Dependency resolutions of null and overt subjects in English speakers’ L2 Chinese: Evidence for the cue-based model

Abstract

This study investigates whether there are different first-language–second-language (L1–L2) dependency resolutions by focusing on less-studied crosslinguistic variances in L2 acquisition of Chinese, a null-subject language, by speakers of English, a non-null-subject language. The overt subject pronoun of a Chinese main clause has free orientation and its antecedent can be the subject or object of the preceding causal subordinate clause, depending on pragmatic biases. The null subject of a Chinese main clause, however, is subject-oriented, and this subject orientation is not affected by any pragmatic bias. English does not allow null subjects and, like Chinese, overt subject pronouns in English have free orientation. An acceptability judgement task and an interpretation task were adopted, and the results suggest that only the free orientation of overt subjects, but not the subject orientation of null subjects, is acquirable for English-speaking learners; they are found to be influenced by the pragmatic bias. This provides evidence for the cue-based model (Cunnings, 2017), which states that L1–L2 differences in dependency resolution can be explained in terms of L1–L2 differences in susceptibility to interference and L2ers’ over-reliance on discourse-based/pragmatic cues. It is also observed that in L1 Chinese, competition between the target antecedent and distractors occurs during the reading of the sentence, while in L2 Chinese, this occurs after the reading of the sentence. These findings add to our growing understanding of different mechanisms in L1 vs. L2 dependency resolutions.

Keywords

cue-based model L2 dependency resolutions null/overt subjects syntactic/pragmatic cues

I Introduction

It is well noted in the literature that subjects in languages like Italian can be null, and that they are considered to be syntactically licensed but distributed on the basis of discourse/pragmatic factors (see Rizzi, 1982). It is reported that although the acquisition of syntactic properties of overt and null subjects can be successful in a second language (L2), L2 learners’ sensitivity to subtle discourse distinctions may emerge late or never reach the same level of determinacy as shown in native speakers, even at near-native levels. This leads to a prediction of the interface hypothesis (IH) proposed by Sorace and Serratrice (2009) that although the acquisition of syntactic properties or those properties involving internal interfaces (such as the syntax–semantics interface) can be successful for L2 learners (L2ers), they may have problems integrating syntactic information with discourse/pragmatic information.

Recently, a growing body of research has implicated a cue-based retrieval mechanism during language comprehension (e.g. Martin and McElree, 2011). Following this mechanism, Cunnings (2017) challenges the IH by stating that L2ers may not necessarily have problems integrating syntactic information with discourse information. Instead, first-language–second-language (L1–L2) different dependency resolutions are probably caused by L2ers using different cue-weighting mechanisms from L1 speakers (L1ers). For example, L2ers may rely more strongly than L1ers on discourse-based/pragmatic cues, and comparatively less on syntactic ones (Cunnings, 2017). In the current study, Cunnings’ claim about the potential L1–L2 different cue-weighting mechanisms is referred to as the cue-based model (CBM).

It should be noted that both the IH and CBM have almost exclusively been informed by data from null and overt subjects in rich-inflection languages like Italian (e.g. Sorace and Serratrice, 2009) and Greek (e.g. Cunnings et al., 2017). In comparison, less attention has been paid to the L2 acquisition of null and overt subjects in languages with no inflection like Chinese. As acknowledged in Felser (2016: 262), ‘further research is needed to explore whether L2 comprehenders’ apparent difficulty computing syntactically mediated referential dependencies extends to other language combinations and other types of syntactic environment.’ The study reported in this article attempts to add a different perspective to our understanding of mechanisms involved in L1–L2 dependency resolutions by focusing on less-studied crosslinguistic variances between a null-subject language, L2 Chinese, and a non-null-subject language, L1 English.

II Different interpretations of null and overt subjects in Chinese complex sentences

Chinese complex sentences are similar to their English counterparts in that they may comprise an adverbial subordinate clause and a main clause. The adverbial subordinate clause may contain a conjunction like yinwei ‘because’, which marks an interclausal relationship between the subordinate clause and the main clause. Both null and overt subjects are allowed in the main clause but their interpretations differ.

If a subordinate causal clause contains both a subject and an object, the null subject of the main clause is obligatorily subject-oriented, and this subject orientation is not affected by any pragmatic bias, which means that the null subject obligatorily refers to the subject of the preceding subordinate clause. For example, in a context such as John gave Ben a hundred pounds, the benefactor, Ben, is the one who is more likely to be happy for being given the money. As shown in (1), however, the null subject in the main clause refers to John, the subject of the subordinate clause, and cannot refer to Ben, the object. When the subject pronoun of the main clause is overt, it has free orientation and its antecedent can be either the subject or the object of the preceding subordinate clause, depending on pragmatic biases; see (2) (see Huang, 1994; Liu, 2014).

(1) Yinwei John_i gei-le Ben_j yibai bang, e_i/*j hen shangxin/gaoxing.

because John give-ASP Ben a hundred pounds very sad /happy

Literal translation: ‘Because John_i gave Ben_j a hundred pounds, e_i/*j was very sad/happy.’

(2) Yinwei John_i gei-le Ben_j yibai bang, ta_i/j hen shangxin/gaoxing.

because John give-ASP Ben a hundred pounds he very sad /happy

Literal translation: ‘Because John_i gave Ben_j a hundred pounds, he_i/j was very sad/happy.’

(adapted from Liu, 2014: 177)

But why can only the subject, but not the object, in the preceding subordinate clause serve as the reference of a null subject in the main clause in Chinese?

One straightforward answer is that subjects, by default, are more salient than objects. Liu (2014) argues that the subject orientation of the null subject in the main clause in sentences like (1) can be due to the fact that a null subject in Chinese can be co-indexed with a topic (Huang, 1984) and that John is actually an A′-topic located in the CP domain (regarding sentences with time adverbials, see also Zhao, 2014). As shown in (3), three Agree relations are manifested in Liu’s configuration. The first is established between the overt topic John and the null subject e in the first clause (Agree Ⅰ); the second is a long-distance agreement between the overt topic John and the null topic TOP in the second clause (Agree Ⅱ). The TOP in turn identifies the null subject e in the second clause (Agree Ⅲ).

Liu’s (2014) analysis is correct in that null subjects can be syntactically bound to topics (Huang, 1984), but it seems that co-indexation relations are more appropriate than the Agree relations for explaining the phenomenon in Chinese. This is because it is not viable to discuss Chinese null subjects, which do not have φ-features or case features (Barbosa, 2019; Roberts, 2009), within the framework of the feature-valuing system in the Minimalist Approach.

In contrast to the null subject condition, Chinese overt subject pronouns can refer to either the subject or the object of the preceding clause; see (2). This is because Principle B of the Government and Binding Theory only stipulates that pronouns must be free in their governing categories (Chomsky, 1981), without any specification on their orientations. Therefore, the pronoun in (2) can be coreferential with the subject or the object in the preceding clause, or even with somebody else depending on the context. In the absence of a discourse context, a sentence in isolation tends to favour coreferential readings within the complex sentence, as shown in (2).

To summarize, differences in the interpretation of overt and null subjects in Chinese complex sentences suggest that dependency resolutions are an important aspect of Chinese. Our study examines to what extent this phenomenon can be acquired by L2ers of Chinese. It is likely that learners can acquire the fact that Chinese allows null subjects, but fail to have the native-like dependency resolutions for null and overt subjects in their L2 Chinese. To the best of our knowledge, no work in the L2 literature has studied the subject orientation of null subjects in the main clause of complex sentences like (1), nor the free orientation of overt subjects in sentences like (2). This study is to fill the lacuna.

III Interface hypothesis and cue-based model

A growing body of research has investigated dependency resolution in L2 null-subject languages with rich inflection, but how to account for L1–L2 differences in dependency resolution is controversial. Sorace and Serratrice’s (2009) IH has been influential. According to the IH, the coordination of information at external interfaces calls for the integration of world knowledge and discourse outside the domain of grammar proper. As such, it adds to the burden on L2ers who may have inadequate processing abilities and sets up barriers to their behaving in a native-like way. Syntactic properties, however, are predicted to be acquirable, as they are internal to the grammar. Sorace and Serratrice (2009) focus on null and overt subjects in English–Italian bilinguals’ and Spanish–Italian bilinguals’ Italian. Italian is a null-subject language where overt subjects are used in a topic-shift context but null subjects are used in a non-topic-shift context, as shown in (4). Spanish has a similar pattern to Italian regarding this null-subject phenomenon. By contrast, English does not allow null subjects. It is found that while both English–Italian and Spanish–Italian bilinguals are successful in the syntactic acquisition of null and overt subjects in Italian, they exhibit an issue where they inappropriately use overt subject pronouns in contexts without topic shifting in Italian. Based on this finding, Sorace and Serratrice (2009) argue that the pattern of over-extension of overt subject pronouns may stem from bilinguals not consistently having the computational resources necessary to coordinate the use of an overt pronoun with the introduction of a new topic, which is at the external syntax–discourse interface. As a result, they are less sensitive to topic shifting discourse situations.

It is worth noting that many cross-sectional L2 studies since 2009 have revealed protracted development of properties at external interfaces despite the successful acquisition of the corresponding purely syntactic properties at low proficiency levels (e.g. Lozano and Mendikoetxea, 2010; Montrul, 2010; Rothman, 2009; Slabakova et al., 2012; Slabakova and Ivanov, 2011).

(4) La vecchietta_i salute la ragazza_j, quando e_i/?j/lei_j attraversa la strada.

the old woman greets the girl when she crosses the road

Literal translation: ‘The old woman greets the girl when e_i/?j/she_j crosses the road.’

(Sorace, 2011: 2)

More recently, Cunnings (2017) challenges the IH by arguing that the L1–L2 different dependency resolution is not necessarily due to L2ers’ difficulty in integrating syntactic information with discourse/pragmatic information. He has used the cue-based retrieval mechanism (e.g. Martin and McElree, 2011) and argued that the L1–L2 difference in dependency resolutions may be attributable to L1–L2 different weighting of retrieval cues.

To be more specific, subject pronouns are sometimes backwards-looking dependencies in that encountering them triggers the retrieval of a previously mentioned antecedent. According to the cue-based retrieval mechanism (see Lewis and Vasishth, 2005), the retrieval process involves comparing a set of retrieval cues against all potential antecedents in parallel. Meanwhile, which referent is retrieved also provides information regarding which cues are utilized during the process of retrieval. For example, when the parser is retrieving a potential antecedent for the pronoun he in (5), the pronoun itself provides important semantics-based retrieval cues, such as [+Male] and [+Singular]. The item that matches this set of retrieval cues becomes highly activated and thus is retrieved. Because Peter, not workers, in the previous discourse in (5) matches the above set of retrieval cues, Peter is retrieved as a potential antecedent for the pronoun he.

(5) The workers_j are in the office. While Peter_i is working, he_i/*j is eating a sandwich.

(6) Peter_i and Hans_j are in the office. While Peter_i is working, he_i/j is eating a sandwich.

(Roberts et al., 2008: 341)¹

However, as the retrieval cues are matched against all potential referents in parallel, a distractor that (partially) matches the retrieval cues may sometimes also be retrieved. For example, Roberts et al. (2008) find that L2ers have a longer reaction time when processing the pronoun he in (6), where two antecedents (Peter and Hans) match the set of retrieval cues, compared to the same region in (5), where only one antecedent (Peter) matches the retrieval cues. The longer reaction time in (6) is interpreted as a result of indexing competition between the retrieval target and distractor when both are male and singular. Phenomena such as this are called ‘interference’ (Badecker and Straub, 2002; Van Dyke and McElree, 2011).

Within the framework of the cue-based retrieval mechanism, Cunnings (2017) derived and applied the CBM which argues that the L1–L2 difference in dependency resolution is possibly due to L2ers’ increased susceptibility to interference, which in part results from L2ers’ over-reliance on discourse-based/pragmatic cues to antecedent retrieval. For example, Pan et al. (2015) find that discourse-based/pragmatic cues are more highly weighted retrieval cues during L2 processing than L1 processing.

IV Interpretations of null subjects in L2 Chinese

To master null/overt subjects in a null-subject L2 requires not only the knowledge that null subjects can be licensed in the language, but also the ability to identify their references (see Sorace and Serratrice, 2009). However, only very limited research has been done on the interpretation of null subjects in L2 Chinese. Zhao (2014) is an exception. Her study focuses on interpretations of null and overt subjects in sentences with temporal subordinate clauses by English speakers who are highly proficient in L2 Chinese. As in (7), if the preceding temporal subordinate clause contains a null subject, the null subject refers to the subject of the main clause; however, if the subordinate clause contains an overt subject pronoun, the pronoun refers to someone else from the discourse. This is different from the dependency resolution in (8), in which the overt or null subject is in the main clause, and either of them can be coreferential with the subject of the temporal subordinate clause. Zhao (2014) finds that English-speaking learners of Chinese can acquire native-like dependency resolutions of null subjects in sentences such as (7) and (8). However they fail to acquire that the overt subject pronoun in the preceding temporal adverbial clause cannot refer to the subject of the main clause.

(7) * ta_i/e_i chi wanfan de shihou, Zhangsan_i dai-zhe yi ding maozi.

he eat dinner DE when Zhangsan wear-PRG one CL hat

Literal translation: ‘When *he_i/e_i was having dinner, Zhangsan_i was wearing a hat.’

(8) Zhangsan_i chi wanfan de shihou, ta_i/e_i dai-zhe yi ding maozi.

Zhangsan eat dinner DE when he wear-PRG one CL hat

Literal translation: ‘When Zhangsan_i was having dinner, he_i/e_i was wearing a hat.’

(Zhao, 2014: 384)

All potential antecedents of null and overt subjects in Zhao’s study are in subject position, and it is unclear whether English-speaking learners of Chinese are aware of the specifications of the subject orientation of null subjects and free orientation of overt subjects in Chinese main clauses. These specifications will be the focus of the investigation in our study.

In addition, Zhao (2014) focuses exclusively on Chinese sentences with temporal subordinate clauses. As Su (2020) points out, the prohibition of the coreference reading of the overt subject with the subject in the main clause is mainly restricted to temporal adverbial clauses in Chinese, but not across the board for all types of clauses including causal subordinate clauses in Chinese. In this study, we focus on Chinese complex sentences with causal subordinate clauses, as the causal relationships between two clauses can help to create biased contexts in favour of certain dependency resolutions.

V The study

1 Research questions and hypothesis

In this study, we investigate L2 Chinese dependency resolutions of null and overt subjects in sentences with causal subordinate clauses and we ask the following questions:

Can English speakers acquire the subject orientation of null subjects in their L2 Chinese? That is, do they refer null subjects back to the subject of the preceding subordinate clause in either a subject-biased or object-biased condition; see (1)?

Can they acquire the free orientation of overt subjects and refer overt subjects to the subject of the preceding clause in a subject-biased condition but refer them to the object of the preceding clause in an object-biased condition; see (2)?

According to the CBM, which referent is retrieved provides information regarding which cues are weighted during the process of retrieval (Cunnings, 2017). Because null and overt subjects in Chinese main clauses have different dependency resolutions, we predict that L1ers may have different weightings of individual cues for the interpretations of null and overt subjects in the following way:

It is suggested that context triggers pragmatic expectations regarding coreference (Rohde and Ettlinger, 2012). The term ‘pragmatic’ here refers to the linguistic notion of ‘what is meant beyond what is said’ (Grice, 1975). We focus on the pragmatics of coreference, a phenomenon that underlies comprehenders’ ability to infer information in order for a sentence to convey a coherent message in the surrounding context and thereby track who is being talked about. Because the overt subject pronoun in a Chinese main clause has free orientation, as shown in (2), and because its interpretation depends on the context, i.e. whether the context is subject-biased or object-biased, L1ers mainly rely on a discourse-based/pragmatic cue when retrieving referents of overt subjects in sentences like (2).

The null subject in a Chinese main clause displays subject orientation, which is independent of the context, as shown in (1). This is because the null subject is syntactically bound to the topic, which is, in turn, co-indexed with the preceding subject. This provides a syntactic cue ([+Subject]/[+Topic]) retrieving an antecedent from a salient subject/topic position, as illustrated in (3). The subject orientation of Chinese null subjects, therefore, suggests that L1ers mainly rely on the syntactic cue when interpreting null subjects in sentences like (1).

Chinese allows both null and overt subjects, whereas English only allows the latter. Like the case in Chinese, overt subjects in English can have free orientation, as shown in (9).

(9) Because John_i gave Ben_j one hundred pounds, he_i/j was very sad/happy.

This study focuses on whether English-speaking learners can have a distinct resolution of Chinese null and overt subjects.

The IH predicts that although L2ers may have problems integrating syntactic information with discourse/pragmatic information during the interpretations of null and overt subjects, they should have native-like sensitivity to pure syntactic properties. As a result, they should rely on syntactic cues more strongly than pragmatic cues for the retrieval of antecedents. This is because syntactic information, which is internal to the grammar, imposes less cognitive demands than pragmatic information, which looks outwards to world knowledge and discourse. Therefore, the acquisition of interpretations of Chinese null subjects, which mainly relies on syntactic information, should be less problematic for L2ers than overt pronouns.

The CBM predicts that discourse-based/pragmatic cues are more heavily weighted during L2 processing than L1 processing. As a result, L2ers may have native-like performance in the interpretations of Chinese overt subjects, which are largely based on discourse and pragmatic information. However, L2ers may over-rely on discourse-based/pragmatic cues during the dependency resolution for Chinese null subjects when the syntactic cue should be more heavily weighted (as is supposedly the case with L1ers). Therefore, it is predicted that L2ers may fail to have native-like dependency resolutions for Chinese null subjects.

2 Participants

To answer the research questions above, English-speaking learners of Chinese were recruited in China and the UK. To examine the trajectory of development of L2 dependency resolutions, an HSK-based² proficiency test (Guo and Yuan, 2020, 2021) was used to divide the participants into five proficiency groups, i.e. beginner learners (BLs), pre-intermediate learners (Pre-ILs), intermediate learners (ILs), advanced learners (ALs), and very advanced learners (VALs). In addition, a control group of 30 Northern Chinese native speakers (CNSs), who had very limited knowledge of an L2, were recruited in China. Detailed information on the participants is given in Table 1. A one-way ANOVA shows a significant difference in scores of the proficiency test between groups (F(5,169) = 545.4, p < .001). Tukey post-hoc tests show significant differences between any two participant groups (all at p < .001) except for the difference between CNSs and VALs (p > .05). In a backward digit test of participants’ working memory spans (see Juffs and Harrington, 2011), no significant difference was found between any two participant groups (all at p > .05). Payments were given to every participant for their participation in the study.

Table 1.

Information of each group (standard deviations in parentheses).

Groups	Numbers of participants	Mean scores of proficiency test (total score = 40)	Average ages	Lengths of staying in China (months)	Lengths of exposure to Chinese (months)
BL	25	9.50 (3.29)***	24 (5.44)	3 (5)	11.5 (6)
Pre-IL	29	18.37 (3.07)***	24 (6.57)	14 (16)	45 (36)
IL	31	25.18 (2.93)***	26 (5.01)	33 (39)	64 (54)
AL	31	32.03 (2.37)***	28 (5.74)	37 (34)	76 (60)
VAL	31	37.69 (1.58)	31 (5.11)	86 (85)	135 (81)
CNS	30	38.20 (1.83)	33 (6.72)	–	–

Note: *** = significantly different from CNSs (Chinese native speakers) at p < .001. AL = advanced learner. BL = beginner learner. IL = intermediate learner. VAL = very advanced learner.

3 Procedures and materials

Participants were required to complete two tasks in the following order: an acceptability judgement task (AJT) and an interpretation task. The purpose of the AJT was to ensure that participants selected for this study allowed null subjects in the main clause. If a participant could not accept sentences with null subjects in the AJT in 3 tokens out of 4 tokens, he/she was removed from the analysis. The data of 7 BLs, 1 Pre-IL, 1 IL, and 4 ALs were consequently excluded.

Participants were then asked to complete an interpretation task on a computer screen. At the start, participants watched a short video on how to do the task. The instruction was given in participants’ L1 English and Chinese respectively. An example, using language structures unrelated to the focus of the experiment, was shown on the screen, and a voice-over, voiced by a native speaker, explained the exact nature of the task. As illustrated in Figure 1, for each test sentence in this task, participants first pressed the space bar, which triggered a statement by someone called Linna appearing at the centre of the screen (also see Table 2 below). After they read the statement, they pressed the space bar again, and Linna’s statement disappeared. Then a question appeared asking them to whom the null/overt subject in Linna’s statement probably referred. On the screen below the question, they were presented with the first possible option and were asked to decide how likely it was that that option was the referent meant by Linna. Participants were expected to make a judgement by pressing the key marked with ‘unlikely’, ‘quite unlikely’, ‘quite likely’ or ‘likely’ on the keyboard.³ If they could not understand Linna’s statement, they pressed the key marked with ‘?’. After they made the choice by pressing a key on the keyboard with the corresponding label, the first option disappeared and a second option appeared. They then repeated the above process for the second option, and continued until the end.

Table 2.

Sample set of experimental materials.

Condition	Test sentence	Interpretation question	Option
A. [Overt subject in a subject-biased context]	林娜说: ‘大卫打了小明, Linna say Dawei hit-ASP Xiaoming 他很高兴。’ he very happy	你觉得谁高兴？ ‘Who do you think is happy?’	1) 大卫 Dawei 2) 小明 Xiaoming 3) 第三个人 A person other than Dawei and Xiaoming
B. [Overt subject in an object-biased context]	林娜说: ‘大卫打了小明, Linna say Dawei hit-ASP Xiaoming 他很生气。’ he very angry	你觉得谁生气？ ‘Who do you think is angry?’
C. [Null subject in a subject-biased context]	林娜说: ‘大卫打了小明, Linna say Dawei hit-ASP Xiaoming 很高兴。’ very happy	你觉得谁高兴？ ‘Who do you think is happy?’
D. [Null subject in an object-biased context]	林娜说: ‘大卫打了小明, Linna say Dawei hit-ASP Xiaoming 很生气。’ very angry	你觉得谁生气？ ‘Who do you think is angry?’

Notes. Participants were informed that 林娜 (‘Linna’) is a common female Chinese name. This was to avoid confusing participants, who might refer the male pronoun 他 (‘he’) to 林娜 (‘Linna’).

The task was preceded by three practice sentences (unrelated to this task) to familiarize the participants with the experimental procedure. Test sentences were presented on the computer screen one at a time. The presentation of the sentences was done with the program E-Prime 2.0, which recorded the time taken by each participant in reading and interpreting the sentence. Participants were asked to make their likelihood ratings as soon as possible and to make decisions according to their first intuition.

As marked in Figure 1, the moment that participants pressed the space bar and a test sentence with a null or overt subject appeared on the screen, the program E-Prime 2.0 started to record time. The moment that participants finished reading the test sentence, they pressed the space bar again. The time interval between these two moments was recorded and regarded as the time taken by participants to read the test sentence (referred to as ‘reading time’). In addition, the program also recorded the time that participants took to make likelihood ratings by pressing a key on the keyboard with a corresponding label (which is referred to as ‘interpretation time’).

Four conditions were used in this study (as shown in Table 2). They were based on the following variables: null vs. overt subject; subject- vs. object-biased context. To exclude the possibility that people from different cultures may have different pragmatic expectations, an English control interpretation task was used. Ten English-native (monolingual) speakers read the English translation of test sentences with overt subjects in this study and were asked to whom the overt subjects refer. It was found that they chose subject reading in the subject-biased context and object reading in the object-biased context. This suggests that the materials in our experiment were valid in triggering pragmatic expectations among English-native speakers. More importantly, this supports the argument that overt subjects in English have free orientation, as shown in (9). Special efforts were made to control other variables, such as sentence structures, types of verbs, types of subordinate clauses and genders of antecedents. This is because previous studies found that such variables could lead to biased dependency resolutions (Brown and Fish, 1983; Holler and Suckow, 2016; Stevenson et al., 2000; Yang et al., 2003). As a result, all critical tokens in this study used a causal subordinate clause followed by a main clause, and only used agent–patient verbs (‘punch’, ‘kick’, etc.) in the subordinate clause.⁴ In addition, the subject and object were made up of two common male names, Dawei and Xiaoming, and participants were informed in the instruction that they are male names. Participants were provided with three possible interpretations of null or overt subjects in each test sentence (e.g. the preceding subject Dawei, the preceding object Xiaoming and a person other than Dawei and Xiaoming). The order of the two interpretations (the preceding subject and object) varied throughout the task (i.e. in half of the tokens, the subject was presented first and then the object; in the other half of the tokens, the object was presented first and then the subject). Participants were asked to judge the likelihood of each interpretation. They could rate the two interpretations as equally likely if they considered that this was the case.

The interpretation task consisted of 144 sentences, out of which 48 were related to the focus of this study as we used 12 critical tokens, each with 4 conditions (as shown in Table 2).⁵ The 48 sentences were embedded in 96 distractors. Two presentation lists were constructed out of these 144 sentences in the following ways: first, we constructed 2 counterbalanced presentation lists out of 96 distractors and 24 test sentences representing Conditions A and B of the 12 critical tokens on the basis of a Latin square design. Second, because Conditions A and D used different psych verbs and differed in the use of null and overt subjects, they had a limited priming effect on each other. The list including Condition A of one critical token also included Condition D of the same token. Conditions B and C were linked in the same way. Third, on each list, we put two test sentences, representing two conditions of the same critical token, in two different blocks. Participants were asked to take a break between the two blocks to minimize any priming effect and to avoid fatigue. The test sentences and distracters in each block were pseudo-randomized. Each participant was assigned to one of the two presentation lists. Therefore, they encountered 6 sentences for each condition.

4 Results of the likelihood ratings

The interpretation task investigates whether English-speaking learners can make a distinction between the subject orientation of null subjects and the free orientation of overt subjects in their L2 Chinese. In the post-hoc scoring, the likelihood ratings (i.e. ‘unlikely’, ‘quite unlikely’, ‘quite likely’, ‘likely’) were converted to 1, 2, 3 and 4 respectively. Table 3 lists the mean scores of the L1 and L2 groups as well as the mean scores of subdivided L2 proficiency groups. Participant groups’ mean scores for the option ‘a person other than Dawei and Xiaoming’ (labelled as ‘Others’ in Table 3) are all below 2.00. This suggests that for L1ers and L2ers, this option is unlikely for all the conditions. In the rest of this section, we will ignore the option of ‘Others’ and only focus on to what degree the subject or the object of the preceding sentence (hereinafter referred to as the ‘preceding’ subject or object) is regarded as a(n) (un)likely antecedent for null/overt subjects in the main clause.

Table 3.

Mean scores for the likelihood of each potential antecedent in four conditions in the interpretation task (standard deviations in parentheses).

Condition	A. [Overt subject in a subject-biased context]			B. [Overt subject in an object-biased context]
Interpretation	Subject	Object	Others	Subject	Object	Others
L1 group (CNS)	3.91 (0.40)	1.91 (0.77)	1.40 (0.62)	2.97 (1.06)	3.53 (0.77)	1.44 (0.69)
L2 group	3.41 (0.89)	2.19 (0.99)	1.54 (0.89)	2.79 (0.99)	3.36 (0.89)	1.60 (0.89)
VAL	3.49 (0.96)	2.13 (0.99)	1.61 (1.04)	2.63 (1.02)	3.48 (0.95)	1.62 (1.00)
AL	3.51 (0.82)	2.04 (0.91)	1.49 (0.77)	2.75 (0.97)	3.39 (0.86)	1.50 (0.74)
IL	3.48 (0.80)	2.16 (0.90)	1.39 (0.74)	2.71 (0.93)	3.49 (0.76)	1.46 (0.75)
Pre-IL	3.41 (0.84)	2.14 (0.96)	1.40 (0.73)	2.85 (0.95)	3.20 (0.96)	1.42 (0.69)
BL	3.24 (0.98)	2.28 (1.14)	1.57 (0.88)	2.99 (1.05)	3.27 (0.84)	1.75 (0.99)
Condition	C. [Null subject in a subject-biased context]			D. [Null subject in an object-biased context]
Interpretation	Subject	Object	Others	Subject	Object	Others
L1 group (CNS)	3.88(0.46)	1.93(0.82)	1.42(0.68)	3.41(0.77)	2.68(1.09)	1.51(0.77)
L2 group	3.41(0.89)	2.19(0.99)	1.74(1.00)	2.86(1.02)	3.25(0.91)	1.81(1.01)
VAL	3.50(0.95)	2.08(1.03)	1.69(1.08)	2.77(1.09)	3.28(0.99)	1.87(1.14)
AL	3.49(0.84)	2.00(0.92)	1.87(1.01)	2.83(1.03)	3.28(0.78)	1.90(0.97)
IL	3.44(0.83)	2.16(0.88)	1.68(1.00)	2.75(0.97)	3.28(0.85)	1.76(1.00)
Pre-IL	3.35(0.89)	2.26(0.98)	1.53(0.81)	2.87(0.99)	3.13(0.98)	1.60(0.89)
BL	3.34(0.91)	2.35(1.04)	1.76(0.88)	3.06(1.03)	3.24(0.87)	1.69(0.88)

Notes. 1 = unlikely, 2 = quite unlikely, 3 = quite likely, 4 = likely. AL = advanced learner. BL = beginner learner. CNS = Chinese native speaker. IL = intermediate learner. L1 = first language. L2 = second language. VAL = very advanced learner.

Because we are interested in whether participants agree that Dawei or Xiaoming is likely to be the antecedent, we collapse both ‘quite likely’ and ‘likely’, and code them as ‘agree’ in our analysis. Similarly, we collapse ‘quite unlikely’ and ‘unlikely’, and code them as ‘disagree’. We then run generalized mixed-effects models, using the function ‘glmer’ with the ‘family = binomial’ argument through the ‘lme4’ package (version 1.1-21) in R (Version 1.0.143).

In our model, response is the dependent variable. We also include fixed main effects of group (L1 vs. L2), antecedent (Dawei vs. Xiaoming), context (subject-biased vs. object-biased), and subject type (overt subject vs. null subject), and all interactions. Subject and item random intercepts and random slopes for each fixed effect are included using the ‘maximal’ random effects structure that converges (Barr et al., 2013). In the following text, we only report significant interactions and main effects.

Our findings show a significant four-way (Group × Antecedent × Context × Subject Type) interaction (estimate = 0.43, z = 2.61, p < .01). To find out how these variables affect participants’ likelihood ratings, Antecedent × Context × Subject Type analyses are conducted on the L1 group and L2 group separately.

a L1 group

Analysis of the L1 data yields a significant three-way interaction between antecedent, context, and subject type (estimate = −0.97, z = −2.68, p < .01). Paired comparisons are carried out in conditions with overt and null subjects respectively to test for preferences for either the subject or object antecedent in each of the two conditions. Analysis of each condition yields a significant Antecedent × Context interaction (overt subject: estimate = −6.63, z = −7053, p < .001; null subject: estimate = −1.22, z = −2.36, p < .05).

Further post-hoc analyses show that in interpreting overt subjects, CNSs rate the subject reading as more likely than the object reading for a subject-biased context (estimate = −0.17, z = −17,975, p < .001), but they rate the preceding object as a more likely antecedent than the preceding subject for an object-biased context (estimate = 1.17, z = 3.59, p < .001).

However, a different pattern is found in CNSs’ interpretation of null subjects. CNSs are significantly more likely to take the preceding subject rather than the object to be the referent in both the subject-biased context (estimate = −3.91, z = −2.01, p < .05), and the object-biased context (estimate = −1.31, z = −2.51, p < .05).

b L2 group

Analysis of the L2 data only yields a significant Antecedent × Context interaction. As such, the effects of context on antecedent are tested, collapsed across subject type. Results show that L2ers as a whole have a preference for the subject reading in a subject-biased context (estimate = −1.81, z = −7.46, p < .001), but the object reading in an object-biased context (estimate = 0.72, z = 4.00, p < .001).

In addition, we also examine the effects of L2 proficiency by conducting an analysis of the L2 data that includes proficiency scores as a continuous predictor. We find a reliable Proficiency × Antecedent × Context interaction (estimate = −0.02, z = −2.28, p < .05), which suggests that these three variables might play different roles in the process whereby the participants judge the likelihoods of interpretations. We then analyse the data of BLs, Pre-ILs, ILs, ALs and VALs separately.

Results show significant Context × Antecedent interactions for all L2 groups (all estimates < −0.95, all z < −4.19, all p < .001). Context effects are then tested, collapsed across subject type. Results show that in a subject-biased context, all the L2 groups rate the preceding subject as a significantly more likely antecedent than the preceding object (all estimates < −1.18, all z < −3.43, all p < .001). However, in an object-biased context, VALs, ALs and ILs rate the preceding object as a more likely referent than the preceding subject (all estimates > 0.69, all z > 2.39, all p < .05). The difference is found not significant for BLs and Pre-ILs. Post-hoc analyses find that in interpreting either null or overt subjects, all the L2 groups have a preference for the subject reading in a subject-biased context, but VALs, ALs and ILs prefer the object reading in an object-biased context, although the difference is marginal for VALs in interpreting null subjects in an object-biased context.

This finding suggests that the subject orientation of null subjects is only available in CNSs’ L1 Chinese grammar, but not in English-speaking learners’ L2 Chinese grammars, where dependency resolutions of null subjects are very much dependent on biased contexts, even at very advanced levels. The findings are summarized in Table 4.

Table 4.

Summary of the interpretations by the groups.

Condition	BL	Pre-IL	IL	AL	VAL	CNS
A. [Overt subject in a subject-biased context]:	SUB	SUB	SUB	SUB	SUB	SUB
B. [Overt subject in an object-biased context]:	?	?	OBJ	OBJ	OBJ	OBJ
C. [Null subject in a subject-biased context]:	SUB	SUB	SUB	SUB	SUB	SUB
D. [Null subject in an object-biased context]:	?	?	OBJ	OBJ	OBJ	SUB

Notes. SUB/OBJ = Participants rate the preceding subject/object as a likely antecedent, ? = No significant distinction is made between the subject and object readings. AL = advanced learner. BL = beginner learner. CNS = Chinese native speaker. IL = intermediate learner. VAL = very advanced learner

5 Results of the time taken for the resolution

In addition to the examination of these participants’ dependency resolutions of overt and null subjects, the time taken by the participants in reading test sentences (reading time), and in making the likelihood ratings for each possible interpretation of test sentences (interpretation time), is also analysed. This is because the time taken in reading and interpreting overt or null subjects can reflect the degree of difficulty or ease that participants have in making the dependency resolution.

Before analysing the reading-time and interpretation-time data, outliers in the data are dealt with first: We use the standard deviation method to identify outliers (Jegerski and VanPatten, 2013) and set the cutoff value following the criterion that data treatment can affect 3% to 7% of the data (Jiang, 2013) although it is acceptable to go as high as 10% (Ratcliff, 1993). As a result, any reading or interpretation time longer than one standard deviation away from a mean (the cutoff value), calculated for each participant group in each stimulus condition, is judged as an outlier. Considering that outliers may not reflect real interpreting difficulty, we replace them with more moderate values that are still relatively high, which are the cutoff values. The percentage of data thus affected is shown in Table 5.

Table 5.

Percentages of data replaced in the data trimming process.

	Reading time	Interpretation time
BL	8	6
Pre-IL	9	7
IL	10	10
AL	11	10
VAL	10	10
CNS	9	9

Notes. AL = advanced learner. BL = beginner learner. CNS = Chinese native speaker. IL = intermediate learner. VAL = very advanced learner.

Results of the L1 and L2 groups’ average reading and interpretation times are shown in Figure 2, together with the average times of different L2 proficiency groups. The reading and interpretation times are log-transformed and analysed in linear mixed-effects regression models.

a Reading time

In the model, reading time is the dependent variable. We also include fixed main effects of group (L1 vs. L2), context (subject-biased vs. object-biased), subject type (null subject vs. overt subject) and all interactions. Subject and item random intercepts are included. Although it is recommended to keep the random structure maximal (Barr et al., 2013), adding the random slopes leads to convergence issues. Thus, only the random intercepts are included.

We find a significant main effect of group (estimate = 0.21, t = 3.91, p < .001) and a significant Group × Context interaction (estimate = 0.05, t = 2.48, p < .001). The simple main effect of context is then explored for L1 and L2 data respectively.

An analysis of L1 data yields no significant effect of context. To test for any possible impact of context information on L1ers’ reading patterns for sentences with null or overt subjects, paired comparisons are conducted separately for the null and overt subject conditions.⁶ Results show that L1ers spend significantly longer time reading test sentences with null subjects in an object-biased context than they do in a subject-biased context (estimate = −0.08, t = −2.01, p < .05). Such a difference is not found between two conditions with overt subjects.

An analysis of L2 data suggests a significant effect of context (estimate = 0.06, t = 3.46, p < .001) with L2ers spending more time in a subject-biased context. To further examine the effects of L2 proficiency, we include proficiency scores as a continuous predictor. Results show that only the main effect of context is significant (estimate = 0.06, t = 3.46, p < .001). Because one aim of this study is to test the IH which is mainly concerned with L2 ultimate attainment, we are particularly interested in the reading pattern of VALs, who are regarded as representatives of an ultimate stage in L2 Chinese. As a parallel for the analysis of L1 data, paired comparisons are then conducted separately for the null and overt subject conditions on the data of the VAL group as well as other L2 groups. No significant difference is found in any L2 group’s reading times between either the two conditions with null subjects or two conditions with overt subjects.

b Interpretation time

We fit a model with the interpretation time as the dependent variable and with antecedent, context, and subject type as fixed variables. Because adding the random slopes leads to convergence issues, only subject and item random intercepts are included. The between-groups analysis only yields a significant main effect of group (estimate = 0.13, t = 3.25, p < .01).

We also conduct an analysis of L2 data that includes proficiency scores as a continuous predictor. It shows that the main effect of proficiency is not significant, but it yields a significant Context × Subject Type interaction (estimate = −0.04, t = −2.17, p < .05). The simple main effect of context is then examined under conditions with overt and null subjects. Results show that the main effect of context is significant only in the condition with null subjects (with L2ers spending more time in an object-biased context) (estimate = −0.08, t = −8.29, p < .001) but not with overt subjects (estimate = 0.01, t = 1.13, p > .05). As a parallel for the reading time analysis, paired comparisons are also conducted on the data of each L2 group. Results show that ALs (estimate = −0.145, t = −2.58, p < .05) and VALs (estimate = −0.126, t = −2.22, p < .05) spend more time judging the likelihood of the preceding subject as the antecedent for null subjects in an object-biased context than they do in a subject-biased context, and that such a difference is not found between conditions with overt subjects. Note that no significant difference is found in CNSs’ interpretation times between the subject-biased and object-biased contexts for either overt subjects or null subjects.

VI Discussion

1 Summary of findings

This study provides evidence for differences in L1 vs. L2 dependency resolution of null and overt subjects in Chinese.

As confirmed by the CNSs’ data, both null and overt subjects are allowed in Chinese, but their co-referential interpretations are different. The null subject of a Chinese main clause is subject-oriented, as in (1), and this subject orientation is not affected by any pragmatic bias. When the subject of the main clause is overt, however, as in (2), it has free orientation, and its antecedent can be influenced by pragmatic biases and can be the subject or object of the preceding subordinate clause. Unlike CNSs, who refer Chinese null subjects back to the subject of the preceding sentence regardless of whether the context is subject- or object-biased, English-speaking learners’ interpretations of both null and overt subjects are dependent on pragmatic information.

CNSs’ and L2ers’ reading or interpretation times for sentences with null subjects reveal a different pattern. CNSs spend a significantly longer time reading test sentences with null subjects in an object-biased context than they do in a subject-biased context. Such a difference is not found in L2ers’ data, although ALs and VALs have significantly longer interpretation times for the former condition than the latter condition. Note that we do not find evidence that CNSs and L2ers spend more time reading or interpreting overt subjects in an object-biased context than in a subject-biased context.

The non-target-like interpretations of null subjects in L2 Chinese shed light on the ongoing debate concerning whether it is L2ers’ lower sensitivity to pragmatic/discourse information, as claimed in the IH, or their different cue weighting mechanism, as argued for in the CBM, that results in a difference between L1 and L2 dependency resolutions.

2 L1 transfer and input

Our findings suggest that English-speaking learners of Chinese, including BLs, have the knowledge that an overt subject in Chinese has free orientation.

One possible explanation is that they transfer the free orientation of overt subjects from their L1 English to their L2 Chinese. As is evidenced in the English control data in Section V.3, overt subject pronouns in English, like in Chinese, have free orientation; see (9).

In addition, the L1 transfer can be confirmed and reinforced by positive evidence of overt subject pronouns referring to either subjects or objects of the preceding clause in their Chinese input. Xie et al. (2020) established a 115,405-word Chinese corpus based on Chinese university students’ compositions, in which – of the complex sentences with an overt subject pronoun in the main clause and two potential referents in the preceding clause – 43.75% and 56.25% of the subjects in the main clause refer back to the subject and object of the preceding clause respectively, and the difference between them is not found to be significant. This suggests that in Chinese native speakers’ production, overt subject pronouns have free orientation. It also indicates that English-speaking learners of Chinese can receive positive evidence in their Chinese input, informing them that overt subjects in Chinese have free orientation. This explains the general native-like performance of English speakers in their dealing with dependency resolutions of overt Chinese subject pronouns in our study.

3 Interface hypothesis

It is found in our study that L2ers have over-reliance on pragmatic information during their interpretation of null subjects in Chinese, which we will argue below may not be fully accounted for by the IH. The explanation for L1 and L2 differences in the IH focuses on real-time processing, which assumes that the interpretation of pronominal forms requires the efficient integration and coordination of grammatical and pragmatic information in real time (Sorace, 2011). In an analysis of null subjects in Chinese on the basis of the IH, one would assume that in a situation where both syntactic and pragmatic information is potentially available for dependency resolutions of, say, a null subject in a Chinese sentence, L2ers would have less difficulty getting access to the syntactic information but more difficulty getting access to pragmatic information because the former is computed within the grammar, but the latter would require additional computational resources to integrate pragmatic information, which is from outside the grammar, in the dependency resolution, and, as a result, L2ers are assumed to be less sensitive to pragmatic information.

It is worth noting that according to the IH, the root cause of L2 problems at external interfaces lies in the additional cognitive demands imposed by the pragmatic or discourse information, which is external to the grammar. Syntactic properties however are within the linguistic computation and are thus less complex to process. Therefore, based on the findings from monolingual speakers that ‘parsing based on syntactic constraints is faster and more automatic’ than other constraints, Sorace (2011: 15) points out that ‘syntactic dependencies are more economical and operate at the very early stages of processing.’ It is important to emphasize that Sorace (2011: 17) also argues that there is no evidence that L2ers, at least near-native speakers, ‘use the processing strategies that privilege semantic, pragmatic and lexical information at the expense of syntactic information’.

Following the IH, it would therefore be more plausible to predict that, in order to relieve processing demands when these become temporarily unmanageable, L2ers may rely on syntactic cues more strongly than discourse-based/pragmatic cues and refer null subjects to the preceding subjects in their L2 Chinese. This is because the use of syntactic cues requires only syntactic computation, which is predicted by the IH to impose fewer demands on processing resources. Recall that previous cross-sectional L2 studies on the IH also find that, developmentally, the syntactic properties appear to be acquirable and stable in L2 (e.g. Slabakova et al., 2012). However, as demonstrated above, L2 learners in our study behave in a manner opposite to what is predicted in the IH. That is, L2ers place over-reliance on pragmatic information and under-use syntactic cues.

4 L1–L2 cue weighting mechanisms

The fact that English-speaking learners interpret Chinese null subjects differently from native Chinese speakers provides supporting evidence for the CBM (Cunnings, 2017), which states that differences in dependency resolutions between L1ers and L2ers can be explained by the fact that L2ers may have different weightings of individual cues from L1ers, and that L2ers are more susceptible to interferences.

Let’s use (1) – repeated in (10) – as an illustration. As explained in Section II, the null subject in (10) is syntactically bound to the topic. This is a syntactic constraint stipulating that the null subject in the main clause of (10) is co-indexed with the preceding subject/topic (John). This provides a syntactic cue ([+Subject]/[+Topic]) retrieving the preceding subject/topic as its antecedent. However, the recipient Ben is more likely to be ‘happy’ than John in the context of (10), which provides a discourse-based/pragmatic cue for the preceding object, Ben, as the antecedent.

a Dependency Resolution by CNSs

Our finding confirms that CNSs refer null subjects back to the subject of the preceding clause in both the subject-biased and object-biased conditions. This suggests that the syntactic cue consistently constitutes a more heavily weighted cue than the discourse-based/pragmatic cue for retrieving an antecedent for the null subject in Chinese. CNSs can resist the pragmatic bias and implement the subject orientation of null subjects in both subject-biased and object-biased contexts.

This is further evidenced in CNSs’ reading-time data, where CNSs are found to slow down when they read test sentences with null subjects in an object-biased context compared to those in a subject-biased context. Such a delay is believed to be due to the fact that the syntactic cue, [+Subject]/[+Topic], is a highly weighted cue during the dependency resolution for null subjects in Chinese. When the null subject in sentences like (11) is processed, the parser of CNSs utilizes the [+Subject]/[+Topic] cue, which retrieves the preceding subject as the antecedent and holds it in working memory while the rest of the sentence is processed. Upon encountering shangxin (‘sad’) in (11), a subject-biased context is created, which is in conformity with the syntactic cue. However, upon encountering gaoxing (‘happy’) in (10), an object-biased context is created, which goes against the syntactic cue. CNSs need to overcome and resist the temptation of the object bias in object-biased sentences like (10), which delays the processing of the sentence and leads the parser to take a longer time to process the object-biased sentences than subject-biased sentences.

In contrast to their reading time, no delay is observed in CNSs’ interpretation time for null subjects in an object-biased condition. This is believed to be related to the point of time when they make the dependency resolution. That is, CNSs make the dependency resolution of referring the null subject to the preceding subject in the object-biased context during their reading time, rather than waiting until the interpretation question is asked. This provides an account for the significant difference between the [Null subject in an object-biased context] and the [Null subject in a subject-biased context] found only in CNSs’ reading time but not in their interpretation time. In the interpretation part of the task, they simply indicate the dependency resolution made during the reading time, and no further dependency resolution is carried out at this stage.

b Dependency resolution by English-speaking learners of Chinese

As reported above, none of the L2 groups has a native-like performance in handling null subjects in object-biased sentences. BLs and Pre-ILs are found to be indeterminate as there is no significant difference between their ratings of the preceding subject and object as the antecedent of the null subject in an object-biased context. This is probably due to the fact that although BLs and Pre-ILs have been exposed to positive evidence of null subjects in their Chinese input, and they have the knowledge that null subjects are allowed in Chinese, the amount of input they are exposed to may not be sufficient enough to enable them to handle sentences involved in the experiment, resulting in indeterminate behaviours in the task.

As their Chinese proficiency improves, ILs, ALs, and VALs are found to be able to retrieve the subject antecedent in the subject-biased condition and at the same time incorrectly retrieve the object antecedent in the object-biased condition. This suggests that L2ers mainly depend on pragmatic information to interpret Chinese null subjects, and that L2ers fail to make a distinction between the subject orientation of null subjects and free orientation of overt subjects. But are English-speaking learners aware of the [+Subject]/[+Topic] cue when they are comprehending null subjects in Chinese sentences like (10)?

Our data suggest that the answer to this question should be positive, as we observe a longer interpretation time during ALs’ and VALs’ dependency resolution of the null subject in an object-biased context than that in a subject-biased context. This suggests that, in the former condition, a competition is taking place between two antecedents, the preceding subject, which is being retrieved by the [+Subject]/[+Topic] cue, and the preceding object, which is being retrieved by the discourse-based/pragmatic cue – see (10) – whereas in the latter condition, both the [+Subject]/[+Topic] cue and discourse-based/pragmatic cue are pointing to the same antecedent, i.e. the preceding subject; see (11). The longer time taken by ALs and VALs in interpreting null subjects in the [Null subject in an object-biased context] condition can be considered a result of the interference effect. That is, ALs’ and VALs’ reactions are slower in the object-biased condition because there is more than one item that matches a set of retrieval cues, which thus leads to a competition where the syntactic cue of [+Subject]/[+Topic] pushes for the preceding subject – e.g. John in (10) – as the antecedent, while the pragmatic information of the object bias promotes the preceding object – e.g. Ben in (10) – as the antecedent. Our likelihood-rating data suggest that the pragmatic-salient antecedent is winning the competition here. This is consistent with the findings of previous studies that L2ers are aware of retrieval cues available during dependency resolution (Cunnings, 2017; Felser, 2016; Felser and Cunnings, 2012; Patterson et al., 2014; Roberts et al., 2008). However, the fact that English-speaking learners incorrectly refer null subjects back to the object of the preceding clause in an object-biased context suggests that L2ers are more susceptible to interference than L1ers and that L2ers over-rely on pragmatic information during the resolution for Chinese null subjects. This may be because pragmatic information is the same across languages and functions as a prominent retrieval cue in L2 dependency resolution.

It is necessary to point out that a significant difference is found between the subject-biased condition and the object-biased condition in ALs’ and VALs’ interpretation time, but not in their reading time. This pattern is opposite to that of CNSs, who spend a significantly longer time reading test sentences with null subjects in an object-biased context than in a subject-biased context, whereas no significant difference is found in their so-called ‘interpretation’ time. We believe that the different times at which competition between antecedents for the dependency resolution of the null subject takes place between L1 and L2 explain this difference. For CNSs, it takes place during the reading time, but for L2ers, after the reading time. In other words, the competition and the dependency resolution occur earlier and probably rather briefly in CNSs’ Chinese, but they occur later and last probably longer in L2 Chinese. This finding enables us to have a better understanding of different mechanisms in L1 vs. L2 dependency resolutions.

VII Conclusions

Our study provides us with evidence for the difference in L1 vs. L2 dependency resolutions. Although L2ers can acquire the free orientation of overt subjects in Chinese, they have difficulty handling the subject orientation of null subjects in Chinese sentences. In the [Null subject in an object-biased context] condition, native Chinese speakers can resist the pragmatic bias and choose the subject as the antecedent, but English-speaking learners of L2 Chinese, including VALs, are influenced by the pragmatic bias and incorrectly take the object as the antecedent. This suggests that in contrast to native speakers, L2ers are more dependent on discourse-based/pragmatic cues during dependency resolution for the Chinese null subject. Although findings from a single study like ours here should not and cannot be used to discredit the IH, the data in the study suggests that it is more plausible to account for the findings with the CBM, which states that L1ers differ from L2ers in their susceptibility to interference and that L2ers over-rely on discourse-based/pragmatic cues to antecedent retrieval (Cunnings, 2017). This implies that the L2 cue weighting mechanism, which is different from that in L1, can influence the success in the L2 acquisition of dependency resolution. Another interesting finding in our study is that the competition between antecedents (target and distractors) takes place in both L1ers’ and L2ers’ dependency resolutions, but it occurs DURING the reading of the sentence in L1 Chinese but AFTER the reading of the sentence in L2 Chinese.

Supplemental Material

sj-docx-1-slr-10.1177_02676583221135189 – Supplemental material for Dependency resolutions of null and overt subjects in English speakers’ L2 Chinese: Evidence for the cue-based model

Supplemental material, sj-docx-1-slr-10.1177_02676583221135189 for Dependency resolutions of null and overt subjects in English speakers’ L2 Chinese: Evidence for the cue-based model by Lilong Xu and Boping Yuan in Second Language Research

Footnotes

Acknowledgements

We are very grateful to all participants and to those who helped us to recruit participants in this research project. Without their participation and support, this project would have been very difficult, if not impossible. Thanks also go to Professor Roumyana Slabakova, editor of Second language Research, and three anonymous Second Language Research reviewers for their insightful comments on and suggestions for earlier versions of this article. The experiment was conducted during the first author’s doctoral study at the University of Cambridge. We thank CSC–Cambridge Scholarship awarded to the first author. The first author would also like to acknowledge the support that she received from the International Postdoctoral Exchange Fellowship Program at Shanghai Jiao Tong University, during which this article was written.

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 authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Lilong Xu

Boping Yuan

Supplemental material

Supplemental material for this article is available online.

Notes

References

Badecker

Straub

(2002) The processing role of structural constraints on interpretation of pronouns and anaphors. Journal of Experimental Psychology: Learning, Memory, and Cognition 28: 748–769.

Barbosa

(2019) pro as a Minimal nP: Toward a unified approach to pro-drop. Linguistic Inquiry 50: 487–526.

Barr

Levy

Scheepers

, et al. (2013) Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language 68: 255–78.

Brown

Fish

(1983) The psychological causality implicit in language. Cognition 14: 237–73.

Chomsky

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

Cunnings

(2017) Parsing and working memory in bilingual sentence processing. Bilingualism: Language and Cognition 20: 659–78.

Cunnings

Fotiadou

Tsimpli

(2017) Anaphora resolution and reanalysis during L2 sentence processing: Evidence from the visual world paradigm. Studies in Second Language Acquisition 39: 621–52.

Felser

(2016) Binding and coreference in non-native language processing. In: Holler

Suckow

(eds) Empirical perspectives on anaphora resolution. Berlin / Boston, MA: De Gruyter, pp. 241–66.

Felser

Cunnings

(2012) Processing reflexives in a second language: The timing of structural and discourse-level constraints. Applied Psycholinguistics 33: 571–603.

10.

Grice

(1975) Logic and conversation. In: Cole

Morgan

(eds) Speech acts. Leiden: Brill, pp. 41–58.

11.

Guo

Yuan

(2020) Facilitative transfer only?: L3 acquisition of Mandarin sentence-final particle clusters by English–Cantonese bilinguals. Journal of Second Language Studies 3: 261–89.

12.

Guo

Yuan

(2021) Asymmetric transfer and development of temporal-aspectual sentence-final particles in English–Cantonese bilinguals’ L3 Mandarin grammars. Linguistic Approaches to Bilingualism. Epub ahead of print 26 Nov 2021. DOI: 10.1075/lab.20116.guo.

13.

Holler

Suckow

(2016) How clausal linking affects noun phrase salience in pronoun resolution. In: Holler

Suckow

(eds) Empirical perspectives on anaphora resolution. Berlin / Boston, MA: De Gruyter, pp. 61–86.

14.

Huang

C-TJ

(1984) On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531–74.

15.

Huang

(1994) The syntax and pragmatics of anaphora: A study with special reference to Chinese. Cambridge: Cambridge University Press.

16.

Jegerski

VanPatten

(2013) Research methods in second language psycholinguistics. New York: Routledge.

17.

Jiang

(2013) Conducting reaction time research in second language studies. New York: Routledge.

18.

Juffs

Harrington

(2011) Aspects of working memory in L2 learning. Language Teaching 44: 137–66.

19.

Lewis

Vasishth

(2005) An activation-based model of sentence processing as skilled memory retrieval. Cognitive Science 29: 375–419.

20.

Liu

C-M

(2014) A modular theory of radical pro drop. Unpublished PhD dissertation. Harvard University, Cambridge: MA, USA.

21.

Lozano

Mendikoetxea

(2010) Interface conditions on postverbal subjects: A corpus study of L2 English. Bilingualism: Language and Cognition 13: 475–97.

22.

Martin

McElree

(2011) Direct-access retrieval during sentence comprehension: Evidence from Sluicing. Journal of Memory and Language 64: 327–43.

23.

Montrul

(2010) How similar are adult second language learners and Spanish heritage speakers? Spanish clitics and word order. Applied Psycholinguistics 31: 167–207.

24.

Pan

H-Y

Schimke

Felser

(2015) Referential context effects in non-native relative clause ambiguity resolution. International Journal of Bilingualism 19: 298–313.

25.

Patterson

Trompelt

Felser

(2014) The online application of binding condition B in native and non-native pronoun resolution. Frontiers in Psychology 5: 147.

26.

Ratcliff

(1993) Methods for dealing with reaction time outliers. Psychological Bulletin 114: 510–32.

27.

Rizzi

(1982) Issues in Italian syntax. Dordrecht: Foris Publications.

28.

Roberts

(2009) A deletion analysis of null subjects. In: Biberauer

Holmberg

Roberts

al.

(eds) Parametric variation: Null subjects in minimalist theory. Cambridge: Cambridge University Press, pp. 58–87.

29.

Roberts

Gullberg

Indefrey

(2008) Online pronoun resolution in L2 discourse: L1 influence and general learner effects. Studies in Second Language Acquisition 30: 333–57.

30.

Rohde

Ettlinger

(2012) Integration of pragmatic and phonetic cues in spoken word recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition 38: 967–83.

31.

Rothman

(2009) Pragmatic deficits with syntactic consequences?: L2 pronominal subjects and the syntax–pragmatics interface. Journal of Pragmatics 41: 951–73.

32.

Slabakova

Ivanov

(2011) A more careful look at the syntax–discourse interface. Lingua 121: 637–51.

33.

Slabakova

Kempchinsky

Rothman

(2012) Clitic-doubled left dislocation and focus fronting in L2 Spanish: A case of successful acquisition at the syntax–discourse interface. Second Language Research 28: 319–43.

34.

Sorace

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

35.

Sorace

Serratrice

(2009) Internal and external interfaces in bilingual language development: Beyond structural overlap. International Journal of Bilingualism 13: 195–210.

36.

Stevenson

Knott

Oberlander

, et al. (2000) Interpreting pronouns and connectives: Interactions among focusing, thematic roles and coherence relations. Language and Cognitive Processes 15: 225–62.

37.

Y-C

(2020) Backward/forward anaphora in child and adult Mandarin Chinese. Language Acquisition 27: 187–216.

38.

Van Dyke

McElree

(2011) Cue-dependent interference in comprehension. Journal of Memory and Language 65: 247–63.

39.

Xie

Chang

Wang

(2020) Anaphoric preferences of Chinese null and overt subject pronouns: Testing and refining the position of antecedent hypothesis. Shijie Hanyu Jiaoxue [Chinese Teaching in the World] 1: 129–40.

40.

Yang

Gordon

Hendrick

, et al. (2003) Constraining the comprehension of pronominal expressions in Chinese. Cognition 86: 283–315.

41.

(1993) Chinese as a paratactic language. Journal of Second Language Acquisition and Teaching 1: 1–15.

42.

Zhao

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

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.03 MB