81 Sentence Processing of Nominative-Genitive Conversion in Japanese by Turkish Speaking Learners and Native Speakers * Barış Kahraman ÇANAKKALE ONSEKİZ MART UNIVERSITY 1. Introduction In the psycholinguistic literature, it has been well-documented that subject relative clauses (SRs), like “the professor that saw the student...” , are easier to process than object relative clauses (ORs), like “the professor that the student saw...” in first language (L1) and second language (L2) (L1: e.g., Dutch: Frazier 1987; English: Staub 2010; French: Cohen and Mehler 1996; German: Schriefers, Friederici and Kühn 1995; Japanese: Ueno and Garnsey 2008; Korean: Kwon, Lee, Gordon, Kluender and Polinsky 2010; Spanish: Betancort, Carreiras and Sturt 2009. L2: e.g., English: Hashimoto 2007, Kashiwagi; Japanese: Mitsugi, MacWhinney and Shirai 2010; Korean: O ’ Grady, Lee and Choo 2003). In order to explain this processing asymmetry, many competing hypotheses, such as linear distance hypothesis (Gibson 1998), structural distance hypothesis (O’Grady 1997, O ’ Grady et al. 2003), and frequency of occurrence hypothesis (Reali and Christiansen 2007) have been put forward. 1 These are universal hypotheses, and their validity is being tested in various languages. In Japanese, Sato, Kahraman, Ono and Sakai (2009) argued that, in addition to universal factors, language-specific factors should also be taken into 82 consideration, and so proposed case-driven expectation hypothesis (CDEH) for Japanese. CDEH assumes that Japanese relative clauses (RCs) are processed more easily if there is an early expectation for another noun before the RC-verb is encountered. In the case of L2 Japanese, the number of online RC processing studies is comparatively few, and it remains unknown to what extent universal factors and/or language-specific factors may affect the RC processing. The present study aims to explore the possible effects of universal and language-specific factors on RC processing, and particularly attempts to distinguish among linear distance hypothesis, frequency of occurrence hypothesis, structural distance hypothesis , and case-driven expectation hypothesis in L2 Japanese. In order to do this, the nominative-genitive conversion was utilized to RCs, and four self-paced reading experiments were conducted with higher-intermediate Turkish speaking learners and native speakers of Japanese. The results suggested that case-driven expectation hypothesis can account for the observed processing patterns of Turkish speaking learners in Japanese. The organization of the paper is as follows. In Section 2, major hypotheses will be explained. Next, in Section 3, the properties of nominative-genitive conversion and predictions of each hypothesis will be briefly explained. In Section 4, the details and the results of self-paced reading experiments will be given. In Section 5, it will be argued that case-driven expectation hypothesis can account for the observed processing patterns of learners, whereas universal hypotheses cannot. Section 6 concludes the paper. 2. Major Hypotheses In this section, linear distance, structural distance, frequency of occurrence 83 and case-driven expectation hypotheses will be explained. Linear Distance Hypothesis (LDH) : Basically, LDH assumes that the number of the discourse elements, namely linear distance between the RC-head and its gap position, determines the processing difficulty of RCs (Gibson 1998). As illustrated below, LDH predicts different results in English and Japanese. (1a) SRs: The professor i that ___ i saw the student... (1b) ORs: The professor i that the student ___ i saw... (2a) SRs: ___ i Gakusei-o mita kyooju i ... Student-ACC saw professor ‘The professor that saw the student...’ (2b) ORs: Gakusei-ga ___ i mita kyooju i ... Student-NOM saw professor ‘The professor that the student saw...’ In English, the number of the words between the RC-head (professor) and the gap is fewer in SRs than ORs. In other words, the linear distance between the RC-head and the gap position is less in SRs. Since the memory load is heavier for ORs than for that of SRs, LDH predicts that SRs are easier to process than ORs. In Japanese, unlike English, the linear distance between the RC-head and the gap is less in ORs than in SRs. Therefore, LDH predicts that ORs should be easier to process than SRs in Japanese. Structural Distance Hypothesis (SDH) : SDH assumes that the number of 84 syntactic nodes between the RC-head and the gap is the main source of the processing difficulty of RCs (O’Grady 1997 , O ’ Grady et al. 2003). According to SDH, the number of syntactic nodes helps the parser to determine the computational complexity of the RCs. SRs t he professor i that ___ i saw the student ORs t he professor i that the student saw ___ i Fig. 1 Structural distance between the RC-head and the gap in SRs/ORs. Fig. 1 shows that the number of syntactic nodes is fewer in SRs than in ORs. In other words, the structural distance between the RC-head and the gap is shorter in SRs. Since the computational complexity can be determined more easily in SRs than in ORs, SDH predicts that SRs should be easier to process than ORs. The position of the RC-head differs in the head-initial languages, like English, and the head-final languages like, Japanese. However, the number of syntactic nodes is always fewer in SRs than in ORs irrespective of typological differences (O’Grady 1997) . Therefore, SDH universally predicts that SRs should always be easier to process than ORs. 85 Frequency of Occurrence Hypothesis (FOH) : Unlike distance based accounts, FOH basically assumes that frequent structures are processed more easily than less frequent structures, because people are more familiar with frequent structures (e.g., Reali and Christiansen 2007). In English, Reali and Christiansen (2007) conducted a corpus analysis and found that SRs are more frequent than ORs. Moreover, they analyzed the noun types within RCs, and found that SRs occur more frequently with proper nouns, whereas ORs occur more frequently with pronouns. Based on these distributional patterns, Reali and Christiansen manipulated the noun types within RCs and conducted a series of self-paced reading experiments. They found that ORs were easier to process than when pronouns were used within RCs. This suggests that frequency would be one of the most important processing constraints. In Japanese, Sato (2011) conducted a corpus analysis and reported that distributions of SRs and ORs do not differ in the corpus. Therefore, according to FOH, there should not be significant processing difference between SRs and ORs in Japanese. However, previous studies have consistently shown that SRs are easier to process than ORs in Japanese (Ishizuka 2005, Miyamoto and Nakamura 2003, Ueno and Garnsey 2008). This suggests two possibilities: that frequency is not the main constraint of RC processing; or the corpus which was analyzed in Sato (2011) does not reflect the real distribution of RCs in Japanese. Moreover, we do not know whether the observed distribution of SRs and ORs in Sato (2011) holds in L2, as well. Therefore, it is hard to test the validity of FOH through the processing of SRs and ORs in L2 Japanese. However, as explained in the next section, nominative-genitive conversion provides a nice test case. Case-Driven Expectation Hypothesis (CDEH) : In Japanese, previous studies have shown that case markers play very crucial roles in the sentence processing 86 (e.g., Kamide, Altmann and Haywood 2003). According to these studies, the parser incrementally utilizes the case marker information and makes predictions about the argument structure and upcoming elements of the sentence in Japanese. Based on these studies, Sato et al. (2009) argued that, in addition to universal factors, language-specific factors should also be taken into consideration and so proposed the case-driven expectation hypothesis (CDEH) for Japanese. CDEH assumes that Japanese RCs are processed more easily if there is an early expectation for another noun before the RC-verb is encountered. (2a) SRs: ___ i Gakusei-o mita kyooju i ... Student-ACC saw professor ‘The professor that saw the student...’ (2b) ORs: Gakusei-ga ___ i mita kyooju i ... Student-NOM saw professor ‘The professor that the student saw...’ In SRs, the nominative-noun is not in its canonical position, and the accusative-noun appears in the sentence initial position. The sentence initial accusative-noun signals that the nominative-noun is missing and elicits an early expectation for that noun before the RC-verb is encountered. In the case of ORs, the nominative-noun appears in its canonical position. Therefore, sentence initial nominative-noun in ORs does not elicit such expectation until the RC-verb is encountered. CDEH assumes that the head noun of SRs is read faster than ORs, because early expectation initiates early start of search for the missing noun, and this makes RC processing easier at the RC-head. In a series of experiments, Sato Early expectation for missing NP Late expectation for missing NP 87 et al. (2009) showed that the early expectation for another noun made the RC processing easier, and argued that CDEH can explain the processing difficulty of RCs in Japanese. In L1 Japanese, many studies compared the processing of SRs and ORs, and attempted to distinguish among various hypotheses (e.g., Ishizuka 2005, Sato 2011, Ueno and Garnsey 2008). However, the number of online RC processing studies in L2 Japanese is comparatively few, and previous studies have mainly attempted to distinguish only between LDH and SDH (Kashiwagi 2011, Mitsugi et al. 2010). In other words, the possible effects of frequency and case-driven expectation have not been directly tested in L2 Japanese. It, thus, can be said that we do not fully understand to what extent universal factors, like distance and frequency , or language-specific factors, like case-driven expectation , may affect the RC processing in an L2. Therefore, the present study aims to explore the possible effects of universal and language-specific factors on RC processing and distinguish among CDEH, FOH, LDH, and SDH in L2 Japanese. In order to do this, the nominative-genitive conversion was utilized to RCs. 3. Nominative-Genitive Conversion In certain environments, the subject-noun can bear either nominative or genitive cases in some languages. This is called nominative-genitive conversion (e.g., Hiraiwa 2001). For example, in Japanese, the subject-noun of ORs can bear both case markers, as shown in (3). The subject-noun kyooju bears the nominative case in (3a), and the genitive case in (3b). However, both sentences have the same meaning. In the present study, the former is called the NOM-RC, and the latter is the GEN-RC. 88 (3a) Kyooju-ga ___ i itta kotoba i -wa wasure-rare-nai Professor-NOM said word-TOP forget-POSS-NEG ‘ I cannot forget the words that the professor said ’ (3b) Kyooju-no ___ i itta kotoba i -wa wasure-rare-nai Professor-GEN said word-TOP forget-POSS-NEG ‘ I cannot forget the words that the professor said ’ The linear distance between the gap and RC-head is equal in the NOM-RC and GEN-RC. Similarly, the structural distance is also equal in the NOM-RC and GEN-RC. On the other hand, the distribution of these structures differs in Japanese. It has been reported that the use of the genitive subject and, hence, the GEN-RC is very limited in modern Japanese (Kim 2009). In the case of case-driven expectation , the nominative-noun in NOM-RC does not elicit an expectation for another noun until the RC-verb is encountered (Sato et al. 2009). In contrast, the sentence initial genitive-noun elicits an expectation for another noun in the GEN-RC, because, when the parser encounters the genitive-noun, it knows there will be another noun in the sentence and it may incrementally start a search for that noun. Therefore, the comparison of the processing of these structures allows us to distinguish among CDEH, FOH, LDH, and SDH because the predictions for processing ease of NOM-RC and GEN-RC differ from that of SRs and ORs, as shown in Table 1. In SRs and ORs, CDEH and SDH predict that SRs should be easier to process than ORs; whereas, LDH predicts that ORs should be easier to process than SRs. On the other hand, FOH ’ s prediction for the processing ease of SRs and OR is unclear because there is no reported large-scale corpus analysis for L2 learners of Japanese. 89 Table 1 Predictions for the processing of SRs/ORs and NOM-RC/GEN-RC Hypothesis SRs vs. ORs NOM - RC vs. GEN - RC CDEH SRs < ORs GEN - RC < NOM - RC FOH ? NOM - RC < GEN - RC LDH ORs < SRs NOM - RC = GEN - RC SDH SRs < ORs NOM - RC = GEN - RC <: easy to process; =: no processing difference In the case of NOM-RC and GEN-RC, LDH and SDH predict that there should not be any processing difference because both the linear and the structural distance between the gap and the RC-head are equal in two sentences. On the other hand, FOH predicts that NOM-RC should be easier to process than the GEN-RC because NOM-RC ’ s frequency is higher than GEN-RC (Kim 2009). CDEH, unlike, FOH, LDH, and SDH, predicts that GEN-RC should be easier to process than NOM-RC because the expectation for another noun takes place earlier in GEN-RC than in NOM-RC. In order to test these predictions, four self-paced reading experiments were conducted with Turkish speaking learners and native speakers of Japanese, and the reading times of SRs vs. ORs and NOM-RC vs. GEN-RC were compared. 4. Experiments 4.1 Experiment 1: Processing of SRs vs. ORs by L2 Learners Aim : The aim of Experiment 1 was to confirm whether Turkish speaking learners of Japanese process SRs and ORs differently in Japanese. Experimental materials : Test sentences consisted of 24 sets of SRs and ORs, as shown in Table 2, with 24 sets of NOM-RC and GEN-RC (for the Experiment 2) and 48 filler sentences. Target sentences started with a locative adverb, and RC 90 followed it. The only difference between the two conditions was the case marker of second word. In the SRs condition, the accusative case, and in the ORs condition, the nominative cases were attached to the noun, as shown in Table 2. Table 2 Target sentences in Experiment 1 1 2 3 4 5 6 SRs Depato - de ryooshin - o sagashiteita kodo mo - wa kyuuni nakidasita dep.store - LOC parents - ACC seeking child - TOP suddenly cried ‘ The child who was looking for the parents at the department store cried suddenly. ’ ORs Depato - de ryooshin - ga sagashiteita kodomo - wa kyuuni nakidasita dep.sto re - LOC parents - NOM seeking child - TOP suddenly cried ‘ The child who the parents were looking for at the department store cried suddenly. ’ Previous studies have shown that the processing asymmetry between SRs and ORs is generally observed at the RC-head (kodomo-wa) in Japanese (e.g., L1: Ishizuka 2005, Miyamoto and Nakamura 2003; L2 Mitsugi et al. 2010). Therefore, it is predicted that the processing asymmetry would be observed at the RC-head in the present study, as well (see Table 1 for the predictions). Participants : Twenty-six undergraduate students at Çanakkale Onsekiz Mart University participated in Experiment 1. They were all native speakers of Turkish, and were studying Japanese Language Education as a major. Although no proficiency test was given to participants, their proficiency was characterized as higher-intermediate, because they had already studied Japanese for more than 3 years (approximately 20 hours per week) at the time of the experiment. Turkish is an SOV language, and the word order of the RCs is also identical to Japanese. Moreover, nouns within RCs bear explicit case markers, as in Japanese, and the subject-noun of ORs obligatorily bears the genitive-case. 91 (4) Profesör-ün ___ i söyle- diğ -i sözler i ... Professor-GEN say-RC-3sg words ‘ The words that the professor said ...’ In order to eliminate possible word order effect, and to make sure that the learners were familiar with the use of case marker information in their L1, Turkish speaking learners were chosen as the participants. Procedure : Experimental materials were divided into 2 lists by a Latin Square design, and each participant was assigned to either one of the lists. Experimental sentences were displayed on the computer screen in a random order, using Linger 2.94 (by Douglas Rohde). The task was a word-by-word non-cumulative self-paced reading. After reading a sentence, participants answered a yes-no question. Results and Discussion : Statistical analyses for reading times were conducted on the sentences wherein yes-no questions were answered correctly. The accuracy rate of SRs was 96%, and of ORs was 94%. This difference was not statistically significant. Before conducting a statistical analysis, all reading time data shorter than 250 milliseconds (ms), or longer than 5000 ms, was discarded from the analysis. Then the data longer than 2.5 SD points was also discarded from the analysis. Mean reading times for each word (region) in two conditions are shown in Fig. 2. 92 The results of ANOVA for repeated measures showed that the reading time difference was significant by subject analysis at the RC-head (region 4: 180 ms) [ F 1 (1,25)=7.74, p< .01; F 2 (1,23)=3.88, p =.06]. In other regions, there was no significant difference. The results demonstrate that the head noun of SRs was read faster than that of ORs, indicating that Turkish speaking learners of Japanese processed SRs more easily than ORs in Japanese. This result is consistent with previous studies in Japanese (L1: Ishizuka 2005, Miyamoto and Nakamura 2003, Ueno and Garnsey 2008; L2 Mitsugi et al. 2010), and confirms that LDH cannot account for the processing difficulty of ORs. On the other hand, the results can be explained by SDH and CDEH. Moreover, frequency might also be responsible for the results. In order to distinguish among these possibilities, reading times of NOM-RC and GEN-RC were compared in Experiment 2. 4.2 Experiment 2: Processing of NOM-RC vs. GEN-RC by L2 Learners Aim : The aim of Experiment 2 was to compare the reading times of NOM-RC and GEN-RC, and examine the validity of CDEH, FOH and SDH. Experimental materials : Test sentences were identical to Experiment 1. Target sentences consisted of 24 sets of NOM-RC and GEN-RC, as shown in Table 3. Target sentences start with a time adverb, and RC follows it. The only difference between the two conditions is the case marker of second word. In the NOM-RC condition, the nominative-case, and in the GEN-RC condition, the genitive-case markers are attached to the noun. CDEH predicts that GEN-RC should be read faster than NOM-RC. In contrast, FOH predicts that NOM-RC should be read faster than GEN-RC, and SDH predicts no difference between two conditions. 93 Table 3 . Target sentences in the Experiment 2 1 2 3 4 5 6 NOM - RC: Sengetsu gakusei - ga kaita sakubun - wa sugoku omoshirokatta l ast month student - NOM wrote essay - TOP very interesting ‘ The essay that the student wrote last month was very interesting ’ GEN - RC: Sengetsu gakusei - no kaita sakubun - wa sugoku omoshirokatta last month student - GEN wrote essay - TOP very interestin g ‘ The essay that the student wrote last month was very interesting ’ Participants and Procedure : Identical to that in Experiment 1. Results and Discussion : The process of data trimming and statistical analysis for reading times were identical to Experiment 1. The accuracy rate of NOM-RC was 97%, and GEN-RC was 96%. This difference was not statistically significant. Mean reading times of each region are as shown in Fig. 3. The results of ANOVA for repeated measures showed that reading time difference was significant by subject analysis at the RC-head (region 4: 104 ms) [ F 1 (1,25)=5.41, p< .05; F 2 (1,23)=2.98, p =.1]. In other regions, there was no significant difference. The results demonstrate that the head noun of GEN-RC was read faster than that of NOM-RC, indicating that Turkish speaking learners processed GEN-RC more easily than NOM-RC in Japanese. This result is consistent with the prediction of CDEH, whereas FOH and SDH fail to explain 94 the observed processing asymmetry. Taken together with Experiment 1, the results suggest that case-driven expectation plays an important role in the processing of L2 Japanese as well. However, it is still unknown whether the learners process these structures in a same manner to native speakers. In order to confirm this, the same experiments were conducted with Japanese native speakers. 4.3 Experiment 3: Processing of SRs vs. ORs by Native Speakers Aim : The aim of Experiment 3 was to confirm whether Japanese native speakers process SRs and ORs differently in Japanese. Experimental materials : It was identical to Experiment 1 (see Table 2.) Participants and Procedure : Twenty-seven undergraduate or graduate students at Hiroshima University participated in Experiment 3. They were all native speakers of Japanese. The procedure was identical to Experiment 1. Results and Discussion : The accuracy rate of SRs was 94%, and ORs was 96%. This difference was not statistically significant. Before conducting a statistical analysis, all reading time data shorter than 250 ms or longer than 2500 ms was discarded from the data analysis. Then the data longer than 2.5 SD point was also discarded from the analysis. Mean reading times for each region are as shown in Fig. 4. Fig. 4 Reading times of SRs and ORs 95 The results of ANOVA for repeated measures showed that reading time difference was significant by subject analysis at the RC-head (region 4: 70 ms) [ F 1 (1,26)=5.29, p< .01; F 2 (1,23)=6.34, p< .01]. In other regions, there was no significant difference. The results demonstrate that the head noun of SRs was read faster than that of ORs by native speakers of Japanese. This suggests that Japanese native speakers processed SRs more easily than ORs. This result is consistent with the Experiment 1 and previous studies in Japanese (Ishizuka 2005, Miyamoto and Nakamura 2003, Ueno and Garnsey 2008), and confirms that LDH cannot account for the processing difficulty of ORs. However, case-driven expectation, frequency, and structural distance might be responsible for the results. In order to examine these possibilities, the processing of NOM-RC and GEN-RC were compared in Experiment 4. 4.4 Experiment 4: Processing of NOM-RC vs. GEN-RC by Native Speakers Aim : The aim of the Experiment 4 was to confirm how Japanese native speakers process NOM-RC and GEN-RC. Experimental materials : Identical to Experiment 2 (see Table 3.) Participants and Procedure : The participants and the procedure were identical to Experiment 3. Results and Discussion : The process of data trimming and statistical analysis for reading times were identical to Experiment 3. The accuracy rate of NOM-RC was 99% and, GEN-RC was 98%. This difference was not statistically significant. Mean reading times of each region are as shown in Fig. 5. The results of ANOVA for repeated measures showed that there was no statistical difference in any region. 96 Fig. 5 Reading times of NOM-RC and GEN-RC The results suggest that the processing difficulty of NOM-RCs and GEN-RCs did not differ for Japanese native speakers. This result is consistent with the findings of Horii (1990), whereas inconsistent with Experiment 2. Therefore, this result can be explained by SDH, but cannot be explained by CDEH and FOH. Overall results will be discussed in more detail below. 5. General Discussion The main question of the present study was to what extent universal and language-specific factors may affect the RC processing in L2 Japanese. The results can be summarized as follows: Turkish speaking learners processed SRs faster than ORs, and they processed GEN-RC faster than NOM-RC in Japanese. On the other hand, native speakers of Japanese processed SRs faster than ORs, whereas they processed GEN-RC and NOM-RC in the same pace. These results show that there is some similarity and difference between the learners and native speakers. As a similarity, the learners and native speakers of Japanese processed SRs and ORs in a similar manner. As a difference, the processing pattern of NOM-RC and GEN-RC differed between the learners and native speakers. This indicates that different factors may have different degrees of impact on the sentence processing of L1 and L2. 97 As for the Turkish speaking learners, neither frequency (Reali and Christiansen 2007), nor distance based accounts, such as LDH (Gibson 1998) and SDH (O ’ Grady 1997) capture the entire results, whereas CDEH can successfully do so (Sato et al. 2009). In other words, language-specific factors can explain the observed processing patterns of RCs by Turkish speaking learners in L2 Japanese, whereas universal factors cannot explain them. The only difference between SRs vs. ORs and NOM-RC vs. GEN-RC were the case markers of the second word in the test sentences, and this difference led to divergent results. This indicates that the learners were sensitive to information obtained from case markers and incrementally utilized this information in the sentence processing of Japanese (Mitsugi et al. 2010, Sato et al. 2009). In SRs, the accusative-noun and in GEN-RC, the genitive-noun appeared after the sentence initial adverb. These nouns would have signaled the existence of another noun (Sato et al. 2009), and learners might have immediately started to search for that noun from these positions. Thus, the processing of the RC-head would have been facilitated by the early start of searching for the missing noun in SRs and GEN-RC. On the other hand, in ORs and NOM-RC, the nominative-noun appeared in its canonical position after the sentence initial adverb, and hence the existence of another noun would have not been signaled until the RC-verb. Therefore, the start of the search for a missing noun would have been delayed in ORs and NOM-RC, and this delay might have caused slower reading times at the RC-head compared to SRs and GEN-RC. One natural question is how the learners could process Japanese RCs in this way? In the present study, I would like to point out two possibilities. As the first and most likely possibility, learners ’ L1 might have influenced their L2 sentence processing in Japanese (Mitsugi et al. 2010). As explained in 4.1, the word order 98 and the use of case markers within RCs in Turkish are very similar to Japanese. Moreover, the subject-noun of ORs always bears the genitive-case in Turkish. Therefore, the processing strategies of Turkish might have been directly transferred to Japanese, and this transfer would have facilitated the processing of SRs and GEN-RC compared to ORs and NOM-RC in Japanese. As the second and less likely possibility, Turkish speaking learners might have learned how to process Japanese sentences independent of their L1. However, at this moment, it is hard to distinguish between these possibilities, and in the future we need to look into other L1 speaking learners’ processing pattern s. In the case of native speakers of Japanese, FOH, LDH, and CDEH cannot account for the entire results whereas SDH can. At first glance, it may seem that the case-driven expectation is not crucial in the processing of Japanese RCs. However, previous studies reported that case markers are one of the most important information sources in Japanese (Kamide et al. 2003). In the present study, a competition between the frequency and the case-driven expectation might have affected the possible effect of the case-driven expectation. In GEN-RC, there is a strong expectation for another noun at the genitive-noun position. On the other hand, the frequency of GEN-RC is very low in Japanese (Kim 2009). The low frequency might have somehow weakened the effect of expectation in GEN-RC; consequently NOM-RCs and GEN-RCs would have been processed at the same pace. Although overall processing patterns of the learners and native speakers of Japanese cannot be explained by the same factors in the current study, there is a very crucial similarity between them: the frequency cannot account for the observed processing patterns of learners and native speakers in Japanese. It can be said that this is one of the most important findings of the present study 99 because this finding shows the possibility that the frequency is not the decisive processing constraint in L1 and L2, as well. This finding also confirms Sato (2011), who argued that the simple frequency, alone, cannot explain the processing difficulty of RCs, in Japanese (cf. Reali and Christiansen 2007). In the present study, I could only investigate the higher-intermediate Turkish speaking learners ’ RC processing in Japanese. In order to examine whether the observed processing pattern was resultant from L1 transfer or L1-indepentend case-driven expectation, other L1 speaking learners’ processing pattern s should also be investigated. Moreover, in order to test how learners ’ proficiency is related to L2 sentence processing, longitudinal sentence processing studies are also necessary. I leave these issues for future studies. 6. Conclusions The main purpose of the present study was to examine to what extent universal and/or language-specific factors may affect relative clause processing, and distinguish among four competing hypotheses, namely case-driven expectation (CDEH), frequency of occurrence (FOH), linear distance (LDH), and structural distance (SDH) hypotheses in L2 Japanese. In order to do this, nominative-genitive conversion was utilized to RCs, and the processing of subject vs. object and nominative vs. genitive RCs were compared by self-paced reading experiments. The results suggested that language-specific CDEH can account for the observed processing patterns of Turkish speaking learners of Japanese, whereas universal accounts of FOH, LDH, and SDH fail to explain learners ’ processing pattern of RCs in Japanese. Moreover, overall results suggested that the simple frequency is not the decisive factor of RC processing in L1 and L2 Japanese. Taken together, the present study suggests that, in 100 addition to universal factors, language-specific factors should also be taken into consideration in L2 sentence processing. Notes * I would like to thank Takuya Goro, Mineharu Nakayama, Duygu Özge, Hiromu Sakai, Atsushi Sato, Patrick Stuart and audience at AMLaP 2011 and TCP 2012 for their constructive comments. All remaining errors are my own. 1 There are many other hypotheses. However, the present study mainly focuses on the examination of the above hypotheses. References Betancort, M., M. Carreiras. and P. Stuart. (2009) The processing of subject and object relative clauses in Spanish: An eye-tracking study. Quarterly Journal of Experimental Psychology, 62 , 1915-1929. Cohen, L. and J. Mehler (1996) Click monitoring revisited: An on-line study of sentence comprehension. Memory and Cognition, 24 , 94-102. Frazier, L. (1987) Syntactic processing: Evidence from Dutch. Natural Language and Linguistic Theory, 5 , 519-559. Gibson, E. (1998) Linguistic complexity: Locality of syntactic dependencies. Cognition, 68 , 1-76. Hashimoto, K. (2007) Subject/object asymmetry in the comprehension of English relative clauses by Japanese learners of English. The University of Queensland Working Papers in Linguistics , 1 Hiraiwa, K. (2001) On nominative-genitive conversion. In E. Guerzoni and O. Matushansky (Eds.), MIT Working Papers in Linguistics 39: A Few from