Brain and Language 68, 218–224 (1999)
Article ID brln.1999.2085, available online at http://www.idealibrary.com on
Decomposition: To What Extent? The Case of Turkish
Ays¸e Gu¨rel
McGill University, Montre´al, Que´bec, Canada
It has been proposed that in agglutinative languages, lexical access of morphologically
complex words must involve decomposition rather than full listing (Frauenfelder
& Schreuder, 1992; Hankamer, 1989). We tested this proposal in Turkish
using a simple lexical decision task. Results show that multimorphemic words that
consist of frequent affixes are processed as fast as monomorphemic words. This
finding suggests that in languages with rich morphology, not all multimorphemic
words are accessed in a decomposed form. To the extent that morphemes are in
frequent use, they may induce whole-word rather than decompositional lexical access.
ã 1999 Academic Press
Key Words: Turkish; lexical access; morphological decomposition; whole-word
access.
INTRODUCTION
Lexical access and representation of complex words has widely been discussed
in the psycholinguistic literature. Existing models of lexical access
of multimorphemic words range from morphological decomposition (Taft &
Forster, 1975) to full-listing (Butterworth, 1983). While the decompositional
account assumes that a morphologically complex form is parsed into its constituent
morphemes prior to lexical access, the full-listing view maintains
that the morphological structure of a complex form has no independent representation,
suggesting that no parsing is involved in word recognition. There
are also hybrid models which include features of both decomposition and
full-listing models (Caramazza, Laudanna, & Romani, 1988; Frauenfelder &
This research was in part supported by an MCRI grant from the Social Sciences and Humanities
Research Council of Canada to Gonia Jarema, Eva Kehayia, and Gary Libben. I thank
all the members of the Mental Lexicon Project, especially Eva Kehayia, Chris Grindrod, Gerald
Rosenau, and Kyrana Tsapkini for their invaluable help in the design and the statistical analysis
of this experiment. I also thank the conference participants in Edmonton for their helpful
suggestions.
Address correspondence and reprint requests to Ays¸e Gu¨rel, Department of Linguistics,
McGill University, 1001 Sherbrooke Street West, Montre´al, Que´bec, H3A 1G5 Canada. Email:
agurel@po-box.mcgill.ca.
218
0093-934X/99 $30.00
Copyright ã 1999 by Academic Press
All rights of reproduction in any form reserved.
DECOMPOSITION IN TURKISH 219
Schreuder, 1992). Among the hybrid models, Augmented Addressed Morphology
(AAM) (Caramazza et al., 1988) postulates that an input can activate
simultaneously both a whole-word representation and constituent morphemes.
When a word is familiar to the subject, whole-word representation
is activated. When the word is novel, however, morpheme activation takes
place. The Morphological Race Model (MRM) proposed by Frauenfelder
and Schreuder (1992) also assumes the existence of two routes which are in
competition. However, in their view, even familiar words can be accessed
through decomposition depending on factors such as transparency and frequency.
According to this model, while the parsing route wins in the recognition
of transparent low-frequency words, the direct route wins in the recognition
of high-frequency opaque words.
The above-mentioned models have been tested using only a few languages
(for the most part English). However, it has recently been acknowledged
that cross-language studies, particularly those using languages with a rich
morphology (such as Turkish and Finnish), allow for experimental confirmation
of the proposed models of lexical access (Laine, Niemi, Koivuselka¨-
Sallinen, Ahlse´n, & Hyo¨na¨, 1995). It has been suggested that due to the
storage efficiency in highly inflected, agglutinative languages like Turkish,
lexical access of morphologically complex words must involve decomposition
rather than full-listing (Hankamer, 1989). A similar claim has come
from Frauenfelder and Schreuder (1992) who suggest that in Turkish, the
morphological parsing route rather than the direct access route must win the
race in the analysis of most complex words. However, they also predict that
depending on the frequency of the root-plus-affix combination, a word can
be recognized by the direct route.
In order to test the above predictions, we investigated word recognition
of morphologically simple and complex words in Turkish using a simple
lexical decision task measuring reaction time (RT). In this study, we also
looked at whether stems that include pseudomorphemes (i.e., units which
are homographic and homophonous with real morphemes) are more difficult
to process than simple nondecomposable stems. There are many Turkish
words (i.e., stems) that can induce ambiguous parsings. That is, assuming a
left-to-right parsing for Turkish (Hankamer, 1989), an input stem that is presented
in isolation can have several alternative parsings. For example, the
word dilim is ambiguous in the absence of any contextual information. Thus,
depending on the access strategy, it can have different readings: it means
‘‘slice,’’ if accessed as a whole or it means ‘‘my tongue,’’ if parsed as
dil 1 im (‘‘tongue’’ 1 first-person singular possessive). We assume that if
every morphemelike representation in a word is activated in word recognition,
then the processing will take longer and this, in turn, will lead to longer
RTs for these pseudomorphemic items.
Given these features of the language, we investigated the following questions:
(1) to what extent does lexical access of multimorphemic words in
220 AYS¸E GU¨ REL
Turkish involve morphological decomposition? and (2) will all possible
substrings of a word be parsed in word recognition? We hypothesize that
(1) if decomposition takes place during word recognition, we anticipate
longer RTs for all multimorphemic words and (2) if every possible morpheme
or morphemelike structure is activated in word recognition, we anticipate
longer RTs not only for multimorphemic but also for pseudomorphemic
items.
METHOD
Participants
Thirty-one native speakers of Turkish, with an average age of 26 years (range 18–36) and
an average number of 19 years of education (range 12–25) participated in this study.
Procedure
Subjects were given a simple visual lexical decision task run on a Power Macintosh using
PsyScope 1.1. Subjects saw a string of letters presented on the computer screen and were
asked to press the ‘‘yes’’ key if they recognized the item as a word of Turkish and the ‘‘no’’
key if they did not. The main experiment was run in a single test of 576 trials. A practice
trial of 10 stimuli preceded the main test. Each stimulus was preceded by a mask (######)
lasting 150 ms followed by a 200-ms interval until the target items appeared on the screen.
The item remained on the screen until the subject pressed ‘‘yes’’ or ‘‘no.’’
Stimuli
The stimuli comprised 130 words, 273 nonwords, and 173 fillers. The nonwords were constructed
by changing the first phoneme of the real-word stimuli. Nonwords included both
simple and inflected types, affixed with a legal suffix of Turkish. Fillers consisted of verbs,
adjectives, and adverbs, both simple and inflected.
The experimental stimuli comprised monomorphemic and multimorphemic items. Monomorphemic
items included nondecomposable and pseudomorphemic items.
Nondecomposable items (NDC) are words that cannot be decomposed in any way (e.g.,
pencere, ‘‘window’’).
Pseudomorphemic items were of three types. (1) pseudostem (PS): this group included
words such as dalga (‘‘wave’’), which consist of a meaningful stem plus a syllable which
has no meaning in Turkish. For example, dal (‘‘branch’’), the first syllable of the word, is
followed by a meaningless syllable, ga. (2) Pseudostem-stem (PSS): this category included
pseudomorphemic items which appear to contain two stems. For example, the first and second
syllables of the word bakkal (‘‘grocery’’) are also meaningful stems in Turkish: bak (‘‘look’’)
and kal (‘‘stay’’). (3) Pseudostem-affix (PSA): this category consisted of pseudostems which
have a stem followed by a legal suffix in Turkish. These were homographic and homophonous
with the possessive suffix in Turkish [e.g., dilim (‘‘slice’’), dil (‘‘tongue’’), and im (the firstperson
singular possessive)].
Multimorphemic items contained one- and two-suffix words inflected for case (ablative,
locative) and number [e.g., stem-ablative (S-AB): deprem-den (‘‘from the earthquake’’); stemlocative
(S-LOC): masa-da (‘‘on the table’’); stem-plural (S-PL): emir-ler (‘‘orders’’); stemplural-
ablative (S-PL-AB): oda-lar-dan (‘‘from the rooms’’); and stem-plural-locative
(S-PL-LOC): resim-ler-de (‘‘in the pictures’’)]. These morphemes differ in frequency in both
DECOMPOSITION IN TURKISH 221
FIG. 1. Mean RTs for one-suffix, two-suffix, and monomorphemic words.
written and spoken Turkish: the ablative suffix has the lowest frequency. This is followed by
the locative suffix. The plural morpheme has the highest frequency (Pierce, 1960).
Multimorphemic and monomorphemic items were matched only for stem frequency. In the
frequency count of Turkish (Pierce, 1960), the stem and the morpheme frequency are given
separately. Therefore, the surface frequency was not available. Nondecomposable items were
matched with one-suffix words for length. Length here can be defined in terms of either the
number of syllables or the number of letters. In either case, one-suffix words were matched
with nondecomposable items in length. Nondecomposable words of three to four syllables in
Turkish are generally of low frequency. Therefore matching these items with one-suffix words
for frequency was not possible. Pseudomorphemic words were inevitably shorter than nondecomposable
monomorphemic items.
Results and Discussion
The error rate across categories ranged from 0.5 to 3.4%. The RTs that
were 62 standard deviations above and below each group mean were
deemed outliers and thus were eliminated from further analysis. Errors and
outliers constituted 6.8% of total responses. The results show that as a whole,
multimorphemic words (687 ms) yield longer RTs than monomorphemic
words (615 ms) (p , .0001). A repeated-measures ANOVA revealed that
overall there is a significant difference among one-suffix, two-suffix, and
monomorphemic words [F(2, 30) 5 24.11, p , .0001]. Post hoc comparisons
using the Fisher’s LSD method showed that monomorphemic words
yield significantly shorter RTs than one-suffix (p,.05) and two suffix words
(p , .005) (see Fig. 1).
Comparisons of the monomorphemic group to each category of the onesuffix
group reveal an important effect of the suffix frequency. A repeatedmeasures
ANOVA for these comparisons revealed a significant difference
[F(3, 30) 5 9.8, p , .0001]. As shown in Fig. 2, a Fisher’s LSD post hoc
analysis showed that the words with the ablative suffix (S-AB) yield significantly
longer RTs than monomorphemic words (p , .005). The interesting
finding is that the difference between words with the locative case marker
(S-LOC), the second most frequent suffix, and monomorphemic words is
not statistically significant. Furthermore, there is no significant difference in
222 AYS¸E GU¨ REL
FIG. 2. Mean RTs for each category in the one-suffix group and monomorphemic words;
S-AB vs MONO, p , .005; S-LOC vs MONO, ns; S-PL vs MONO, ns.
RT for words with the frequent plural suffix (S-PL) and monomorphemic
words. That is, words with the plural morpheme are accessed as fast as monomorphemic
words.
The individual analysis within one-suffix words shows an effect of frequency
of suffixes in lexical access. A repeated-measures ANOVA revealed
a significant difference among the categories in this group [F(2, 30) 5 7.01,
p , .005]. As Fig. 3 shows, words with the ablative suffix, which has the
lowest frequency, yield the longest RT. This is followed by the words with
the locative suffix. The words which contained the plural morpheme, however,
yield significantly faster RTs. A Fisher’s LSD post hoc analysis showed
that words with the ablative suffix yield significantly longer RTs than the
words inflected with the plural morpheme (p , .05).
Within monomorphemic words, a repeated-measures ANOVA yielded a
significant difference [F(3, 30) 5 7.9, p , .0001]. However, a Fisher’s LSD
post hoc analysis did not show any significant difference between the nonde-
FIG. 3. Mean RTs for each category in the one-suffix group; S-AB vs S-LOC, ns; S-AB
vs S-PL, p , .05; S-LOC vs S-PL, ns.
DECOMPOSITION IN TURKISH 223
FIG. 4. Mean RTs for nondecomposable (NDC) and pseudomorphemic words (PS, PSS,
and PSA).
composable group and any of the categories in the pseudomorphemic group
(Fig. 4). In this analysis, the only difference that was close to reaching significance
was between NDC and PS (p 5 .07). However, it was not statistically
significant.
As mentioned earlier, in a left-to-right parsing in Turkish, there may be
several alternative parsings in the lexical access of certain words and this is
expected to cause a delay in processing, and, in turn, longer RTs. However,
it seems that not every ‘‘possible’’ substring of a word is parsed. If this had
been the case, we would have obtained longer RTs for pseudomorphemic
items.
A comparison among multimorphemic words reveals a significant interaction
between the number of suffixes and the speed of word recognition. We
find that overall, two-suffix words yield significantly longer RTs than onesuffix
words (p , .05). A repeated-measures ANOVA for the multimorphemic
group yielded a significant difference [F(4, 30) 5 6.03, p , .0005].
However, as can be seen in Fig. 5, the two-suffix group (both S-PL-AB and
S-PL-LOC) are accessed as fast as one-suffix words (S-AB and S-LOC). A
Fisher’s LSD post hoc analysis revealed that the only significant difference
is between one-suffix words with the plural morpheme (S-PL) and two-suffix
words with the plural plus ablative morpheme (S-PL-AB) (p , .05). This
suggests that the plural morpheme in two-suffix words does not contribute
any extra processing load to the access of these words; it is accessed as part
of the stem.
CONCLUSION
This study has shown that not all multimorphemic words are accessed in
a decomposed form in Turkish. Words with frequent suffixes seem to be
accessed through a whole-word access procedure. Depending on the frequency
of the suffix, a word can be accessed via the direct access route or
the parsing route. The higher frequency of a suffix makes processing easier
and faster. It appears, then, that in a highly inflected language like Turkish,
224 AYS¸E GU¨ REL
FIG. 5. Mean RTs for multimorphemic words; S-AB vs. S-PL-AB, ns; S-AB vs S-PLLOC,
ns; S-LOC vs S-PL-AB, ns; S-LOC vs S-PL-LOC, ns; S-PL vs S-PL-AB, p , .05;
S-PL vs S-PL-LOC, ns.
the recognition of morphologically complex words is not as ‘‘costly’’ as it
may be in languages with little inflection (such as English), at least for words
that contain frequent suffixes. In other words, lexical access in languages
with complex morphology involves the direct route where possible in order
to save time in processing. This suggests that the parser in Turkish needs to
be much more effective than the parser in English in order to handle the
complexity of word forms during lexical access.
REFERENCES
Butterworth, B. 1983. Lexical representation. In B. Butterworth (Ed.), Language production
(Vol. 2). New York: Academic Press.
Caramazza, A., Laudanna, A., & Romani, C. 1988. Lexical access and inflectional morphology.
Cognition, 28, 297–332.
Frauenfelder, U. H., & Schreuder, R. 1992. Constraining psycholinguistic models of morphological
processing and representation: The role of productivity. In G. E. Booij & J. van
Marle (Eds.), Yearbook of morphology 1991. Dordrecht: Kluwer.
Hankamer, J. 1989. Morphological parsing and the lexicon. In W. Marslen-Wilson (Ed.), Lexical
representation and process. Cambridge, MA: The MIT Press.
Laine, M., Niemi, J., Koivuselka¨-Sallinen, P., & Hyo¨na¨, J. 1995. Morphological processing
of polymorphemic nouns in a highly inflected language. Cognitive Neuropsychology,
12(5), 457–502.
Pierce, J. 1960. A frequency count of Turkish words. Ankara: Milli Egitim Mudurlugu.
Taft, M., & Forster, K. I. 1975. Lexical storage and retrieval of prefixed words. Journal of
Verbal Learning and Verbal Behavior, 14, 638–647.
Hiç yorum yok:
Yorum Gönder