suffer any interference syllable plays the same kind of role when
: .!:
polysyl
phemic
Ifac
ing it s
of occr
inf luen
t hough
STAN
eflCe
itt
count,
t han i
words
first cr
wordsi
( Ruber
Forster
t97s),
t ake lo
STAN
A s i
pected
DUST
items r
the ba:
that wa
it need
t hat is
possibl
that fo
possibl
one lex
for exa
for the
BAG , .
bead
for the
fore, ir
when it
entry I
constitr
to forn
the firs
before
reliablS
POLYMORPHEMIC AND POLYSYLLABIC WORDS
polysyllabic words (in this case, polymorphemic words) are used as test stimuli.
615
search.Since the searchcontinues to the same
point regardless of when the interruption to
the search occurred, the frequency ofthe first
constituent should not affect reaction times.
ExpnnrurNr V
This sort of argument was also usedby Forster
If a compound word is recognizedbyaccess- and Bednall (1976).
Thus, the predictions for this experiment
ing its first constituent, then the frequency
of occurrence of the first constituent should are that for words,the frequency of occurrence
influence reaction times. To illustrate: A1- of the first constituent will control reaction
time, but that for nonwords, no such effect
though the words LOINCLOTH and HEADSTAND have the same frequency of occurr- will be observed.More specifically,compound
ence according to the Kudera-Francis word words that have low-frequency words as
count, the word LOIN is much less frequent first constituents (LF word condition, €.9.,
than HEAD. Therefore, since compound LOINCLOTH) should take longer to classify
words are recognized on the basis of their than compound words that have high-frefirst constituents, and since high-frequency quency first constituents (HF word condition,
words are accessedbefore low-frequency words e.g.,HEADSTAND), and, second,compound
(Rubenstein, Garfield, & Millikan, 1970; nonwords that have low-frequency words as
Forster & Chambers, 1973; Taft & Forster, first constituents(LF nonword condition, e.g.,
1975), one would expect LOINCLOTH to STALEGRIP) should take no longer to
classify than compound nonwords that have
take longer to classifyas a word than HEADhigh-frequencyfirst constituents(HF nonword
STAND.
exwould
be
condition, e.8., STONEFOIL).
effect
A similar frequency
(e.9.,
WN
nonwords
WW
and
pected for
DUSTWORTH, FOOTMILGE), if these
Method
items were classified as nonwords solely on
Materials. Twenty LF words (e.9., LOINthe basis of information in the lexical entry
were matched for length and overall
that was found for their first syllable. However, CLOTH)
with 20 HF words (e.g., HEADfrequency
it need not necessarilybe true that the entry
items were divided into the two
The
that is found for the first syllable is the only STAND).
basis
of the frequency of occurthe
classes
on
possible entry for that syllable. If the letters
The first confirst
constituent.
of
their
rence
that form the first syllable can have several
of low freas
being
considered
was
stituent
possiblemeanings,theremaywell be more than
of l0 or
value
frequency
quency
had
a
if
it
one lexical entry for that letter combination;
Kudera-Francis
the
to
according
below
for example, the entry SAND that is accessed
word count.l The high-frequency constituents
for the recognition of SANDMAN, SANDfrequencies of 50 and above. SimiBAG, and SANDPAPER would presumably were of
20
LF nonwords (e.g., STALEGRfP)
larly,
be a different entry SAND to that accessed
for length with 20 HF nonmatched
were
for the recognition of SANDWICH. There(e.g.,
STONEFOIL).
words
fore, in the case of a WW or WN nonword,
All items are listed in the appendix.
when it is found from information in the lexical
entry for the first syllable that the second
I A few words, like HENCHMAN and LUKEconstituent does not combine with the first
whichhavea nonwordfirst constituent,were
WARM,
to form a word, a further lexical search for
asLF words,sincea nonwordis equivalent
considered
the first constituent would have to take place to an extremelyrare word. In fact, Forsterand Chambefore classification as a nonword could be bers(19?3)found that very rarewordsweretreatedin
reliably made. This would be an exhaustive thesamewayasnonwordsin a namingtask.
TAFT AND FORSTER
616
Resultsand Discussion
rates over
Mean reaction times and error
5'
Table
all
- subjectsare presentedin
showed
A; analysis tt tttt reaction times
(LOINCLOTH)
that, as pr.di.t.d, LF words
recognize than
;;;k significantly longer to
F'(l ' 25) :
min
Hr *oias (HEADSTAND) ,
(STALE5-.82,p < .05. The LF nonwords
nonwords
HF
the
Cnipi did not differ from
in their classificationtimes'
tiioNerolL)
though the
*i, F'(1, 26):l'13, P>'05'
F
significant
a
yield
r"i.i.", analysis did
H
o
w
e
v
e
r'
<
'
0
5
'
valut, Fr(l, 14):6'08' P
recognized
quickly
more
LF nonwords were
this is unthan HF nonwords and, hence'
effect' The
frequency
likely to be a normal
F2(l' 19) :
significant'
not
itenr analysiswas
1 . 3 5P, > . 0 5 .
TABLE 5
MraNCt-lssIrtcnrIoNTn,ars(Mtt-t-tsrcoNos)'
Rnrrs
EnRons, lNo PsncTNTAGE Ennon
StlNolno
rNo LF NoNwoRos
io" fff rNo LF Wonns rNo HF
(ExpEnrrvmNr V)
Condition
Words
HF
LF
Example
HEADSTAND
LOINCLOTH
Nonwords
STONEFOIL
HF
STALEGRIP
LF
RT
SE
598 27
629 28
827 60
808 56
Percentage
error
the
This, however, does not Preclude
a
as
word
the
of
frequency
the
possibility that
time
the
on
influence
an
have
*ttot. can also
decision
taken to recognize the word' The
go
together
can
constituent
that the second
be inwith the first constituent is likely to
conthat
of
commonness
the
fluenced by
frequency
the
is,
that
stituent combination,
the items
of the word as a whole' For example'
were
HEADSTAND
and
HEADACHE
former
the
with
experiment,
this
both used in
the latter
taking 547 msec to recognize and
be
probably
would
This
msec.
taking 640
by
becauie HEAD is more often followed
of
ACHE than bY STAND, the frequencY
of
HEADACHE being 5 and the frequency
HEADSTAND being l '
the first
The fact that the frequency of
times
constituent does not affect classification
the
to
of compound nonwords adds support
is
it
after
claim that the search must continue
the
for
discovered from the lexical entry
constituent
first constituent that the second
result'
This
first'
the
to
partner
valid
a
is not
and
Forster
along with those reported by
the
for
case
Bedriall (1976), make a strong
of
terms
in
interpretation of frequency effects
the order in which entries are searched'
s
9
7
4
signiThe analysis of errors produced no
word
the
between
either
ficant differences,
: 1'94' P > '05'
30)
F'(1,
min
conditions,
min
or between the nonword conditions'
:
0
.
5
9
,
.
0
5
'
>
P
F'(1, 30)
ihe resrrltsobtained again strongly support
recogthe notion that compound words are
nized on the basis of their first constituents;
varied'
if the frequency of the first constituent is
as a
then the recognition time for the word
whole also varies' even though the frequency
of the word as a whole is held constant'
Gnurnel DrscussloN
paper
The five experimentsdiscussedin this
makethe following Points'
polysyllabicword
itl The lexicatentiy for a
first
is accessedvia a representationof its
this
not
or
whether
syllable irrespectiveof
lexical
the
in
,yttuUt. is a word. Information
of
*try stipulatesthe possiblecombinations
syllable'
first
letterstliat can follow the
(2) Storageof polysyllabicwords is the same
,.gutJr.tt ir how many morphemet th'
(however,seelater for a disword possesses
cussion
--i,tt" of Prefixedwords)'
determinethe first syllableof a word'
of the
a leiical searchis undertaken for all
until
righ{
to
left
from
letter combinations
found'
is
entry
an appropriatelexical
IS
TI
ct
in
(\
th
m
l(
pr
pl
rI
(r
rt
v
tl
tt
\l
s'
s'
tl
S
It
ft
a
6
a
a
v
u
S
F
S
s
1
\
C
I
(
I
I
I
POLYMORPHEMIC AND POLYSYLLABIC WORDS
6t7
An important questionthat should be raised and that words
are not stored on the basis of
is the generalone of the definition of a syllable. their pronunciation,
as Hansen and Rodgers,
The difficulties in defining a syllable are dis- and Spoehr
and Smith, would hold.
cussed by Hansen and Rodgers (1973) and,
It should be noted that, even if a reader is
instead,they propose a Vocalic Center Group not consciously
aware of what is stored in his
(VCG). A VCG is a syllable-like strucrure lexicon,
which would be the case if abstract
that is defined in phonological terms (as are representations
are stored, he should nevermost definitions of syllables, e.g., Bolinger, theless have no
difficulty in accessingappro1968; Mackay, Ig74) as being the smallest priate lexical
entriesaccordingto the model of
pronunciationalunit within which all rules of word recognition
proposed in this paper.
phonemic co-occurrence,that is, phonotactic Since
a searchis conductedon letter combinarules, can be specified.Hansen and Rodgers tions frorn
left to right, the reader will eventu(1968)and Spoehr and Smith (1973)present ally
use the correct letter combination that
rules by which a word can be parsed into its encounters
the appropriate lexical entry.
VCGs. However, problems arise if we define
The notion that only the first syllable of a
the syllable or VCG in purely phonological polysyllabic
word is used in the accessingof
terms. For instance, while the word SIGN lexical information,
however, is seeminglyin
would be considered phonologically as one conflict with
the proposal put forward by
syllable, in the related word SIGNIFY, the Taft and Forster (1975)
that a prefixed *ord
syllablebreak would occur betweenthe G and (e.g., REJUVENATE)
is recognized on the
the N. Thus, it would haveto be that SIGN and basisof its stem (JUVENATE).
By definition,
SIGNIFY are accessed through different it is the prefix,
not the stem, that is the first
lexical entries, namely, SIGN and SIG, syllable of
a prefixed word. However, the
respectively. However, if one subscribesto conflict is easy
to resolve. Obviously, prefixes
a generative phonological theory (Chomsky must be listed in
the lexicon or elsethey could
& Halle, 1968),such a state of affairs is to be not be recognized
as prefixes. Hence, we can
avoided, sincethe relationship betweenSIGN assumethat the left-to-right
search technique
and SIGNIFY would be lost. Instead, one will first isolate
the preflx as the initial syllwould have to postulate that the important able, but when it
is discovered that the first
unit of reading is some sort of abstract syllable is a prefix,
the search begins again
syllable that is not necessarilymanifest in the from the first letter
after the prefix. That is,
pronunciation of the word. Thus, SIGN an& the prefix is "stripped
off." Thus, the only
SIGNIFY would both have the same first modification to the
theory required to acsyllable,which might be either SIG or SIGN. commodate the
findings for
prefixed
The same problem arises with many other words with those for polysyllabic
words is
words, like THIRSTY (syllabifiedby Hansen that the term "first syllable"
should be taken
and Rodgers' rules as being THIR + STy), to mean "first syllable
after the prefix (if
FASTING (FAS + TING), and ACTOR
any)." This would permit items such as
(AC + TOR). If thesewords were stored in the MISHAP and
HAppEN, INSPECT and
Iexicon on the basis of their phonologically SPECTATOR to be stored
together, namely,
defined first syllable, then the relation- in the lexical entries HAp
and SpECi,
ship between such pairs as THIRSTy and though there has beenno
evidenceproduced to
THIRST,
FASTING
and FAST, and say that this is actually the case.
ACTOR and ACT would be lost. Thus, it is
Throughout this report, we have avoided
likely that there is an abstract level of syllable extensive discussion of alternative
explanarepresentation which maintains the relation- tions of the data. This has
been mainly in the
ship betweena suffixed word and its baseform interests of clarity and should
not be taken to,
23
618
TAFT AND FORSTER
imply that there are no viable alternatives sample of nonwords
was also present in the
or that we have not consideredthem. Most sample of words. Given
the very stringent
of the theoreticalclaimsmadewould, however, criteria for selecting
words in these experibe unchangedif we were to switch to a totally ments (e.g., being the
first syllable of a more
different model of word recognition,that is, frequent word), this seems
most unlikelv.
a direct-accesscontent-addressablememory.
The three major claimsmade at the beginning
ApprNorx
of t his d i s c u s s i o ns e c ti o nw o u l d s ti l l stand,
althoughthe wording of the third claim would
Belowarelistedthe itemsusedin eachexperihaveto be alteredto eliminateany referenceto ment, together
with the mean lexicaldecision
"search." In our view, the only result that
time for the item.
might commit one to a search model is the
finding in Experiment IV that frequency of Experiment I
the first syllableis relevant only when the item
The items are arranged in quadruplets,
is a wo rd .
with a WW nonword followed by a matching
It is possible that, in constructing items WN, NW,
and NN item:
for each of the experiments,we have uninbrieftax 755, cleanmip 924, thernlow 632,
tentionally confounded some totally extran_
spilkwut693;topdrug738,redbtin751.homrank
eous variable with the treatment variable.
670,radmosh681; hearfew7lg, hallwub722,
In the case of nonwords, for example, we
gurmday767,vashpon682;flowgun745,flatbew
725,&lfhit 689,haldne
g 674 ; lotcool67l, asktarp
may have accidentally selected items that
734, nasfund627, erkfand679; oddhard6gj,
varied systematically in their .,wordness',
sumwoll720,orktype624,bixmook67g; toastproperties. The nonwords of one condition
pull 805,spellcung
706,flurbpair642,thrimnade
may have been more pronounceable than
697; formmind805,bestpiltg00,toopcase792,
the nonwords of another condition. It is,
gindtrem673; dustworth7g2, footmilge ggg,
trowbreak 667, mowdflisk 661; gasbay 770,
of course, extremely difficult to show that
oilrad756,lisfat683,rogchy660.
the variable selectedby the experimentersis
the only relevant variable. One can only
Experiment II
subject the hypothesis to as many different
Nonword items. The items are arranged in
tests as possible, in each case relying on the
triplets, with a Condition I nonword followed
min F' test to demonstrate generality across 'by
its matched Condition 2 and Condition 3
items, thereby making it improbable that the
control
item, the letters in parentheses being
effectsare due to extraneousvariables.How_
the letters that have been removed:
ever, this procedure will not eliminate the
hench883 (MAN), splot 710 (CHED), shult 702;
effects of any systematically confounded
tad 712 (POLE), faw 701 (N), ged 642; trom 649
variable. Until we have a reliable method for
(BONE), traw 647 (LER), trup 593; cran 693
evaluating the properties of nonwords, the
(BERRY), scam 7ll (PERING), btet 587; chim
possibility of there being systematicallycon_
639 (NEY), plig 612 (HT), chon 596; scound
founding variableswill remain.
728 (REL), draugh 642 (TS), spoard 700; plat
699 (FORM), brot 686 (HER), pren 565; shuff
The same difficulties are not nearly as
756 (LE), twing 774 (E), shink 841; yester 740
problematical when the test items are words,
(DAY), patien 590(T), yetian 617; bund 714(LE)
since there is much less freedom in selecting
nint 673 (H), murt 691; strug 795 (LE), maint
words than is the case for nonwords. In this
707 (AIN), slift 837; voy 679 (AGE), rop 209
respect, it should be noted that several of the
(E), lom 651; fash 688 (ION), Iibe 705 (RAL),
fave 694; cust 794 (OMERS), mank 662 (IND),
experiments also used words as stimuli and
denk 688; samp 730 (LE), welf 652 (ARE), selp
it would have to be argued that the system_
634; ambass 727 (ADOR), archit 669 (ECT),
atically confounded variable present in the
almith 630; pract 663 (ICE), healt762(H), thart
6t0;
&3:
cava,
Wot
triplet
itsCor
the let
have I
neigl
shan'
nrotl"
(LE)
sr.rpp
pier
577 r
(cul
(N),
thud
6t3(l
sight
egg 5
624.
Exper
Non
pairs,
by its
parenr
remo!
cule
tepe I
(cAl
639:
tork (
627 (
630;I
lir 57
(U),r
l(or
pairs,
its Co
enthes
movec
cape
&2;
(sov
521;l
dawn
586(t
kin 6(
lain 5
soak.
POLYMORPHEMIC AND POLYSYLLABIC WORDS
tg
It
ie
610; chall 677 (ENGE), worsh 659 (IP), crish
643; ath 669 (LETE), awf 645 (UL), arn 673;
caval797(RY), dilem 667(MA), covim 604.
Word items. The items are arranged in
triplets, with a Condition 4 word followed by
its Condition 5 and Condition 6 control words.
the letters in parentheses being the letters that
have been removed:
neigh 754 (BOUR), shrew 627 (D), scoff 734;
sham 733 (ROCK), bran 741 (CH), fern 597;
moth 615 (ER), boar 556(D), frog 517;stab628
(LE), toot 603 (H), loot 571; thresh812 (OLD),
supple612 (MENT), maggot650; tick 593 (ET),
pier 617 (CE), turf 622; colon 775 (EL), locus
577 (T), venom 640; tack 602 (LE), mole 587
(CULE), Ioaf 555; candid 607 (ATE), patter 633
(N), morsel 621: garb 77O(AGE), clot 577 (H),
thud 644; bib 675(LE), fir 627(ST),nun 626;mast
613(ER),brow577(N), loom622;cult576(URE),
sigh646(T), calf 542;log 543(IC), beg544(AN),
egg 541; muff 709 (LE), dial 560 (OGUE), romp
624.
Experiment
III
Nonword items. The items are arranged in
pairs, with a Condition I nonword followed
by its Condition 2 control, the letters in
parentheses being the letters that have been
femoved:
cule 630 (MOLE), sune 633; lete 574 (ATH),
tepe590;strone706(MINE), splike775;flage7l7
(CAMOU), chipe 710; toise 702 (TOR), fouse
639; pult 677(CATA), bift 686; wark 663(BUL),
tork 639;marm 718(SCHOOL),durd 620;drome
627 (SYN), brone 681; ine 665 (MACH), ake.
630;lum646(HOOD),rom 626;ner543(PART),
lir 578; thon 568 (MARA), chom 605; surp 628
(U), solf 621; nique654(TECH), bogue661.
Word items. The items are arranged in
pairs, with a Condition 3 word followed by
its Condition 4 control, the letters in parentheses being the letters that have been removed:
cape 550 (ES), crop 5261'fare 589 (WAR), lure
602; ledge499 (KNOW), probe 578; reign 530
(SOVE), plank 524; graph 563 (PHOTO), grasp
(THOU),
521; band514(HUS),seat512;sand492
dawn 497; mite 680 (DYNA), moot 675; sphere
586(ATMO), thrust 559;ice482(SERV),odd 565;
kin 605 (NAP), fry 525; pet 498 (CAR), pad 512;
lain 575 (CHAP), pail 605; hood 562 (LIKELI),
soak 562; chant 612(MER), prowl 565.
619
Experiment IV
The items are arranged in quadruplets,
with a WNC nonword followed by a matching
NNC nonword, WNU nonword, and WNC
nonw ord:
joinlird654,doinrint613,pagelont
620,tagerald
613; warmvoost662, woamfeard627, flataunch
649, flodainst637; thawherk 560, trawyunt 563,
sofanulk565, tafomunt 568; latchmun635, bonchmip 593, trucerin 588, glicerax 594; chefpite
606, thifbure 586, trionade 723, driotane 627;
slowhent672,spowhink6l 6, treepold679,sneepisk
622; quietpeld 747, quindbilt 639, framelume
694,frinetoke593; howlboop655,morlboin 596,
hikelork 693,nipelond652; hotelrin 577,satelrop
593, radiocag 572,lodionug 549; tartconce626,
mormsarse 641, stiroarse 661, clurierce 609;
rockfeft 709, leckpept 599, votemimp 607,
wikenink 646; sendbosh622, fardbith 617, richoast 661, nishount 562; limbeand 659, fambtald
567, lushorld 644, fothanch 610; bustfoar 619,
duntvair 569, dametimp 693, ronetunk 564;
dognala 650, jegmara 570, barotta 684, horissa
596; padlato 663, tidnilo 541, rotaspo 583,
dopilto619.
Experiment V
Word items. The items
pairs, with an LF word
matching HF word:
are arranged in
followed by its
warehouse587, milestone 576; loincloth 646,
headstand640; ribcage 724, eyelash552; grapevine 629,classmate532; henchman623,shortcut
589; skullcap643,deathbed633; poppyseed620,
riverboat 558; treadmill 660, bloodshot 590;
soothsayer669, housepaint619; frostbite 625,
checklist610; teenage543,wildcat 541; dockside
663, handbook 546; junkyard 593, timebomb
648; lukewarm 602,headache
547; thumbnail 690,
spacesuit623; soybean647,seaweed599; tugboat
597, warlord 666; swineherd646, stockpile 603;
yesterday 537, everyone 599; pawnshop 650,
woodwind641.
Nonword items. The items are arranged in
pairs, with an LF nonword followed by its
matching HF word:
grindpath764,grasscast
969; pigbulk 807,farsoup
807; cavetap 814, caselip 906; swapnewe 767,
stopspark 750; stalegrip 862, stonefoil 891;
cloakbud 787, sweetrag 868; heaverace 828,
judgerise773; hikedrug 819, finerank 864; crisplobe 775, monthgasp907; hugcalf 808, askdash,
620
TAFT AND FORSTER
Journal of Experimental Psychology, 1969, 80,
392-394.
Honownz, L. M., WuITe,M. A., & Arwoon, D' W'
Word fragmentsas aids to recall: The organization of a word.Journal of ExperimentolPsychology,
1968,76,219-226.
J. G. Seeingring in rang
JnRvrlln, R. J., & SNoDGRASs,
putflag727.
and retain in retention: On recognizing stem
RrrrnrNcrs
morphemesin printed words. Journal of Verbal
and VerbalBehauior,1974,13, 590-598'
Learning
York:
New
language'
of
Aspects
BoLtxceR, D.
Abstract nouns: Imageryversuslexical
W.
KrNtscs,
Harcourt,Braceand World, 1968'
. Journal of VerbalLearning and Verbal
complexity
informathe
on
note
D.
A
O'DowD,
BnuNEn,J. S., &
ll, 5945.
1972,
Behauior,
Speech'
tivenessof parts of words' Languageand
W. N. Computatiorulanalysis
H.,
&FR^lNCts,
Kudru,
1 9 5 81. . 9 8 - 1 0 1 .
English' Providence,
American
present-day
of
word
during
CHauneRs,S. M. Orthographic analysis
Brown UniversityPress,1967'
R.
I.:
Experimental
the
presented
at
Paper
recognition
MecK.nv, D. G. Aspectsof the syntax of behavior:
Psychology Society Conference, University of
Syllable structure and speech rate' Quarterly
1975MaY
Sydney,
of Experimental Psychology, 1974, 26,
Journal
pattern
of
sound
Cnor'rixv, N., & HALIE, M. The
@2457.
English.New York: Harper and Row, 1968'
MuRRElr, G. A., & MonroN, J. Word recognitionand
DoLlNsxY,R. Word fragrnentsas recall cues: Role of
morphemic structure. Journal of Experimental
syllables. Journal of Experimental Psychology'
sychologY, 1974, 102,963-968.
P
1973,97,272-274.
A. (Ed.). Models of human memory'
Fonsrrn, K. L, & BroNrll-r, E. S. Terminating and NonuAN, D.
AcademicPress,1970'
York:
New
and
Memory
access'
exhaustive search in lexical
RusENsrnN, H., GIRTIELD,L., & Mrrr-rxlN, J' A'
Cognition,1976, 4, 53-61.
Homographic entries in the internal lexicon'
FoRsrEn,K. I., & Cunuarns, S. M. Lexical accessand
of Verbal Learning and Verbal Behauior'
Journsl
naming time. Journal of Verbal Learning and
1970,9,487494.
-635'
627
12,
3,
Verbal Behauior,197
J. G., & Jlnvrlu, R. J' Some linguistic
GlnsoN,E. J., & GutNrl, L. Perceptionof inflections SNoocnass,
of word classificationtimes' Psychodeterminants
of
Journol
words'
of
presentations
in brief visual
1972, 27, 220-222'
mic
Science,
no
l0'
Verbal Learntng and Verbal Behauior, 1971,
E- E. The role of syllablesin
Surrn,
T.,
&
K.
SeorHR,
182-189.
Cognitiue Psychology'
processing.
perceptual
of
HANSTN,D., & Roocrns, T. S' An exploration
1973,5,71-89.
psycholinguisticunits in initial reading' In K' S'
M., & FonsrER, K. I. Lexical storage and
Goodman (Ed.), Thepsycholinguisticnatureof the TArt,
' retrieval of prefixed words. Journal of Verbal
University
State
Wayne
Detroit:
process.
reading
Learning and Verbal Behauior, 1975, 14, 638-&7 '
Press.1968.
Honowrrz, L.M., CrutllN, P. C., &DuNNtcaN, K' P'
Decemberl, 1975)
Word fragmentsand their redintegrativepowers' (Received
812; hedgedate915, watchrole922; dumptail
894, carejump806; clownhip 837, cleandieT5O;'
smashboss794, smilecrop, 831; cheerturf 798
piecetick884;hogtide840,hotsigh792;slopharsh
i76, testshelf818; cagefan 788, needrod 724;
glintmeat 725, provecoat 810; petdamp 777'
JOURNAL O
T
This st
of conte
prehensir
and Joh
prerequir
a precur
investiga
nonsensi
improve'
by appr,
tures or
context
compreh
ready le
Thispa
pleted at
held a G
grateful I
chairman
comment!
drafts of
help of Ri
Walker d
collabora
of the i
should b
Universit
Reading
Illinois 6
CoPYright t
All rights o
Printcd in (