The dynamics of changes in linguistic complexity and writing scores in timed argumentative writing among beginning-level EFL learners

2.1 Dynamic patterns of linguistic complexity development in L2 writing

In essence, CDST perceives language learning as an emergent, non-linear process within a dynamic and interconnected system. Two main principles govern the CDST framework: the relational principle and the adaptive principle (Hiver et al. 2022). The relational principle assumes that language systems consist of multiple subsystems, which are interconnected and interact with each other. This means that any changes in one subsystem may lead to changes in other subsystems, and even the entire system (Larsen-Freeman 2006). According to the adaptive principle, language systems exhibit both stability and dynamic change, and this relationship is characterized by non-linear development (Hiver et al. 2022). The non-linear nature of language development reflects that changes may emerge unexpectedly, rather than occurring in a predictable, linear manner. For example, L2 learners may undergo rapid progress in using certain linguistic features, followed by periods of slower development or even regression.

A wealth of CDST-oriented L2 writing research has examined written output using the CAF framework (e.g. Bulté and Housen 2020; Larsen-Freeman 2006; Verspoor et al. 2008; Vyatkina 2012). This research specifically centers on linguistic complexity, which has been extensively explored using the CAF dimensions of linguistic performance. In the remaining text, we use the term “linguistic complexity” in an absolute sense, referring to the objective quantitative aspects of a language feature (Bulté and Housen 2012). The main components of linguistic complexity include lexical, phraseological, and syntactic complexity (e.g. Bulté and Housen 2020; Paquot 2019). In the subsections that follow, we explore how CDST-oriented research has examined lexical, phraseological, and syntactic complexity in L2 writing.

2.2 Interconnected dynamics of lexical and syntactic complexity in L2 writing

Although lexical, phraseological, and syntactic complexity in L2 writing display dynamic changes over time, it is important to recognize that they are intricately interconnected (e.g. Bulté and Housen 2020; Larsen-Freeman 2006). Research from a CDST perspective has explored possible interactions among various dimensions of language subsystems, uncovering four distinct longitudinal relationships (i.e., the connections between at least two variables of interest observed over an extended period of time): support (features evolving together, also called connected growers), competition (features developing in opposite directions), precursor (one feature’s growth preceding another’s), and asymmetry (changing relationships between features over time; Verspoor and van Dijk 2011).

Prior studies in the context of CDST and L2 writing have documented longitudinal connections between lexical and syntactic complexity (e.g. Bulté and Housen 2020; Caspi 2010; Larsen-Freeman 2006; Spoelman and Verspoor 2010; Verspoor et al. 2008, 2012; Vyatkina 2012). Multiple studies have reported a supportive relationship between lexical and syntactic complexity (e.g. Spoelman and Verspoor 2010; Vyatkina 2012). Spoelman and Verspoor (2010) conducted an analysis of L2 (Finnish) academic writing produced by a Dutch learner over three years. Their findings reported a supportive relationship between lexical complexity, quantified by the number of morphemes per word, and NP complexity, assessed by the number of words per NP. Vyatkina (2012) delved into a cross-sectional corpus of L2 (German) essays produced by L2 beginning-level learners, supplemented by the writings of two focal students with four-semester longitudinal data. The findings indicated a positive relationship between lexical diversity, measured using the corrected type-token ratio, and global syntactic complexity, measured by the mean length of sentence. This finding suggests that, over time, L2 learners at the beginning level tend to produce longer sentences while also incorporating a more diverse range of vocabulary, thus reflecting a pattern of connected growth.

A competitive relationship between lexical and syntactic complexity has also been documented in the study conducted by O’Leary and Steinkrauss (2022). Their research examined L2 (English) essays produced by three Dutch-speaking advanced-level university students over a span of four years. Their findings revealed a negative correlation between lexical diversity (measured using Vocd-D and the Guiraud index) and phrasal complexity (assessed through MLC and CN per sentences). This observation suggests a competitive dynamic where resources are allocated between lexical diversification and phrasal sophistication.

2.3 Improvement of L2 holistic writing scores over time

In the context of L2 writing, holistic or analytic writing scores evaluated by expert raters are important because they can be indicative of an individual’s L2 writing ability, which often hinges on “the selection of appropriate words and phrases; on facility with the conventions of grammar, punctuation, and spelling; and on the competent use of logic and rhetorical devices to sustain a reader’s attention and direction” (Cumming et al. 2000: 14). Consequently, improvements in writing scores, as determined by scoring rubrics that present “implicit assumptions … about the development of L2 writing skills” can signify the advancement of L2 writing ability (Valdes et al. 1992: 334–335).

From a CDST perspective, prior studies have explored longitudinal changes in L2 holistic (rather than analytic) writing scores (e.g. Huang et al. 2021; Lowie and Verspoor 2019). Lowie and Verspoor (2019) examined timed L2 narrative writing among 22 Dutch learners of English throughout one academic year in a CLIL setting in a secondary school, noting an increase in holistic writing scores, which was expected due to the students’ immersion in a CLIL environment. Huang et al. (2021) investigated untimed L2 writing assignments of various genres completed by 22 Chinese EFL learners at an intermediate level over an academic year at a university. The results showed a significant improvement in L2 holistic writing scores over time, even though the learners had limited exposure to English and did not receive writing-focused instruction. These collective findings suggest that repetitive writing practice can play a pivotal role in enhancing L2 writing scores.

2.4 Current study

Collectively, a growing body of CDST-oriented studies have revealed the dynamic development of lexical, phraseological, and syntactic complexity in L2 writing (e.g. Bulté and Housen 2020; Duan and Shi 2021; Larsen-Freeman 2006; Verspoor et al. 2008; Vyatkina 2012), the interrelated dynamics between lexical and syntactic complexity in L2 writing (e.g. Huang et al. 2021; Lowie and Verspoor 2019), and longitudinal changes in L2 holistic writing scores (e.g. Huang et al. 2021; Lowie and Verspoor 2019).

However, there exist at least four research gaps in our understanding of the development of linguistic complexity requiring more scrutiny. First, the scope of complexity measures used in previous studies and the scope of contexts in which linguistic complexity has been examined appear to be somewhat limited. For instance, although phraseological complexity constitutes an important component of linguistic complexity (e.g. Paquot 2019), the developmental patterns in phraseological complexity features among beginning-level L2 learners have not been extensively explored. In addition, there have not been thorough examinations of the developmental patterns of linguistic complexity in the context of argumentative essays produced by L2 beginning-level learners in a university-level general English course, an area of importance within the academic setting (e.g. Gardner and Nesi 2013) that is investigated in this study. Second, it remains uncertain whether scores in timed L2 argumentative writing, often deemed more challenging than narrative or descriptive writing, would exhibit improvement over time when produced by beginning-level L2 learners with limited L2 exposure and limited writing instruction. Third, it appears that further research into the longitudinal connections among linguistic complexity features, including the relationships of lexical and phraseological complexity with syntactic complexity, is essential to gain a comprehensive understanding of the interconnected development of language subsystems. Finally, the relationship between linguistic complexity features, which can reflect a learner’s language systems, and L2 holistic writing scores, which can reflect a learner’s L2 writing ability, over time, has not been systematically examined. Such examination may be of particular interest among writing assessment researchers and writing teachers.

To fill these gaps, the current study extends the existing research on CDST-oriented L2 writing by examining (1) the longitudinal changes in L2 lexical, phraseological, and syntactic complexity, as well as L2 holistic writing scores, (2) the relationships among lexical, phraseological, and syntactic complexity over time, and (3) the relationships of lexical, phraseological, and syntactic complexity with L2 holistic writing scores in beginning-level Japanese university EFL learners in the context of argumentative writing. The following three questions are addressed:

1. How do lexical, phraseological, and syntactic complexity features and holistic writing scores in EFL beginning-level learners’ argumentative writing change over two academic semesters?

2. How are lexical and phraseological complexity features related to syntactic complexity features in EFL beginning-level learners’ argumentative writing that is produced over time?

3. How are lexical, phraseological, and syntactic complexity features related to holistic writing scores in EFL beginning-level learners’ argumentative writing that is produced over time?

In addressing the research questions, it is important to note that our focus is on examining “the ‘grand sweep of development’ where global structure and similarities across participants can be seen” (Larsen-Freeman 2006: 613). We operate on the assumption that, at the group level, development over time may exhibit relatively regular trends leading towards target-like language use (Bulté and Housen 2018). While this perspective is considered essential in CDST-oriented research (e.g. Bulté & Housen 2020; Han et al. 2022), we are aware of the limitations of applying group-level statistics to individual learners (Lowie and Verspoor 2019). In addition, to address the research questions, we employed generalized additive mixed models (GAMMs), capable of modeling not only linear, but also nonlinear and more dynamic developmental trajectories.

3.1 Research context and participants

Data were collected from two intact English-medium classes in a business-oriented university in Japan over two semesters (nine months). The course was mandatory for students majoring in International Studies in the university. The course covered all four language skills, with a particular emphasis on grammar, listening, and reading. Its goal was to develop students’ English proficiency to reach intermediate levels. Each week, the class gathered for one meeting lasting 200 min (two 100 min sessions in a day). Due to the Covid-19 pandemic, all but one class met fully online; the university partly allowed one class to have face-to-face class meetings during the last seven weeks of the second semester.

Participants were 42 Japanese beginning-level EFL students (M = 19.10, SD = 0.55) who were majoring in International Studies at a university in Japan and consented to participate in this study. All of the participants had taken the institutional TOEIC Listening and Reading test as a requirement from the department. Their TOEIC scores ranged from 275 to 560 out of 990, with a mean of 410.56 (SD = 71.38). Given that TOEIC scores ranging from 226 to 550 are mapped onto the A2 level (beginning level) of the Common European Framework of Reference (Tannenbaum and Wylie 2008), participants were considered beginning-level English learners.^⁠[2] As for English learning backgrounds, participants learned English for 9.00 years (SD = 3.03), and most of them aimed to improve their English skills and TOEIC scores. Acknowledging that labeling L2 learners who had studied English for an average of nine years as beginners might be less appropriate, this classification was deemed unavoidable based on their TOEIC scores and their corresponding proficiency level mappings (Tannenbaum and Wylie 2008).

In relation to writing practice, most students reported that prior to taking the course, they had not learned how to write an argumentative essay. During the two semesters, writing was not a primary focus. However, the instructor placed an emphasis on the genre of argumentative writing, using prompts similar to those for the writing test of the TOEIC. This choice was motivated by the recognition of the significance of argumentative writing within the academic context of higher education, along with the potential for students to take the TOEIC writing test in the future. To facilitate this, students received concise instructions on the organization of English argumentative essays at the beginning of the first semester. It is important to note that these instructions were provided solely at the commencement of the data collection. Subsequently, students were asked to write an argumentative essay throughout approximately the two semesters.

3.2 Instruments for data collection: writing prompts

Students produced six argumentative essays in response to prompts similar to those of the TOEIC writing opinion test over a span of nine months in class. TOEIC prompts were used because the participants’ course focused on business-related English. Writing directions were taken from the TOEIC writing opinion task. Each prompt asked students to write an essay in response to a controversial statement by stating, explaining, and supporting their opinion. Students were given 30 min to write an essay. While no specified word limit was given, following the TOEIC writing test instructions, our prompt stated that a well-crafted essay should typically contain a minimum of 300 words. A total of nine prompts with three different topics (i.e., business, education, and technology) were used (see Table 1). The topics were counterbalanced by using three different prompts (one prompt from each topic) at each writing occasion to minimize potential prompt effects.

3.3 Data collection procedure

As shown in Table 2, using the first week of the first semester as a starting time point, a total of six essays were collected on the second, fifth, tenth, fifteenth, twenty-fourth, and thirty-eighth weeks. Across the six writing occasions, participants responded to six different prompts. Between the first and second writing occasions, the instructor gave a 40 min in-class lecture on how to structure argumentative essays in English because many of the students had not learned how to write essays before. A total of 259 argumentative essays were collected. We removed 11 essays that did not meet a minimum requirement of 50 words needed to calculate lexical diversity measures (Zenker and Kyle 2021). Then, the number of essays that each participant produced was counted. Two students produced only three essays, which we considered insufficient to track changes over time, and the essays were thus removed. Also, one essay was off topic and was consequently removed. The remaining 241 essays produced by 42 students were included in the analysis. On average, students produced 5.74 essays (SD = 0.59), ranging from four to six. The number of essays produced on each occasion is provided in Table 2. All the essays were written on computers.

3.4 Data analysis

4.1 Changes in linguistic complexity features and holistic writing scores

Research question 1 addressed how holistic L2 writing scores and lexical, phraseological, and syntactic complexity features in EFL beginning-level learners’ argumentative writing changed over time. The results of the six GAMMs, in each of which a fixed-effect variable was writing score or one of linguistic complexity features and a response variable was time in weeks, are presented in Table 5. To illustrate, Table 5 first presents a fitted intercept for the response variable. Next, estimates for three terms are presented: (1) the fixed effects, s(fixed effect); (2) the random intercepts, s(participant); and (3) the random smooths, s(fixed effect, participant). For each of these terms, each model provided an EDF value along with the maximum degrees of freedom (in the Ref.df column), which were used to calculate the F- and p-value. The two most important pieces of information to answer research question 1 (as well as research questions 2 and 3) were the EDF and its p-value for the fixed effects (which were shown in bold in Table 5). This was because EDF values indicated whether the fixed effects were linear (EDF = 1.00) or curvy (EDF > 1.00), and p-values showed whether the estimated effects (line or curve) were statistically significant. On the other hand, the estimates of random intercepts and smooths indicate the variance of random effects. The significant p-values for random effects indicate that there is evidence in support of the random effects. However, we do not further discuss random effects because we included them to consider inter-individual variation in models, but whether random effects were evident or not was beyond the scope of this study.

Based on the GAMM results and the corresponding visualizations for research question 1 (see Figure 1), the following observations were made. Writing scores significantly increased over time in a linear manner (EDF = 1.00, Ref.df = 1.00, F = 25.84, p < 0.001). In terms of lexical complexity, no longitudinal change was found in MTLD (EDF = 1.30, Ref.df = 1.53, F = 1.03, p > 0.05), while CW age-of-acquisition scores increased significantly in a quite linear manner (EDF = 1.57, Ref.df = 1.91, F = 5.22, p < 0.01). For phraseological complexity, bigram MI scores showed no significant change (EDF = 1.00, Ref.df = 1.00, F = 2.20, p > 0.05). With respect to syntactic complexity, CN/T significantly increased over time in a linear manner (EDF = 1.00, Ref.df = 1.00, F = 16.75, p < 0.001); MLC showed no significant change (EDF = 1.95, Ref.df = 2.36, F = 2.30, p > 0.05); and DC/T showed significant non-linear changes (EDF = 3.48, Ref.df = 4.01, F = 3.93, p < 0.01), such that DC/T tended to increase until week 10 (Writing 3) and stagnate afterwards.

Figure 1:

GAMM plots for showing changes in writing scores and linguistic complexity features. Notes. Red lines indicate significant fitted lines or curves of time effects, while red shades represent their corresponding confidence intervals. Dashed blue lines connect mean scores for each time point. Thin colored lines connect each participant’s scores across time.

To illustrate the linear increases in writing scores, Figure 2 presents writing samples from one student (Student A) alongside their corresponding linguistic complexity and holistic L2 writing scores across three occasions: the first essay, second essay, and seventh essay. In the initial writing session, limited content development and language use were observed, resulting in a holistic score of one. Following a brief instruction on English essays between the first and second writing occasions, the second essay showed enhanced organization by incorporating three reasons to support the ideas, although content development and language use remained limited. Consequently, the second essay received a holistic score of two. The seventh essay exhibited progress in content, language use, and organization, resulting in a holistic writing score of four. Hence, it appears that the student improved their writing ability, evident in the increasing holistic writing scores over time. Notably, this improvement occurred despite the L2 courses taken, which did not specifically emphasize L2 argumentative writing.

Figure 2:

Student A’s writing samples with linguistic complexity and writing scores over time.

4.2 Relationships between lexical and phraseological complexity features and syntactic complexity features

Research question 2 addressed how lexical and phraseological complexity features were related to syntactic complexity features in EFL beginning-level learners’ argumentative writing. Three GAMMs (in which one of the three syntactic features [CN/T, MLC, and DC/T] was the response variable) were created. Fixed-effect variables were MTLD, CW age of acquisition, and bigram MI. The results of the GAMM in which a response variable was CN/T are presented in Table 6 and Figure 3. Results indicated that CN/T was significantly related to MLTD in a quite linear manner (EDF = 1.18, Ref.df = 1.33, F = 7.15, p < 0.01), such that higher MTLD scores tended to relate to lower CN/T. Results also indicated that CN/T was significantly related to CW age-of-acquisition scores in a non-linear manner (EDF = 2.39, Ref.df = 3.03, F = 4.99, p < 0.01), such that until CW age-of-acquisition scores reached around 5.7, they were not associated with CN/T, but when CW age-of-acquisition scores reached around 5.7 or higher, higher CW age-of-acquisition scores tended to relate to greater CN/T. On the other hand, CN/T was not related to bigram MI scores (EDF = 1.58, Ref.df = 1.98, F = 0.58, p > 0.05).

Figure 3:

GAMM plots to predict CN/T using lexical and phraseological complexity features. Notes. Red lines indicate significant fitted lines or curves of the fixed effects, while red shades represent their corresponding confidence intervals. Thin colored lines connect each participant’s scores across time.

The results of the GAMM whose response variable was MLC are presented in Table 7 and Figure 4. Results indicated that MLC was not related to MLTD (EDF = 1.00, Ref.df = 1.00, F = 2.25, p > 0.05). On the other hand, MLC was significantly related to CW age-of-acquisition scores in a linear manner (EDF = 1.00, Ref.df = 1.00, F = 10.41, p < 0.01), such that higher CW age-of-acquisition scores tended to relate to longer MLC. MLC was also significantly related to bigram MI scores in a linear manner (EDF = 1.00, Ref.df = 1.00, F = 14.78, p < 0.01), but inversely, such that higher MI scores tended to relate to shorter MLC.

Figure 4:

GAMM plots to predict MLC using lexical and phraseological complexity features. Notes. Red lines indicate significant fitted lines or curves of the fixed effects, while red shades represent their corresponding confidence intervals. Thin colored lines connect each participant’s scores across time.

The results of the GAMM in which a response variable was DC/T are presented in Table 8 and Figure 5. Results indicated that DC/T was not related to either MTLD (EDF = 1.00, Ref.df = 1.00, F = 1.12, p > 0.05) or CW age of acquisition (EDF = 1.00, Ref.df = 1.00, F = 0.00, p > 0.05). Results, however, indicated that DC/T was significantly related to bigram MI scores in a non-linear manner (EDF = 2.30, Ref.df = 2.89, F = 4.91, p < 0.01), such that greater DC/T tended to relate to higher bigram MI scores until MI scores reached around two, but afterwards, as MI scores increased, DC/T tended to decrease slightly.

Figure 5:

GAMM plots to predict DC/T using lexical and phraseological complexity features. Notes. Red lines indicate significant fitted lines or curves of the fixed effects, while red shades represent their corresponding confidence intervals. Thin colored lines connect each participant’s scores across time.

4.3 Relationships between linguistic complexity features and writing scores

Research question 3 addressed how lexical, phraseological, and syntactic complexity features were related to writing scores in EFL beginning-level learners’ argumentative writing. The results of the six GAMMs, in each of which a response variable was writing score and a fixed-effect variable was one of the six lexical, phraseological, and syntactic complexity features, are presented in Table 9 and Figure 6. With respect to lexical complexity, writing scores were not related to MTLD (EDF = 1.00, Ref.df = 1.00, F = 0.85, p > 0.05). On the other hand, writing scores were significantly related to CW age-of-acquisition scores in a slightly non-linear manner (EDF = 1.95, Ref.df = 2.47, F = 3.61, p < 0.05), such that overall, writing scores tended to increase as CW age-of-acquisition scores increased, but the degree of increases in writing scores became slightly lessened after CW age-of-acquisition scores reached around 5.70. For phraseological complexity, bigram MI scores were significantly associated with writing scores in a non-linear manner (EDF = 3.30, Ref.df = 4.13, F = 5.03, p < 0.001), such that writing scores tended to increase until bigram MI scores reached around 1.80 but decrease afterwards. In terms of syntactic complexity, writing scores were significantly related to CN/T (EDF = 1.00, Ref.df = 1.00, F = 20.45, p < 0.001) and MLC (EDF = 1.00, Ref.df = 1.00, F = 7.84, p < 0.01), respectively, in a linear manner, such that as CN/T and MLC increased, writing scores also tended to increase. Writing scores were also significantly related to DC/T (EDF = 2.54, Ref.df = 3.29, F = 2.47, p < 0.05), but in a non-linear manner, such that writing scores tended to increase until DC/T reached around 0.40, stagnate when DC/T was between around 0.40 and 0.70, and increase again after DC/T reached around 0.70.

Figure 6:

GAMM plots to predict writing scores using linguistic complexity features. Notes. Red lines indicate significant fitted lines or curves of the fixed effects, while red shades represent their corresponding confidence intervals. Thin colored lines connect each participant’s scores across time.

5.1 Changes in linguistic complexity and writing scores over time

The first research question aimed to understand the changes in lexical, phraseological, and syntactic complexity features and writing scores in the argumentative writing of EFL beginning-level learners over two academic semesters (nine months). Despite the considerable intra- and inter-variability and the fluctuations in the changing patterns of linguistic complexity measures, our primary aim was to identify shared developmental patterns among L2 beginning-level learners. Overall, significant changes were observed in holistic writing scores, CW age-of-acquisition scores, CN/T, and DC/T, while no changing patterns were noted in MTLD, bigram MI scores, or MLC.

The observation that L2 holistic scores for argumentative writing tended to increase over time, signifying the progression of their L2 writing ability (e.g. Cumming et al. 2000; Valdes et al. 1992), is noteworthy, given that this study was conducted in a general English class with limited instruction about argumentative writing over time. It is possible that, through repetitive engagement in the same genre of writing, L2 learners may have become more familiar with the task and gained more knowledge in English words and structures in class, thereby challenging their output to produce more well-written essays with more logical arguments and more complex linguistic features. This could explain the observed increases in holistic scores for argumentative writing, broadening the scope beyond prior studies that focused on improvements in a different genre, such as L2 narrative writing, achieved through repetitive practice (e.g. Lowie and Verspoor 2019).

Regarding lexical complexity, L2 beginning-level students displayed improvements in their lexical sophistication, as indicated by increases in CW age-of-acquisition scores. This observation aligns with prior CDST-oriented research (e.g. Verspoor et al. 2008; Zheng 2016), suggesting that L2 beginning-level learners tended to use more sophisticated words as they produced L2 argumentative writing over time. It is possible that L2 learners aimed to convey more intricate ideas in their arguments by employing more sophisticated words, although further research is needed to substantiate this hypothesis. On the contrary, L2 beginning-level students did not exhibit improvement in their lexical diversity, as evidenced by the absence of changes in MTLD over time. This finding contrasts with Zheng (2016), who reported a general increase in lexical diversity over one year in L2 intermediate-level learners’ argumentative writing. One plausible explanation for this divergence is that L2 beginning-level learners might not prioritize diversifying their vocabulary in their texts due to their focus on other aspects of writing and their limited L2 lexicon. As their proficiency advances to an intermediate level, L2 learners may become more capable of incorporating a wider range of words over time, as observed in Zheng (2016) and Pfenninger (2020). Therefore, potentially, the notion that L2 beginning-level learners are “especially busy learning words” (Verspoor et al. 2012: 257) could be more specifically framed as them being occupied with complexifying their lexicon rather than diversifying it.

In terms of phraseological complexity, as measured by bigram MI scores, there was no improvement observed. This suggests that L2 beginning-level learners did not increasingly employ more strongly associated bigrams. This finding complements earlier research indicating that the development of phraseological complexity tends to occur in at least intermediate-level learners (Duan and Shi 2021; Vandeweerd et al. 2022). Therefore, given that the “growth of phraseological sophistication is very gradual” (Vandeweerd et al. 2022: 18), it is possible that L2 beginning-level learners may not yet be developmentally prepared to notice and use more sophisticated phraseological units.

In terms of syntactic complexity, three measures employed to track the development of syntactic complexity, namely CN/T, MLC, and DC/T, exhibited distinct developmental trajectories over time. Firstly, CN/T, which assesses NP complexity, a crucial aspect of academic writing (Biber et al. 2020), demonstrated a linear increase over time. This observation of a progressive trend in NP complexity aligns with Bulté and Housen (2018), but not with Wang (2022) where changes were not observed. It is worth noting that while Bulté and Housen (2018), Wang (2022), and the present study focused on L2 beginning-level learners’ writing, they differed in terms of writing genres. Bulté and Housen (2018) and the present study delved into school-based writing genres, specifically creative writing and argumentative writing, respectively. These genres fall within the domain of formal academic writing (Gardner and Nesi 2013). In contrast, Wang (2022) investigated informal diaries that might not necessarily encourage the use of academic language, including complex NPs. Thus, it is conceivable that L2 NP complexity may increase even in the writing of L2 beginning-level learners, but this trend seems to be pronounced in the context of academic writing.

Next, MLC, serving as an indicator of phrasal complexity, which is another crucial aspect of academic writing (Biber et al. 2020), exhibited stability over time. This observation of a lack of change in MLC aligns with prior research that investigated the writing of L2 beginning-level learners (Bulté and Housen 2018; Wang 2022). It is plausible that, due to their limited L2 proficiency, L2 beginning-level learners’ use of phrasal elaboration remains within “attractors, regions in state space” (Larsen-Freeman 2019: 64).

Lastly, DC/T, used as a measure of subordination, displayed noteworthy non-linear changes in this study. Specifically, it exhibited a trend of increasing subordination up to the first two months, followed by a plateau. This finding partially aligns with earlier studies that reported an increase in subordination over time (Bulté and Housen 2018; Köylü et al. 2023; Wang 2022), but this study contributes to the existing body of research by highlighting the non-linear nature of this developmental trajectory. Possibly, L2 beginning-level learners may have initially made efforts to incorporate more subordination into their early writing attempts. Still, over time, they may have become trapped within an attractor state (Larsen-Freeman 2019), potentially lacking the need to displace the system from this attractor.

5.2 Relationships between lexical and phraseological complexity and syntactic complexity

The second research question investigated the relationship between lexical and phraseological complexity features and syntactic complexity features in argumentative writing over time by EFL beginning-level learners. To achieve this, the three syntactic complexity measures, CN/T, MLC, and DC/T, were predicted, respectively, with the predictors being MTLD, CW age-of-acquisition scores, and bigram MI scores.

The findings unveiled that CN/T, as a measured of NP complexity, exhibited a negative linear relationship with MTLD, which aligns with the findings of a prior case study (O’Leary and Steinkrauss 2022). The competition observed between MTLD and CN/T suggests a trade-off effect between lexical diversity and NP complexity. In their texts, if L2 beginning-level learners incorporated a wider range of words (resulting in an increase in MTLD), they might have used fewer CNs (leading to a decrease in CN/T). This could be attributed to the possibility that L2 beginning-level learners may encounter challenges in employing both diverse words and CNs simultaneously in their writing, potentially due to limitations in the available cognitive resources at their disposal. Conversely, CN/T had a positive non-linear relationship with CW age-of-acquisition scores. Specifically, the connection between CW age-of-acquisition and CN/T exhibited a plateau until a certain point, followed by a positive relationship. This finding indicates that, as L2 beginning-level learners employed more sophisticated CWs in their texts, they also tended to use more CNs once they surpassed a particular threshold. This finding underscores the partially supportive relationship between lexical sophistication and NP complexity, corroborating the results of a previous case study (Spoelman and Verspoor 2010). On the other hand, CN/T did not exhibit a significant relationship with bigram MI scores. This absence of a relationship may indicate that phraseological sophistication and NP complexity represent relatively independent aspects within the linguistic phenomena of L2 beginning-level learners’ argumentative writing.

The findings demonstrated that MLC, as a measure of phrasal complexity, exhibited no relationship with MTLD. The absence of a relationship between MTLD and MLC implies that the utilization of diverse words and phrasal elaboration are distinct and separate aspects of linguistic complexity within the context of argumentative writing by L2 beginning-level learners. Next, MLC had a positive linear relationship with CW age-of-acquisition scores. This observed positive relationship suggests that, among L2 beginning-level learners, incorporating more sophisticated CWs into their writing was associated with a tendency to produce longer clauses via phrasal elaboration. Furthermore, since CW age-of-acquisition scores also showed a positive relationship with CN/T, it implies that using more sophisticated CWs may facilitate the creation of more complex NPs as well as longer clauses. This underscores a pattern of “connected growth” between lexical sophistication and phrasal complexity (Verspoor and van Dijk 2011). Conversely, MLC showed a negative linear relationship with bigram MI scores. This competitive relationship suggests that when L2 beginning-level learners strived to use more strongly associated bigrams, it may have led to the creation of shorter clauses with limited additions of phrase-level elements. There could potentially be a trade-off effect between phraseological sophistication, assessed through bigram MI scores, and phrasal complexity, evaluated via MLC. However, considering that, to our knowledge, there is no previous research on the relationship between phraseological sophistication and phrasal complexity, further research is needed to substantiate this finding.

The findings revealed that DC/T, a measure of subordination, showed no relationship with MTLD or CW age-of-acquisition scores. This absence of a relationship suggests that, in the context of L2 beginning-level learners’ argumentative writing, subordination, which reflects clause-level complexity, is not associated with lexical complexity. On the other hand, DC/T demonstrated a non-linear relationship with bigram MI scores. Among L2 beginning-level learners, using more strongly associated bigrams was associated with greater subordination, but beyond a certain point, this relationship reached a plateau. It is possible that employing more strongly associated bigrams may facilitate increased subordination up to a certain threshold. However, the precise mechanisms underlying this finding are beyond the scope of our study, and further research is warranted. In any case, this study contributes to our understanding by highlighting a supportive relationship between phraseological sophistication and subordination in the argumentative writing of L2 beginning-level learners.

5.3 Relationships between linguistic complexity and writing scores

The third research question aimed to explore the relationship between lexical, phraseological, and syntactic complexity features and holistic writing scores in argumentative writing by EFL beginning-level learners over time. Overall, the results indicated that L2 holistic writing scores exhibited significant associations with CW age-of-acquisition scores, bigram MI scores, CN/T, MLC, and DC/T, but not with MTLD.

In terms of lexical complexity, our findings indicate that over time, higher-rated L2 texts tended to incorporate more sophisticated words, but not necessarily a more diverse range of words. This observation concerning lexical sophistication aligns with prior studies that have reported a positive relationship between lexical sophistication and L2 writing scores (e.g. Kim et al. 2018). However, the finding related to lexical diversity presents a perplexing contrast, especially considering that the scoring rubric emphasizes the use of a wide range of vocabulary. Our findings did not support the notion, as suggested in previous research (e.g. Crossley and McNamara 2012), that greater lexical diversity is associated with higher-rated L2 texts. One possible explanation for this divergence could be that, in line with our finding showing no improvement in lexical diversity among L2 beginning-level learners, the limited diversification of their lexical choices may not have had a noticeable impact on the raters’ assessments of their texts.

In the realm of phraseological complexity, the significant association between bigram MI scores and holistic writing scores displayed a non-linear pattern, characterized by an initial increase until reaching a certain point followed by a decrease, forming an omega-shaped relationship. This finding is intriguing, particularly in light of previous research that has established a positive relationship between the utilization of phraseological units with higher MI scores and higher L2 writing scores (e.g. Bestgen and Granger 2014; Paquot 2019). The unexpected negative relationship between MI scores and L2 writing scores, identified in our study after a certain threshold, prompts questions. As L2 beginning-level learners made efforts to use more strongly associated bigrams in their writing, it is conceivable that beyond a certain point, this effort might have introduced a trade-off with other aspects of writing, such as syntactic complexity and idea development. Another possible explanation, more specifically, is that given the negative relationship between bigram MI scores and MLC, it is plausible that as L2 beginning-level learners introduced more strongly associated bigrams into their texts with a certain threshold MI score surpassed, they might have produced shorter clauses. This, in turn, could have negatively impacted raters’ evaluation of their texts. However, verifying these hypotheses would require additional research.

Regarding syntactic complexity, our study found that holistic writing scores exhibited a relatively linear increase as CN/T, MLC, and DC/T increased. This finding is consistent with prior research that has demonstrated that L2 writing scores have a positive relationship with NP complexity, phrasal complexity, and subordination (e.g. Lu 2010).