1 Interpretations of kare/kanojo by Turkish Speaking Learners of Japanese Bar ı ş Kahraman [1] and Mineharu Nakayama [2] The University of Tokyo/Japan Society for the Promotion of Science [1] and The Ohio State University/National Institute for Japanese Language and Linguistics [2] 1. Introduction Japanese has null pronouns ( pro ) and kare ‘he’ and kanojo ‘she’, which are often called overt pronouns (e.g., Kanno 1997). They are called pronouns because they share pronominal characteristics However, kare / kanojo are not overt counterparts of null pronouns and cannot be considered as pronouns similar to English pronouns he/she. 1 As Hoji (1991) argues, they are more like demonstratives, and not allowed to be bound by quantifier antecedents (Saito & Hoji 1983). 2 Consider examples in (1). 3 (1) a. Yamada-san i -ga [ pro i / kare i -ga hon-o katta] to itta. Yamada-san-NOM he-NOM book-ACC bought that said ‘Yamada i said that he i bought a book.’ b. Daremo i –ga [ pro i / kare *i –ga hon-o katta] to itta. everyone-NOM he-NOM book-ACC bought that said ‘Everyone i said that he i bought a book.’ c. Dare i -ga [ pro i / kare *i -ga hon-o katta] to itta-no? who-NOM he-NOM book-ACC bought that said-Q ‘Who i said that he i bought a book?’ Example (1a) shows that both pro and kare at the embedded subject position can refer to Yamada. On the other hand, (1b) and (1c) indicate that pro can be bound by the quantifier daremo ‘everyone’ and wh -word dare ‘who’, but kare cannot. When a pronoun is bound by a quantifier antecedent, it becomes a bound variable (BV) and we call this reading the BV reading/interpretation. As the grammaticality of (1b) and (1c) 1 Historically speaking, they did not appear in Spoken Japanese until the 19 th century (Okumura 1954). 2 Chomsky (1981) on “ be bound ”: Where α is bound by β if and only if α and β are coindexed, β c-commands α and β is in an A-position. Reinhart (1976) on “ c-command ”: Node A c-commands node B if neither A nor B dominates the other and the first branching node dominating A dominates B. 3 Abbreviations: ACC: Accusative case; GEN: Genitive case; NML: Nominalizer; NOM: Nominative case; Q: question marker. 2 suggests, pro can take daremo / dare as its antecedent, having the BV reading. However, kare cannot take the quantifier as its antecedent and does not permit the BV reading. 4 Masumoto (2008) and Pimentel & Nakayama (2012a) investigated whether English speaking learners of Japanese know that kare cannot be bound by a quantifier. Masumoto’s instrument used minna ‘all people/everyone’ while Pimentel & Nakayama’s included dono X ‘every X’ as the antecedents of kare . The test sentences were all grammatical, but their interpretations did not match the situations. In that way, the universal quantifier situations were examined with the kare and pro sentences. If the learners had the knowledge that kare cannot take the quantifier antecedent, then they were expected to reject the kare sentence with the situation in which it took the universal quantifier as its antecedent. They found that the learners accepted BV interpretations with kare/kanojo more than half of the time. Pimentel & Nakayama interpreted these results as either there was an L1 transfer strategy or the learners had a default strategy that all pronominals allow the BV reading. Because English has only overt pronouns, their study cannot answer which of these interpretations is correct. We need to test this knowledge among learners of Japanese whose native tongue is not like English, but rather like Japanese. In this respect, Turkish is such a language, in which both null and overt pronominals are allowed and overt pronominals cannot have the BV reading while pro can. Consider Turkish counterparts to Japanese examples in (1), which are listed in (2). (2) a. Yamada-san [ pro i /o i -nun kitap al-d ı ğ ı -n ı ] söyledi. Yamada-san-NOM s/he-GEN book buy-NML-ACC said ‘Yamada i said that he i bought a book.’ b. Herkes i [ pro i /o *i -nun kitap al-d ı ğ ı -n ı ] söyledi. everybody-NOM s/he-GEN book buy-NML-ACC said ‘Everyone i said that s/he i bought a book.’ c. Kim i [ pro i /o *i -nun kitap al-d ı ğ ı -n ı ] söyledi? who-NOM s/he-GEN book buy-NML-ACC said ‘Who i said that s/he i bought a book?’ As (2a) shows, Turkish allows both the null pronoun and the overt pronoun (e.g., 3 rd person singular pronoun ‘ o ’) in the embedded subject position and both pronouns can 4 This difference is captured by Montalbetti’s (1984: 94) Overt Pronoun Constraint: Overt pronouns cannot link to formal variables iff the alternation overt/empty obtains. Independently of Montalbetti’s OPC, Hong (1984) also reports this characteristic in Korean, which has both overt and null pronouns like Japanese (cf. Sheen 2000). 3 refer to Yamada, similar to the Japanese example (1a). 5 Note that the Turkish overt pronoun o is gender-non-specific, used for both ‘he’ and ‘she’, and it generally bears genitive case instead of nominative case in the embedded clause. Since the genitive form is more natural than the nominative form and expresses the meaning of the Japanese counterpart more unambiguously, we use the genitive case marked pronouns in the Turkish examples here. Irrespective of these surface differences, however, Japanese and Turkish are very similar with respect to the interpretations of pronouns. As shown in (2b) and (2c), pro can take the quantifier herkes ‘everyone’ and wh -word kim ‘who’ as the antecedent, respectively, whereas o cannot. Since there is no report on Turkish speaking learners of Japanese with respect to their pronominal interpretations, this study focuses on them, specifically, on adult Turkish speaking learners’ acquisition of the BV reading in Japanese. By doing so, we can provide an important answer to the question regarding L1 transfer vs. default strategies as mentioned in Pimentel & Nakayama (2012a, b). Given the similarities between the two languages with respect to the BV interpretation, the following predictions can be made about Turkish speaking learners’ acquisition of the BV reading in Japanese: If there is an L1 transfer strategy, Turkish speaking learners of Japanese would demonstrate the knowledge that kare/kanojo cannot have the BV reading from an early stage of their learning. On the other hand, if Turkish speaking learners have a default strategy that pronouns allow the BV reading (Pimentel & Nakayama 2012a, b), they would have the BV interpretation with kare/kanojo initially, and as their language skills improve, they would realize that kare/kanojo cannot take quantifier antecedents in Japanese. This study tests these predictions. Our experimental results show that not only the learners at the lowest proficiency level, but also some advanced learners erroneously accepted the BV reading of kare/kanojo . This was similar to the results found among American English speaking learners of Japanese in Pimentel & Nakayama (2012a). Considering Pimentel & Nakayama (2012a, b), we suggest that some L2 learners are likely to follow the default 5 Gürel (2003) states o tends to be extra-sentential while kendi (‘self’) is intra-sentential. For instance, the following sentence is ungrammatical when the embedded subject overt pronoun refers to the matrix subject. (i) Elif i [o-nun *i/j çok inatç ı ol-du ğ u]-nu bil-iyor Elif s/he-GEN very stubborn be-NML-ACC know-PROG ‘Elif knows that s/he is very stubborn.’ However, as in (2a), o can refer to the matrix referential subject, suggesting that the embedded clause subject o can refer to the matrix referential subject depending on the construction. 4 strategy during the early stage of the acquisition of Japanese pronouns, irrespective of their L1. The organization of the paper is as follows: We will briefly review previous studies including Pimentel & Nakayama’s (2012a) findings in the next section. Then, our experiment will be described and its results will be reported in Section 3. Finally, Section 4 provides a discussion of our results and theoretical implications. 2. Previous studies Studies such as Kanno (1997, 1998), Yamada (2005), Masumoto (2008), and Pimentel & Nakayama (2012a, b) investigated the BV interpretation of kare among English speaking learners of Japanese. 6 Kanno (1997), for instance, conducted an experiment with a coreference-judgment task in a written questionnaire, in which the English speaking learners of Japanese were asked to read test sentences and answer questions about the referents of the subject noun phrases (NPs) in the embedded clauses. There were four types of test sentences with null and overt pronouns. The pronouns appeared at the embedded subject positions with the dare ‘who’ antecedents in matrix clauses. Another two sets of test sentences contained referential NPs as their antecedents. Examples are given below. Correct answers are bold faced. (3) a. Null pronoun with a wh /quantified NP as antecedent Dare-ga senshuu waapuro-o tsukatta-to itteiru-n desu-ka. ‘Who says that he used the word processor last week?’ Q: Dare-ga waapuro-o tsukatta-n deshoo-ka. ‘Who do you suppose used the word processor?’ A. same as dare B. another person A&B b. Overt pronoun with a wh /quantified NP as antecedent Dare-ga kyoo kare-ga osokunaru-to itteiru-n desu-ka. ‘Who says that he is going to be late today?’ Q: Dare-ga kyoo osokunaru-n deshoo-ka. ‘Who do you suppose is going to be late today?’ A. same as dare B. another person A&B 6 See also Kellerman & Yoshioka (1999), Perez-Leroux & Glass (1999), Sheen (2000), Gürel (2003, 2006), and Pimentel (2014) on the Overt Pronoun Constraint. We did not review these studies in the text due to space limitation and the relevance to the current study (i.e., the interpretations of kare/kanojo ). Note that Kanno’s (1997, 1998) studies were to argue for the UG accessibility hypothesis in L2 acquisition, but the UG accessibility is not a main concern in this paper. Our interest lies in whether kare and kanojo are believed to have the BV readings by L2 learners. 5 c. Overt pronoun with a referring NP as antecedent Yamamoto-san-wa ashita kare-ga buchoo-ni au-to itteimasu-yo. ‘Yamamoto says that he will meet the division manager tomorrow.’ Q: Dare-ga ashita buchoo-ni au-n deshoo-ka. ‘Who do you suppose will meet the division manager tomorrow?’ A. Yamamoto B. someone other than Yamamoto A&B d. Null pronoun with a referring NP as antecedent Tanaka-san-wa raishuu Kyooto-e iku-to itteimashita-yo. ‘Tanaka was saying that he is going to Kyoto next week.’ Q: Dare-ga raishuu Kyooto-e iku-n deshoo-ka. ‘Who do you suppose is going to Kyoto next week?’ A. Tanaka B. someone other than Tanaka A&B Kanno reports that when wh words were the antecedents of the null pronouns like (3a), native speakers of Japanese answered correctly 83% of the time, but when wh words were the antecedents of the overt pronouns ( kare ) like (3b), they took the same referents as the wh words referred to only 2% of the time. This indicates their strong preference for null pronouns to take the wh word antecedents. English speaking learners of Japanese in her study showed a similar trend. They accepted the wh words as the antecedents of the null pronouns 78.5% of the time while accepting those for kare only 13% of the time. Because her learners were in an early stage of language development (i.e., in the fourth-semester of a Japanese course, and had not been instructed on the BV reading), this is evidence against L1 transfer, but supports the theory of innate knowledge of the null vs. overt difference regarding the BV reading (i.e., the Overt Pronoun Constraint, see footnote 4). Kanno (1998) follows up this study by examining the English speaking learners of Japanese twice, at the beginning of the semester and after approximately 12 weeks of instruction. She found similar results to her 1997 study in the two sessions, i.e., the learners tended not to allow the wh words as the antecedents of kare/kanojo . From this, she considered the finding as support for the UG accessibility hypothesis in L2 acquisition, i.e., accessing innate knowledge during the course of L2 acquisition. Yamada (2005) investigated the interpretations of pro and kare as well as zibun ‘self’, which is often considered as a reflexive and can take the quantifier antecedent, among British English speaking learners of Japanese at various levels of Japanese proficiency. She employed a questionnaire with a similar format to Kanno’s. Her results were different from Kanno’s in that the learners selected (A) as the answer to (3b) 50.8% of the time for kare ‘he’ and 43.9% of the time for kanojo ‘she’. 6 Unlike Kanno and Yamada, both Masumoto (2008) and Pimentel & Nakayama (2012a) employed a different methodology, a truth value judgment task, that provided situations with the test sentences. They asked American learners of Japanese whether they could say kare/kanojo and pro sentences in the particular situations described in English in a questionnaire. Masumoto’s used minna ‘all people/everyone’ while Pimentel & Nakayama’s included dono X ‘every X’ as the antecedents of the pronouns (see Appendix). Unlike Kanno’s and Yamada’s test sentences, they used simplex sentences like (4a) below to reduce the processing/memory load so that the stimuli were at an appropriate level for learners with low proficiency. As (4b) and (4c) show, the pro and kare grammaticality difference is similar to those in (1b) and (1c). 7 (4) a. Yamada-san i -ga pro i / kare i -no okaasan-o yonda. Yamada-san-NOM his mother-ACC called ‘Yamada i called his i mother.’ b. Daremo i –ga pro i / kare *i – no okaasan-o yonda. everyone-NOM ‘Everyone i called his i mother.’ c. Dare i -ga pro i / kare *i -no okaasan-o yonda-no? who-NOM -Q ‘Who i called his i mother?’ They found that the learners accepted BV interpretations with kare/kanojo more than half of the time. Since these studies differed in experimental methodology, Pimentel & Nakayama (2012b) employed Kanno’s methodology with her wh test sentences as well as Pimentel & Nakayama’s (2012a) dono X test sentences. The results 7 Note that Turkish counterparts are slightly different from Japanese in that o at the possessive position cannot refer to the clausemate subject. That is, it cannot refer to Yamada in (i). However, the BV readings are not permissible as in (ii) and (iii), the same as (2b) and (2c), respectively. (i) Yamada-san i pro i /o *i /j -nun annesi –ni ça ğ ı rd ı Yamada-san-NOM s/he-GEN mother-ACC called ‘Yamada i called his *i /j mother.’ (ii) Herkes i pro i /o *i –nun annesi –ni ça ğ ı rd ı everybody-NOM ‘Everyone i called his i mother.’ (iii) Kim i pro i /o *i -nun annesi –ni ça ğ ı rd ı ? who-NOM ‘Who i called his i mother?’ 7 were similar to those of Masumoto and Pimentel & Nakayama (2012a). The learners accepted the BV reading more than half of the time (see Table 3 in the Discussion section). Based on these results Pimentel & Nakayama (2012a, b) suggested that either L1 transfer or a default strategy that all pronouns can have the BV reading is at work from an early stage of L2 development. Then, they learn demonstrative characteristics of kare/kanojo . Like a demonstrative ano hito ‘that man/person’, kare designates a very specific person (referentially higher than null pronouns in the referential hierarchy) and cannot become a formal variable. Furthermore, they also need to learn other anaphoric uses of words like zibun ‘self’ and title/occupation nouns like sensei ‘teacher’, that are also used like overt pronouns. After they learn these anaphoric uses, they can demonstrate clear judgment of pronominal interpretations. Presumably, this acquisition process takes time. This stance is different from Kanno’s because she claims that L2 learners know that the overt pronoun cannot have the BV interpretation from the beginning of L2 acquisition thanks to the innate knowledge (e.g., the Overt Pronoun Constraint). Pimentel & Nakayama (2012b) explain the confounding data by conjecturing that Kanno’s learners were more exposed to Japanese than theirs (i.e., higher proficiency) because those in the “one level above” class in their study could show their knowledge that kare/kanojo cannot take the quantifier antecedent. There was likely a proficiency level difference. Since Masumoto and Pimentel & Nakayama’s studies cannot differentiate L1 transfer from the default strategy, it is necessary to study those learners whose native tongue behaves like Japanese. Turkish speaking learners of Japanese provide an ideal test case because the interpretation of the overt pronoun in Turkish is similar to that in Japanese. Below, we now discuss our experiment. 3. The present study 3.1 Participants The participants of this study were 32 Turkish speaking learners of Japanese with varying Japanese language proficiency. Because the length of the study varied and the participants’ proficiency differed, we divided them into three groups based on the Japanese Language Proficiency Test (JLPT) proficiency level. The three groups are called Superior, Advanced, and Intermediate, conventionally. The Superior Group had nine students who passed the JLPT N1, i.e., the proficiency level that allowed them to study in regular university curricula in Japan. All but one learner had studied or been studying in Japan at the time of the experiment. Their mean length of study was 120 months, ranging from 44 months to 248 months. The Advanced Group included eight Turkish university students who passed the JLPT N2. Two students had studied in Japan for one year. Their mean length of study was 44 months (32-56 months). The 8 Intermediate Group had 15 students who had not taken the JLPT, but their language abilities were considered by one of the experimenters and their Japanese instructors as between the JLPT N2 and N3. Their mean length of study was 52.8 months (32-80 months). In their university, one semester is 14 weeks, and students study Japanese approximately 20 hours per week, taught by both native and non-native Japanese instructors. 8 Among the participants, one in the Intermediate and two in the Advanced were studying Japanese as the second language, i.e., for most learners, it was the third language following Bulgarian, English, French, or German. 9 Most of the participants were recruited from a Japanese-Turkish translation class and given the questionnaire during the class. Among them a few participants who could not attend the class were given the questionnaire as homework. 10 They all received extra credit for their participation in the course. Seven participants in the Superior Group were recruited individually and given the questionnaire via e-mail from Japan. 3.2 Materials The experiment employed a Turkish translation of Pimentel & Nakayama’s 2012a questionnaire to make a direct comparison. Pimentel & Nakayama’s task was a truth value judgment task with narratives (cf. Crain & McKee 1985). Situations were described in Turkish, and names in narratives were also given in Turkish to avoid a possible gender ambiguity. Each participant was asked to judge whether the Japanese sentence below the situation narrative matched the situation just read. If it matched the situation, they selected TRUE, and if not, they chose FALSE (in Turkish, Do ğ ru and Yanl ı ş , respectively). The questionnaire included 37 pairs of the situation-sentence stimuli (three practice sentence pairs, 24 test sentence pairs, and 10 filler sentence pairs). The Turkish version used in the present study is in (5) with English gloss and translation 8 During the first year of their study (i.e., the remedial course), Minna no Nihongo Honsatsu I and II (published by 3A Network) were used as main textbooks with supplemental textbooks, such as the Basic Kanji Book (Kano, Shimizu, Takenaka & Ishii, 1990). 9 Note that one of the Advanced L2 Japanese learners was not included in the final analysis because of her poor correct response rate on fillers. An Ars Linguistica reviewer questioned if the current results may have been affected by the learners’ L2 because Japanese is actually their L3. See our discussion below. 10 The homework group and seven Superior Group learners could have access to their dictionaries. Even when the data were collected in the classroom, the experimenter told the participants that they could ask the meanings of words unknown to them. Since, the words used in the questionnaire were very easy and they knew every word in the task, nobody asked the experimenter about unknown words. Therefore, vocabulary knowledge was not an issue and even if they had access to dictionaries, they do not tell them that kare/kanojo cannot have the BV interpretations. 9 (See Pimentel & Nakayama’s test sentences in Appendix). Test sentences used Chinese characters ( kanji ) with their readings ( yomigana ) as in (5). The correct answer is boldfaced. (5) Kemal, Tugay ve Ş aban, k ı z karde ş lerini yanlar ı na alarak bir al ı ş veri ş merkezine giden kuzenlerdir. Dedeleri de onlarla birlikteydi. Onlar büyük bir elektornik ma ğ azas ı na girdiler. Onlar televizyonlar ı incelerken, k ı z karde ş leri, di ğ er ma ğ azalar ı gezmeye karar verdiler. Televizyonlar ı inceledikten sonra, Kemal, Tugay ve Ş aban k ı z karde ş lerine bakmaya gittiler. Tugay karde ş ini kozmetik ürünler satan ma ğ azada makyaj malzemelerine bakarken buldu ve yan ı na ça ğ ı rd ı Kemal ise karde ş ini bir cafede buldu ve yan ı na ça ğ ı rd ı Ş aban da karde ş ini kad ı n elbiseleri satan bir ma ğ azada buldu ve yan ı na ça ğ ı rd ı Herkes karde ş ini bulduktan sonra hep birlikte en üst kattaki restoranda ö ğ le yeme ğ i yediler. どのいとこも彼 かれ の 妹 いもうと を呼 よ んだ。 Do ğ ru / Yanl ı ş Kemal, Tugay, and Ş aban are cousins who went to a department store with their little sisters. Their grandfather accompanied them as well. They went to the department store in order to check out a large sale on electronics. While they were looking at televisions, their little sisters decided to go off in different areas of the store to see what else might be on sale. After looking at the televisions, Kemal, Tugay and Ş aban went off to try to find their little sisters. Tugay found his little sister at the cosmetics counter looking at makeup so he called her. Kemal found his little sister in a café, so he called her. Ş aban found his little sister, in the ladies apparel section so he called her. After they all found their sisters, they all got together and had lunch at a restaurant on the top floor of the department store. Dono itoko-mo kare-no imooto-o yonda. True/ False every cousin also his sister-ACC called ‘Every cousin called his sister.’ Sentences in (6) are example test sentences. (6a) is a crucial BV sentence with the overt pronoun. The pronouns in the parentheses indicate the referents the corresponding situations. In (6a), however, kare cannot refer to his own cousin in the situation (5). (6) a. Dono itoko-mo i kare *i -no imooto-o yonda. ( kare - cousin) every cousin also his sister-ACC called 10 ‘Every cousin called his sister.’ b. Dono hisho-mo i pro j meeru-o kaita. ( pro - secretary) every secretary also his mail-ACC wrote ‘Every secretary wrote his mail.’ c. Dono ruumumeito-mo i kare j -no asagohan-o tabeta. ( kare - one roommate) every roommate also his breakfast-ACC ate ‘Every roommate ate his breakfast.’ d. Dono kookoosei-mo i pro j bideo-o mita. ( pro - one high school student) every high school student also her video-ACC watched ‘Every high school student watched his video.’ e. Dono onnanohito-mo i kanojo j -no hanashi-o kiita. ( kanojo - discourse referent) every woman also her story-ACC heard ‘Every woman heard her story.’ f. Dono onnanoko-mo i pro j arubamu-o wasureta. ( pro - discourse referent) every girl also her album-ACC forgot ‘Every girl forgot her album.’ Kare cannot be bound by dono itokomo ‘every cousin’ in (6a), but null pronoun pro can be bound by dono hishomo ‘every secretary’ in (6b). The pronouns in (6c) and (6d) refer to one of the roommates dono roomumeetomo includes and one of the high school students dono kookooseimo describes in the situations, respectively. The pronouns in (6e) and (6f) refer to other persons dono onnano hitomo and dono onnanokomo do not refer to in the situations, respectively. For clarification of the situations, please see the Appendix. Since Pimentel & Nakayama’s questionnaire was employed, each sentence type had a different number of actual sentences. Their unbalanced numbers of test sentences were not adjusted in this study in order to directly compare two studies, with the same stimuli context. In addition, we found one translation error in the (6d) TRUE type sentences after the experiment was conducted: one (6d) TRUE sentence became a FALSE sentence. Thus, the final sentence numbers for the sentence types became as follows: (6a) four FALSE sentences, (6b) four TRUE and two FALSE sentences, (6c) one TRUE and one FALSE sentence, (6d) one TRUE and three FALSE sentences, (6e) three TRUE and three FALSE sentences, and (6f) two TRUE sentences. The total number of sentences did not change from that of Pimentel & Nakayama (2012a): 24 test sentences and 10 fillers. As discussed above, our predictions are as follows: If there is an L1 transfer strategy, even those at the lowest proficiency level would show that kare/kanojo cannot have the BV reading in Japanese. However, if they employ the default strategy that all pronouns can take the BV interpretation, then those with the lowest proficiency level 11 would allow the BV reading on (6a), but those at a higher proficiency level would correctly demonstrate their knowledge that kare/kanojo cannot take the quantifier as its antecedent. 3.3 Results One student each from the Intermediate and the Advanced groups did not answer correctly on the fillers more than 80% of the time. Thus, their data were removed from the data for statistical analyses. For the sake of comparison, 20 native speakers’ data from Pimentel & Nakayama (2012a) were brought in as the Control Group data, but 5 speakers’ data were removed because they did not answer correctly on the fillers or/and the test sentences 80% of the time. Therefore, the data of the following number of the participants were considered as more reliable and further analyzed: 14 in the Intermediate Group, 7 in the Advanced Group, 9 in the Superior Group, and 15 in the Native Control Group. Table 1 shows the numbers and percentages of the correct responses to the BV readings of both kare and pro. TRUE and FALSE in the parentheses are the correct answers. That is, kare (FALSE, 6a) means sentences like (6a) that contain kare incorrectly depicted the situations described. Table 1. Numbers and percentages of correct responses for the BV reading Group kare (FALSE, 6a) pro (TRUE, 6b) pro (FALSE, 6b) Intermediate ( n =14) 26/56 (46%) 54/56 (96%) 26/28 (93%) Advanced ( n =7) 19/28 (68%) 26/28 (93%) 14/14 (100%) Superior ( n =9) 36/36 (100%) 32/36 (89%) 18/18 (100%) Natives (PN , n= 15) 53/60 (88%) 57/60 (95%) 29/30 (97%) Table 2 shows the numbers and percentages of the correct responses for the referential noun antecedents. TRUE and FALSE are the correct answers, i.e., kare (6c) TRUE means sentences like (6c) that contain kare correctly depicted the situations in the narratives. 12 Table 2. Numbers and percentages of correct responses for the test sentences with common noun antecedents Group kare (6c) pro (6d) kare (6e) pro (6f) TRUE FALSE TRUE FALSE TRUE FALSE TRUE Intermediate 13/14 (93%) 11/14 (79%) 14/14 37/42 (100%) (88%) 33/42 34/42 (79%) (81%) 27/28 (96%) Advanced 6/7 (86%) 7/7 (100%) 7/7 19/21 (100%) (90%) 19/21 19/21 (90%) (90%) 12/14 (86%) Superior 7/9 (78%) 8/9 (89%) 9/9 27/27 (100%) (100%) 20/27 23/27 (74%) (85%) 16/18 (89%) Natives 14/15 (93%) 14/15 (93%) 30/30 30/30 (100%) (100%) 40/45 41/45 (89%) (91%) 27/30 (90%) We analyzed the number of correct responses in the TRUE and FALSE conditions separately among the four groups, using the generalized linear mixed-effects model (GLMM) with a binomial function (Baayen 2008). 11 We included the responses as a dependent variable, the group as a fixed effect, and participants and items as a random effect in the model. The analyses were conducted based on the Maximal Model (Barr, Levy, Scheepers & Tily 2013). The native speakers were the baseline group. The results showed that there were no significant differences among the four groups on the test sentences with the common noun antecedents, shown in Table 2 ( p > .17).This suggests that all participants judged the sentences with common noun antecedents relatively accurately as a whole, though the pro sentences evoked better responses from all groups. 12 In contrast, however, as Table 1 shows, the BV reading of the overt pronoun was correctly rejected at different rates by the groups. The Intermediate Group correctly rejected the kare sentence (6a) only 46% of the time while the null pronoun sentence (6b) was responded correctly 96% of the time. Even the Advanced Group answered correctly on the kare sentence only 68% of the time, though they did well on the null pronoun sentence like the Intermediate Group. The Superior Group evoked the correct response to (6a) 100% of the time. The results of GLMM showed that there was no 11 Since the numbers of participants in each group were relatively small, the power of the statistical analysis is arguable. However, we conducted a statistical analysis to demonstrate the tendency among the different learner groups. The results showed that there was a significant difference among the three learner groups in the (6a) BV reading sentences, despite the small data set. 12 Note that the Superior Group’s (6c) and (6e) TRUE condition types were lower than other learner groups, but this was largely due to one learner’s responses. 13 significant difference between the Native Speaker Group and the Superior Group in the BV reading of kare/kanojo , although the performance of the Superior Group was higher than the native speakers. ( β = 0.47, SE = 1.7, z = 0.28, p = .78). On the other hand, the differences among the Native Speaker Group, the Advanced Group and the Intermediate Group were statistically significant (Native Speaker Group vs. Advanced Group: β = -2.00, SE = 0.97, z = -2.06, p = .04; Native Speaker Group vs. Intermediate Group: β = -3.37, SE = 1.08, z = -3.13, p = .002; Advanced Group vs. Intermediate Group: β = 1.37, SE = 0.69, z = 1.98, p = .05). As for the null pronouns, there was no significant difference among the four groups ( p > .26). These results suggest that the learners gradually acquire the knowledge that the overt pronoun cannot take the BV interpretation. On the other hand, the null pronouns evoked relatively consistent performances. The varying performances on the BV reading of kare/kanojo by proficiency were also seen in the number of the individuals who performed well. All participants in the Superior Group, two in the Advanced and one in the Intermediate understood that kare/kanojo cannot take the quantifier antecedents (i.e., 100% correct). 13 On the other hand, 10 out of 14 Intermediate learners and two out of seven Advanced learners instead considered that kare/kanojo can take the quantifier antecedent half or more than half of the time. The participants at the lowest proficiency level, i.e., those in the Intermediate Group, did not seem to have understood this restriction fully. The difference between the common noun and the quantifier antecedents seems similar to those found among English speaking learners of Japanese in Pimentel & Nakayama (2012a, b). 4. Discussion Our results showed that Turkish speaking learners of Japanese could appropriately interpret the sentences with common noun antecedents and null pronouns. However, most of the learners in the Intermediate Group and some of the learners from the Advanced Group erroneously accepted the BV reading with kare/kanojo 14 When we consider the fact that Turkish permits both null and overt pronouns in the subject/possessive position and overt pronouns cannot be bound by quantifier 13 These three learners in the Intermediate and the Advanced groups studied English as their L2, which suggests that there was no L2 influence over L3 Japanese. 14 One may argue that the results may simply be due to the use of particular curriculum and teaching materials. However, similar curricula and textbooks to this university’s are used in all universities where Japanese language is taught as a major field of study in Turkey (Ay ş e Nur Tekmen and Volkan Erdemir, personal communication, April 26, 2014). Thus, it is unlikely that the particular curriculum and/or textbooks are the main source of our participants’ performance level. 14 antecedents as in Japanese, this finding suggests that most of our participants did not seem to adopt an L1 transfer strategy. 15 Moreover, the learners at the lowest proficiency level are likely to consider that kare/kanojo can be bound by quantifier antecedents and learn the lexical meaning of each anaphoric expression. This supports Pimentel & Nakayama’s default strategy that all pronouns can have the BV reading. Here let us look at the following table, an adjusted version of Pimentel’s (2014) table, a summary of the performances of comparable learners in previous studies (roughly speaking, ACTFL OPI Intermediate Low level) and the current study. Our lowest proficiency group, Intermediate, may have a higher proficiency level than the 2 nd year students in Kanno, Masumoto, and Pimentel & Nakayama because they have studied Japanese for more than 3 years. However, since they are our lowest level, we list them as a comparable group here. Table 3. Error rate among the comparable learners in different studies Kanno (1997) Kanno (1998) S1 S2 Masumoto (2008) Pimentel & Nakayama (2012a) Pimentel & Nakayama (2012b) The current study Intermediate 13% 29% 34% 61% 56% 58% 54% Table 3 shows that Kanno’s two studies (S1 means session 1 in her 1998 study) have lower error rates than Masumoto and Pimentel & Nakayama (2012a, b). Our error rate is comparable to Pimentel & Nakayama’s despite the fact that our learners may have had higher proficiency than others and our learners’ L1 is like Japanese. How do we explain this? Since Pimentel & Nakayama (2012b) employed Kanno’s methodology with her test sentences (cf. Sheen 2000), they conjectured the difference may have come from the difference in learning sites (Hawaii vs. Ohio). It might be the case that the learners in Hawaii were exposed to Japanese much more than those in Ohio, and thus, their level was higher than Masumoto’s and Pimentel & Nakayama’s. If so, how can we explain our Intermediate participants’ performance? None of our learners had stayed in Japan longer than two weeks, and thus, their exposure to Japanese in Turkey was quite 15 When Pimentel & Nakayama’s questionnaire (English version) was given to eight Chinese speaking learners of Japanese (including advanced students) at their institution, most of them accepted the BV reading of the overt pronoun. Chinese also has both null and overt pronouns like Turkish. 15 limited. 16 In addition, the characteristics of kare/kanojo were not explicitly taught in the classroom. It is likely that the participants were not exposed to Japanese long enough to acquire this restriction of kare/kanojo . Therefore, the Intermediate Group’s performance on the interpretation of kare/kanojo would have been poor. Despite the L1 similarity with Japanese, it does not appear that most Turkish students had L1 transfer, but instead the current finding appears to support the idea that L2 learners start learning Japanese with the default strategy that all pronouns can have the BV interpretation. However, there is an alternative account we need to consider. There were L1 and L2 transfers as many of the current Turkish learners had English as their second language. 17 That is, they allowed the null pronoun with the BV reading (positive transfer) from their L1 Turkish, and then, they permitted the overt pronoun with the BV reading (negative transfer) from their L2 English. The two transfers led to the erroneous BV responses. And then they would retrieve the second transfer as they learn that Japanese is like Turkish. This is a possibility for the current Turkish speaking learners. In order to examine this possibility, we compared the correct response rates for the BV sentences by L2 English speaking learners (N = 9) and non-L2 English speaking learners (N = 5) in the Intermediate Group. The results showed that the L2 English speaking learners’ correct rejection rate was 50%, while that of the non-L2 English learners was 40%. 18 This suggests that it is difficult to attribute the poor performance of our participants on the interpretation of kare/kanojo to negative transfer from L2 English. If this were the case, the L2 English speaking learners’ performance would have been worse than that of the non-L2 English speaking learners. This multiple transfer account does not provide a strong explanation of the current results, although we cannot deny its possibility for some learners. Thus, it is more plausible to assume that a more general account, the default strategy, is at work. Note that this strategy is a universal heuristic strategy. Thus, it applies to L1 and predicts that children apply it to a pronominal and have the BV reading with it when the configuration is met. However, this strategy does not predict how children select null or overt pronouns to start their L1 with nor does it decide which item is pronominal. To obtain the BV interpretation, one must label a noun as pronominal and the structural 16 Another possibility is that they may have been less proficient students because 11 of the 14 students in this group were so called “repeaters” who did not pass the remedial course during the first year. 17 We thank the Ars Linguistica reviewer and Noriko Yoshimura, who brought this possibility to our attention. For L3 acquisition, see Yoshimura et al. (2013). 18 One of the non-L2 English speakers was an L2 Japanese learner. Her correct rejection rate was 25%, and she seemingly did not adopt an L1 transfer strategy. Another L2 Japanese learner, who was in the Advanced Group, correctly responded only 50% of the time. 16 configuration with the antecedent (i.e., c-command and Binding condition B) must be satisfied. Chien and Wexler (1990) report young children have the BV readings of overt pronouns in L1 English, but we do not know what interpretation English speaking children in the root infinitive stage would have for null pronouns, assuming that they have null pronouns in their very early grammar. Kare is acquired late in L1 Japanese (i.e., school age) and since it is not really a pronoun, it is not strange at all even if they do not have the BV reading of kare by the time they acquire it. These are the predictions that the default strategy makes and they have to be investigated in the future. Now, given this strategy, how do Turkish participants come to know that kare/kanojo cannot have the BV reading? Suppose the learners in the current study were in a situation similar to the English speaking learners of Japanese in Masumoto and Pimentel & Nakayama studies. Then, as Pimentel & Nakayama (2012b) claim, they may come to understand that kare/kanojo cannot have the BV reading when they acquire all anaphoric expressions in Japanese, especially, when they realize that kare/kanojo are demonstratives and zibun can become a bound variable. In order to acquire this, learners need to assemble relevant (lexical) features (cf. Feature assembly hypothesis, Lardiere 2008). As Kano & Nakayama (2004) found, English speaking learners of Japanese had a perfect understanding of the BV reading of zibun from an early stage in their learning. This seems to support that learners acquire the BV interpretation once the anaphoric nature of the t