The acquisition of a native language in natural informal settings takes place within a relatively short period in childhood. In contrast, late learning of a new language provides a different challenge for an individual, because the conditions that promote holistic language acquisition often are no longer present or possible. The late language learner probably relies more on linguistic awareness and cognitive processes such as analytical thinking, reasoning, as well as various learning strategies (for a review, see, Harley & Wang, 1997). However, the notion of a critical period for language acquisition should not be overly emphasized because older children and adults are also able to learn new languages and can achieve high levels of proficiency. However, it seems that the acquisition of certain aspects of language such as phonology and syntax are more difficult for late learners (Johnson & Newport, 1989; Mayo, Florentine, & Buus, 1997; Yeni-Komshian, Flege, & Liu, 2000). It is also possible that the acquisition of receptive language skills in L2 differs from the acquisition of language production skills.

Vocabulary acquisition is a key element in foreign language learning. Part and parcel of word learning is the acquisition of the morphological structure of words. Morphemes, the smallest meaning bearing units in a language, carry crucial semantic-syntactic information. How do late learners acquire morphologically complex words¹

It is generally presumed that we have two alternative ways to recognize morphologically complex words. First, word forms can be accessed and recognized via the mental representations of their constituent morphemes. This is the so-called decomposition route, which is slower and more error-prone but spares storage space in long-term memory (e.g., Taft, 1994; Taft & Forster, 1975). Second, morphologically complex words can be accessed and recognized via mental representations that correspond to whole word forms. This is the so-called full-form route, which is faster but requires more storage space (e.g., Butterworth, 1983). More recent models of morphological processing maneuver between these two opposing principles and emphasize the role of psycholinguistic factors such as affix properties, word formation patterns, and lexical frequency in the choice of processing routes for a morphologically complex word (Caramazza, Laudanna, & Romani, 1988; Frauenfelder & Schreuder, 1992; Niemi, Laine, & Tuominen, 1994; Schreuder & Baayen, 1995; for a review, see Bertram, 2000).

The lexical decision task is one of the main experimental methods in the study of lexical processing. In this task the participants have to decide whether the presented letter or phoneme string is an existing word or not. The speed and accuracy of this decision varies systematically between lexical items and is considered to reflect the way these words are represented and accessed in the mental lexicon. The employment of the two morphological processing routes (decomposition vs. full-form access) can be studied with lexical decision task by, for example, contrasting inflected words and monomorphemic words (e.g., Lehtonen & Laine, 2003; Lehtonen, Niska, Wande, Niemi, & Laine, 2006). The two word groups should be carefully matched for all major nonmorphological factors that are assumed to have an effect on the speed of word recognition. If the inflected words exhibit a processing load, that is, elicit longer reaction times and more errors than the matched monomorphemic control words, it is generally concluded that the inflected words have undergone morphological decomposition during recognition. If, in contrast, the reaction times and error rates do not differ between the two word types, it is generally assumed that recognition of inflected words has occurred via full-form representations. However, one should note that this lexical decision design does not provide the means for refuting the possibility that morphological decomposition is always applied for inflected words but that its speed and accuracy can vary considerably. Moreover, as with any chronometric measure, the evidence on the underlying mental architecture is, of course, indirect. Therefore, it is important to note that the experimental evidence for the two hypothesized morphological processing routes is not based on the above-mentioned lexical decision tasks only. In addition, other experimental designs with lexical decision have been used, such as manipulating stem and surface frequency separately or studying stem priming. The processing cost for the inflected words has also been observed with other techniques tapping the perception and production of single words both with normal participants and aphasic patients (see, e.g., Ahlsén, 1994; Bertram, Schreuder, & Baayen, 2000; Clahsen, Hadler, & Weyerts, 2004; Hyönä, Laine, & Niemi, 1995; Laine, Niemi, Koivuselkä-Sallinen, Ahlsén, & Hyönä, 1994).

Word frequency is one of the main factors that affects the speed of word recognition. High-frequency words are in general processed faster than low-frequency words, and this has been observed both with morphologically simple and complex words (e.g., Gardner, Rothkopf, Lapan, & Lafferty, 1987; Taft, 1979). It is also assumed that word frequency influences the selection of the processing route. Pinker (1999, p. 138) suggested that high-frequency regular inflected words may be coded into long-term memory as whole units. Thus, high-frequency inflected words would be accessed and recognized via the faster full-form route. Evidence for this argument was found by Alegre and Gordon (1999), who studied visual lexical decision performance in English-speaking individuals. They found that full-form representations start to develop for morphologically complex words when surface frequency of the word is higher than 6 occurrences per million. Lehtonen et al. (2006) obtained similar results in Swedish: Native monolingual speakers showed a processing cost indicative of the use of the decomposition route with low-frequency inflected nouns that had a surface frequency range of 0.04–4.09 per million. Conversely, medium-frequency inflected words with a surface frequency range of 8.8–38.9 per million appeared to be processed via the full-form route. The authors conclude that their results are in line with Alegre and Gordon (1999), although the exact frequency value at which full-form representations start to develop for native speakers of Swedish remains uncertain.

In addition to lexical frequency, the morphological richness of a language is a potentially important factor in the organization of the mental lexicon (Hankamer, 1989). Finnish, for example, is a non-Indo-European language that uses morphology to a far greater extent than most of the other languages that have been explored in psycholinguistic research. A Finnish noun can have as many as 2000 inflectional forms (Karlsson, 1983), whereas Swedish, a Germanic language, includes only eight possible forms of each noun. Given the morphological richness of the Finnish language, it is not surprising that Finnish inflected nouns are apparently processed via the decomposition route more often than in morphologically limited languages like Swedish and English. Lehtonen and Laine (2003) found that Finnish monolinguals employed the decomposition route even for recognition of medium-frequency inflected nouns that had an average surface frequency range of 8.77–36.4 per million.

Along with the psycholinguistic factors, the language background of the participants can affect the word recognition process. Word token counts from lexical corpora provide a general measure of the frequency of lexical items in a language, but individuals' actual exposures to a language vary considerably. In this respect, late learners of a language form an interesting group. Due to their late learning they have received less language exposure compared to native speakers. However, it is impossible to determine each individual's exposure rates to specific word forms. Thus, we make the assumption that the exposure rates of both native speakers and late learners correlate highly with corpus frequencies, but for the late learners the exposure rates are substantially lower for all words in the L2. Thus, the relative frequency difference between high- and low-frequency word forms should be approximately the same for native speakers and late learners.

Lexical–morphological processing in bilinguals can differ from that of monolinguals. First, bilinguals are often slower in speeded verbal tasks even when one tests early simultaneous bilinguals whose proficiency levels are high in both languages (Grosjean, 1998; Lehtonen & Laine, 2003; Portin & Laine, 2001; Ransdell & Fischler, 1987). Second, it appears that especially in a morphologically rich language, bilinguals employ morphological decomposition more than monolinguals. Lehtonen and Laine (2003) found that Finnish–Swedish early bilinguals employed the decomposition route with Finnish inflected words throughout the frequency range (low, medium, and high), while Finnish monolinguals decomposed only low- and medium-frequency inflected words. However, this difference in morphological processing between mono- and bilinguals was attenuated in a morphologically limited language, Swedish. According to the study by Lehtonen et al. (2006), Finnish–Swedish early bilinguals decomposed Swedish low- and even some medium-frequency inflected nouns, whereas Swedish monolinguals used decomposition only with low-frequency inflected words. The authors assume that the rather restricted morphological structure of the Swedish language promotes the development of full-form representations at lower frequencies than in Finnish. Conversely, the authors suggest that due to less exposure to the Swedish word forms (lower subjective frequencies), early bilinguals used the decomposition route at least for part of the medium-frequency inflected words as well.

The way an individual processes morphologically complex words may thus emerge from the interplay among language background of the individual, word frequency, and the morphological structure of the language. Although there is some evidence for this from early simultaneous bilinguals, data on morphological processing in late learners of a foreign language is very limited. Accordingly, this study sought to investigate how late learners of Swedish recognize inflected words in that language. We examined whether late bilinguals process Swedish inflected words in the same way as the native speakers did in a previous study by Lehtonen et al. (2006), or whether late bilinguals decompose Swedish inflected words to a higher degree. We expected to see a processing cost for low-frequency and medium-frequency inflected words, indicative of morphological decomposition on these items. Given the late bilinguals' lower overall exposure to Swedish it might even be possible that they decompose Swedish inflected words regardless of frequency.

Concerning late learners of Swedish, those with Finnish as their native tongue are a particularly interesting group. A theoretically interesting situation is created in a native speaker of a morphologically complex language (Finnish) who has later learned a morphologically limited language (Swedish). However, although the Swedish morphology is limited in comparison with Finnish morphology, Swedish does include some inflectional aspects that do not exist in Finnish. In addition, there is also relatively little vocabulary overlap between Swedish and Finnish.

We carried out the same experiment with two Finnish participant groups. All participants had been monolingual speakers of Finnish up to school age. Finland is officially a bilingual country, and it is obligatory to study Swedish as a foreign language in comprehensive school as well as in high school. The learning of Swedish often begins at the age of 13, but it is also possible to start earlier at the ages of 11 or 9, and lasts 6, 8, or 10 years, respectively. At the time of the present study the participants were graduate students of Nordic philology²

EXPERIMENT 1

Method

Materials

The target words for the visual lexical decision task were selected from an unpublished Göteborgs-Posten lexical database with 24.2 million word tokens by using a computerized search program (Laine & Virtanen, 1999). Altogether, six target word lists were included, with two lists from each frequency range: low, medium, and high. Each list pair included 20 bimorphemic inflected Swedish nouns contrasted with 20 monomorphemic Swedish nouns. The contrasted word lists were matched at each level of frequency for the following factors: average word length in letters,⁴

In visual processing of single-word forms, the length in letters is an essential factor: Fovea provides details for about a 2° visual angle, which corresponds to the word length of approximately eight letters or less. Therefore, short words (length less than eight letters) can be recognized by one fixation, whereas long words need two fixations (for a survey see, e.g., Rayner, 1998).

surface, lemma and bigram frequency, morphological family size, and abstractness/concreteness estimates (see Table 1 for the properties of the target words and Appendix A for the target word lists). The inflectional form used was the fully transparent and productive definite singular form that has two alternative surface forms depending on the gender of the stem (e.g., bok + en = “book” + definite suffix article, “the book”; hus + et = “house” + definite suffix article, “the house”).⁵

In the Göteborgs–Posten lexical database about 70% of the nouns are of uter gender (-en) and about 30% are of neuter gender (-et).

Note that this suffix is very common in Swedish, but the Finnish language does not include either gender or definiteness of nouns. These particular target words were chosen because they were exactly the same as the ones used in the study by Lehtonen et al. (2006). However, the present experiment was longer than the previous one because it included derived words as fillers to achieve more variation in the suffixes. Second, the nonwords were also partly different. Existing suffixes were added to the nonwords to have a similar proportion of different suffixes both in real words and in nonwords.

In addition to the 120 target words, the task included 496 filler words (188 real words, 308 nonwords), yielding altogether 616 stimuli. Of the nonwords, 189 were “monomorphemic” whereas 119 were “bimorphemic,” that is, they had endings that were identical to real suffixes. The nonwords were created by changing between one and three letters of the stems of the real words. The nonwords followed the phonotactic rules of the Swedish language.

Participants

Twenty-five university students (24 women)⁶

The fact that almost all participants are women reflects the gender distribution in language studies. According to the unpublished comparisons between male and female participants' performances in our earlier lexical decision studies, the gender of the participants does not appear to have an influence on the morphological processing.

representing senior students participated in Experiment 1. However, the final number of participants entering the result analysis was 21 (20 women) because 2 participants had to be removed due to a technical failure and a further 2 because of their high overall error rates (>20%) in the experimental task. The remaining 21 participants had studied Nordic philology for 2.5 to 7 years before the experiment (mean = 4.02, SD = 1.16). Age varied between 22 and 31 years (mean = 24.38, SD = 2.01). The participants were native speakers of Finnish and monolingual in the sense that they had acquired only the Finnish language at home before school age (before the age of 7). The language at school had been Finnish for all participants. Sixteen participants (76.2%) had begun studying Swedish at school at the age of 13, while five participants (23.8%) had begun to study it at the ages of 11 or 9. Thus, all the participants had studied Swedish at school for at least 6 years (mean = 6.81, SD = 1.54) before entering the university. Given that their major was Nordic philology at the university, all participants had continued their Swedish studies up to the time of testing.

The language skills of the participants were assessed by a questionnaire that tapped talking, writing, listening, and reading comprehension in everyday context in Finnish and in Swedish by using a 4-point scale (1 = deficient, 2 = satisfactory, 3 = good, 4 = excellent). The overall average for language skills in Finnish was 3.90 (SD = 0.24) and in Swedish 3.36 (SD = 0.35). The language skill estimates are presented in detail in Table 2. The Finnish skills were estimated to be significantly better than the Swedish skills in talking, t(20)=8.00, p<.001, listening comprehension, t(20)=3.16, p=.005, reading comprehension, t(20)=3.51, p=.002, and writing, t(20)=7.07, p<.001.

The participants were also asked to estimate the relative proportion of the different languages they had used during the last years. The mean value for Finnish was 79% and for Swedish 14%.⁷

The use of other languages was also included in the question, but some participants mentioned only the use of Finnish and Swedish. Therefore, we report only these two.

We will return to these results when describing Experiment 2.

Procedure

We used a standard visual lexical decision task in which the participants were instructed to decide as fast and as accurately as possible whether a letter string shown at the center of the computer screen was a real Swedish word or not. The experiment was run by using a reaction time program (SuperLab Experimental Laboratory Software, Version 2.0, Cedrus Corporation) that recorded the participants' reaction times (milliseconds) and the correctness of their responses. The reaction time was measured from the letter string onset to the pressing of the reaction time key. The setup for the visual lexical decision task was as follows: a centrally presented fixation point (asterisk) preceded each stimulus to alert the participant of the forthcoming item. After 500 ms the stimulus appeared at the center of the screen. The stimulus was visible for a maximum of 2 s or until the participant pressed the reaction time key. After giving their response (yes to a word, no to a nonword) the participants pushed a third button with their nondominant hand to make the next stimulus appear. The participants began with a practice session of 30 items to familiarize themselves with the task. The words and nonwords in the practice session were not used in the experiment itself. The actual task was divided in three sections. Section one included 215 items (42 target words: 7 monomorphemic and 7 inflected words from each frequency range, 67 fillers, and 106 nonwords). Section two yielded 209 items (42 target words: 7 monomorphemic and 7 inflected words from each frequency range, 63 fillers, and 104 nonwords). Section three included 192 items (36 target words: 6 monomorphemic and 6 inflected words from each frequency range, 58 fillers, and 98 nonwords). Each participant performed all the three blocks, with a short pause between the sections. For counterbalancing purposes, there were six different presentation orders of the sections (1-2-3, 2-3-1, 3-1-2, 1-3-2, 2-1-3, 3-2-1), that is, every seventh participant received the same presentation order. The presentation order of the individual items within each section was randomized for each participant. It took about 45 min to complete the entire experiment. The participants were tested individually in a closed room. All communication during the experiment was in Finnish.

Results

All incorrect responses and reaction times that differed more than 3 SDs from the individual mean latency per stimulus category were removed from the reaction time data. One inflected target word from the low-frequency list (kulmen) was removed because the word is ambiguous and can be considered both as a monomorphemic word and as an inflected word form. Moreover, one inflected target word from the medium-frequency list (loket) was also removed as a concordance analysis with the Göteborgs–Posten corpus showed that most of the occurrences of this word form referred to a local celebrity. It thus turned out to be a low-frequency word for our Finnish participants. The average reaction times and error percentages per target word condition can be found in Table 3.

Two-way analyses of variance (Frequency × Morphological Structure) were performed on reaction times and error rates. The analysis of variance (ANOVA) for reaction times revealed a significant main effect for frequency both in the by-participant, F1 (2, 40) = 86.0, p<.001, and in the by-item analysis, F2 (2, 112) = 100.7, p<.001. The main effect for morphological structure was significant in the by-participant analysis, F1 (1, 20) = 10.4, p= .004, and in the by-item analysis, F2 (1, 112) = 4.14, p= .044. The main effects indicate that high-frequency words were processed faster than low-frequency words, and that the inflected words yielded longer reaction times than the monomorphemic words. There was also a significant interaction between frequency and morphology, F1 (2, 40) = 16.8, p<.001; F2 (2, 112) = 5.80, p= .004, indicating that the reaction time difference between the inflected and monomorphemic words varied between the frequency levels.

The ANOVA for error rates showed a significant main effect for frequency, F1 (2, 40) = 152.7, p<.001; F2 (2, 112) = 53.6, p<.001. This indicates that lower frequency items elicited more errors than higher frequency items. The main effect for morphological structure was also significant, F1 (1, 20) = 68.7, p<.001; F2 (1, 112) = 5.09, p= .026. This is due to the fact that the inflected words elicited more errors than the monomorphemic ones. In addition, a significant interaction between frequency and morphology was found, F1 (2, 40) = 106.3, p<.001; F2 (2, 112) = 5.74, p= .004. This is because the difference in error rates between inflected and monomorphemic items was far greater in the low-frequency condition than in the medium- or high-frequency condition.

The observed interactions prompted the comparisons of the reaction times and error rates between inflected and monomorphemic words in the three frequency ranges separately. Two-tailed paired samples t-tests for participants and two-tailed independent samples t-tests for items were used for this purpose. In the low-frequency range, the inflected words elicited significantly longer reaction times than the monomorphemic words both in the by-participant, t1(20)=4.03, p=.001, and in the by-item analysis, t2(37)=2.43, p= .020. Here, the inflected items also received significantly more errors than the monomorphemic items, t1(20)=10.6, p<.001; t2(37)=2.35, p=.024. In the medium-frequency range the reaction time difference between inflected and monomorphemic items did not reach significance, t1(20)=0.09, p= .931; t2(37)=0.30, p=.765. There were no differences between error rates by item types, t1(20)=0.21, p=.839; t2(37)=0.22, p=.825. In the high-frequency range, it was surprising that the inflected items were recognized significantly faster than the monomorphemic words in the by-participant analysis, t1(20)=2.41, p=.026, but the difference was not significant in the by-item analysis, t2(38)=1.04, p=.306. The error rates in these two analyses did not differ from each other, t1(20)=1.37, p=.186; t2(38)=1.18, p=.249.

Discussion

The results of the experiment can be summarized as follows: first, there was a significant difference in reaction times and error rates between inflected and monomorphemic words in the low-frequency list, suggesting morphological decomposition for the low-frequency inflected words. Second, there was no difference in reaction times or error rates with regard to the medium-frequency inflected and monomorphemic words. The lack of a processing cost would indicate full-form processing of the medium-frequency inflected words. Third, in the high-frequency list, we observed a difference in reaction times that unexpectedly favored the inflected words. However, this finding could be spurious as it did not reach significance in the by-item analysis and was not reflected in the error rates. A post hoc item-by-item analysis revealed no particular features that would separate the word forms with the fastest as opposed to the slowest reaction times from the rest of the high-frequency stimuli.⁸

The high-frequency target words were examined individually but no outliers were found. With regard to the reaction times, the major part of the monomorphemic and half of the inflected words were recognized in 600–700 ms. Yet, four monomorphemic words (parti, söder, område, medlem) and three inflected words (perioden, världen, bollen) were recognized slightly slower, in 709–730 ms. Moreover, we found three monomorphemic words (vatten, teater, sommar) and seven inflected words (natten, huset, skolan, målet, klockan, livet, mötet) that were recognized faster, in 555–596 ms. There are no apparent features that would differentiate the targets with fastest versus slowest reaction times from the rest of the high-frequency stimuli.

One could also speculate that the inflected word forms with the fastest reaction times appeared particularly often in the Swedish school books that our participants used but it is impossible to verify this. In addition, we did not observe a similar result in our previous studies. As a consequence, the data would support the view that the high-frequency inflected word forms enjoyed full-form processing.

In sum, the overall pattern of the results appeared to be largely similar to those obtained with native speakers in the study by Lehtonen et al. (2006). However, as our participants were advanced students in Nordic philology, it is interesting to see whether another group of late bilinguals with less exposure to Swedish, namely junior university students with the same major, showed a similar performance pattern. This was the objective of Experiment 2.

EXPERIMENT 2

Method

Materials and procedure

The stimulus materials and the experimental procedure were the same as in Experiment 1.

Participants

Twenty-one university students (19 women) who were junior students participated in this experiment. Only 16 participants (15 women) were included in the final analysis. One participant was removed due to a technical failure and four because of their high overall error rates (>20%) in the experimental task. The remaining 16 participants had studied Nordic philology as their major for 3 months before the experiment. The junior students had thus significantly less language studies in Swedish at the university level than the senior students, t(35)=13.0, p<.001. The junior students' age varied between 18 and 26 years (mean = 20.38, SD = 1.96), yielding a statistically significant age difference between the junior and senior students, t(35)=6.06, p<.001.

The junior students had the same kind of monolingual Finnish background as the senior students in Experiment 1. Fourteen participants (87.5%) had begun studying Swedish at the age of 13, and 2 participants (12.5%) had begun their studies at the ages of 11 or 9. Mean length of Swedish studies was 6.50 years (SD = 1.32). There was no statistically significant difference between the junior and senior students when Swedish studies at school were compared, t(35)=0.65, p=.523.

The junior students also estimated their own language skills in Finnish and in Swedish by using the 4-point scale (1 = deficient, 2 = satisfactory, 3 = good, 4 = excellent). The mean estimate for Finnish was 3.86 (SD = 0.27) and for Swedish 2.73 (SD = 0.37). The language skill estimates are presented in detail in Table 4. They estimated their Finnish skills to be significantly better than their Swedish skills in talking, t(15)=11.2, p<.001, listening comprehension, t(15)=5.51, p<.001, reading comprehension, t(15)=7.00, p<.001, and writing, t(15)=6.33, p<.001.

There was no statistically significant difference between the junior and senior students in their Finnish skill estimates: talking, t(35)=0.35, p= .727; listening comprehension, t(35)=0.19, p= .848; reading comprehension, t(35)=0.84, p=.407; writing, t(35)=0.84, p=.406. As expected, the Swedish skill estimates of the two groups differed from each other as the senior students rated their Swedish skills as significantly better compared with the junior students: talking, t(35)=5.15, p<.001; listening comprehension, t(35)=3.34, p=.002; reading comprehension, t(35)=3.81, p<.001; writing, t(35)=2.70, p=.011.

In addition, the junior students assessed the relative proportion of the different languages they had used during the last years. The mean value for Finnish was 89% and for Swedish 7%. There was a significant difference between the junior and the senior students on these estimates, as the junior students had used more Finnish, t(35)=2.13, p=.040, and less Swedish, t(35)=2.12, p=.041.

Results

The statistical analyses were performed in the same way as for Experiment 1. The average by-participant reaction times and error percentages per target word condition can be found in Table 5.

The ANOVA for reaction times revealed significant main effects both in the by-participant and the by-item analysis for frequency, F1(2, 30)=73.3, p<.001; F2(2, 112)=72.2, p<.001, and for morphological structure, F1(1, 15)=7.24, p= .017; F2(1, 112)=8.51, p=.004. These main effects indicate that higher frequency words were processed faster than lower frequency words, and that morphological structure affected reaction times by slowing reaction times for inflected words. There was also a significant interaction between frequency and morphological structure, F1(2, 30)=6.39, p=.015; F2(2, 112)=5.77, p=.004. This shows that the reaction time difference between the inflected words and the monomorphemic words varied between the frequency levels.

The ANOVA for error rates yielded significant main effects both for frequency, F1(2, 30)=218.6, p<.001; F2(2, 112)=87.9, p<.001, and for morphological structure, F1(1, 15)=48.7, p<.001; F2(1, 112)=9.94, p= .002. The main effects indicate that the lower frequency items elicited more errors than higher frequency items, and that the inflected words yielded more errors than the monomorphemic ones. There was also a significant interaction between frequency and morphological structure, F1(2, 30)=68.2, p<.001; F2(2, 112)=10.5, p<.001, confirming that the difference in error rates between the inflected and the monomorphemic items varied between the frequency levels.

The statistically significant morphology by frequency interactions prompted a closer analysis where each frequency range was analyzed separately by using two-tailed paired samples t tests for participants and two-tailed independent samples t tests for items. In the low-frequency range, the inflected words elicited significantly longer reaction times than the monomorphemic words both in the by-participant, t1(15)=2.78, p=.014, and in the by-item analysis, t2(37)=2.72, p=.010. In this frequency range the inflected items also had significantly more errors than the monomorphemic ones, t1(15)=8.53, p<.001; t2(37)=3.23, p=.003. In the medium-frequency range the reaction time difference between the inflected and monomorphemic items did not reach significance in the by-participant analysis, t1(15)=1.54, p=.144, or in the by-item analysis, t2(37)=1.89, p=.067. The error rates did not differ between the item types, t1(15)=0.57, p=.576; t2(37)=0.75, p=.459. In the high-frequency range there was no significant difference in reaction times between the inflected and monomorphemic words, t1(15)=1.15, p=.270; t2(38)=0.48, p=.638. Regarding error rates, there were again no differences between the two item types, t1(15)=1.29, p=.216; t2(38)=1.13, p=.268.

Discussion

With regard to the three frequency ranges that we studied, the low-frequency words revealed a significant difference between the inflected and monomorphemic words both in reaction times and error rates. This result suggests morphological decomposition for low-frequency inflected words. In the medium- and high-frequency ranges, there were no differences between the inflected and monomorphemic words in reaction times or error rates. This indicates that the junior students applied full-form processing for the medium- and high-frequency inflected words. In sum, the results are mostly consistent with those of Experiment 1 and Lehtonen et al. (2006).

APPENDIX A