Brain and Language 68, 324–332 (1999)

Article ID brln.1999.2114, available online at http://www.idealibrary.com on

Lexical Morphology and Lexical Access

Jennifer Vannest
The Ohio State University

and
Julie E. Boland
Rutgers University

Research on morphology in word recognition has been plagued by conflicting

results (McQueen & Cutler, 1998, give a recent review). Some findings suggest that
words are accessed as full forms, while others suggest that words are accessed in
terms of their component morphemes. The answer may lie in the properties of the
affixes themselves: Kiparsky’s (1982) Lexical Phonology and Morphology assigns
affixes in English to different ‘‘levels’’ of attachment, based on their productivity,
order of attachment, and phonological interaction with roots. We present data sug-
gesting that productive, phonologically neutral, semantically transparent ‘‘Level 2’’
suffixes are ‘‘decomposed’’ for analysis in some cases, but that words with idiosyn-
cratic, structure-changing, semantically opaque ‘‘Level 1’’ suffixes are not.  1999
Academic Press

Key Words: morphology; lexical morphology; lexical access; lexical decision;

mental lexicon.

Historically, two competing models have characterized morphological

structure during lexical access. Taft and Forsters (1975) ‘‘Affix-Stripping’’
model, argues that listeners must break words into their component mor-
phemes for analysis, and that each morpheme is listed individually in the
lexicon. Butterworth’s (1983) Full Listing Hypothesis maintains that words
are available for recognition in the lexicon with their morphology complete.
More recently, several mixed models have been proposed. For example,
Marslen-Wilson, Tyler, Waksler, and Older (1994) and Wurm (1997) have
maintained that decomposition is more likely when the relationship between
an affixed word and its root is semantically transparent. Ito, Sugioka, and

Jennifer Vannest’s work was supported by NIH Grant T- 32DC 00051-02. The authors
thank Dr. Keith Johnson and Dr. Elizabeth Hume for their comments.
Address reprint requests to Jennifer Vannest, 222 Oxley Hall, 1712 Neil Ave., Columbus,
OH 43210. E-mail: [email protected].
324
0093-934X/99 $30.00
Copyright  1999 by Academic Press
All rights of reproduction in any form reserved.
 MORPHOLOGY AND LEXICAL ACCESS 325

Hagiwara (1996) suggested that only regular prefixes are attached (and there-
fore decomposed) by rule. Alternatively, Colé, Segui, and Taft (1997) sug-
gested a modification of the full-listing model where suffixed words and their
roots compete based on lexical frequency.
We introduce a mixed model loosely based on Kiparsky’s (1982) Lexical
Phonology and Morphology, which divides affixes into different levels of
attachment based on proximity to the root and interaction with phonological
processes (including stress assignment). More idiosyncratic, structure-chang-
ing derivational affixes are attached at the earliest levels of composition
(Level 1), and more productive, phonologically neutral derivational affixa-
tion takes place at Level 2 (Katamba 1990). The motivation for this arrange-
ment was initially distributional: a Level 1 affix cannot be attached to a stem
containing a Level 2 affix, although a Level 2 affix can be attached to a stem
containing a Level 1 affix, (e.g., grievousness). Both levels of affixes may
be attached to stems containing other affixes of the same level, but only
Level 1 affixes can attach to bound stems (Mohanan 1986). In addition, there
are phonological processes that apply to words containing Level 1 affixes
but do not apply to those with Level 2 affixes.
We test the hypothesis that readers access words with Level 1 and Level
2 morphology differently: suffixed words are stored in the lexicon with Level
1 morphology intact, so morphological decomposition takes place only when
words carry Level 2 suffixes.

EXPERIMENT 1
We used a lexical decision paradigm adapted from Taft (1979) and Brad-
ley (1979): participants decided whether a string of letters is a word and
responded accordingly with a button-press. This paradigm is grounded in
the assumption that frequency effects in lexical decision primarily reflect
lexical access time. We manipulated the root and whole-word frequency val-
ues (Kucera & Francis, 1967) of suffixed forms. (Root frequency is the total
frequency count of all forms including this root). Thus, there were two types
of stimuli pairs. In Root-Contrast pairs, words have very different root fre-
quencies, but are matched for whole-word frequency. For example, in the
pair worthless/seamless, worth is much more frequent than seam, but worth-
less and seamless are equally frequent. Thus, all other things being equal,
decision times should be equivalent if the words are accessed in their suffixed
form, but worthless should be faster than seamless if the words are accessed
by their roots. In Whole-Word (WW) Contrast pairs, whole-words vary in
frequency, but are matched for root frequency. For example, in endless/
pointless, endless is much more frequent than pointless, but end and point
are about equally frequent.
These pairs were constructed using suffixed words with three different
suffixes. The first was the Level 2 suffix, -less and the Level 1 suffixes:
326 VANNEST AND BOLAND

-ity and -ation. We included both -ity and -ation to examine the effect of
phonological change within Level 1 suffixation: for the majority of -ity
words, suffixation induced a segmental phonological change (i.e. trisyllablic
shortening) to the root, such as in severe/severity. However, none of the
-ation stimuli in this study involved a segmental change, though there was
a stress shift from a monosyllabic root to the suffix in some cases, for exam-
ple tax/taxation.
Bradley (1979) found that in words containing Level 2 suffixes, there was
a frequency-based difference in response-time for the Root-Contrast pairs.
For the WW-Contrast pairs, there was no frequency effect. The fact that root
frequency contributed to RT suggests that these words were decomposed.
For words with Level 1 suffixes, Bradley (1979) found a frequency effect
for WW-Contrast pairs, but not for Root-Contrast pairs. Because only
Whole-word frequency influenced RT, there is no evidence of decomposi-
tion.
Our results should be similar to those of Bradley (1979): for -less pairs,
there should be a frequency effect for Root-Contrast pairs. For -ity and
-ation pairs, there should be a frequency effect for the Whole-Word Contrast
pairs. This pattern would indicate that words with different levels of suffix-
ation are accessed differently.

Method
Materials. Eight Root-Contrast and eight WW-Contrast pairs using -less and -ity were se-
lected. Due to frequency matching constraints, only six root-contrast and only six WW-contrast
pairs were designed using -ation. Different roots were used with each suffix. Frequency con-
trasts were maximized: root frequency contrasts were generally larger than WW contrasts, but
this was consistent across the three suffixes. Log frequency means for each set are given in
Table 1.
All pairs were matched as closely as possible for number of characters (mean length 9.2).
Some words were included in more than one stimulus pair for a given suffix, again due to
frequency matching, though each word was presented only once. Included in the stimulus list
were 140 nonsuffixed pronounceable nonwords (70 of which were nine characters long to
match the mean length of the suffixed words) and 70 monomorphemic real words.
Design and procedure. Targets were presented on a computer screen. Before each one, a
visual warning signal of three asterisks appeared. This remained for 500 ms, then the visual
target word appeared. Participants made a lexical decision by pressing one response key for
real words, and another for nonwords. The target word disappeared upon response.

TABLE 1
Mean Log Frequency for Stimuli in Experiment 1

High frequency Low frequency High frequency Low frequency

Suffix roots roots whole words whole words

-less 2.36 1.32 1.29 0.15

-ity 2.32 1.16 1.82 0.60
-ation 2.23 1.03 1.48 0.26
 MORPHOLOGY AND LEXICAL ACCESS 327

FIG. 1. Mean Response times for Root-Contrast Pairs in Experiment 1.

Participants. Fifty Ohio State University students participated in the experiment for course
credit or a nominal fee. All were native speakers of English with normal or corrected-to-
normal vision.

Results
Mean RTs across participants and items were calculated for each suffix
and are shown in Figs. 1 and 2. Incorrect responses were removed from the
data set prior to analysis (5.2% of responses to -less words, 5.4% to -ity
words, and 5.6% to -ation words). Participant and item means were submit-
ted to a 2 (Whole-Word or Root-Contrast pair) ⫻ 2 (high or low frequency)
⫻ 3 (suffix group) ANOVA. By participants,1 there was a three-way suffix
⫻ pair type ⫻ frequency interaction, [F1(2, 98) ⫽ 8.5, p ⬍ .05; F2 ⬍ 1.0]
(there was also a main effect of suffix group [F1(2, 98) ⫽ 13.4, p ⬍ .05;
F1(2, 76) ⫽ 3.07, p ⬍ .052] and of frequency [F1(1, 49) ⫽ 42, p ⬍ .05;
F2(1, 76) ⫽ 7.22, p ⬍ .05]. To explore the nature of this interaction, t tests
were performed on each set of Root-Contrast and Whole-Word Contrast
pairs.
-less words. In Root-Contrast pairs: words with high-frequency roots
were accessed more quickly [t1(49) ⫽ ⫺2.06, p ⬍ .05 though not by items,
t 2(14) ⫽ ⫺1.66 p ⫽ .11]. Whole-Word Contrast pairs showed the same
pattern of results, [t1(49) ⫽ ⫺2.14, p ⬍ .05 though not by items, t2(14) ⫽
⫺1.67 p ⫽ .11]. In these words, both Root and Whole-Word frequency af-
fected RT.
1
While we report the ANOVA using the items means for both experiments, its interpretation
is problematic because the same words were sometimes used in Root and WW-Contrast pairs
for a given suffix.
328 VANNEST AND BOLAND

FIG. 2. Response times for Whole Word-Contrast Pairs in Experiment 1.

-ity and -ation words. No frequency effect was observed in Root-Contrast

pairs, but a frequency effect was evident in Whole-Word Contrast pairs [for
-ity, t1(49) ⫽ ⫺2.49, p ⬍ .05; t2(14) ⫽ ⫺1.99, p ⫽ .06. For -ation, t1(49)
⫽ ⫺2.98, p ⬍ .05; t 2(10) ⫽ ⫺1.99, p ⫽ .06]. Only Whole-Word frequency
affected RT to these words.

Discussion
Consistent with Bradley (1979), these results suggest that suffixes of dif-
ferent lexical levels may indeed have different status in the lexicon, as pre-
dicted by the level-based model. In words with the Level 2 suffix, root fre-
quency influenced recognition time, indicating that the roots themselves are
accessed in these cases. However, in words with the Level 1 affixes, root
frequency did not affect RT. It was only the frequency of the full suffixed
forms that affected response times, indicating that the full form is the only
one the reader is accessing.

EXPERIMENT 2
Kiparsky (1982) separated affixes into level of attachment based on prox-
imity to the root and interaction with phonological processes. While there
is this linguistic motivation to group a variety of derivational suffixes into
these levels, it is an open question whether Level 2 affixes, as a group, will
be decomposed like -less above. Experiment 2 investigated response times
to three groups of suffixed words, instead of examining many pairs of words
with the same suffix. The first group of words had the Level 2 suffixes,
-ship, -ness, -less, -hood, -er. The second group had Level 1 suffixes, -ous,
-ory, -ity, -ian, -ation, but no segmental phonological change in the root.
The third group included words with a variety of Level 1 suffixes (-ous,
-ity, -ian, -ary or -ion) that did induce a phonological change in the root.
 MORPHOLOGY AND LEXICAL ACCESS 329

TABLE 2
Mean Log Frequency for Stimuli in Experiment 2
High frequency Low frequency High frequency Low frequency
Suffix roots roots whole words whole words

Level 2 2.24 1.00 1.66 0.10

Level 1/no phonological 2.22 1.21 1.77 0.99
change
Level 1/with phonological 2.34 1.23 1.77 0.96
change

Method
Eight Root-Contrast and eight WW-Contrast pairs were designed for each group. No suffix
was used more than three times in a contrast set, and suffixes were matched across the two
contrast sets in each suffix group. Frequency means for each set are given in Table 2.
The design, filler trials, and procedure were exactly as in Experiment 1, and participants
were 44 Ohio State University students with qualifications as in Experiment 1.

Results and Discussion

Mean response times across participants and items were calculated for
each suffix and are displayed in Figs. 3 and 4. Incorrect responses were
removed from the data set prior to analysis (2.0% of Level 1 words; 2.1%
of Level 2 words). A 2 (Whole-Word or Root Contrast pair) ⫻ 2 (high or
low frequency) ⫻ 3 (suffix group) ANOVA revealed a pair type ⫻ frequency
interaction, by participants only [F1(1, 86) ⫽ 15.1, p ⬍ .05; F2(1, 84) ⫽
2.2, p ⬎ .10], with a frequency effect for Whole-Word Contrast pairs, but
not for Root-Contrast pairs. This pattern was confirmed by t tests on each

FIG. 3. Mean Response times for Root-Contrast Pairs in Experiment 2.

330 VANNEST AND BOLAND

FIG. 4. Mean Response times for Whole Word-Contrast Pairs in Experiment 2.

set of Root-Contrast and Whole-Word Contrast pairs (α ⫽ .05) [For Whole-

Word Contrast Pairs: For Level 2, t1(43) ⫽ ⫺3.86, p ⬍ .05; t2(14) ⫽ ⫺1.89,
p ⫽ .07. For Level 1 without change, t1(43) ⫽ ⫺3.785, p ⬍ .05; t2(14) ⫽
⫺1.92, p ⫽ .07. For Level 1 with change, t1(43) ⫽ ⫺2.77, p ⬍ .05; t2(14)
⫽ ⫺1.687, p ⫽ .11]. (Though the three pairs of -less words included in this
group showed a pattern of results similar to those in Experiment 1).
In contrast with the results of Experiment 1, these results did not support
the claim that Level 1 and Level 2 suffixes have different lexical status. Root
frequency did not impact response time to any of these words, suggesting
that the roots themselves were not accessed, but each suffixed word was
accessed from the lexicon as its own full form. Evidence for morphological
decomposition of only Level 2-suffixed words that was clear with -less in
Experiment 1 was not evident with a variety of suffixes.

GENERAL DISCUSSION
In two experiments, we tested the hypothesis that morphological level
determines whether a suffixed word is decomposed for lexical access. While
this hypothesis was supported in Experiment 1, the hypothesis cannot, by
itself, account for the pattern of results in Experiment 2, where we examined
a larger set of affixes from each level. We believe that some Level 2 affixes,
such as -less, do require decomposition. However, other Level 2 affixes may
not.
How might a listener acquiring these morphemes learn to organize the
lexicon so that some Level 2 suffixes induce decomposition? Perhaps we are
sensitive to certain properties of some suffixes which suggest they be stored
in the lexicon like individual words. For example, the lack of phonological
 MORPHOLOGY AND LEXICAL ACCESS 331

TABLE 3
Mean Semantic Transparency Rating for Each
Suffix

Mean semantic
Suffix transparency rating

Level 1 -ary 5.380

-ation 6.674
-ian 6.510
-ion 6.154
-ity 5.964
-ory 5.743
-ous 6.264
Level 2 -er 6.600
-hood 6.535
-less 5.708
-ness 6.536
-ship 6.095

effects in words that include the suffix may be an initial cue to its lexical
status. Level 2 suffixes, as discussed earlier, never induce a phonological
change in roots. Second, the distributional properties of the suffix might be-
come apparent to the learner: Level 2 suffixes are able to appear further away
from the root than Level 1 suffixes. Third, the productivity of a suffix may
be a further cue. Some Level 2 suffixes like -ness are consistently more
productive than other Level 2 suffixes used in Experiment 2, and also more
productive than Level 1 suffixes like -ity, if we calculate productivity using
Baayen’s (1989) method (as cited in Lieber, 1992).
Another property of suffixed words that may give cues to lexical status
is the semantic transparency of these words (Marslen-Wilson et al., 1994,
for example, used this as a variable). In order to examine this factor, we
asked 20 Ohio State University students to judge the semantic relatedness
of each of the suffixed words (from the earlier experiments) to the corre-
sponding root word. Additional suffixed words were included so that ten
words with each suffix were rated; filler pairs were included as well. The
participants rated the semantic relatedness of the pairs on a scale from 1(un-
related) to 8 (very related); mean ratings for each suffix are in Table 3.
While the level 2- suffixed words (-ship, -ness, -less, -hood, -er) had gener-
ally higher ratings than those with Level 1 suffixes (-ous, -ory, -ity, -ion,
-ian, -ation, -ary) these two groups were not significantly different. Interest-
ingly, -less, which showed evidence of decomposition in Experiment 1, had
one of the lowest semantic transparency ratings. Thus, semantic transparency
itself cannot explain these results, but its interaction with some of the other
factors described above may help listeners determine how to treat specific
suffixes.
Phonological interaction, distribution, productivity, and semantic transpar-
332 VANNEST AND BOLAND

ency, may all encourage language learners to store some Level 2 suffixes,
like -less, separately from roots, and decompose them when they are encoun-
tered. Words with other suffixes, however, would be stored as whole-words
because these suffixes do not exhibit the motivating properties for decompo-
sition.
The lexical status of an affix depends on the properties of the affix itself,
and how it interacts with roots. Lexical phonology and morphology makes
an initial division, between affix types. The results described above suggest
that while Level 1 morphology is intact in the lexicon, Level 2 morphology
may in some cases be separated from its root. This model of the lexicon
requires the integration of several types of information about an affix for
effective storage and access.

REFERENCES
Bradley, D. 1979. Lexical representation of derivational relation. In M. Aronoff & M. Kean
(Eds.), Juncture. Saratoga, CA: Anna Libri.
Burani, C., & Caramazza, A. 1987. Representation and processing of derived words. Language
and Cognitive Processes, 2, 217–227.
Butterworth, B. 1983. Lexical representation. In B. Butterworth (Ed.), Language production,
Vol. 2. London: Academic Press.
Colé, P., Segui, J., & Taft, M. 1997. Words and morphemes as units for lexical access. Journal
of Memory and Language, 37, 312–330.
Ito, T., Sugioka, Y., & Hagiwara, H. 1996. Psychological status of rules in derivational mor-
phology: Evidence from Japanese nominal suffixation. Metropolitan Linguistics, 16, 10–
40.
Katamba, F. 1990. Morphology. New York: St. Clair.
Kiparsky, P. 1982. Lexical phonology and morphology. In I. Yang (Ed.), Linguistics in the
morning calm. Seoul: Hanshin.
Kucera, H., & Francis, W. 1967. Computational analysis of present day American English.
Providence: Brown University Press.
Leiber, R. 1992. Deconstructing morphology. Chicago: University of Chicago Press.
Marslen-Wilson, W., Tyler, L., Waksler, R., & Older, L. 1994. Morphology and meaning in
the mental lexicon. Psychological Review, 101(1), 3–33.
McQueen, J., & Cutler, A. 1998. Morphology in word recognition. In A. Zwicky and A.
Spencer (Eds.), The handbook of morphology. Oxford: Blackwell.
Mohanan, K. 1986. The theory of lexical phonology. Dordrecht: Reidel.
Taft, M. 1979. Recognition of affixed words and the frequency effect. Memory and Cognition
7, 263–272.
Taft, M., & Forster, K. 1975. Lexical storage and retrieval of prefixed words. Journal of Verbal
Learning and Verbal Behavior, 14, 638–647.
Wurm, L. 1997. Auditory processing of prefixed English words is both continuous and decom-
positional. Unpublished manuscript, SUNY Stony Brook.