A Corpus-Based Study of Syntactic Complexity in L2 Japanese Writing: Insights from Usage-Based Approaches

Abstract

Recent research on second language (L2) writing has increasingly emphasized syntactic complexity as a key indicator of L2 writing proficiency. From an emergentist usage-based view, drawing on data from the B-JAS corpus, the study conducted both a longitudinal comparison across proficiency levels and a cross-sectional comparison between advanced learners and native speakers to investigate the features and development trends of syntactic complexity in L2 Japanese writing. Two indices were employed to measure syntactic complexity: the clause ratio (a large-grained index) and the ratio of different types of subordinate clauses (a fine-grained index). The results showed that learners had a higher clause ratio and greater use of te-form clauses at the advanced level than at the intermediate level. However, their use of relative clauses was lower, and their use of te-form clauses was significantly higher than that of native speakers. These findings reveal a syntactic usage tendency among learners, marked by an underuse of relative clauses and an overuse of te-form clauses. From a usage-based perspective, attentional biases from Chinese, frequency-based entrenchment, and the semantic ambiguity of te-form structures may primarily drive the observed syntactic usage patterns.

1. Introduction

Second language (L2) writing, as a form of comprehensible output, supports language use and acquisition, reflecting both compositional capability and linguistic proficiency (Tsubone, 2020). While compositional skills involve content and structure, linguistic proficiency covers grammar, vocabulary, and features like lexical richness. Due to its complexity, L2 writing is more challenging to enhance than basic skills like listening or speaking. In recent years, syntactic complexity—referring to structural complexity and variety (Foster & Skehan, 1996)—has become a key metric in evaluating L2 writing (Lee, 2006; Lu, 2011; Ortega, 2003; Lei, 2017; Jiang et al., 2019; Guo & Zhu, 2022; O’Leary & Steinkrauss, 2022; Gurmani et al., 2025).

Writing is used to convey information or ideas, and the goal of second language acquisition (SLA) is to achieve fluent language use. From this perspective, usage-based approaches provide a valuable framework, highlighting actual language use as the primary force behind linguistic form and the foundation of language learning (Tyler, 2010; Wulff & Ellis, 2018). Adopting this view, the present study analyzed the syntactic complexity of writing samples produced by learners with different levels of Japanese proficiency and native speakers, drawing on data from the B-JAS corpus (Beijing Corpus of Japanese as a Second Language) (NINJAL, 2023). Through a longitudinal comparison across proficiency levels and a cross-sectional comparison between advanced learners and native speakers, the study examined the features and development trends of syntactic complexity in L2 writing as evidenced by Chinese learners of Japanese.

2. Literature Review

2.1. Syntactic Complexity and Its Indices

In the field of SLA, the definition and evaluation of language proficiency have long been central topics of interest. Recently, language proficiency has been increasingly viewed as a composite capability comprising various elements, with three key evaluation criteria: complexity, accuracy, and fluency (referred to as “CAF”) (Craig & Judit, 2014; Vercellotti, 2017). Among these, research on complexity has become a significant focus in SLA. Language complexity refers to the range and degree of complexity of language output forms (Ortega, 2003; Lu, 2011; Bi & Jiang, 2020), encompassing both lexical complexity and syntactic complexity. Syntactic complexity refers to the ability to employ more complex language structures and a broader range of syntactic forms (Foster & Skehan, 1996), which is considered a critical standard for assessing the proficiency of L2 learners and the development of their target language skills (Norris & Ortega, 2009; Vyatkina, 2012). Some scholars have emphasized that this ability is manifested in writing through increases in sentence length and the use of complex clause structures (Chafe, 1982; Brown & Yule, 1983). Others have emphasized the significance of phrasal complexity—such as noun phrase modification, pre- and postmodifiers, and appositive structures—particularly in written and academic settings, highlighting its importance as a key component of syntactic complexity (Biber & Gray, 2010, 2016; Biber et al., 2011). In the domain of Japanese language acquisition, syntactic complexity has been defined as the ability to produce a wide range of clause types and sentence constructions that reflect increasing structural and functional sophistication. This includes, for example, conditional forms (to, ba, tara, nara), explanatory expressions such as ~n desu (Fan, 2017), and quotative constructions like ~to omou (Fuse & Suzuki, 2021).

In existing research, “the minimal terminable unit (T-unit)” is a core concept in current indices for measuring the syntactic complexity in L2 writing. It was first introduced by Hunt (1966) and defined as the main clause along with all subordinate clauses embedded within it, representing the smallest complete grammatical unit that can exist independently. The specific definition of a T-unit is illustrated in example sentences (1) to (3). The T-unit is the smallest complete grammatical unit to represent a sentence. Based on this, a simple sentence constitutes one T-unit (example sentence (1)). In the case of complex sentences, a main clause serves as the core to integrate subordinate clauses and modify elements to form a single T-unit. For example, sentence (2) is a single T-unit, comprising two clauses: a subordinate clause (underlined once) and a main clause. Moreover, if the clauses are syntactically equal in a compound sentence, each clause will independently satisfy the T-unit condition, thereby forming a multi-T-unit structure. For instance, sentence (3) contains two coordinated clauses (underlined twice), forming two T-units.

(1)【私は春が好きだ。】	1 T-unit
【I like the spring. 】	1 T-unit
(2)【[忙しかったため，][飲み会に参加できなかった。]】	1 T-unit
【[Due to being busy], [I was unable to attend the social gathering. ]】	1 T-unit
(3)【彼は部屋に入って➀】，【テレビを付けました。➁】	2 T-units
【He entered the room➀ 】【and switched on the TV➁.】	2 T-units

Current research on syntactic complexity based on the T-unit has produced over ten indices, which can be classified into large-grained and fine-grained indices depending on the level of detail in the examined language forms (Jiang et al., 2019). Large-grained indices capture overall syntactic usage, such as “the number of clauses per T-unit,” and “the number of T-units per sentence” (Wolfe-Quintero et al., 1998; Iwashita, 2006; Beers & Nagy, 2009; Yang et al., 2015; Bulté & Housen, 2014). Fine-grained indices analyze specific structures, such as coordinated phrases, complex noun phrases, and types of subordinate clauses (e.g., adverbial and relative clauses), offering more detailed insights (Biber et al., 2011; Lu, 2011; Parkinson & Musgrave, 2014; Kyle & Crossley, 2018; Wu & Lu, 2021; Nasseri, 2021; Peng et al., 2024). These two kinds of indices emphasize different aspects, and there is no clear superiority between them. Large-grained indices offer a broad, macro-level view of syntactic structures but lack transparency due to limited detail. In contrast, fine-grained indices provide detailed analysis of specific structures but may overlook overall writing patterns. Therefore, this study adopts a comprehensive approach combining both index types to more accurately assess learners’ syntactic complexity and provide deeper insight into their writing features.

2.2. Research on Syntactic Complexity in Second Language Writing

In research on syntactic complexity in L2 writing, the selection of measurement indices has evolved from focusing on single-dimensional measures to adopting multidimensional approaches. Early studies mainly used large-grained indices (e.g., T-unit length, clause ratios) to capture broad syntactic features and assess overall language proficiency (Casanave, 1994; Lee, 2006; Taguchi et al., 2013; Bulté & Housen, 2014). For example, Bulté and Housen (2014) analyzed sentence length and the frequency of simple structures in the writing of adult English learners. Lee (2006) found that Chinese learners produced lower clause ratios in Japanese argumentative writing compared to native speakers, even at the JLPT N1 level. However, the study lacked longitudinal data and did not investigate the underlying causes. More recent research has begun to adopt more fine-grained indices to capture specific and micro-level syntactic features (Jiang et al., 2019; Polat et al., 2020; Díez-Bedmar & Pérez-Paredes, 2020; X. Zhang & Lu, 2022; Li et al., 2022; Chen et al., 2024). This trend is particularly evident in studies on English as a foreign language (EFL) writing. For instance, X. Zhang and Lu (2022) applied Lu’s (2010) 14 syntactic complexity measures and additional dimensions from the TAASSC tool (Kyle, 2016) to analyze Chinese EFL learners’ application letters and argumentative essays. They found that fine-grained indices more effectively predicted writing quality and that genre influenced the predictive power of different measures. Similarly, Chen et al. (2024), along with Li et al. (2022), used multiple indices, including the frequency of complex phrasal structures and the ratio of subordinate clauses, to examine EFL learners’ writing. In contrast, the syntactic complexity measurements used in studies on Japanese L2 writing remain relatively limited. Most existing research has focused on dimensions such as the number or ratio of clauses and subordinate clauses per T-unit or sentence. While this reflects a shift from length-based to ratio-based metrics, the analyses still tend to be macro-level and lack detailed categorization of specific syntactic structures (Horiba & Matsumoto, 2008; Nishi, 2011). Given that Japanese attributive structures used to modify nouns are generally long, flexible in form, and highly variable, recent studies have also begun to examine the use of complex noun phrases in Japanese (Guo & Zhu, 2022). Sakoda & Hosoi (2020) focused on the ratio of different subordinate clause types in learners’ spoken Japanese. Nevertheless, fine-grained categorization of subordinate clause types has yet to be widely extended to studies of written Japanese.

In terms of research perspective, recent studies on syntactic complexity have gradually shifted from static, cross-sectional analyses to longitudinal investigations that track dynamic changes in syntactic development over time (Zheng & Feng, 2017; Guo & Zhu, 2022; Lei et al., 2023). Based on a large-scale learner corpus, Lei et al. (2023) conducted a longitudinal analysis of EFL learners’ writing performance over two academic years. The findings revealed that, as learners’ language proficiency improved, measures related to length, coordination, and phrasal complexity increased significantly, while the use of subordinate clauses declined. This pattern suggests a shift toward a more academic writing style. In comparison, a longitudinal study by Guo and Zhu (2022) on Japanese language learners found a steady increase in the use of complex structures in writing as learners’ proficiency improved. However, the absence of a native speaker comparison group in their study limited the interpretation of learners’ syntactic features.

The review of current research on syntactic complexity in L2 writing highlights two key issues that need further investigation. First, the selection of syntactical complexity indices remains inadequate. Despite the use of multiple indices for multidimensional analysis in most studies, the focus on fine-grained indices remains insufficient, particularly regarding the under-exploration of various types of subordinate clauses. While many studies report the overall ratio of subordinate clauses, few have examined how internal structural differences among these clauses contribute to syntactic complexity. Future research should aim to integrate large and fine-grained indices in a complementary manner to provide a more comprehensive account of learners’ syntactic abilities. Second, there is a lack of integration between longitudinal and cross-sectional approaches. Combining developmental perspectives with group comparisons can yield a more comprehensive understanding of how syntactic complexity evolves over time and provide empirical support for enhancing instructional practices in L2 writing.

2.3. Emergentist Usage-Based View

The two issues identified in previous studies correspond closely with the central concerns of the emergentist usage-based view. The term “usage-based” was first introduced by Langacker (1987), who argued that speakers’ linguistic systems are grounded in actual usage events or instances of speech. According to Langacker (1987), the language system emerges gradually through repeated symbolic units in specific usage events. With accumulated linguistic experience, more abstract patterns may develop; however, even these are considered fundamentally rooted in language usage. Both usage-based and emergentist theories of language acquisition maintain that language is acquired through use itself, positing that linguistic structure is an emergent property of language use (Behrens, 2009).

Regarding the first issue—the lack of detailed analysis of subordinate clause types—the usage-based perspective holds that linguistic knowledge is derived from recurring usage events. Each structural type (e.g., adverbial clauses, relative clauses) is gradually entrenched through repeated exposure to frequent and salient instances (Bybee, 2010). Therefore, a fine-grained analysis of specific structures, including their frequency, functional distribution, and input sources, can reveal the micro-level mechanisms of language acquisition. This approach exemplifies the construction of linguistic knowledge through use and corresponds with the emergentist usage-based view’s emphasis on “micro-pathways” of structural development (Winter-Froemel, 2018). As for the second limitation—the insufficient integration of longitudinal and cross-sectional comparisons—the usage-based framework highlights the processual and context-dependent nature of language learning (N. C. Ellis & Larsen-Freeman, 2009). It conceptualizes language acquisition as a temporally dynamic construction process. Additionally, differences in learners’ input environments and familiarity with linguistic structures can influence construction trajectories. Integrating longitudinal and cross-sectional analyses can thus reveal how particular structures evolve across learning stages and learner groups, providing a more comprehensive understanding of the developmental and variable nature of language acquisition.

In sum, the emergentist usage-based view offers a solid theoretical foundation for addressing these issues. Accordingly, this study adopted this perspective to guide further investigation.

2.4. Current Study

Building on the two aforementioned issues, this study investigates the development of syntactic complexity in the Japanese writing of Chinese university students from a usage-based perspective. To capture both general and specific aspects of syntactic performance, two indices were used: the clause ratio (the number of clauses per T-unit) and the ratio of different types of subordinate clauses (the number of each type of subordinate clause per T-unit). The study adopts a mixed-methods design, combining longitudinal and cross-sectional analyses, and aims to address the following research questions:

(1)

To what extent does the clause ratio differ across learning stages and between learners and native speakers?

(2)

How do different types of subordinate clauses develop in learners’ writing as language proficiency increases? Do their usage patterns differ significantly from those of native speakers?

(3)

What are the usage-based factors underlying the observed patterns of syntactic complexity?

3. Research Methods

3.1. Research Design

This study employed a comparative mixed-methods design that integrates longitudinal within-group analysis of learners and cross-sectional between-group comparison of learners and native speakers to investigate the development of syntactic complexity in L2 Japanese writing. Following previous research (Lee, 2006; Muraki, 2007; Sakoda & Hosoi, 2020), the clause ratio (the number of clauses per T-unit) was used as the large-grained index, while the ratio of different types of subordinate clauses (the number of each type of subordinate clause per T-unit) served as the fine-grained index to capture more detailed syntactic features. Statistical analysis was conducted using SPSS 27.0 (IBM SPSS 27.0, Chicago, IL, USA), with paired-sample t-tests applied to assess developmental changes within the learner group and independent-sample t-tests used for cross-group comparisons. To complement the quantitative analysis, textual examples were analyzed to interpret syntactic patterns from a usage-based perspective.

3.2. Corpus and Materials

The writing materials analyzed in this research were drawn from the “Beijing Corpus of Japanese as a Second Language (B-JAS)”. This corpus was jointly developed by the National Institute for Japanese Language and Linguistics (NINJAL), Beijing Center for Japanese Studies at Beijing Foreign Studies University, and the School of Foreign Languages at Beijing Normal University. Data collection was conducted from January 2016 to April 2019, compiling Japanese oral and written data from Chinese university students whose first language is Mandarin. The corpus was officially released on the NINJAL website in March 2023. The present study focused on the publicly available written component of the corpus to compute and analyze syntactic complexity. Ethical approval was obtained from the institutional ethics committee. All data were anonymized prior to analysis, and no personally identifiable information was accessed or used.

A full-sample analysis was conducted on all publicly available compositions in the corpus that met the research criteria. The dataset includes writing samples regularly collected from 18 Japanese major students in a Chinese university over a four-year period (twice in the first year and three times per year thereafter, totaling eleven tasks), along with writings by 19 native Japanese speakers produced under the same topics and conditions. The learners included in this corpus began studying Japanese from scratch upon entering university and followed the same Japanese major curriculum. Learners participated in the J-CAT (Japanese Computerized Adaptive Test) annually from their first to fourth years. Among the 18 learners, one individual was excluded due to incomplete J-CAT data, resulting in a final sample of 17 learners. Their average J-CAT scores increased significantly from 194 in the second academic year to 264 in the fourth year, indicating a clear improvement in language proficiency (t (16) = 6.22, p < 0.001, r = 0.841). The standard deviations and coefficients of variation for the two time points were relatively low (CV = 0.243 in the second year; CV = 0.165 in the fourth year), suggesting that learners’ proficiency levels were relatively homogeneous within each stage and that no extreme individual differences were observed. According to Imai et al. (2009), their proficiency was classified as intermediate in the second year and advanced in the fourth year. Therefore, this study conducted a longitudinal analysis of the 17 learners’ second- and fourth-year writings, as well as a cross-sectional comparison with the native speakers’ writings on the same topics and under the same conditions. (see

Table 1. Writing Topics and Word Count.

Groups	Writing Topic	Word Count
Second-year learners	私の一番欲しいもの (My Most Wanted Thing)	700 words
Fourth-year learners	私の好きな有名人 (My Favorite Celebrity)	1500 words
Native Japanese speakers	私の好きな有名人 (My Favorite Celebrity)	1500 words

Note. The writing samples from second- and fourth-year learners correspond to the 4th and 10th writing tasks, respectively, in the B-JAS corpus.

for detailed information).

3.3. Indices for Measurement

This study employed the clause ratio and the ratio of different types of subordinate clauses as indices to measure and analyze the syntactic complexity of written texts.

The measurement method of the clause ratio is based on Lee (2006) and Sakoda and Hosoi (2020), defined as “the number of clauses per T-unit (number of clauses/number of T-units).” Clauses conclude both the main clauses and the subordinate clauses. Subordinate clauses are identified based on forms such as “カラ(kara),” “ノデ(node),” “ト(to),” “トキ(toki),” and “テ(te) (expressing causes or reasons)” (refer to example sentence (4)). The criteria for T-unit division are based on linguistic forms such as “テ(te) (expressing coordination),” “タリ(tari),” “ナガラ(nagara),” and “ガ(ga) (expressing coordination)” (refer to example sentence (5)). In the example sentences, [➀]indicates the T-unit division, and {(i)} indicates the clause division.

(4)[{お腹が空いたから(i)}，{ご飯を作りました(ⅱ)}➀]
	2 clauses, 1 T-unit
[{Because of hunger(i)}, {I made a meal. (ⅱ)}➀]
	2 clauses, 1 T-unit
(5)[{庭には灯篭があって(i)}➀]，[{茶室まで細い道が続いていた(i)}➁]
	2 clauses, 2 T-units
[{A lantern was in the garden(i)}➀], [{and a narrow path led to the tea room(i)}➁]
	2 clauses, 2 T-units

This study used the subordinate clause classification proposed by Sakoda and Hosoi (2020) and Muraki (2007) to measure the ratio of different types of subordinate clauses. Sakoda and Hosoi (2020) identified seven types commonly appearing in learners’ spoken data: quotation clauses, adnominal clauses, conditional clauses, toki-form clauses (expressing the time), te-form clauses1 (showing causal reasoning with a slight sense of temporal sequence), kara/node-form clauses (indicating causality), and contrastive clauses. Muraki (2007) further subdivided adnominal clauses into “genuine adnominal clauses” (a clause followed by an independent, meaningful noun, which can be referred to as a relative clause) and “pseudo adnominal clauses” (a clause followed by formal nouns such as ‘こと(koto)’ or ‘ところ(tokoro)’, which can be referred to as a nominal clause). Although the initial annotation process identified all eight types of subordinate clauses proposed in previous studies (Sakoda & Hosoi, 2020; Muraki, 2007), three of them—kara/node-form clauses, toki-form clauses, and contrastive clauses—occurred at very low frequencies across both learner and native texts. Given their limited occurrence and negligible statistical contribution, these three types were excluded from subsequent quantitative analyses. Therefore, this study focused on the five most frequently used types in written content: relative clauses, nominal clauses, quotation clauses, te-form clauses, and conditional clauses (see

Table 2. Definitions and Example Sentences of Different Types of Subordinate Clauses.

Type	Definition	Example Sentence (Subordinate Clause in [ ])
Relative Clause	A clause that modifies a noun, providing additional information about it.	[健常者と障害者が時間を共有する]空間を創造する。 (Create a space [where able-bodied and disabled people share time].)
Nominal Clause	A clause followed by formal nouns that act as noun-like elements in the sentence.	[目標や希望がある]ことは私の心の支えとなりました。 ([Having goals and hopes] became the support of my heart.)
Quotation Clause	A clause expressing a quoted opinion or idea, using forms like “…と(to)” or “…よう(に) (yōni)”.	母親が子供に[花瓶を割らないように]注意した。 (The mother warned the child [not to break the vase].)
te-form Clause	A clause expressing a reason or cause, typically with a te-form verb, where the first action temporally causes the second.	[階段から落ちて]骨折した。([Having fallen down the stairs], I broke my bone.)
Conditional Clause	A clause that expresses a condition, using forms such as “…ば(ba),” “…たら(tara),” or “…なら/のなら(nara/nonara)”	[秋になれば]もみじが赤く染まります。 ([If autumn comes], the leaves turn red.)

Note. Definitions are adapted from Muraki (2007) and Sakoda and Hosoi (2020). All example sentences are created by the author for illustrative purposes.

for definitions and example sentences). This selection ensures both the reliability of the coding process and the statistical interpretability of the results.

3.4. Data Analysis

This study employed the clause ratio and the ratio of different types of subordinate clauses as indices of syntactic complexity. The clause ratio was calculated as “total number of clauses/total number of T-units”, and each subordinate clause type ratio was calculated as “the number of a specific subordinate clause type/total number of T-units”. For each text, the total number of clauses, T-units, and each subordinate clause type was manually annotated by the researcher and verified by two trained coders to ensure reliability.

Descriptive statistics (mean, standard deviation, coefficient of variation) were first computed to examine the distribution and variability of the syntactic indices. Inferential statistics were then conducted using SPSS 27.0 (IBM SPSS 27.0, Chicago, IL, USA). Paired samples t-tests were conducted to compare syntactic complexity between the second and fourth years within the same group of learners, and independent samples t-tests were conducted to compare fourth-year learners with native speakers. The significance threshold was set at p < 0.05. Textual examples were also analyzed qualitatively to supplement and interpret the quantitative findings.

4. Results

4.1. Large-Grained Index: The Clause Ratio

Table 3. The Clause Ratio for Learners and Native Speakers.

	Intermediate Level (n = 17)	Advanced Level (n = 17)	Native Speakers (n = 19)
The Clause Ratio	1.86	2.07	2.15
SD	0.26	0.28	0.29

presents the clause ratio measured for learners with intermediate and advanced Japanese proficiency, as well as native speakers.

In

Table 4. Paired-Sample t-Test of the Clause Ratio Between Proficiency Levels.

Mean	Std. Deviation	Std. Error Mean	95% CI Lower	95% CI Upper	T	Df	Sig. (2-Tailed)
−0.22	0.35	0.09	−0.40	−0.03	−2.51	16	0.02

, a paired-sample t-test was applied to compare the clause ratio of 17 intermediate to advanced learners. Sig (0.02) is lower than 0.05, indicating a significant difference between the two stages of Japanese learning, with a significantly higher ratio at the advanced level compared with the intermediate level (t(16) = −2.51, p = 0.02, r = 0.53).

In

Table 5. Independent-Sample t-Test of Clause Ratio Between Advanced Learners and Native Speakers.

T	Df	Sig. (2-Tailed)	Mean Difference	Std. Error Difference	95% CI Lower	95% CI Upper
−0.79	34	0.44	−0.08	0.10	−0.27	0.12

, an independent-sample t-test was conducted to examine whether there was a considerable difference in clause ratios between 17 advanced learners and 19 native speakers. Sig (0.445) higher than 0.05 indicates unremarkable differences in the clause ratio between the groups (t (34) = −0.79, p = 0.44, r =0.13).

4.2. Fine-Grained Index: The Ratio of Different Types of Subordinate Clauses

Table 6. The Ratio of Different Types of Subordinate Clauses for Learners and Native Speakers (SD in Parentheses).

	Relative Clause	Nominal Clause	Quotation Clause	te-Form Clause	Conditional Clause
Intermediate learners (n = 17)	0.26 (0.15)	0.12 (0.09)	0.15 (0.09)	0.18 (0.11)	0.08 (0.07)
Advanced learners (n = 17)	0.34 (0.16)	0.15 (0.07)	0.13 (0.07)	0.27 (0.15)	0.07 (0.04)
Native speakers (n = 19)	0.47 (0.13)	0.18 (0.07)	0.16 (0.09)	0.15 (0.07)	0.07 (0.05)

illustrates the ratio of different types of subordinate clauses measured for learners and native speakers, examining subordinate clauses within T-units to provide a more detailed analysis of syntactic complexity.

Table 7. Paired-Sample t-Test of the Ratio of Different Types of Subordinate Clauses Between Proficiency Levels.

	Mean	Std. Deviation	Std. Error Mean	95% CI Lower	95% CI Upper	T	Df	Sig. (2-Tailed)
Relative Clause	−0.09	0.23	0.06	−0.20	0.03	−1.59	16	0.13
Nominal Clause	−0.02	0.10	0.02	−0.08	0.03	−1.00	16	0.34
Quotation Clause	0.02	0.12	0.03	−0.04	0.08	0.63	16	0.54
te-form Clause	−0.09	0.16	0.04	−0.16	−0.01	−2.27	16	0.04 *
Conditional Clause	0.01	0.08	0.02	−0.04	0.05	0.30	16	0.77

(* p < 0.05).

shows the results of paired-sample t-tests. According to the paired-sample t-tests, there is a significant difference in the ratio of te-form clauses between the two stages of Japanese learning, with a significantly higher ratio of te-form clauses at the advanced level than at the intermediate level (t (16) = −2.27, p = 0.04, r = 0.49). However, since multiple t-tests were conducted across the five types of subordinate clauses, the risk of Type I error may be inflated. To address this issue and enhance the robustness of the findings, a Bonferroni correction for multiple comparisons was applied (adjusted α = 0.01). Under this more conservative threshold, the observed result did not remain statistically significant and should thus be interpreted as marginally significant. Nevertheless, given the exploratory nature of this study and the medium effect size observed (r = 0.49), the result is retained for discussion, with a recommendation for further validation in future studies with larger samples. No other clause types showed statistically significant differences between the two levels.

Table 8. Independent-Sample t-Tests of the Ratio of Different Types of Subordinate Clauses Between Advanced Learners and Native Speakers.

	T	Df	Sig. 2-Tailed	Mean Difference	Std. Error Difference	95% CI Lower	95% CI Upper
Relative Clause	−2.60	34	0.01 *	−0.13	0.05	−0.23	−0.03
Nominal Clause	−1.50	34	0.14	−0.04	0.02	−0.08	0.01
Quotation Clause	−0.83	34	0.41	−0.02	0.03	−0.08	0.03
te-form Clause	3.22	34	0.003 **	0.12	0.04	0.05	0.20
Conditional Clause	−0.36	34	0.72	−0.01	0.01	−0.04	0.02

(* p < 0.05, ** p < 0.01).

summarizes the results of independent-sample t-tests comparing the ratios of different subordinate clause types between advanced learners and native speakers. Significant differences were found in the use of relative clauses and te-form clauses: advanced learners produced a significantly lower ratio of relative clauses (t (34) = −2.60, p = 0.01, r = 0.41) and a significantly higher ratio of te-form clauses (t (34) = 3.22, p = 0.003, r = 0.48) compared to native speakers. As multiple comparisons were conducted across five subordinate clause types, a Bonferroni correction was applied (adjusted α = 0.01) to control for inflated Type I error. Under this correction, the difference in te-form clause usage remained statistically significant, while the result for relative clauses fell exactly at the corrected threshold and should be interpreted with caution. Nevertheless, given the exploratory nature of the study and the moderate effect size (r = 0.41 for relative clauses), the finding of relative clauses is retained to highlight potential usage tendencies that warrant further investigation.

5. Discussion

5.1. Discussion Based on Statistical Results (RQ1 & RQ2)

Regarding RQ1, based on the results of the large-grained index, it is evident that the clause ratio at the advanced level was significantly higher than at the intermediate level, indicating that as language exposure increases and L2 proficiency improves, the syntactic complexity in learners’ writing also increases. From the usage-based view, it can be said that the complexity of language is gradually constructed over time as learners are exposed to and interact with linguistic forms in various contexts. Additionally, no significant difference was found in the clause ratio between advanced learners and native speakers, suggesting similar syntactic complexity in writing regarding large-grained measures. The results contrast with Lee (2006), who found that even international students who had passed the N1 level of the JLPT used fewer complex syntactic structures than Japanese students. The differing results may stem from the writing difficulty and time constraints. Lee (2006) used a more complex topic (“日本では本当に国際化が実現しているかどうか” [Has Japan truly achieved internationalization?]) and shorter texts (around 400 words), which likely led learners to prioritize content over syntactic complexity. Considering the time factor, Lee imposed a 30 min limit, whereas the B-JAS corpus allowed unlimited time, providing learners with more opportunities to refine both content and structure, thereby narrowing the syntactic gap with native speakers. The results of this study at least indicate that, under more generous time conditions, when advanced learners write about familiar topics in Japanese, their syntactic complexity in writing may be comparable to that of native speakers, which means that with increased language exposure and refined understanding of language use, learners can produce more complex syntactic structures, supporting the emergentist usage-based view that language complexity develops gradually through sustained and cumulative usage.

Regarding RQ2, the fine-grained index revealed a marginally significant increase in the use of te-form clauses from the intermediate to the advanced stage. Compared with native speakers, advanced L2 learners also showed a significantly higher ratio of te-form clauses and a marginally significant lower ratio of relative clauses. From a more specific perspective, investigating how specific syntactic structures are constructed through language use, these results indicate that learners become more proficient in using te-form clauses as their Japanese proficiency improves. At the same time, although relative clauses were used more frequently than the other four types of subordinate clauses at both learning stages (see Table 6), their overall usage ratio remained lower than that of native speakers, suggesting a continued gap in learner performance. This indicates that relative clauses may still pose complexity and difficulty for learners in the context of L2 writing.

By incorporating both large-grained and fine-grained indices, this study offers a comprehensive analysis of learner writing and its development across proficiency levels. On the one hand, from the intermediate to the advanced stage, learners showed a significant improvement in overall clause-level complexity, gradually approaching native speaker levels. Notably, the use of te-form clauses increased substantially and even surpassed that of native speakers at the advanced stage. This suggests that learners’ development in syntactic complexity may rely heavily on their proficient use of te-form constructions. On the other hand, although advanced learners demonstrated a similar overall level of clause complexity to native speakers, the patterns of subordinate clause usage differed between the two groups. Both groups used relative clauses most frequently among the five subordinate clause types (see Table 6), yet learners’ relative clause ratio was still slightly lower than that of native speakers. In contrast, their use of te-form clauses was significantly higher. These results suggest that learners may compensate for the less frequent use of relative clauses by relying more on te-form constructions, thereby achieving an overall level of syntactic complexity comparable to that of native speakers. Further analysis of syntactic development from the intermediate to the advanced stage revealed no significant increase in the use of relative clauses, while the ratio of te-form clauses showed a notable rise. This suggests that learners’ frequent use of te-form clauses may represent a structural preference gradually developed during the intermediate-to-advanced stages of language acquisition.

5.2. Discussion Based on Writing Content (RQ3)

The data analysis revealed overall trends in the development of syntactic complexity in learners’ writing, addressing RQ1 and RQ2. According to emergentist usage-based theories, language is acquired through use; therefore, further analysis of actual writing samples can provide deeper insight into the syntactic usage of specific subordinate clauses. As previous findings highlight relative clauses and te-form clauses as key contributors to observed learners’ syntactic features, this section will analyze their practical usage in the actual writing materials from the B-JAS corpus and explore the underlying reasons behind these usage tendencies, aiming to address RQ3.

5.2.1. Use of Relative Clauses

Examples 6 to 8 are authentic sentences extracted from the B-JAS corpus, produced by learners at different levels of Japanese learning and by native speakers. All three sentences appear at the beginning of the composition to introduce the topic, and each describes a person or thing the writer likes. While the content is comparable across the examples, they differ in their use of syntactic structures ([] indicates the relative clause).

(6) いろんなゲームに興味があって一番欲しいものはソニーのプレイステーション４代、即ちPS4です。 (Learner Intermediate Level B-JAS)

I am interested in various games, and what I want the most is Sony’s PlayStation 4, that is, the PS4.

(7) 今私の大好きな有名人は1つのアイドルグループのメンバー村上信五さんである。

(Learner Advanced Level B-JAS)

The celebrity I like the most right now is Shin-go Murakami, a member of an idol group.

(8) 私が好きな有名人は［30歳を過ぎてもなお現役の第1線を行く］アイススレッジホッケーのオリンピック選手の上原大佑さんです。 (Native Speaker B-JAS)

The celebrity I like is Daisuke Uehara, an Olympic ice sledge hockey player [who is still active in the first line of competition even after turning 30].

The above examples show that native Japanese speakers tend to use longer relative clauses as attributive structures to modify nouns (e.g., “30歳を過ぎてもなお現役の第1線を行く”, “who is still active in the first line of competition even after turning 30”), reflecting stronger skills in information integration and syntactic organization. In contrast, learners often opt for direct narration to convey straightforward content, suggesting that they have not yet developed the habitual use of relative clauses as standard modifiers, as native speakers do. This contrast is also evident in Examples 9 and 10: learners present information in separate sentences, while native speakers integrate details into a single, complex structure via relative clauses. Notably, however, relative clauses remain the most frequently used type of subordinate clause among learners (see Table 6). This suggests that while learners have not fully acquired the ability to use structurally complex relative clauses as native speakers do, they have at least begun to make more frequent use of such constructions in their subordinate clause production.

(9) 私の好きな有名人は宮野真守という人である。アダ名はマモあるいはマもちゃん。

(Learner Advanced Level B-JAS)

My favorite celebrity is Mamoru Miyano. He is nicknamed Mamo or Mamo-chan.

(10) ［世の中の暗い所に焦点を当てた］風刺的な作品が多く。

(Native Speaker B-JAS)

There are many satirical works [that focus on the darker sides of the world].

From the usage-based view, L2 experience is influenced by attentional biases rooted in the first language (L1), as well as other types of interference. Transfer is a widespread phenomenon in SLA (Odlin, 1989; Flege, 2002; Jarvis & Pavlenko, 2008). N. C. Ellis et al. (2014) noted that learners’ prior experience with particular cue dimensions can influence the focus of their explicit attention during language processing, which in turn gives rise to implicit attentional biases during comprehension and production. Consequently, compared to native speakers, learners’ usage pattern of relative clauses in written production may stem from the influence of their L1 experience.

Noriko (2013) observed that learners’ native language features can affect the production of relative clauses. The pattern of using relative clauses among Chinese learners of Japanese, compared to native speakers, may be due to the linguistic characteristics of Chinese. Qin (2000) argued that the syntactic structure of Chinese is relatively fixed and uses fewer long modifiers. Furthermore, scholars such as Wang (1984), Pan (1997), and L. Zhang (2024) have classified Chinese as a paratactic language. In parataxis, clause connections in complex sentences are typically established through semantic relations rather than formal linking devices. In other words, syntactic relations are not marked by overt grammatical forms but are instead constructed through the logical relationships among lexical meanings. This language type contrasts with the hypotactic structure commonly found in Western languages, where grammatical relationships are indicated through morphological means (L. Zhang, 2024). As a paratactic language, Chinese presents clauses sequentially without clear linguistic markers indicating their connections. The relationship between clauses can only be understood through a holistic semantic analysis (Xu, 2011). This results in a lower frequency of long relative structures in Chinese, and consequently, reduced perceptual salience for such structures among native Chinese speakers.

During SLA, all target language input is filtered through the L1 cognitive framework, meaning not all input is equally noticed or processed (Corder, 1967). As Schmidt’s (2001) Noticing Hypothesis suggests, conscious attention to linguistic forms in the input is essential for learning: “people learn about the things they attend to and do not learn much about the things they do not attend to” (p. 30). From this perspective, the typological characteristics of Chinese may exert negative transfer effects, suppressing learners’ sensitivity to Japanese relative clause structures and reducing both the efficiency of acquisition and frequency of use. Nevertheless, since long attributive constructions are a prominent feature of Japanese, they are often emphasized repeatedly in instructional settings. As a result, learners may still achieve a certain degree of acquisition and use through explicit teaching and practice. This helps explain why relative clauses still have a high usage frequency among learners. However, compared to native speakers, learners have not yet developed the habitual use of relative clauses as default modifying structures. Especially when expressing straightforward, everyday content, they still tend to unconsciously revert to the paratactic expression patterns characteristic of their first language.

5.2.2. Use of te-Form Clauses

Compared to native speakers, learners tend to overuse te-form clauses. Examples 11–14 illustrate typical misuse resulting from this overreliance, with the following analysis exploring the underlying causes.

(11) ［地域が違って］、文化は違う。 (Learner Intermediate Level)

[Because the regions differ], the culture differs.

(12) ［最初の夢を諦めなくて］、難しくても頑張り続けます。 (Learner Advanced Level)

Even though it is difficult, I will continue to work hard [because I have not given up on my original dream].

(13) 大学で、［日本語を専攻にして］、その夢にもっと近づけることが分かっています。 (Learner Intermediate Level)

[By majoring in Japanese], I realized I could get closer to my dream.

(14) 普段，［高く澄んだ声は聞き飽いて］、アデルのやや低めで、しゃがれた声が新鮮感をもたらしてくれるのだ。 (Learner Advanced Level)

[I am usually tired of hearing high, clear voices], but Adele’s slightly lower, husky voice brings a refreshing feeling.

The above examples all involve misuses of te-form clauses. The corrected versions of these sentences are shown below.

(11’) ［地域が違うと］、文化も違う。

(12’) ［最初の夢を諦めずに］、難しくても頑張り続けます。

(13’) 大学で、［日本語を専攻にしたら／すれば］、その夢にもっと近づけることが分かっています。

(14’) 普段、［高く澄んだ声は聞き飽きたので］、アデルのやや低めで、しゃがれた声が新鮮感をもたらしてくれるのだ。

As shown above, learners often use te-form clauses in place of other grammatical structures in their writing—for instance, replacing the “to” for expressing “generic conditionals” (11’), “nai de” for expressing “ attendant circumstances “ (12’), “tara/ba” for “hypothetical conditionals “ (13’), and “node/kara” for “cause/reason” (14’).

From a usage-based perspective, the overuse of te-form clauses may be attributed to frequency-based entrenchment and misuse due to semantic ambiguity. Entrenchment refers to the way linguistic structures are stored and organized in the mental inventory, enhancing the strength of their mental representation. As Langacker (1987) noted, increased frequency deepens entrenchment, describing a continuum of entrenchment within cognitive organization. Each instance of processing and using a particular structure positively reinforces its degree of entrenchment. Through repeated use, a novel structure gradually becomes entrenched and forms a psychological unit, enhancing its automaticity and fluency in comprehension and processing. In Japanese, the te-form serves as a common inflectional and conjunctive form that expresses various meanings, including coordination, causation, temporal sequence, and ellipsis, which is frequently used in both spoken and written language (Katō, 2006). Consequently, the te-form has a high input frequency in the course of Japanese language learning. Moreover, it is typically introduced at an early stage in Japanese language education, and continued practice and repeated use throughout the learning process further enhance learners’ fluency with the te-form, leading to a relatively high frequency of te-form clauses in their output. In summary, the high input and output frequency of te-form clauses leads to stronger entrenchment, facilitating their automatic comprehension and processing, and thus contributing to their overuse in learners’ writing.

At the same time, some scholars have pointed out that many high-frequency grammatical structures exhibit a high degree of semantic ambiguity, which poses significant challenges for language learners’ acquisition and use (Goldschneider & DeKeyser, 2001; DeKeyser, 2005; Gries, 2015; N. C. Ellis, 2008). In Japanese, the te-form as a conjugation contains multiple meanings. However, as it is introduced at the beginning of Japanese education, it is difficult to offer a comprehensive and in-depth explanation of all meanings of te-form due to the limited proficiency of learners at that stage. As a result, learners may misunderstand its usage and focus solely on its versatility, overlooking the subtle differences between the te-form and other similar grammatical structures. In fact, Liao (2022) noted that due to the widespread use of the te-form, learners tend to believe that errors can be minimized using te-form clauses. It can also be inferred from the findings of the present study that since te-form clauses are relatively easy to use, they are typically selected by learners who are unsure about the most suitable structure for subordinate clauses. However, Katō (2006) pointed out that while the te-form is often used to express causality or reasons, it typically carries a sense of sequence, making its causal relationship less clear than structures like “node” or “kara.” Additionally, the te-form’s grammatical function is somewhat ambiguous, contributing to its misuse by learners.

6. Conclusions

From the emergentist usage-based view, this study examined the syntactic complexity of Chinese learners’ Japanese writing across different proficiency levels, with a focus on the use of five types of subordinate clauses. The results show that although learners’ overall clause-level syntactic complexity gradually approached native speaker levels with increasing proficiency, significant differences remained in the use of specific subordinate clause types. Relative clauses, while the most frequently used among the five types by learners, were still used at a lower ratio compared to native speakers. In contrast, the frequent use of te-form clauses by learners appears to compensate for this disparity, resulting in overall syntactic complexity levels that are comparable to those of native speakers. Notably, this structural preference began to emerge at the intermediate level and became more pronounced at the advanced level. Its development may be driven by attentional biases rooted in the L1, frequency-based entrenchment, and semantic ambiguity. These findings point to an important general implication: surface-level measures of syntactic complexity, such as clause counts, may obscure underlying differences in structural choices. In other words, learners may rely heavily on familiar or multifunctional constructions like te-form clauses to offset the limited use of cognitively demanding structures such as relative clauses. While this strategy may yield comparable levels of complexity in terms of quantity, it does not reflect native-like structural use.

These findings offer important pedagogical implications for Japanese language education. Previous research has shown that to acquire specific structures in L2 effectively, learners must engage in conscious and selective attention to the target forms (Wulff & Ellis, 2018). Due to the paratactic nature of Chinese, learners may overlook relational structures in Japanese during early stages of exposure. Although later instruction may partially address this gap, learners generally have not developed the fluency or automatization necessary for the frequent and natural production of relative clauses. Therefore, further emphasis on form-focused instruction (FFI) is recommended. By implementing FFI, instructors can help learners direct their attention to grammatical structures that might otherwise go unnoticed (R. Ellis, 2012), ultimately supporting more effective use in production. As for the overuse of te-form clauses, it is important not only to introduce the basic functions of te-form conjugation at the early stages of Japanese acquisition, but also to provide more detailed instruction on the specific meanings of its various usages as learners’ proficiency increases and their exposure to te-form structures accumulates. Such instruction should help learners distinguish te-form clauses from similar structures, focus on their specific meanings and functions in high-frequency contexts, and reduce misuse caused by semantic ambiguity.

In conclusion, this study demonstrates that learners’ syntactic choices in writing reflect underlying cognitive processing and language experience patterns. Future Japanese language instruction should pay closer attention to these deeper representational mechanisms, providing targeted guidance to enhance learners’ sensitivity to syntactic structures and their ability to use them effectively, thereby improving overall L2 writing proficiency. At the same time, to further validate and extend these insights, future research should involve larger samples and adopt more robust statistical methodologies to enhance the generalizability of the results.