Interpretation of Mandarin pronouns and reflexives by L1-Korean and L1-English learners of Mandarin

Abstract

This study investigates (1) whether second language learners (L2ers) acquire the Mandarin system of pronouns and reflexives despite differences from their first languages (L1s) and (2) whether L1-English and L1-Korean L2ers differ due to L1-transfer. Unlike English, Mandarin and Korean allow long-distance (LD) reflexives. While himself/herself and the Mandarin equivalent taziji both require local readings, Mandarin ziji ‘self’ allows both local and LD readings. In Korean, caki and casin allow LD readings while caki-casin requires local readings. For pronouns, English him/her and Mandarin ta both disallow local readings while Korean ku/kunye allows them. These cross-linguistic differences lead to different transfer-based predictions for L1-Korean and L1-English L2-Mandarin learners. Sixty-two Mandarin native speakers, 42 L1-Korean L2ers, and 32 L1-English L2ers completed a picture-based Truth Value Judgment Task and a Mandarin proficiency test. Results show that proficiency-matched L2ers from both L1s predominantly allowed only local readings of ziji, suggesting that the local reading is the default option, and that Korean speakers do not transfer the properties of Korean simplex reflexives onto ziji, possibly because LD ziji is logophoric, while caki is not. Regarding ta, L1-Korean L2ers under-accepted LD readings and over-accepted local readings, a result that can be attributed to transfer, per Kim’s analysis that Rule I, posited by Grodzinsky and Reinhart, that regulates the distribution of pronouns is weak or absent in Korean.

Keywords

binding Korean Mandarin pronouns reflexives

I Introduction

The effect of first language (L1) transfer is well attested in second language (L2) acquisition (e.g. Lardiere, 2008, 2009; Schwartz and Sprouse, 1994, 1996). Given that the properties of reflexives like himself/herself differ across languages, reflexives have attracted much attention in L2-acquisition literature with regard to L1-transfer, with a particular focus on reflexives that allow long-distance (LD) antecedents. One such LD reflexive is the Mandarin ziji ‘self’, which we analyse as logophor, following Huang and Liu (2001) (but see Reuland et al., 2020). However, previous studies have not clearly disentangled the role of L1-transfer from a possible universal preference for local antecedents. The first objective of this study is to determine whether L2-Mandarin learners’ preference for local antecedents of ziji results from L1-transfer or is transfer-independent.

Unlike reflexives, object pronouns are traditionally considered to behave the same cross-linguistically, and therefore to be unproblematic for L2-learners. However, Kim (2019; see also Kim, 2018, under review) showed that Korean native speakers allow local antecedents of pronouns; Kim attributed this finding to pragmatic constraints on local co-reference being weak or absent in Korean. Kim (2019, under review) shows that local interpretations of Korean pronouns are transferred into the L2-English of Korean speakers. The second objective of the present study is to test whether similar transfer takes place when Korean speakers acquire L2-Mandarin.

The present study investigates (1) whether L1-English and L1-Korean L2-Mandarin learners can become target-like in their interpretations of Mandarin object anaphors with regard to (im)possible antecedents, and (2) whether the learners transfer the properties of object anaphors from their L1s. Mandarin and English (but not Korean) have a similar pronoun distribution, while Mandarin and Korean (but not English) have a similar distribution of reflexives. These differences allow us to make specific predictions for how L1-transfer influences L2-acquisition in this domain. We test these predictions using a picture-based Truth Value Judgment Task (TVJT).

While many prior studies have looked at reflexives in L2-acquisition, this is the first study to examine both pronouns and reflexives in the acquisition of Mandarin by English and Korean native speakers; this comparison allows us to tease apart the role of L1-transfer from universal preferences, and to determine whether transfer affects reflexives and pronouns equally. Our findings point to the conclusion that transfer affects the L2-acquisition of Mandarin pronouns but not of Mandarin reflexives, a finding that we explain by appealing to particular theoretical analyses of Mandarin anaphors and logophors.

The article is structured as follows. Section II reviews the relevant literature. The research questions are raised in Section III. The methods and results are presented in Sections IV and V, respectively. After the discussion (Section VI), we conclude the article with limitations and future directions (Section VII).

II Background

In this section, we first discuss the distribution of pronouns and reflexives in English, Mandarin, and Korean, with experimental and corpus data from Mandarin and Korean. Prior studies on L2-Mandarin reflexives and L2-English pronouns are reviewed; we are unaware of any studies on overt object pronouns in L2-Mandarin.

1 The distribution of pronouns and reflexives in English, Mandarin and Korean

We define LD and local readings as follows: a reflexive or pronoun has a LD reading when it has an antecedent outside its immediate clause, and it has a local reading when its antecedent is inside the immediate clause. We first discuss reflexives in all three languages, and then move on to pronouns.

a Reflexives in English

In (1), the reflexive himself can refer to Peter, the local antecedent, but not to John, the LD antecedent. According to Binding Principle A (Chomsky, 1981), a reflexive must be bound in its binding domain, roughly a clause.¹

(1) John₁ thinks [Peter₂ trusts himself_*1/2].

Binding Principle A was once viewed as parameterized (Manzini and Wexler, 1987). However, given the presence of different reflexive types in a single language (see Sections II.1.b and II.1.c), this approach is no longer viable and will not be discussed further (for a critique, see Cole et al., 2006).

Binding Principle A has been challenged (e.g. Pollard and Sag, 1992; Reinhart and Reuland, 1993). For example, Principle A cannot explain why himself is able to take the LD antecedent Max in (2). According to Reinhart and Reuland (1993), antecedents and anaphors need to be co-arguments (for the purpose of this article, the subject and direct object) of a predicate. Thus, himself in (2) is not a syntactic anaphor but a logophor because it is contained inside a direct co-argument. Unlike syntactic anaphors, logophors impose a consciousness requirement on the antecedent, illustrated in (3) (Kuno, 1987: 123, cited in Charnavel et al., 2017). In (3a), John is clearly aware of the claim in the embedded clause (since he is the one making it); in contrast, John in (3b) is not aware of it, resulting in degraded acceptability.

(2) Max₁ boasted that the queen invited Lucie and himself₁ for a drink.

(3) a. John₁ said to Mary that physicists like himself₁ were a godsend.

b. ? Mary said to John₁ that physicists like himself₁ were a godsend.

b Reflexives in Mandarin

Unlike English, Mandarin has two types of reflexives, the morphologically complex reflexive taziji ‘himself/herself’ and the morphologically simplex reflexive ziji ‘self’. While taziji requires a local antecedent, as in (4), following Binding Principle A, ziji can take either a LD or a local antecedent, as in (5), apparently violating Binding Principle A.²

(4) Zhangsan₁ renwei [Lisi₂ xiangxin taziji_*1/2].

Zhangsan think Lisi trust himself/herself

‘Zhangsan thinks that Lisi trusts himself.’

(5) Zhangsan₁ renwei [Lisi₂ xiangxin ziji_1/2].

Zhangsan think Lisi trust self

‘Zhangsan thinks that Lisi trusts himself/him.’

There are a variety of approaches to LD reflexives in Mandarin. Currently, no consensus exists as to whether all LD reflexives are logophors (for an overview, see Charnavel et al., 2017, an updated version of Cole et al., 2006). We consider two main approaches: the non-uniform approach specific to ziji (Huang and Liu, 2001) vs. the co-argument approach (Reinhart and Reuland, 1993) applied to ziji (Reuland et al., 2020; Su, 2017). Under Huang and Liu’s (2001) approach, which is more dominant in the literature on ziji, ziji comes in two types: a syntactic anaphor when it is locally bound and a pragmatic logophor when it has a LD antecedent. The syntactic anaphor involves only syntax, while the pragmatic logophor involves the syntax-discourse interface. As with English logophors (see example (3)), the logophoric antecedent of ziji denotes ‘an individual conscious of the relevant event being reported’ (Huang and Liu, 2001: 159). Similarly, Charnavel (2019) advances a logophoricity-based hypothesis of LD ziji (e.g. LD interpretation is impossible without being the perspective center).

An alternative analysis is that ziji in (5) is a syntactic anaphor on both local and LD readings, following Reinhart and Reuland’s (1993) co-argument analysis; on this approach, ziji is a logophor only when it is not in a co-argument position (e.g. inside a possessive phrase). Wong (2021: 246) explicitly states that ‘there is no qualitative difference in the mechanisms used for local and nonlocal binding, since both are part of narrow syntax’ (see also Reuland et al., 2020).

In experimental studies, native speakers’ acceptance of LD readings of ziji reveals much variability, with the mean acceptance rates ranging from under 40% (Chien et al., 1993) to over 90% (Kong, 2011). Even the acceptance of local readings of ziji varies, ranging from under 70% (Chen, 2019) to 90% (Kong, 2011). Such speaker variability likely results from native speakers’ placing different weights on syntax vs. discourse factors (on Korean, see Kim and Yoon, 2020), as well as from different methodologies across studies, the verbs tested, dialects and/or contact with English (e.g. L1 attrition, Zhang, 2018); see Chen (2020b) for more discussion. For taziji, which is uncontroversially a syntactic anaphor, there is much less variability, with the acceptance of LD readings ranging from 9% (Chen, 2019) to 36.5% (Chien et al., 1993), and the acceptance of local readings typically higher than 90% (Chen, 2019; Chien et al., 1993; Kong, 2011).

In online/psycholinguistic studies, local readings of ziji are generally found to be the default option for Mandarin native speakers, since processing local antecedents is easier than processing LD antecedents (Chen et al., 2012; Dillon et al., 2016; Dillon et al., 2014; Jäger et al., 2015; Li and Kaiser, 2009; Li and Zhou, 2010; but, for an opposite finding, see Lu, 2011). This could be due to memory retrieval (since LD antecedents are linearly and structurally further away from the reflexives than local antecedents) (see Dillon, 2014) or to frequency. While Lu (2015) found that local readings of ziji are more frequent than LD readings (in a Taiwan-based corpus after 1980), Liu (2010) found the opposite (in two Chinese novels written in mainland Mandarin before 1950). The divergent results might result from different criteria applied to the corpus search (e.g. whether mono-clausal sentences are considered), different genres, dialects, time periods, etc. Given that mono-clausal sentences are presumably more frequent than bi-clausal sentences, and, by definition, only local readings are available in mono-clausal sentences, we assume that local readings of ziji should be more frequent when all occurrences of ziji are considered.

In sum, while much variability exists with regard to the behavior of ziji in judgments and corpus data, there is evidence suggesting that local readings are easier to process as well as more frequent than LD readings. These experimental results are in principle compatible with different accounts of ziji: whether LD-ziji is a logophor or a syntactic anaphor, it is still possible to explain the relative difficulty of processing LD-readings by appealing to memory and/or frequency considerations. For the remainder of the article, we adopt Huang and Liu’s (2001) non-uniform approach, given that this is the most influential account in the literature. This account is also best positioned to explain the variability in speakers’ acceptance of ziji, which may depend on the relative weights native speakers place on syntax vs. discourse factors (on Korean, see Kim and Yoon, 2020).

c Reflexives in Korean

Korean patterns with Mandarin in having LD reflexives, but has a three-way distinction: caki, casin, and caki-casin. In (6), caki can refer either to Peter or John, but prefers LD antecedents; furthermore, caki can only take third person antecedents.³ In (7), casin can refer either to Peter or John and does not have a strong preference; it can take antecedents in any person. While caki is not subject to blocking effects, casin seems to be like ziji (see footnote 2). In contrast, caki-casin requires a local antecedent, which is Peter in (8).

(6) John₁-un [Peter₂-ka caki_1/2-lul misnunta-ko] sayngkakhanta.

John-TOP Peter-NOM self-ACC trust-COMP thinks

‘John thinks that Peter trusts him/himself.’

(7) John₁-un [Peter₂-ka casin_1/2-ul misnunta-ko] sayngkakhanta.

John-TOP Peter-NOM self-ACC trust-COMP thinks

‘John thinks that Peter trusts him/himself.’

(8) John₁-un [Peter₂-ka caki-casin_*1/2-ul misnunta-ko] sayngkakhanta.

John-TOP Peter-NOM himself-ACC trust-COMP thinks

‘John thinks that Peter trusts himself.’

Like ziji, caki and casin apparently violate Binding Principle A, and no consensus regarding their status has been reached yet (for a review, see Madigan, 2015). However, unlike with ziji, the logophority analysis of caki no longer seems to be dominant (Kim and Yoon, 2020; Madigan, 2015). In experimental studies with native Korean speakers, the mean acceptance of LD vs. local readings of caki is over 75% vs. below 40%, respectively (Kim and Yoon, 2020; Kim et al., 2009; Lee, 2012), with a clear LD preference. The few existing analyses of casin treat it as a true LD reflexive, but as Madigan (2015) points out, many problems remain unsolved. With mean acceptance rates of LD vs. local readings of casin being 65% vs. 49% (Kim et al., 2009), the lack of an obvious preference toward either LD or local readings makes casin seems more similar to ziji than caki is. The existence of caki-casin as a local anaphor which obeys Binding Principle A is less controversial, with the mean acceptance of LD vs. local readings being 15% vs. 94% (Kim et al., 2009), similar to that with taziji. While the status of caki and casin is debated, crucially, caki and casin allow LD readings while caki-casin typically only allow local readings; and, unlike ziji, recent analyses do not treat caki as a logophor (Han and Storoshenko, 2012; Kim and Yoon, 2020).

Table 1 summarizes the distribution of reflexives in the three languages involved in this study and the dominant analyses of each form that we adopt for this study.

Table 1.

The distribution of reflexives in English, Mandarin, and Korean.

	English	Mandarin	Korean
Simplex reflexive: ziji, caki/casin	non-existent	both LD and local readings (LD ziji is a logophor, Huang and Liu, 2001; LD caki/casin are anaphors, Madigan, 2015)
Complex reflexive: himself/herself, taziji, caki-casin	only local readings

d Pronouns

We now move on to an overview of pronouns in the three languages. In (9), the pronoun him can refer to John, the LD antecedent, or someone else not mentioned in the sentence, but not to Peter, the local antecedent. Like English him/her, the Mandarin pronoun ta in (10) may refer to Zhangsan or someone else not mentioned in the sentence, but not to Lisi. The English and Mandarin facts are consistent with Binding Principle B (Chomsky, 1981), on which a pronoun must not be bound in its binding domain.

(9) John₁ thinks [Peter₂ trusts him_1/*2/3].

(10) Zhangsan₁ renwei [Lisi₂ xiangxin ta_1/*2/3].

Zhangsan think Lisi trust him/her

‘Zhangsan thinks that Lisi trusts him.’

As with standard Binding Principle A, standard Binding Principle B does not explain some well-known linguistic differences, e.g. binding vs. co-reference (Reinhart and Reuland, 1993). While the former is constrained by the syntax, the latter is regulated by discourse or pragmatics. The bound variable and coreferential readings of pronouns are indistinguishable with referential antecedents like proper names (11a), but only bound variable readings are available to pronouns with quantificational antecedents like every/no boy (11b), because such antecedents do not refer to specific individuals.

(11) a. John₁ said [that he₁ is smart].

b. Every/No boy₁ said [that he₁ is smart].

Grodzinsky and Reinhart (1993) propose Rule I, a discourse-based rule which requires syntax to take priority over pragmatics: a coreferential reading is impossible if a bound variable reading would result in the same interpretation. Thus, in (9), for example, him cannot be bound by Peter because this would violate Principle B; and it cannot be coreferential with Peter because Rule I prevents a coreferential reading that is identical to a bound variable reading. There are certain discourse contexts where the two readings can be teased apart, however, and where apparent violations of Principle B occur, but only with referential antecedents; on this account, these are not violations of Principle B, but rather coreferential readings. For example, the second pronoun him in (12) (taken from Reinhart, 1983, cited in Reuland, 2001: 448) refers to Bill, the local antecedent. The bound variable reading (x adores x, i.e. ‘self-adoration’) and the coreferential reading (x adores Bill, i.e. Bill-adoration) result in different interpretations. While the bound variable reading is ruled out by Principle B, the co-referential reading is not blocked by Rule I, since it is not identical to the bound variable reading. See also Chien and Wexler (1990), and much subsequent literature, on support for the binding/coreference distinction from child language.

(12) I know what Mary and Bill₁ have in common. Mary adores him₁ and Bill₁ adores him₁, too.

The bound variable/coreference distinction is also relevant for Korean pronouns (Kim, under review). Unlike other languages, Korean pronouns allow local readings (in addition to non-local ones), apparently violating Binding Principle B. In (13), ku/kunye ‘him/her’ can refer to Peter (the local antecedent) as well as to John (LD antecedent) or someone not mentioned in the sentence.

(13) John₁-un [Peter₂-ka ku_1/2/3-lul misnunta-ko] sayngkakhanta.

John-top Peter-nom him-acc trust-comp thinks

‘John thinks that Peter trusts him/himself.’

Kim (2019) experimentally showed that Korean pronouns have both LD and local readings, and that acceptance of neither reading is at ceiling: with referential antecedents, pronouns were accepted 77% of the time with LD antecedents and 69% of the time with local antecedents. This is a novel finding, since pronoun distribution has traditionally been found to not vary cross-linguistically. Interestingly, Kim (2019) found a referential/quantificational asymmetry for Korean pronouns: the local readings of kunye were allowed predominantly with referential rather than quantificational antecedents. Kim (under review) interprets this in the framework of Grodzinsky and Reinhart (1993), and argues that while Principle B is operative in Korean, Rule I is weak or inoperative. As a result, Korean permits coreferential readings for pronouns with referential antecedents; with quantificational antecedents, pronouns must be bound variables, and hence are subject to Principle B.

2 L2-acquisition of Mandarin LD reflexives

In this section, we first summarize and critique the most relevant offline/behavioral L2-studies on ziji. If a given study tested ziji in multiple contexts, we discuss only conditions where ziji is used in neutral bi-clausal sentences. We first present the findings, then move into the theoretical explanations.

a Reflexives in L2-Mandarin

So far, most L2-studies have found that L2-learners do not accept LD readings of ziji, at least not to the same degree as native speakers (e.g. Chen, 1995; Chen, 2018; Chen, 2019; Christie and Lantolf, 1998; Dugarova, 2007; Ying, 1999; Zeng, 2010; but, for some exceptions, see Kong, 2011; Sperlich 2013; Yuan, 1998). For example, Chen (1995) found that neither English nor French speakers acquired LD readings of ziji, though French soi is usually said to have LD readings (Everaert, 1991); however, the native Mandarin speakers in Chen (1995) also predominantly allowed only local readings of ziji, making it difficult to interpret the L2-results. Dugarova (2007) (a replication of Yuan, 1998) found that English speakers outperformed Russian speakers in accepting LD readings of ziji at every proficiency level, though neither English nor Russian has LD reflexives. Relatedly, Tsang (2009) found that L1-Tagalog L2-English L3-Cantonese learners disallowed LD readings of Cantonese reflexives even though Tagalog allows LD reflexives.

In contrast, Yuan (1998) found that Japanese speakers accepted LD readings of ziji and outperformed proficiency-matched English speakers, which is consistent with L1-transfer, since Japanese has LD reflexives. However, Yuan (1998) only had three tokens per condition, which is a methodological limitation. Sperlich (2013, see also 2016, 2017) found that Korean speakers outperformed English speakers in acquiring LD readings of ziji, attributed to the fact that Korean has LD reflexives while English does not. However, in the condition where ziji was used in neutral bi-clausal sentences, while the Korean group was native-like and the English group was not, the two L2-groups did not differ significantly. Given that Sperlich (2013) only had two tokens per condition, the results need to be interpreted with caution. We will revisit this study in Section VI, as the present study includes the same L1-groups as Sperlich (2013).

b Theoretical explanations: L1-transfer vs. default local readings with reflexives

While there are some differences among the findings of different studies, one basic pattern that emerges for the literature is that L2-Mandarin learners tend to overaccept the local readings of ziji and underaccept its LD readings; in contrast, native Mandarin speakers do allow both, despite variability, as discussed earlier.

One possible explanation is that local readings of ziji are the default option in L2-acquisition, independent of the L1. There are several reasons for local readings being the default. The first is that ziji is a syntactic anaphor when it has a local antecedent, but a logophor on its LD readings (Huang and Liu, 2001). Logophors fall at the syntax-discourse interface, which is particularly problematic for L2-learnеrs under the Interface Hypothesis (IH; Sorace, 2011; Tsimpli and Sorace, 2006). Studies on LD reflexives that explain their findings in terms of the IH include Chen (2018) on ziji as well as Umeda et al. (2017) and Yoshimura et al. (2012) on L2-Japanese zibun, and Kim (2013) on caki-casin and pronoun-casin in heritage Korean.

Additionally, the local readings of ziji may be preferred for reasons of processing and/or frequency. As discussed earlier, local readings of ziji appear to be more frequent relative to LD readings, and processing local readings of ziji is easier than processing LD readings (see Dillon, 2014). Related to the ease of processing is Hawkins’s (2004) proposal of ‘Minimize domains’, according to which the parser chooses the nearest antecedent for both syntactic and processing considerations (for work on LD reflexives in Turkish and Icelandic under this framework, see Gračanin-Yuksek et al., 2020; Putnam and Arnbjörnsdóttir, 2015). A similar concept, ‘minimal distance’, was proposed by Tsang (2009) to account for her finding that learners acquiring L3-Cantonese do not benefit from having an L1 that has LD reflexives. While the present study uses offline methodology, processing considerations may still influence learners’ judgments.

In sum, many different reasons (syntactic and processing-based preferences, frequency, the status of ziji as a logophor) may lead to L2-learners’ preference for local readings of ziji. These factors are not necessarily mutually exclusive, but may in fact conspire to make acquisition of the LD readings of ziji particularly difficult in the L2, regardless of the learners’ L1.

Alternatively, it is possible that the preference for local readings is affected by L1-transfer. On the Full Transfer/Full Access (FT/FA) Hypothesis (Schwartz and Sprouse, 1994, 1996), L2-learners fully transfer their L1-grammar to their L2 in the initial state and restructure their interlanguage grammar based on L2-input, and via access to Universal Grammar. In order to examine the role of L1-transfer with regard to ziji, two L2-Mandarin groups with different L1 reflexive systems need to be compared. Indeed, Yuan (1998) and Sperlich (2013) both found patterns consistent with L1-transfer in that L1-Japanese learners and L1-Korean learners outperformed their English-speaking counterparts in acquiring LD readings of ziji. White (2003) interpreted Yuan’s (1998) finding of L1-transfer with reflexives as supporting the FT/FA, but argued that the ambiguity of ziji makes L2-input insufficient to trigger the restructuring of L2-grammar. Given that recent analyses on caki do not treat it as a logophor constrained by pragmatics, we will assume lexically-based transfer instead of pragmatic transfer on reflexives.

c Interim summary

To sum up, prior studies yield inconsistent evidence regarding whether L1-transfer plays a role in the L2-acquisition of ziji. There are also reasons to treat local readings of ziji as the default, independently of the learners’ L1. Prior studies that looked at LD readings of ziji often lacked a baseline condition (failing to test taziji, e.g. Sperlich, 2013; Ying, 1999; Yuan, 1998; and/or local readings of ziji, e.g. Chen, 2018) and often had a very small number of tokens per condition (e.g. two in Chen, 2018 and Sperlich, 2013; three in Yuan, 1998). By comparing L1-Korean and L1-English L2-Mandarin learners on both ziji and taziji, we are able to tease apart L1-transfer from a default preference for local readings. Given that caki and casin in Korean are simplex reflexives that allow LD readings, just like ziji, L1-transfer should lead L1-Korean L2-Mandarin learners to allow LD readings of ziji. In contrast, if local readings are the default independently of the L1, then we expect the same performance from both L2-groups.

3 L2-acquisition of overt pronouns: a role for L1-transfer

For pronouns, most L2-studies focus on null vs. overt pronouns (e.g. L2-Japanese: Okuma, 2012), especially subject pronouns (e.g. L2-Mandarin: Li, 2014; Zhao, 2014). Given that there are no L2-Mandarin studies on overt object pronouns, we discuss some L2-English studies on object pronouns with local vs. LD antecedents. Both White (1998), a study with L1-French and L1-Japanese L2-English learners, and Slabakova et al. (2017), a study with L1-French and L1-Spanish L2-English learners, found these learners were largely target-like, disallowing local antecedents for her/him (one exception was the lower-proficiency learners in Slabakova et al., 2017) Testing pronouns (and reflexives) in various structures, Cook (1990) found L2-English learners from Romance, Japanese, and Norwegian L1s were all native-like on pronouns in bi-clausal sentences, correctly allowing only LD and not local antecedents.

Crucial to the present study is the finding that L1-Korean L2-English learners seem to differ from learners with other L1s in over-accepting local readings of pronouns (Kim, 2019, under review; Kim et al., 2015; Lee and Schachter, 1997). Testing both pronouns and reflexives, Lee and Schachter (1997) and Kim et al. (2015) found that these learners have more difficulty with pronouns than with reflexives, and overaccept local antecedents of English pronouns. Testing pronouns with both quantificational and referential antecedents, Kim (2019, under review) shows that L1-Korean L2-English learners over-accepted local readings of English pronouns. As discussed in Section II.1.d, Kim (2019) proposes that Korean pronouns allow local referential antecedents due to a weak or absent Rule I in Korean. Kim’s findings with L1-Korean L2-English learners indicate that the availability of local readings for pronouns transfers from Korean to English.

Thus, if L1-transfer is at work in the acquisition of L2-Mandarin object pronouns, we expect L1-Korean L2-Mandarin learners, but not L1-English L2-Mandarin learners, to overaccept local readings, if they assume that Mandarin, like Korean, lacks Rule I.

III Research questions and hypotheses

In light of the above discussion, we formulate the following two research questions:

• Can L2-learners acquire the properties of Mandarin pronouns and reflexives?

(a) Do they allow only LD readings for the pronoun ta?

(b) Do they allow both LD and local readings for the simplex reflexive ziji?

• Do L1-English and L1-Korean L2-learners of Mandarin differ in which readings they allow for Mandarin pronouns and reflexives, due to L1-transfer?

Starting with object pronouns, we follow Kim (2019) in assuming that Korean speakers allow local readings of Korean pronouns due to lack of Rule I. Kim (2019, under review) found that Korean speakers transfer the possibility of local antecedents for pronouns from Korean to English. We hypothesize that transfer of the possible readings of pronouns occurs in L2-Mandarin as well. Given that Korean, but not English or Mandarin, allows local readings of pronouns with referential antecedents, we predict that L1-English L2-Mandarin learners should have no difficulties accepting only LD readings of ta, whereas L1-Korean L2-Mandarin learners should accept both local and LD readings of ta.

For reflexives, we consider two possible hypotheses. On the L1-transfer hypothesis, L1-Korean L2-Mandarin learners should map complex reflexives in their L1 (caki-casin) to complex reflexives in their L2 (taziji), and simplex reflexives in their L1 (caki, casin) to simplex reflexives in their L2 (ziji). Such mapping would be made based on the superficial similarity in the reflexive forms in the two languages (simplex vs. complex). In contrast, L1-English L2-Mandarin learners have only complex reflexives (herself/himself) in their L1, and should therefore map this to both reflexive types in their L2. Such L1-transfer should lead L1-Korean L2-Mandarin learners to correctly allow LD readings for ziji, while L1-English L2-Mandarin learners should disallow such readings.

An alternative hypothesis is that local readings of reflexives are the default in L2-Mandarin, regardless of the L1. Under this hypothesis, no difference is expected between the two L2-groups, and both should allow only local readings of ziji. There are several reasons to expect the local readings of ziji to be the default. First, local readings have higher frequency and are easier to process. Second, if LD readings of ziji are indeed logophoric, the difficulty in acquiring them is predicted by the IH. Moreover, assuming that LD readings of caki (and perhaps also casin) are not logophors (see Madigan, 2015; for caki, see Kim and Yoon, 2020), L1-Korean L2-Mandarin learners might not map caki/casin to ziji. Admittedly, it is possible that both L1-transfer and local readings as the default work together. In the case of L1-Korean learners, this might result in some degree of transfer (in that this group would allow LD readings of ziji more than L1-English L2-Mandarin learners) but an overarching preference for local readings.

For taziji, given that the distribution of complex reflexives is cross-linguistically the same and receives similar analyses across languages, both L2-groups should correctly allow only local readings.

While L1-transfer is likely to occur with both pronouns and reflexives, we note the possibility of selective L1-transfer. That is, L1-transfer with pronouns does not necessarily entail L1-transfer with reflexives, or vice-versa. For pronouns, transfer of local readings from Korean to Mandarin results from Rule I not operating in Korean. In contrast, transfer of LD readings of reflexives from Korean to Mandarin results from mapping the particular meaning of caki and/or casin onto the meaning of ziji. Given the different processes, transfer of the meaning of pronouns could occur without corresponding transfer of the meaning of reflexives, or vice-versa.

Finally, as discussed in the next section, nearly all L1-Korean learners of Mandarin in our study learned English prior to learning Mandarin, making Mandarin their L3 (and in some cases, their L4). Thus, we address the possibility of transfer from the learners’ L2 to their L3 (for a recent review of morphosyntactic transfer in L3 acquisition, see Puig-Mayenco et al., 2020). Tsang (2009), as discussed above, found that L1-Tagalog L2-English L3-Cantonese learners disallowed LD readings of Cantonese reflexives, thus showing no L1-transfer from LD reflexives in Tagalog. Tsang (2009) interpreted the results as not involving transfer from L2-English, but this possibility cannot be discounted. If L1-Korean learners of Mandarin transfer the properties of pronouns and reflexives from English (rather than from Korean) to Mandarin, they should behave just like L1-English learners of Mandarin, allowing only LD readings for ta, and only local readings for taziji and ziji.

IV Methods

Participants completed a language background questionnaire in their native languages (Mandarin, Korean, and English), a picture-based Mandarin TVJT, and a Mandarin proficiency test; all tasks were presented on a computer screen.⁴ The Mandarin proficiency test was a multiple-choice test (five options per question), consisting of 24 vocabulary items (in which participants chose the correct English translations for Chinese words) and 16 cloze items (in which participants chose appropriate words to fill the blank in isolated sentences), for a maximum score of 40. The test items were taken from the unused questions in a question pool in a college-level Chinese placement test.

The experiment was untimed. For all tasks, native speakers of Taiwanese Mandarin were provided with test versions in traditional Chinese characters, while native speakers of mainland Mandarin were provided with test versions in simplified Chinese characters; this ensured that native speakers would complete the tasks in the script with which they were most familiar. To minimize effects of (low) literacy, the L2-groups were provided with four options: in addition to the two native-speaker versions with traditional and simplified Chinese characters, there were two versions in which traditional and simplified characters, respectively, were accompanied with Pinyin (Romanized script). While most learners selected to complete one of the Pinyin versions, a few L1-Korean L2-learners chose a version without Pinyin, likely due to familiarity with Chinese characters or higher reading proficiency.

1 Participants: Background and Mandarin proficiency

The participants included in the analysis were 62 Mandarin native speakers, 42 L1-Korean L2-Mandarin learners, and 32 L1-English L2-Mandarin learners, for a total of 136 participants; 13 additional participants were excluded based on language background and/or proficiency. To have a proficiency-matched comparison, only L2-learners who scored at least 25 out of 40 in the proficiency test were included. While the Korean group scored higher than the English group on average (34.4 vs. 32.8), the difference between the two L2-groups is only marginally significant (p = .08). To keep more L2-learners, we did not further exclude participants. Table 2 summarizes the participants’ background information and scores on the Mandarin proficiency test.

Table 2.

Participants’ demographic information and scores on the Mandarin proficiency test.

	Mandarin native speakers (n = 62)	L1-Korean L2-Mandarin learners (n = 42)	L1-English L2-Mandarin learners (n = 32)
Mean age (range)	22 (19–37)	25 (18–43)	28 (19–46)
Mean age of acquisition of Mandarin (in years) (range)	n/a	20. 2 (7–41)	18.8 (6–31)
Average years of Mandarin classes (range)	n/a	4.3 (0–15)	4 (1–13)
Mandarin proficiency test (max: 40) (range)	n/a	34.4 (25–40)	32.8 (25–40)

The L2-learners were either tested by the first author or completed the study remotely on their own. All Korean speakers were in China or Taiwan at the time of testing; all English speakers were in Taiwan or the U.S. We tested L2-learners in different locations to maximize the chance of recruiting more (advanced) L2-learners. Given that L2-learners were exposed to different varieties of Mandarin, we recruited Mandarin native speakers in China (n = 22) and Taiwan (n = 40) in case there are dialectal differences (see footnote 7). They were all born and raised in China or Taiwan, had not lived abroad for more than one year, had not been immersed in a bilingual environment like English-speaking international schools, and were not students of linguistics or Chinese pedagogy.

As indicated earlier, nearly all Korean speakers learned English before Mandarin, so Mandarin is their L3 (after L2-English). In contrast, no English speakers learned Korean before Mandarin (although one learned Korean at age 18 after learning Mandarin at age 14). For simplicity, we refer to the Korean group as L1-Korean L2-Mandarin learners; however, given that Mandarin is their L3 or L4, we will address the possibility of L2-English transfer to L3-Mandarin for this group.

2 TVJT materials

The picture-based TVJT was adapted from Chen (2019), which in turn was adapted from Kim et al. (2009). For each TVJT item, participants saw a picture and a sentence and had to indicate whether the picture and sentence matched or not by selecting TRUE or FALSE. Three anaphors (ta, ziji, and taziji) were tested. In Chinese characters, ta (with ‘human’ radical) is the default third-person singular pronoun that means both ‘he/him’ and ‘she/her’ while ta (with ‘female’ radical) can only mean ‘she/her.’ To avoid orthographical influence, the two antecedents (matrix subjects and embedded subjects) in each target sentence always had the same gender (by using English names that are typically male or female).⁵ Half of the sentences used male names and ta with ‘human’ radical; half used female names and ta with ‘female’ radical. The target sentence frames were always of the form ‘Name1 say Name2 Verb ta/ziji/taziji’. The matrix verb say was represented by a speech bubble in the picture. All the embedded verbs were transitive verbs, so that the pronoun/reflexive was always the direct object. Twelve verbs were used in the materials: kan ‘see’, qiao(dao) ‘hit/knock’, mo ‘touch’, yao ‘bite’, bao ‘hug’, guancha ‘observe’, xiao ‘laugh at’, geshang ‘cut’, zhi ‘point at’, tang ‘burn’, da ‘hit/slap’, and hua ‘draw’. All 12 verbs are included in the 5,000-word China-based New HSK vocabulary list (2012), a standardized Chinese proficiency test for non-native speakers.

The six target conditions were created by crossing two factors: ‘anaphor type’ (three levels) and ‘picture type’ (two levels: LD vs. local readings). A sample token set is in Table 3, along with predictions for native speakers. Pictures A and B have the same two characters (Ann and Betty), and Ann is the speaker outside of the speech bubble. Picture A shows Betty looking at Ann, while Picture B shows Betty looking at herself (in the mirror).

Table 3.

Sample token set, with predicted responses from native Mandarin speakers.

	Picture A (LD readings)	Picture B (local readings)

Ann shuo Betty zai kan ta. Ann say Betty progressive look.at her	TRUE	FALSE
Ann shuo Betty zai kan ziji. Ann say Betty progressive look.at self	TRUE (with some variability)	TRUE (with some variability)
Ann shuo Betty zai kan taziji. Ann say Betty progressive look.at herself	FALSE	TRUE

With six conditions and six tokens per condition, there were 36 target sentences in each experimental list. Each list also contained 24 distractors (items for a different experiment on relative clauses) and 24 filler sentences, for 84 items total; four practice items preceded the test. None of the distractors or fillers contained anaphors. The overall number of expected TRUE vs. FALSE responses from native speakers was balanced across all test items.

Two lists were created. In each list, each picture was repeated three times, with ta, ziji, and taziji, respectively; for each token set, the three anaphors showed up with the ‘local’ picture in one list and the ‘LD’ picture in the other, with full counterbalancing across token sets. To avoid trials from a single condition from occurring consecutively, items were blocked and pseudorandomized within each block. A potential concern with this study is that, in each list, each picture was repeated three times.⁶ This could potentially give rise to priming effects, with participants choosing TRUE responses more often after reading similar sentences. The blocks were randomized for each participant to lessen priming effects and to discourage participants from comparing different anaphor forms. Participants saw one item at a time on the computer screen and could not return to the previous page. Therefore, they could not change their answers in the previous items.

3 Predictions

For native speakers, the predicted responses are shown in Table 3. Given prior literature, some variability with ziji was expected.

The main predictions for L2-learners are based on L1-transfer, summarized in Table 4. For pronouns, only Korean speakers, but not English speakers, would incorrectly give TRUE responses to local readings of ta. For ziji, only Korean speakers would correctly give TRUE responses to LD readings while English speakers would not. For taziji, both L2-groups would give TRUE responses only to local readings, as there are no cross-linguistic differences. Given that most (possibly all) of the Korean speakers in our study acquired English before Mandarin, transfer from English to Mandarin is possible; if that is the case, Korean speakers should pattern with English speakers on both pronouns and reflexives in Mandarin.

Table 4.

Predictions for L2-learners under L1-transfer.

	L1-Korean learners of Mandarin under transfer from Korean	L1-English learners of Mandarin under transfer from English (and L1-Korean learners of Mandarin under transfer from their L2-English)
LD readings of ta	TRUE	TRUE
Local readings of ta	TRUE*	FALSE
LD readings of ziji	TRUE	FALSE*
Local readings of ziji	TRUE	TRUE
LD readings of taziji	FALSE	FALSE
Local readings of taziji	TRUE	TRUE

Notes. * Non-target-like responses.

Alternatively, if local readings of ziji are the default, both L2-groups would perform similarly on ziji, accepting local and rejecting LD readings; however, the two groups may still differ on ta in that case, since transfer of pronouns is unaffected by the default readings of reflexives. Finally, higher-proficiency L2-learners who have recovered from L1-transfer and have become target-like on both reflexives and pronouns are expected to pattern with native speakers (see Table 3).

V Results

1 Group results

Following Kim et al. (2009) and Chen (2019), a TRUE response was assigned a score of ‘1’ and a FALSE response a score of ‘0’. Then, the raw scores were averaged across the participants in each group and converted to percentages. Figure 1 shows the descriptive group results on the mean acceptance of LD and local readings (error bars show standard error).⁷

Figure 1.

Mean acceptance of long-distance (LD) and local readings by group (in percentages).

A logistic mixed-effects model (Jaeger, 2008) was conducted via glmer in R (R Core Team, 2019) since the dependent variable was binary. The fixed effects are the independent variables, i.e. group, anaphor, and antecedent, and their interactions. The dependent variable is the response (coded as 1 vs. 0). Dummy coding was used. The reference level is the native speakers for group, taziji for anaphor. The random effects include random intercepts for subjects and items. The overall main effect was assessed via ANOVA. There were significant main effects of anaphor (χ² (2) = 288.90, p < .0001), antecedent (χ² (1) = 138.57, p < .0001), and group (χ² (2) = 8.22, p = .02); the two-way and three-way interactions were all significant; see Table 5 for the model output.

Table 5.

Results from the logistic mixed-effects model.

	Chisq	df	Pr(>Chisq)
(Intercept)	141.5622	1	< .0001*
Anaphor	288.8961	2	< .0001*
Antecedent	138.5726	1	< .0001*
Group	8.2156	2	.016*
Anaphor × antecedent	363.7818	2	< .0001*
Anaphor × group	71.5869	4	< .0001*
Antecedent × group	20.5944	2	< .0001*
Anaphor × antecedent × group	103.3020	4	< .0001*

Note. * p < .05.

Significant interactions were followed up with pairwise comparisons via emmeans (Lenth et al., 2019); the p-values are significant at the Tukey-adjusted alpha level of .05. The between-group differences in the six target conditions are reported below. On both LD and local readings of ta, native speakers pattern with English speakers, but differ from Korean speakers (native vs. Korean on LD readings: z = 5.01, p = .0001; native vs. Korean on local readings: z = −7.54, p < .0001). Compared to English speakers, Korean speakers allow significantly more local readings of ta (z = 5.78, p < .0001) and numerically (but not significantly) fewer LD readings. For ziji, compared to native speakers, both L2-groups accept significantly fewer LD readings (native vs. Korean: z = 6.87, p < .0001; native vs. English: z = 7.87, p < .0001) and accept significantly more local readings (native vs. Korean: z = −5.77, p < .0001; native vs. English: z = −5.38, p < .0001). The two L2-groups pattern similarly on both LD and local readings of ziji, with Korean speakers accepting numerically (but not significantly) more LD readings than English speakers. Finally, on both LD and local readings of taziji, native speakers pattern with Korean speakers, and the two L2-groups also pattern together. Yet, English speakers differ from native speakers in accepting significantly fewer local readings (z = 3.51, p = .048) and accepting numerically (but not significantly) more LD reading of taziji.

2 Proficiency effects

In order to examine whether learners become more target-like with proficiency, we conducted additional analyses with the L2-data only (excluding native speakers). An initial logistic mixed-effects model that included the fixed effects of anaphor, group, antecedent, and proficiency as a covariate, and their interactions, as well as both participant and item as random effects, resulted in convergence issues. The model converged after both participant and item random effects were removed. The output of a logistic model (via glm in R) indicates that group (χ² (1) = 4.47, p = .04), antecedent (χ² (1) = 14.58, p = .0001), and proficiency (χ² (1) = 6.69, p = .01) (but not anaphor) are significant, and that all the two-way, three-way, and four-way interactions are significant except for the three-way interaction among anaphor, group and proficiency; see Table 6. Significant interactions were followed up with pairwise comparisons via emtrends (since proficiency is a continuous predictor variable). Results show that there is a significant change with proficiency to the Korean-English contrast for four of the conditions: LD readings of ziji (z = −2.68, p = .007), LD readings of taziji (z = −2.42, p = .016), local readings of ziji (z = 2.34, p = .019), and local readings of taziji (z = 2.32, p = .020); for the other two conditions (LD and local readings of ta), the changes with proficiency affect the two groups about the same.

Table 6.

Results from the logistic model (L2-groups only, with proficiency as a covariate).

	Chisq	df	Pr(>Chisq)
Anaphor	3.363	2	0.186
Antecedent	14.584	1	0.0001*
Group	4.466	1	0.035*
Proficiency	6.692	1	0.010*
Anaphor × antecedent	20.810	2	p < .0001*
Anaphor × group	6.259	2	0.044*
Antecedent × group	9.695	1	0.002*
Anaphor × proficiency	9.755	2	0.008*
Antecedent × proficiency	29.431	1	p < .0001*
Group × proficiency	5.782	1	0.016*
Anaphor × antecedent × group	19.436	2	p < .0001*
Anaphor × antecedent × proficiency	41.484	2	p < .0001*
Anaphor × group × proficiency	4.517	2	0.105
Antecedent × group × proficiency	11.042	1	0.001*
Anaphor × antecedent × group × proficiency	14.006	2	0.001*

Note. * p < .05.

A visualization of these contrasts is given in Figure 2, which plots the proficiency scores against the probability of allowing LD vs. local readings for each anaphor type and for each group. Proficiency effects are most evident with ta in the Korean group (more proficient learners allow local readings less and allow LD readings more), ziji in the English group (more proficient learners allow LD readings more and allow local readings less), and taziji in the Korean group (less proficient learners, those scoring below 30 out of 40, allow local readings less and allow LD readings more, relative to high-proficiency learners).

Figure 2.

L2-learners’ trends on the probability of having long-distance (LD) and local readings of anaphors by proficiency.

VI Discussion

We discuss the results by native speakers (only on ziji due to variability) before the results by L2-learners. We focus on the interpretation of ziji by both L2-groups, ta by the Korean group, and taziji by the English group, as these are the areas where we see non-target performance.

1 Interpretations of ziji by native speakers

Consistent with prior studies (e.g. Chen, 2019; Chien et al., 1993), the present study finds below-ceiling performance on both local and LD readings of ziji among native speakers. The patterns across studies look somewhat different regarding which readings are more acceptable for native speakers. These differences are probably traceable to methodological differences and/or dialectal variation (see Chen, 2020b for more discussion). Crucially, while variability exists, native speakers do allow both local and LD readings of ziji, and treat ziji differently from ta (only LD readings) and taziji (predominantly local readings).

2 Interpretations of ziji by L2-learners

Compared to native speakers, both L2-groups in this study allowed significantly fewer LD readings of ziji, consistent with some previous studies (e.g. Chen, 2019), but differing from the Japanese speakers in Yuan (1998) and the Korean speakers in Sperlich (2013). The two L2-groups also allowed local readings of ziji significantly more than native speakers.

While Korean speakers accepted more LD readings of ziji than English speakers, the difference was not significant. Thus, there is no clear evidence of L1-transfer of LD reflexives for the Korean group, despite LD reflexives existing in Korean (both caki and casin are accepted with LD readings by native Korean speakers: 93% for caki vs. 65% for casin, Kim et al., 2009). Furthermore, the Korean group had a very high acceptance rate for local readings of ziji (89.7%) even though the local readings of Korean simplex reflexives tend to be quite low (around 40% for caki; Kim et al., 2009; Lee, 2012; 49% for casin; Kim et al., 2009). The Korean learners of Mandarin in our study did not appear to map caki or casin to ziji.

The highly similar performance of the English and Korean groups on ziji speaks against L1-transfer from Korean LD reflexives, and in favor of local binding being the default option for reflexives in L2-Mandarin. That is, both groups appear to start with the assumption that ziji, like taziji, requires local antecedents. While English speakers start to acquire LD readings with increased proficiency, Korean speakers do not; see Figure 2. We do not have an explanation as to why only English (and not Korean) speakers improve with proficiency. As for why L2-learners in general fail to acquire LD readings of ziji, it may be because LD readings are fairly infrequent or at least not consistently available (since native speakers have variable judgments). Another explanation is that L2-learners have difficulty acquiring LD ziji because it is a logophor at the syntax-discourse interface (compare with the IH; Sorace, 2011).

Results from the present study differ from Sperlich (2013), which found that Korean speakers outperformed English speakers in acquiring LD readings of ziji. However, it is possible that the transfer effects were smaller than what Sperlich (2013) claimed. As mentioned before, the two L2-groups in Sperlich (2013) did not differ significantly on the condition which corresponds most closely to our study, i.e. where ziji is used in neutral bi-clausal embedded sentences, even though the Korean group was native-like and the English group was not. Additionally, Sperlich (2013) only had two tokens per condition.

As discussed before, most Korean speakers in our study learned English before Mandarin, while no English speakers learned Korean before Mandarin. The similar behavior of the two groups could therefore potentially be due to transfer from English in both groups. However, if both groups transfer primarily from English (and not Korean in the case of the Korean group), we would expect English transfer for pronouns and reflexives. As discussed next, L1-Korean L2-Mandarin learners do show clear Korean transfer for pronouns. We cannot rule out the possibility that the Korean speakers transfer the properties of pronouns from Korean, yet transfer the properties of reflexives from English. However, we believe that this is not very likely, considering that Korean is closer to Mandarin than English is in having both simplex and complex reflexives. We therefore conclude that the Korean group’s performance on ziji is better explained by local readings being the default, and not by transfer from their previously learned L2-English. However, this conclusion remains tentative, given that we did not test this group in their L2-English.⁸

3 Interpretations of ta by L1-Korean L2-Mandarin learners

As predicted under L1-transfer, only Korean speakers (but not English speakers) differed from Mandarin native speakers on local (and LD) readings of ta. Compared to English speakers, Korean speakers allowed significantly more local readings of ta. The current finding is fully consistent with the findings of Kim (2019, under review) that Korean native speakers allow local readings of pronouns in both their native Korean and their L2-English. The present study finds a similar pattern in that L1-Korean L2-Mandarin learners allow local readings of pronouns in Mandarin; since no such pattern is found in the L1-English group, it is most likely due to transfer from Korean. Note that Korean speakers’ non-target behavior on ta cannot be due to low proficiency in Mandarin, since the L2-groups are proficiency-matched, with Korean speakers having slightly higher proficiency. The Korean group’s performance on ta suggests that the L1-Korean group was transferring from Korean rather than English, despite having studied English prior to Mandarin.

4 Interpretations of taziji by L1-English L2-Mandarin learners

As predicted, both L2-groups were quite target-like on taziji. However, English speakers accepted numerically more LD readings of taziji than native speakers (and Korean speakers) and accepted significantly fewer local readings than native speakers. Indeed, previous studies have shown that some English speakers accept LD readings of taziji (Chen, 2019; Zeng, 2010) and ziji more than native speakers (Kong, 2011); these authors speculated that this is possibly due to (mis)analysis of reflexives as the pronoun ta. However, given that the mean acceptance for local readings of taziji was above 90% in all groups in our study, it seems unlikely that the learners misanalysed taziji as ta, at least not at the group level. There are at least two potential reasons for the divergent findings between studies. First, our L2-learners may have higher Mandarin proficiency and therefore did not misanalyse reflexives as ta, given that it was the less proficient group(s) that accepted LD readings of taziji (Zeng, 2010) and ziji (Kong, 2011); this said, however, we did not see proficiency effects in Figure 2 with taziji. Second, the inclusion of ta in the present study may help learners to differentiate reflexives from pronouns, preventing misanalysis. These two possibilities remain speculative due to differences in proficiency tests and data collection methods across studies.

VII Conclusions, limitations, and future direction

The inclusion of Mandarin, Korean, and English in our study allowed us to fully investigate the role of L1-transfer: Mandarin behaves like English on pronouns, but like Korean on reflexives. For pronouns, compared to English speakers, Korean speakers over-accepted local readings due to L1-transfer. For reflexives, both L2-groups predominantly allowed only local readings. For English speakers, the performance with reflexives can be explained by either L1-transfer or by local readings being the default option, but for Korean speakers, L1-transfer of reflexives is not operative, or is at least overridden by the default local readings. In short, results show clear transfer of Korean pronouns, but not of Korean LD reflexives.

We now address why the local readings of pronouns, but not the LD readings of reflexives, are transferred from Korean. While pronouns and reflexives are in complementary distribution in classic Binding Theory (Chomsky, 1981), they might actually be different phenomena, subject to language-specific constraints. While local binding of reflexives is dictated by syntax, whether LD readings of simplex reflexives are constrained by syntax, semantics, and/or pragmatics is still subject to debate (for overviews, see Charnavel et al., 2017; Madigan, 2015). In contrast, pronouns are governed by Binding Principle B in syntax and Rule I in pragmatics, with Rule I being weak or absent in Korean (Kim, under review). If these recent analyses are on the right track, then there is no principled reason to expect L1-transfer to work similarly for LD reflexives vs. for pronouns. With regard to the latter, our study provides clear evidence that Korean speakers transfer the local readings of pronouns from Korean to Mandarin; following Kim (under review), we argue that what they transfer is at the level of pragmatics (weak or inoperative Rule I), which leads learners to permit local referential readings that do not violate Principle B. One possibility is that this prolonged transfer of local readings of pronouns is due to the lack of negative evidence in L2-input. The fact that local readings of pronouns are disallowed in Mandarin is unlikely to be taught in class and cannot be derived from positive evidence alone.

At the same time, positive evidence did not help either Korean speakers or English speakers to successfully acquire LD readings of Mandarin ziji, possibly due to low frequency of LD readings in the input (e.g. Lu, 2015). As for lack of L1-Korean transfer with reflexives, one possibility is that the Korean caki receives a different analysis than the Mandarin ziji (e.g. Cole et al., 1990; Han and Storoshenko, 2012: 787), and learners therefore do not map them to each other. As discussed earlier, ziji is often analysed as a logophor when it takes a LD antecedent (Charnavel, 2019; Huang and Liu, 2001; but see Reuland et al., 2020; Wong, 2021) while caki is not (Han and Storoshenko, 2012; Kim and Yoon, 2020). Casin remains understudied (see Kim and Yoon, 2020: footnote 6; Madigan, 2015). Many scholars have noted differences between caki and ziji, with ziji behaving much more like casin than like caki (e.g. caki can only take third-person antecedents while casin, ziji, and Japanese zibun can take antecedents in any person; caki is not subject to blocking effects while ziji is, and casin seems to be). Thus, it is conceivable that Korean speakers map casin rather than caki (despite caki being much more common) onto ziji. Given that both casin and ziji lack a strong preference toward LD readings (unlike caki), it is possible that transfer from casin is largely overridden by the frequency of local readings of ziji in the input.

One limitation of the present study is that most Korean speakers learned English before Mandarin, while none of the English speakers learned Korean before Mandarin. We have argued that English transfer with reflexives is not very likely to be at work in the Korean group, given their performance on Mandarin pronouns that shows clear L1-Korean transfer. However, in future research, it would be fruitful to test Korean learners of Mandarin on both English and Mandarin, in order to more definitively examine the possibility of English transfer. Another direction would be to compare Japanese and Korean learners of Mandarin, as both learner groups would likely have learned English before Mandarin. Under L1-transfer, the Japanese group, unlike the Korean group, should be target-like on ta, since Japanese disallows local readings of pronouns; furthermore, prior studies with L1-Japanese L2-English learners have found target-like performance on English pronouns (Cook, 1990; White, 1998), unlike the corresponding studies with L1-Korean L2-English learners (Kim, under review; Lee and Schachter, 1997). As for reflexives, if L1-transfer is at work, we would expect Japanese speakers to acquire LD readings of ziji due to the existence of zibun, which allows LD readings (like Yuan, 1998). If there is a difference between Japanese and Korean speakers on ziji, we should further investigate why there is transfer from zibun but not from caki/casin to ziji.

To sum up, the present study examined whether L1-English and L1-Korean L2-Mandarin learners acquire the properties of Mandarin pronouns and reflexives despite differences from their L1s, and whether the two L2-groups differ due to L1-transfer. The results show that L1-Korean L2-Mandarin learners transfer local readings of pronouns to Mandarin, but not LD readings of reflexives. For ziji, local readings are strongly preferred by both groups, suggesting that the default status of local readings overrides L1-transfer. This study contributes to our understanding of L2-interpretations of pronouns and reflexives and shows that transfer is selective and does not operate across the board even in seemingly similar phenomena.

Footnotes

Acknowledgements

We are grateful to the audiences of the 2019 Generative Approaches to Second Language Acquisition Conference (GASLA) and the 2019 Boston University Conference on Language Development (BUCLD), where parts of this work were presented, for their comments and suggestions. Thanks to Joy Jen and Watson Chen for help with written stimuli creation, and Ajin for visual stimuli creation. We thank the three anonymous reviewers and the editors for their detailed comments and suggestions. All remaining errors are our own.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Science Foundation (BCS-1844447), a UIUC Graduate College Dissertation Travel Grant, and a UIUC SLATE Doctoral Research Award.

ORCID iDs

Chung-yu Chen

Tania Ionin

Notes

References

Baker

(1995) Contrast, discourse prominence, and intensification, with special reference to locally free reflexives in British English. Language 71: 63–101.

Charnavel

(2019) Locality and logophoricity: A theory of exempt anaphora. Oxford: Oxford University Press.

Charnavel

Huang

CTJ

Cole

Hermon

(2017) Long-distance anaphora: Syntax and discourse. In: Everaert

van Riemsdijk

(eds) The Wiley Blackwell companion to syntax. 2nd edition. Oxford: Wiley Blackwell.

Chen

(1995) Mandarin reflexive ziji in second language acquisition. University of Pennsylvania Working Papers in Linguistics 2: 37–52.

Chen

(2019) The acquisition of Mandarin reflexives by heritage speakers and second language learners. In: Rispoli

Ionin

(eds) Three streams of generative language acquisition research. Amsterdam: John Benjamins, pp. 225–51.

Chen

(2020a) The acquisition of Mandarin relative clauses and binding by heritage speakers and second language learners. In: Brown

Kohut

(eds) Proceedings of the 44th Boston University Conference on Language Development. Somerville, MA: Cascadilla Press, pp. 77–90.

Chen

(2020b) An experimental investigation of Mandarin reflexive ziji. Unpublished paper presented in the 32th North American Conference on Chinese Linguistics (NACCL-32), 18–20 September, University of Connecticut, Storrs, CT, USA.

Chen

(2020c) The acquisition of Mandarin by heritage speakers and second language learners. Unpublished PhD thesis, University of Illinois at Urbana-Champaign, USA.

Chen

(2018) Comprehension and production of reflexives by second language learners of Chinese and English. Unpublished MA thesis, National Tsing Hua University, Taiwan.

10.

Chen

Jäger

Vasishth

(2012) How structure-sensitive is the parser? Evidence from Mandarin Chinese. In: Stolterfoht

Featherston

(eds) Empirical approaches to linguistic theory: Studies of meaning and structure. Berlin: Mouton de Gruyter, pp. 43–62.

11.

Chien

Wexler

(1990) Children’s knowledge of locality conditions in binding as evidence for the modularity of syntax and pragmatics. Language Acquisition 1: 225–95.

12.

Chien

Wexler

Chang

(1993) Children’s development of long-distance binding in Chinese. Journal of East Asian Linguistics 2: 229–59.

13.

Chomsky

(1981) Lectures on government and binding. Dordrecht: Foris.

14.

Christie

Lantolf

(1998) Bind me up bind me down: Reflexives in L2. In: Flynn

Martohardjono

O’Neil

(eds) The generative study of second language acquisition. Mahwah, NJ: Erlbaum, pp. 239–60.

15.

Cole

Hermon

Sung

(1990) Principles and parameters of long-distance reflexives. Linguistic Inquiry 21: 1–22.

16.

Cole

Hermon

Huang

CTJ

(2006) Long-distance binding in Asian languages. In: Everaert

van Riemsdijk

(eds) The Blackwell companion to syntax. Oxford: Blackwell Publishing, pp. 21–84.

17.

Cook

(1990) Timed comprehension of binding in advanced L2 learners of English. Language Learning 40: 557–99.

18.

Dillon

(2014) Syntactic memory in the comprehension of reflexive dependencies: An overview. Language and Linguistics Compass 8: 171–87.

19.

Dillon

Chow

Xiang

(2016) The relationship between anaphor features and antecedent retrieval: Comparing Mandarin ziji and ta-ziji. Frontiers in Psychology 6: 1966.

20.

Dillon

Chow

Wagers

, et al. (2014) The structure-sensitivity of memory access: evidence from Mandarin Chinese. Frontiers in Psychology 5: 1025.

21.

Dugarova

(2007) Acquisition of the Chinese reflexive ‘ziji’ by Russian and English speakers. In: Hilton

Arscott

Barden

, et al. (eds) Proceedings of CamLing 2007. Cambridge: Cambridge Institute of Language Research, pp. 48–55.

22.

Everaert

(1991) Contextual determination of the anaphor/pronominal distinction. In: Koster

Reuland

(eds) Long-distance anaphor. Cambridge: Cambridge University Press, pp. 77–118.

23.

Gračanin-Yuksek

Lago

Şafak

Demir

Kırkıcı

(2020) The interpretation of syntactically unconstrained anaphors in Turkish heritage speakers. Second Language Research, 36: 475–501.

24.

Grodzinsky

Reinhart

(1993) The innateness of binding and coreference. Linguistic Inquiry 24: 69–101.

25.

Han

Storoshenko

(2012) Semantic binding of long-distance anaphor caki in Korean. Language 88: 764–90.

26.

Hawkins

(2004) Efficiency and complexity in grammars. Oxford: Oxford University Press.

27.

Huang

CTJ

CSL

Liu

(2001) Logophoricity, attitudes, and ziji at the interface. In: Cole

Hermon

Huang

CTJ

(eds) Long-distance reflexives. New York: Academic Press, pp. 141–95.

28.

Jaeger

(2008) Categorical data analysis: Away from ANOVAs (transformation or not) and towards logit mixed models. Journal of Memory and Language 59: 434–46.

29.

Jäger

Engelmann

Vasishth

(2015) Retrieval interference in reflexive processing: experimental evidence from Mandarin, and computational modeling. Frontiers in Psychology 6: 17.

30.

Kehler

Rohde

(2019) Prominence and coherence in a Bayesian theory of pronoun interpretation. Journal of Pragmatics 154: 63–78.

31.

Kim

(2018) Pronoun interpretation with referential and quantificational antecedents in SLA. In: Fukuda

Kim

Park

(eds) Proceedings of the 25th Japanese/Korean linguistics conference. Stanford, CA: CSLI Publications.

32.

Kim

(2019) Interpretation and processing of overt pronouns in Korean, English and L2-acquisition. Unpublished PhD thesis, University of Illinois at Urbana-Champaign, Champaign, IL, USA.

33.

Kim

(under review) L1-transfer effects and the role of computational complexity in second language pronoun interpretation.

34.

Kim

Yoon

(2020) Experimental evidence supporting the overlapping distribution of core and exempt anaphors: Re-examination of long-distance bound caki-casin in Korean. Linguistics 58: 1775–1806.

35.

Kim

Montrul

Yoon

(2015) The on-line processing of binding principles in second language acquisition: Evidence from eye tracking. Applied Psycholinguistics 36: 1317–74.

36.

Kim

(2013) Testing the interface vulnerability hypothesis with Korean heritage speakers. Korean Journal of Applied Linguistics 29: 49–72.

37.

Kim

Montrul

Yoon

(2009) Binding interpretations of anaphors by Korean heritage speakers. Language Acquisition 16: 3–35.

38.

Kong

(2011) Monomorphemic and polymorphemic anaphors in L2 Chinese. Journal of Chinese Linguistics 39: 451–77.

39.

Kuno

(1987) Functional syntax: Anaphora, discourse and empathy. Chicago, IL: University of Chicago Press.

40.

Lardiere

(2008) Feature assembly in second language acquisition. In: Liceras

Zobl

Goodluck

(eds) The role of formal features in second language acquisition. New York: Lawrence Erlbaum, pp. 106–40.

41.

Lardiere

(2009) Some thoughts on the contrastive analysis of features in second language acquisition. Second Language Research 25: 173–227.

42.

Lee

Schachter

(1997) Sensitive period effects in binding theory. Language Acquisition 6: 333–62.

43.

Lee

(2012) The acquisition of Korean reflexive caki by heritage speakers and L2 learners. Korean Journal of Linguistics 37: 383–400.

44.

Lenth

Singmann

Love

Buerkner

Herve

(2019) Emmeans: Estimated marginal means, aka least-squares means: R Package Version 1.3.4 [software]. Available at: https://cran.r-project.org/web/packages/emmeans (May 2022).

45.

DCH

Kaiser

(2009) Overcoming structural preference: Effects of context on the interpretation of the Chinese reflexive ziji. In: Devi

Branco

Mitkov

(eds) Proceedings of the 7th Discourse Anaphora and Anaphor Resolution Colloquium (DAARC 2009), pp. 64–72.

46.

Zhou

(2010) Who is ziji? ERP responses to the Mandarin reflexive pronoun during sentence comprehension. Brain Research 1331: 96–104.

47.

(2014) Variation in subject pronominal expression in L2 Chinese. Studies in Second Language Acquisition 36: 39–68.

48.

Liu

(2010) A pragmatic account of anaphora: the cases of the bare reflexive in Chinese. Journal of Language Teaching and Research 1: 796–806.

49.

(2011) The effects of verb bias, context and tasks on Mandarin reflexives. Unpublished PhD thesis, University of Illinois at Urbana-Champaign, Champaign, IL, USA.

50.

(2015) ‘Self in reality’: A corpus study of Mandarin reflexive ‘ziji’. Unpublished manuscript, National Taipei University of Education, Taipei, Taiwan.

51.

Madigan

(2015) Anaphora and binding. In Brown

Yeon

(eds) The handbook of Korean Linguistics, pp. 137–54.

52.

Manzini

Wexler

(1987) Parameters, Binding Theory, and learnability. Linguistic Inquiry 18: 413–44.

53.

New HSK vocabulary [Xin Hanyu Shuiping Kaoshi (HSK) cihui] (2012) Chinese Testing International. Available at: http://www.chinesetest.cn/godownload.do (accessed May 2022).

54.

Okuma

(2012) L2 acquisition of discourse constraints on the use of Japanese pronouns. In: Biller

Chung

Kimball

(eds) The 36th Annual Boston University Conference on Language Development (BUCLD36) Online Proceedings Supplement. Boston, MA: BUCLD.

55.

Pan

(1998) Closeness, prominence, and binding theory. Natural Language and Linguistic Theory 16: 771–815.

56.

Pollard

Sag

(1992) Anaphors in English and the scope of binding theory. Linguistic Inquiry 23: 261–303.

57.

Puig-Mayenco

González Alonso

Rothman

(2020) A systematic review of transfer studies in third language acquisition. Second Language Research 36: 31–64.

58.

Putnam

Arnbjörnsdóttir

(2015) Minimizing (interface) domains: The loss of long-distance binding in North American Icelandic. In: Page

Putnam

(eds) Moribund Germanic heritage languages in North America: Theoretical perspectives and empirical findings. Leiden: Brill, pp. 203–23.

59.

R Core Team (2019) R: A language and environment for statistical computing [software]. Vienna: R Foundation for Statistical Computing. Available at: https://www.r-project.org (accessed May 2022).

60.

Reinhart

(1983) Anaphora and semantic interpretation. London: Croom Helm.

61.

Reinhart

Reuland

(1993) Reflexivity. Linguistic Inquiry 24: 657–720.

62.

Reuland

(2001) Primitives of binding. Linguistic Inquiry 32: 439–92.

63.

Reuland

Wong

SCH

Everaert

(2020) How the complexity of Mandarin zi-ji simplifies the grammar. Linguistic Inquiry 51: 799–814.

64.

Schwartz

Sprouse

(1994) Word order and nominative case in non-native language-acquisition: A longitudinal study of (L1 Turkish) German interlanguage. In: Hoekstra

Schwartz

(eds) Language acquisition studies in generative grammar: Papers in honor of Kenneth Wexler from the 1991 GLOW workshops. Amsterdam: John Benjamins, pp. 317–68.

65.

Schwartz

Sprouse

(1996) L2 cognitive states and the full transfer/full access model. Second Language Research 12: 40–72.

66.

Slabakova

White

Guzzo

(2017) Pronoun interpretation in the second language: Effects of computational complexity. Frontiers in Psychology 8: 1236.

67.

Sorace

(2011) Pinning down the concept of ‘interface’ in bilingualism. Linguistic Approaches to Bilingualism 1: 1–33.

68.

Sperlich

(2013) The acquisition of long-distance reflexives in Chinese as an interlanguage: An experimental study. Unpublished PhD thesis, University of Auckland, Auckland, New Zealand.

69.

Sperlich

(2016) Pragmatic or syntactic ziji? Evidence from language transfer. Journal of Cognitive Science 174: 607–52.

70.

Sperlich

(2017) The instantiation of binding through pragmatic and syntactic processes. In: Kecskés

Sun

(eds) Key issues in Chinese as a second language research, pp. 188–213.

71.

(2017) Rethinking parameter setting from locality and orientation constraints for ziji in Mandarin. Unpublished paper presented in the 1st International Conference on Theoretical East Asian Psycholinguistics, Chinese University of Hong Kong, Hong Kong, China.

72.

Tsimpli

Sorace

(2006) Differentiating interfaces: L2 performance in syntax-semantics and syntax-discourse phenomena. In: Bamman

Magnitskaia

Zaller

(eds) Proceedings of the Annual Boston University Conference on Language Development 30. Boston, MA: BUCLD.

73.

Tsang

(2009) The L3 acquisition of Cantonese reflexives. In: Leung

YKI

(ed) Third language acquisition and Universal Grammar. Bristol: Multilingual Matters, pp. 192–219.

74.

Umeda

Takeda

Hirakawa

, et al. (2017) Acquiring antecedents for reflexives when both L1 and L2 permit long-distance binding. Journal of the European Second Language Association 1: 38–48.

75.

White

(1998) Second language acquisition and Binding Principle B: Child/adult differences. Second Language Research 14: 425–39.

76.

White

(2003) Second language acquisition and Universal Grammar. Cambridge: Cambridge University Press.

77.

Wong

SCH

(2021) Reflexivization in Mandarin: The role of zi-ji and its components. Unpublished PhD thesis, Utrecht University, Utrecht, The Netherlands.

78.

Ying

(1999) Access to UG and language transfer: A study of L2 learners’ interpretation of reconstruction in Chinese. Second Language Research 15: 41–72.

79.

Yoshimura

Nakayama

Shirahata

Sawasaki

Terao

(2012) Zibun and locality in L2 Japanese. Journal of Japanese Linguistics 28: 90–111.

80.

Yuan

(1998) Interpretation of binding and orientation of the Chinese reflexive ziji by English and Japanese speakers. Second Language Research 14: 324–40.

81.

Zeng

(2010) A cross language study of L2 acquisition of Chinese and English reflexives by English and Chinese speakers. Unpublished PhD thesis, City University of Hong Kong, Hong Kong, China.

82.

Zhang

(2018) The effect of L2 exposure on processing and perceiving aspect marking and reflexive binding in Mandarin–English bilinguals. Unpublished PhD thesis, University of Reading, Reading, UK.

83.

Zhao

(2014) Ultimate attainment of anaphora resolution in L2 Chinese. Second Language Research 30: 381–407.