Student perceived effectiveness of task-based instructional design of data-driven synonym learning featuring “mini-lecture”

2.1 Traditional approaches to synonym learning

In the EFL context, learners primarily gain synonym knowledge from teacher instruction and/or dictionary consultation. The former, mainly based on teachers’ experience and intuition, is form-focused instruction and may fall short in providing learners with accurate and abundant information on word recognition and collocation (Tsai, 2019), while the latter seldom represents lexical regularity in extended texts (Quinn, 2014), and may even lead to incorrect substitution between synonyms in production (Murphy, 2003; Thienthong, 2020; Yeh et al., 2007). Both methods lack sufficient nuance information and may present confusing, overlapping interpretations of synonyms (Song, 2021), resulting in learners’ shortage of collocational knowledge, which makes synonyms differentiation difficult.

2.2 Data-driven synonym learning

Synonym learning requires far more than just providing the implied equivalence; the knowledge of collocational patterns, semantic features, and stylistic guidance is essential (Harvey & Yuill, 1997; Phoocharoensil, 2021). With the development of cognitive linguistics and corpus linguistics, synonym research has gained attention and progress (Liu & Zhong, 2016). Corpus linguistics proposes lexical semantic theory (Firth, 1957; Halliday, 1966; Sinclair, 1966, 1991), arguing that the meaning of words is determined by their collocates and other contextual features that accompany them, and thus collocation is a powerful indicator for synonym discrimination (Kilgarriff et al., 2014; Liu & Zhong, 2016). In this vein, many studies have investigated specific sets of synonyms with online tools such as COCA (Corpus of Contemporary American English), Sketch Engine, BNC Web, and so forth., by examining distribution across genre, collocations, and semantic preference (e.g., Divjak & Gries, 2006; Gu, 2017; Liu, 2010, 2013; Phoocharoensil, 2020, 2021, 2022). These studies show that synonyms differ in terms of collocational patterns despite their similar conceptual meaning (Murphy, 2009), and the collocational interchangeability between synonyms is rare (Xiao & McEnery, 2006). Research in this area yields abundant information about corpus-based synonym differentiation. However, studies have conventionally focused on certain sets of synonyms, and this seems to be too narrow in terms of synonym learning. What matters most is to develop learners’ independent synonym query skills and consultation literacy, yet how they can achieve this is still not clear.

Some corpus-based studies focused on learners’ synonym acquisition. Ahmadian and Darabi (2012) investigated the relationship between EFL learners’ knowledge of near synonyms and their performance on a corpus-driven test of collocational behavior. They found students were unaware of the subtle distinctions among synonyms, and their knowledge of synonymy was related to their performance on the test of collocational behavior. Liu and Zhong (2016) used a forced-choice question instrument and conducted a comparative analysis of synonym usage of native speakers and ESL/EFL learners, and they found that salience of target pattern and learners’ construal of the communication context/goal were two major factors influencing synonym acquisition. Wongkhan and Thienthong (2021) employed a forced-choice collocation test and examined the relationships between learners’ academic experience at university and their knowledge of academic collocation and synonymy. The results revealed that students with more experience significantly outperformed those with less experience in most collocation questions. These studies provide evidence for learners’ difficulty in synonym acquisition and the importance of corpus-based synonym information (i.e., collocational patterns, distribution, and language formality), but they focused on learners’ synonym knowledge and used corpus data only as a reference. How corpus can be applied in L2 synonym teaching remains unknown.

Yeh et al. (2007) investigated the effectiveness of the use of corpora in enhancing university students’ synonym learning. They designed five online units for increasing students’ awareness of underused specific adjectives for EFL college writing. The results showed that students’ synonym knowledge not only improved in their immediate post-test but was retained in the two-month delayed post-test. Students also reported positive attitudes toward inductive learning despite the difficulty in verbalizing the semantic differences. Jafarpour et al. (2013) compared the effect of the corpus-based approach and traditional approach in teaching 90 university students the collocation of near synonyms. The results revealed that the group with the corpus-based approach performed significantly better than the traditional group in the comprehension and production of collocations of synonyms. Similarly, Soruç and Tekin (2017) conducted a study with 72 secondary school students to examine the effect of the DDL approach on vocabulary learning. They found that the DDL group achieved significantly higher scores on both the immediate and the delayed post-test, and they displayed favorable attitudes toward the DDL practice. Boontam (2022) examined the effectiveness of 30 Thai second-year EFL university students’ paper-based data-driven learning of three synonymous English adjectives, “naughty,” “disobedient,” and “rebellious.” The instructional materials were designed according to the meanings, collocations, formality, and grammatical patterns from BNC and Sketch Engine. The results showed that learners’ vocabulary knowledge developed after this four-week learning through DDL activities. The participants held positive attitudes toward this DDL practice. Notably, DDL research has been conducted with languages than other English. For example, Yao (2019) implemented a three-week quasi-experimental design to examine the difference between the DDL approach and the dictionary-based approach to vocabulary learning with 32 Chinese learners of Spanish. The study provided further evidence for students’ positive attitudes toward DDL practice. The DDL group significantly outperformed the control group. This is “meaningful for the popularization and acceptance of DDL” in foreign language learning other than ELT (Yao, 2019, p. 38).

While these studies offer examples of how to design concordance-based materials with electronic referencing tools, the treatment span was relatively short (i.e., about four weeks in Yeh et al.’s and Boontam’s studies, and only four classroom hours for each instructional group in Soruç & Tekin’s study). Therefore, the conclusions have to be considered carefully, and the long-term effect should be tested further. Besides, these studies are fundamentally empirical, and the teaching materials design was purely instructional but without “theoretical positioning” (Pérez-Paredes, 2019), which underscores one of the problems in the 30 years of DDL research (Boulton & Vyatkina, 2021).

2.3 Learners’ perceptions of DDL

Existing studies are rich in investigating learners’ perceptions of DDL implementation (e.g., Bernardini, 2004; Kennedy & Miceli, 2010; Lee et al., 2020; O’ Sullivan & Chambers, 2006; Yoon, 2008; Yoon & Hirvela, 2004), but no conclusive results have been identified. Some studies found learners’ positive attitudes toward DDL (e.g., Mizumoto & Chujo, 2015) while others found negative attitudes (e.g., Quan, 2016) or mixed attitudes within the same cohort of learners (e.g., Kennedy & Miceli, 2001). Given the differences in conceptualization, operationalization, and research paradigm, most studies were small-scale and short-term with different research questions and teaching contexts, thus failing to ensure generalizable conclusions (Geluso & Yamaguchi, 2014; Tsai, 2019). In addition, DDL has little impact concerning the real problem of “how to implement big data in language teaching” (Boulton, 2009) for it has not been normalized in language teaching (Conrad, 2005; Flowerdew, 2012; Römer, 2006). Therefore, more empirical evidence, especially studies generated from cutting-edge teaching practices, is needed to support DDL in the teaching context (Chambers, 2019). Longitudinal research or long-term DDL treatment is also needed (Boulton, 2009; Boulton & Vyatkina, 2021).

Against this backdrop, the present study examines the effectiveness of a tentative data-driven synonym learning model to investigate learners’ perceptions (changes) in the two rounds of DDL practice. It addresses the following research questions (RQs):

3.1 Theoretical framework

Schmidt (2001) argued that the noticing hypothesis is intentional guidance for noticing some features of words in context, which promotes language acquisition, thus supporting the use of the inductive method in language learning (Kavaliauskiene, 2003; Shaffer, 1989). As learners read concordances in KWIC (keyword in context) or other input-enhanced forms presented by corpus tools, the salient input helps the noticing necessary for meaningful input processing to occur (VanPatten & Benati, 2010, as cited in Crosthwaite & Boulton, 2022).

By “noticing” the corpus data, learners can be stimulated to use inductive learning strategies, especially those of “perceiving similarities and differences and of hypothesis formation and testing” (Johns, 1994, p. 297). These high-order cognitive skills can be practiced with process-oriented constructivist pedagogy. Within this framework, DDL highlights exploratory language learning that requires learners to employ cognitive mechanisms and background knowledge to process new language data and discover salient new language input.

Nonetheless, learners with varying learning styles react differently to the constructivist approach; many learners are reluctant to use it, and some may even get lost in language data because of ineffective independent exploration (O’Keeffe, 2021), which calls for scaffolding teaching (Cobb & Boulton, 2015). As stated by Yoon and Hirvela (2004), “it may be pedagogically unwise to simply let students browse through a tremendous amount of sample texts or corpora without proper guidelines” (p. 278). Similarly, Smart (2014) argued that DDL “is not necessarily characterized by direct interaction with language corpora or by total learner autonomy” (p. 186). Instead, the implementation of DDL relies on “carefully designed and scaffolded activities” to help learners explore language data, and these activities “place the learner at the center of the learning task” (p. 187). Johansson (2009) proposed “a guided inductive approach or a combination of an inductive and deductive approach where the elements of explanation and corpus use are tailored to the needs of the student” (p. 42).

The present study adopts this guided inductive approach (Flowerdew, 2009; Mizumoto & Chujo, 2016; Smart, 2014) and conducts an instructional design, which takes the form of student-centered “mini-lecture” tasks, with teacher training and feedback before, during, and after the tasks. Through these scaffolding activities, learners would engage more as they can gain support and interaction in the process of discovering language.

3.2 Participants and settings

Participants in this study were 23 students (business English majors, 19 women and 4 men, aged 20–21) from a university in China. Three participants did not fully engage in the whole process; ultimately data from 20 students were collected for analysis. Using convenience sampling, this research was conducted in an “Advanced English (I/II)” course, with the first author serving as both researcher and instructor. This course was a required course for junior English majors. Altogether, learners underwent two semesters (each 18 weeks) with four sessions (180 min) per week. Within each unit text, many synonymous words posed great challenges. As such, a corpus-based student-centered “mini-lecture” task was designed to enhance learners’ vocabulary learning vision. Arguably, junior English majors may have the potential to work more independently with cognitively load corpus query tasks, which benefits them in both language improvement and cognitive development.

Each “mini-lecture” presentation of the synonyms was synchronous with one unit of learning. Participants were divided into seven groups (five groups of three and two groups of four). To fully promote learners’ self-regulated learning, DDL should avoid conformity when possible, and it was better to start in the most comfortable way for learners (Gilquin & Granger, 2022; Hunston, 2002). The “mini-lecture” task allowed students to independently choose 3 to 5 new words from the text and the pairing synonyms. Thus, the task catered to their learning needs, and the selection of the follow-up quiz items also aligned with their language proficiency. By increasing the task relevance, this freedom of choice will motivate the students in DDL practice.

The research was conducted with two rounds (each one semester) of DDL. The first round was the running-in stage where learners were familiarized with the basic query skills, and the second round was the enhanced stage in which we adjusted the teaching design based on the learners’ first perceptions of DDL. The emphasis was placed on optimizing the task procedure and providing the scaffolding worksheet to streamline the process.

3.3 “Mini-lecture” instructional design

We used COCA as the platform for data-driven synonym learning. COCA provides a web-based corpus-query interface. Specifically, the “Compare” function makes discriminating synonyms more accessible. Figure 1 displays the flowchart for the task-based instruction design.

Figure 1:

The flowchart for the data-driven “mini-lecture” instructional design.

The design involved three main stages for learners:

Stage 1: Before the task, the teacher introduced the purpose of integrating corpus into language learning and conducted a 90-min training about how to use COCA for synonym differentiation, especially on the function “COMPARE” and “COLLOCATES.”

Stage 2: During the task, groups selected 3–5 pairs of target synonyms; then, they conducted a corpus query and made an in-class report of their findings and conclusions. In this process the students fully engaged to learn how to use different query skills for information gathering and participated in more high-order cognitive activities by observing the data, forming hypotheses, and verifying the conclusions. Scaffolding and immediate feedback were provided to help students better acquire query skills and high-order cognitive skills.

Stage 3: After the task, the group distributed a self-designed synonym quiz based on their presentation and collected their peers’ answers. Feedback was provided in the next class. This step is very crucial: the quiz helped them verify the effectiveness of their “mini-lecture”; it also helped to consolidate all the students’ understanding of target synonyms.

3.4 Research procedures and instruments

This study was well-grounded in the classroom setting and the research procedures are displayed in Figure 2; data were collected through three questionnaires and two interviews (see Figure 2).

Figure 2:

The research procedure.

Questionnaire A (see Appendix A) was distributed before the first-round DDL practice to gain learners’ basic information, such as English learning experiences, their attitudes toward English learning, habits of using online resources, tools of vocabulary learning, the evaluation of their learning effect, and background knowledge of corpus use.

Questionnaire B (see Appendix B) was distributed at the end of the first round of the DDL practice. It was adapted from Yoon and Hirvela (2004) and the Cronbach Alpha coefficient was 0.799. Questionnaire C (see Appendix C) was distributed it at the end of the second-round DDL practice. The Cronbach Alpha coefficient was 0.743. Questionnaire C maintained the consistency in most of the items but was modified in line with the optimized measures in Round 2.^[1] Table 1 presents the constructs of the two questionnaires.

In Questionnaire C, to counterbalance the testing effect, many items were displayed in random order (see Table 1). Both questionnaires aimed to collect the students’ perceptions of the DDL practice in three aspects: (1) the student-centered “mini-lecture”; (2) the practice of DDL; (3) the difficulties in the process. However, since they served different purposes, differences existed as well: Questionnaire B attempted to tentatively investigate learners’ first perceptions of DDL in a more tentative way, while Questionnaire C focused more on learners sustained efforts for DDL practice and possible perception changes.^[2]

Both Questionnaires B and C investigated learners’ perceptions from three dimensions, namely, how they responded to the “mini-lecture” task (the Cronbach Alpha coefficient were 0.610 and 0.640), corpus-aided synonymous learning in general (0.813 and 0.802), and the difficulties in corpus consultation (0.648 and 0.551^[3]). Additionally, we maintain that this study focuses on designing the learning model of DDL classroom operationalizations and examining its effectiveness, so the learners’ perceptions of the “mini-lecture” task itself would be different from their overall perceptions of the DDL approach, namely, the use of the corpus, and the difficulty perception was helpful to reveal the potential factors influencing DDL in practice.

Semi-structured interviews were conducted at the end of each semester. To follow and investigate the perceptions (changes), we selected the same three focus groups (10 students) for the two interviews. The topics covered learners’ use of corpus, the experience of preparing for the “mini-lecture”, their comments on their presentation, and also the evaluation of their learning habits and difficulties.

4.1 Learners’ perceptions of DDL in the first round

A cohort of 20 learners participated in this study (17 females, 3 males). The results of Questionnaire A show that participants were positive toward English learning, and most frequently used electronic resources (88 %), including online websites (25 %), dictionary APPs (49 %), and electronic dictionaries (12 %). However, 94 % of them were not satisfied with their learning effects. The majority (91 %) used COCA mainly to query collocations (41 %), word meaning (16 %), and stylistic features (16 %), but none used it to discriminate synonyms.

To calculate and display the results of Questionnaire B and C more explicitly, we merged the responses of 1 and 2, 4 and 5 in the Likert scale respectively as “disagree” and “agree” in general. Results of Questionnaire B show that learners seemed to be more positive toward the “mini-lecture” task (M = 4.33) than data-driven synonym learning (M = 3.76); the difficulties encountered were also at a moderate level (M = 3.24).

Moreover, no significant correlation between difficulty perceptions and the other two perception dimensions was found, indicating that the corpus consultation was not beyond learners’ ability. Therefore, the positive responses paved the way for the second-round DDL implementation.

To further investigate the possible sources of learners’ difficulties, Questionnaire B included items (B-20, 23, 24) about task motivation. Table 2 shows that both active and passive learners were negatively correlated with the difficulty perceptions, but the correlations were more significant for passive learners at the individual task level; this indicates some relevance between learners’ task motivation and their perceptions of DDL.

Questionnaire B investigated learners’ needs and willingness to continue this learning model (see Figure 3). Overall, learners were enthusiastic about DDL while proposing suggestions for improvements, especially on the need to increase corpus use. DDL in the first round was a unit-based learning process, and provided only one opportunity for students to present their query outcomes. Limited corpus use hampered learners’ understanding of DDL and its effectiveness. In addition, in the open questions, some learners responded that “the presenting forms and query skills used in each group were different, and this made some conclusions confusing and unreliable.” The collected responses revealed learners’ need for continuing this model and optimizing the operationalizations in the next round.

Figure 3:

Suggestions for improving the teaching design.

In the second round, we standardized and condensed the procedures into a worksheet with clear steps for the students to prepare and present the “mini-lecture” task, which provided scaffolding for learners to better understand and grasp the skills. The worksheet highlighted the process of observing concordance, formulating hypotheses, and verifying hypotheses, to promote cognitive development. In addition to breaking up tasks into smaller segments, the optimized instruction encouraged the teacher and peers to provide immediate comments and feedback as learners presented “mini-lectures.” The activity of after-task vocabulary quiz design remained while corpus-aided peer-review writing practice^[4] was added to increase the corpus use frequency. In short, the optimized steps and enhanced scaffolding further promoted learners’ synonym learning by developing their cognitive ability and query skills in DDL.

4.2 Exploring learners’ perceptions (changes) in the longer term practice

4.2.1 Comparison of learners’ perceptions during two rounds of DDL

As aforementioned, Questionnaire C also investigated learners’ perceptions from the three dimensions. Table 3 shows the comparison results between the two rounds DDL practice.

Table 3:

Comparison between two-round perceptions of three dimensions.

Dimension	Round	Case number	Mean	Std.	Sig.
“Mini-lecture” perceptions	First	N = 20	4.33	0.617	0.899
	Second	N = 20	4.11	0.330
DDL approach perceptions	First	N = 20	4.13	0.448	0.756
	Second	N = 20	4.12	0.471
Difficulty perceptions	First	N = 20	4.08	0.518	0.428
	Second	N = 20	4.16	0.600

In the second round, it seems that learners still had positive responses to the “mini-lecture” task, and there was no significant difference between the two rounds; the standard deviation in the second round was smaller, indicating that learners’ attitudes seemed to be consistent after the reinforcement. In the first round, 80 % of the students acknowledged their improvement in vocabulary knowledge; in the second round, 75 % of students still held this perception (B-21, C-25). With the after-task vocabulary quiz, positive supporters increased from 70 % to 90 % (B-22, C-48); this supports the effectiveness of the optimized learning model.

As for the DDL approach, learners also had positive responses. Similar to “mini-lecture” task perception, no significant difference was found between the two rounds.

Nevertheless, it is worth noting that learners’ attitudes toward future corpus use present a mixed and even contradictory picture. They acknowledged the effectiveness of corpus (see Figure 4) while still prioritizing dictionaries in vocabulary learning; this trend increased by 20 % in the two rounds. Also, the answers (“would not take DDL as a main learning method”) to the open question (C-53) echoed this result although they tended to describe COCA as “accurate,” “professional,” or “good” (C-54).

Figure 4:

Learners’ willingness of future corpus use.

As for difficulty perceptions (see Figure 5), the results show that learners encountered network instability, high time and energy cost, and inadequate query skills, but there was no significant difference between the two rounds.

Figure 5:

Learners’ difficulty perceptions.

The perceptions of network instability did not change, but those of time and energy costs increased (from 90 % to 100 %); the relative complexity brought by the optimized and standardized procedures in the second round may contribute to the results. Meanwhile, the majority (90 %) perceived increased query skills development (C-6), and 80 % reported decreased anxiety (C-17). Accordingly, their demand for technical training also decreased, indicating that the optimized instruction provided efficient scaffolding for learners’ DDL practice. This inference finds further evidence from the answers to the question (C-52) concerning difficulties in Questionnaire C, namely, the decreased perceptions of “network operation” (19 %), “time- and energy-consuming task” (18 %), “selecting proper synonyms” (15 %), and “generalization of results” (13 %). Generally, after two rounds of DDL, learners’ skills of corpus consultation greatly improved. In the first round, 85 % admitted to having technical problems (C-5) while in the second round, 90 % realized that their query skills had improved (C-6).

Similar to the first round, no significant correlations were found between difficulty perceptions and the other two dimensions. The correlational analysis covers comparable dimensions in the two-rounds, while the new dimensions that emerged in the second round will be elaborated.

Also, we investigated the correlation between learners’ motivation and difficulty perceptions.

In the first round, active learners did not display significant correlations with their difficulty perceptions at all levels, while passive learners did in terms of network stability (see Table 2). Therefore, in the second round, efforts were focused on further verifying this result. Table 4 indicates that learners who considered the “mini-lecture” task complex tended to be more sensitive about the time and energy cost, but not about the need for training, whereas learners who “only used corpus to complete tasks” showed significant negative correlations with “need for technical training.” The results imply that the training before the DDL practice and the standardized procedures seem to have guaranteed learners’ completing the “mini-lecture” tasks, which evidenced the effectiveness of the learning model.

Table 4:

Correlations between difficulty perceptions and negative task motivation in the 2nd round.

Difficulty perceptions	Motivation
	Passive: tasks are complex (C-26)	Passive: use corpus only for task (C-38)
Network instability (C-41)	−0.309	0.142
Time-consuming (C-42)	0.492*	0.371
Need more training (C-43)	−0.031	−0.511*

1. *p < 0.05.

4.2.2 The effectiveness of optimized instruction in the second round

Figure 6 shows learners’ acknowledgment of the effectiveness of the standardized procedure. The majority believed that they became more skillful in using corpus (90 %, C-6), and the “mini-lecture” was better presented (95 %, C-19), and they highly appreciated the additional steps added in the second round.

Figure 6:

Perceptions of scaffolding provided.

Figure 7 displays how learners responded to knowledge construction and cognitive ability development. In the second round, 60 % of the learners valued peer collaboration, and 95 % valued teacher-student interaction and scaffolding. The majority were aware of cognitive development in themselves (90 %) and their peers (75 %), especially in language awareness (90 %). Meanwhile, they believed that the cognitive skills acquired in the second round could be transferred to other courses (85 %).

Figure 7:

Perceptions of knowledge construction and cognitive development.

4.3 Potential factors influencing learners’ perceptions

4.3.4 The working language in DDL

The cognitive load of data-driven synonym learning is fairly high, and may be greater if learners are required to use English for expressing technical terms in their “mini-lecture.” Therefore, the working language for DDL is likely to be a potential factor affecting DDL effectiveness.

During the two-round DDL practice, learners predominantly used English to present the “mini-lecture,” and the mother tongue (Chinese) was occasionally interspersed in peer feedback or teacher comments. However, Figure 8 shows that Chinese was far more favorable for explicit process explanations.

Figure 8:

Learners’ responses to the working language in DDL.

5.1 Learners’ perceptions of the DDL featuring task-based “mini-lecture”

This study conducted a task-based “mini-lecture” learning model with a guided approach and examined learners’ perceptions from the dimensions of the “mini-lecture” tasks, the DDL approach, and the difficulties. In the first round, learners generally held a more positive attitude toward the “mini-lecture” tasks and the DDL approach, but they also encountered some difficulties. These findings echo the existing literature (see Boulton & Vyatkina, 2021). However, no significant correlations between learners’ difficulty perceptions and the other two dimensions were found. Thus, we can infer that the learners acknowledged the “mini-lecture” teaching design and the DDL approach was within their reach in this context. However, learners suggested more practice with query skills to help improve their analysis of the query results.

5.2 Learners’ perceptions (changes) of the long duration DDL practice

Accordingly, the second round of DDL sought to optimize the procedures and provide learners with more adequate scaffolding. Learners maintained positive responses to the “mini-lecture” tasks and the DDL approach. Again, no significant correlations were found between difficulty perceptions and the other two dimensions. Importantly, learners’ perceptions of the three dimensions did not show significant differences between the two rounds.

There were some slight fluctuations in learners’ perceptions during the two rounds. In the first round, many learners intuitively took DDL only as a compulsory task (only one-time practice for each student); in the second round learners viewed DDL with a wider perspective and the longer duration allowed them to reflect on its compatibility with their own learning needs. This change was evidenced in their lower willingness of future use. More explanations can be found with the task itself. Synonym discrimination poses a great challenge to language learners (Liu & Zhong, 2016), and the threshold for the data-driven approach would be much higher. In the interviews after the second round, some learners mentioned that “the whole process was time/energy-consuming, and they were even not sure whether the conclusions were convincing.” They also commented “the synonym discrimination task was more suitable for learners of higher language proficiency because of their richer vocabulary knowledge to observe, hypothesize and testify language patterns from the concordances.” Indeed, learners had different needs for vocabulary learning; this was confirmed with interviews and the answers to the open questions in Questionnaire C. Some studies have also shown that learners no longer use corpus for data-driven learning outside classroom training or activities (e.g., Crosthwaite & Cheung, 2019; Meunier, 2019).

5.3 Potential factors influencing learners’ perceptions of the task-based DDL

This study also attempted to explore the potential factors (i.e., task complexity, task motivation, learners’ language proficiency, and the working language) that may influence learners’ perceptions. No significant correlation was found between task complexity and learners’ perceptions of the three dimensions, implying the practicality of the teaching design for synonym learning. But learners with passive motivations were more likely to perceive difficulties in DDL (e.g., time/energy cost), and were less willing to use it in the future (see Table 7) despite their acknowledgment of query improvements (see Figure 5). This supported the scaffolding effect of the instructional design. Besides, learners’ perceptions did not seem to be influenced by L2 proficiency. This differs from Quan’s (2016) findings, which showed advanced learners were the only suitable learners of DDL. However, we maintain that the main difference may not lie in learners’ L2 proficiency but in the setting. To be specific, in Quan’s study, the learners employed DDL in a self-directed manner, without support from peers or teachers, which constituted a big challenge for intermediate learners, who tended to view themselves as language learners, “thereby creating more interest in this new technique for improving their language skills” (Yoon & Hirvela, 2004, p. 277). Generally, if learners are equipped with the necessary and sufficient scaffolding, lower-intermediate learners could also integrate corpus into effective language learning and hold positive attitudes toward DDL. Additionally, most learners would welcome the use of their native language during the presentation of “mini-lecture,” which proves the need for translanguaging (Li, 2018) in EFL context, for translingual practices in language classrooms help students to support one another, build rapport, assert their cultural identity, and enable meaningful access to in-depth knowledge construction (Cenoz & Gorter, 2015; Rajendram, 2021; Stavrou, 2020).

5.4 Further discussions

First, learners need to be trained to practice DDL and this takes more time. This finding is further supported in the current study. First, before implementing DDL, learners need training in reading concordances and induction skills (Yeh et al., 2007), which helps provide learners with directions for knowledge construction and avoids large deviations in the learning process. But these skills cannot be acquired at once and learners need more time to get familiar with the corpus tools (Ädel, 2010; Alsolami & Assrar, 2020). In the current study, learners’ first response to DDL also showed the needs and the effectiveness was well illustrated by the learners’ decreased training needs in the second round even though some were passive in completing the tasks. The significance was echoed in other studies; for example, Ädel (2010) commented that an isolated experiment with no follow-up work constituted a serious limitation of her research; Götz and Mukherjee (2006) also noted that their learners would have preferred a longer introduction. This is consistent with the literature (Cobb, 1997; Kennedy & Miceli, 2001; Yoon & Hirvela, 2004), and also supports the conclusion that “DDL can take many forms to meet the needs and proficiency of learners” (Mizumoto & Chujo, 2016, p. 62).

Second, sustainable DDL calls for the constructive alignment of the instruction design and language assessment. The results of this study reveal the need for sustainability of DDL practice. Learners showed mixed and even contradictory attitudes toward future corpus use. This echoed previous studies that found the auxiliary role of a corpus in language learning (Boulton, 2009; Chambers, 2019); the interviewees in Meunier’s study (2019) said they did not need corpora except for the class. This can be explained by the educational system and cultural traditions (Chan & Liou, 2005), namely, “the more authoritative roles of teachers in Chinese culture… (and learners) are more accustomed to deductive learning in which teachers presented rules in order to save time” (Yeh et al., 2007, p. 136). Lee and Lin (2019) also found that students raised in deductive and teacher-led reasoning educational settings may struggle with and even personally reject them. Similar findings in the studies indicate the need to increase the sustainability of the DDL practice. Revising some language assessment practices to prompt learners’ motivation would be a solution (Meunier, 2019).

Third, the theoretical framework provides solid underpinnings for the DDL instruction design. Theoretical positioning is lacking in the DDL research (O’Keeffe, 2021; Pérez-Paredes, 2019). Using noticing hypothesis in SLA, constructivism, and sociocultural theories (Flowerdew, 2015) as the theoretical framework, this study offers an innovative example of how guidance can be fully employed in DDL, filling a gap “between researchers’ claims about the importance of and the need for more guidance and the small number of attempts to empirically support these claims” (Boulton & Vyatkina, 2021, p. 79). The key features are: this instruction design well integrates DDL and TBLT, emphasizing more of teacher’s guidance and learners’ interactions in the process. This also highlights learners’ problem-based learning and promotes their autonomy (Templeton & Timmis, 2023). Besides, learners’ needs were taken into consideration, namely, the universal difficulty in learning synonyms and the freedom of choosing targeted synonymous words, which greatly trigger learners’ motivation, creating conditions for cultivating self-regulated learning. It remains consistent with the zone of proximal development from sociocultural theory and represents differentiated teaching (Tomlinson, 2001). After the second round, learners indicated that the “mini-lecture” task expanded their understanding of target words from various perspectives, and scaffolding provided clues for reading concordances, which alleviated anxiety and improved the effect of the class presentation. This finding is consistent with previous studies (Jansen et al., 2019; Nelson, 1990). Further, this instructional design allowed for learner differences (i.e., L2 proficiency, and task motivation, among others), and displays the feasibility and generalizability in the EFL context. According to Templeton and Timmis (2023), these considerations make DDL more accessible to learners; they are also in line with Tomlinson’s (2011) three basic criteria for designing language learning materials: be motivating and relevant to the learners; present an achievable level of challenge; draw attention to selected features of the input.

5.5 Task-based data-driven synonym learning model

Accordingly, we present the flowchart in Figure 9 to put forward the task-based data-driven synonym learning model with meticulous pedagogical details, which is uncommon in studies and can create replication challenges (Vyatkina, 2020). The instruction process consists of pre-task, in-task, and post-task stages. The dotted part represents the standardized procedures added in the second round. Overall, the three-stage procedure takes a guided inductive approach, which is “a hybrid of an inductive DDL approach and a deductive teaching method” (Mizumoto & Chujo, 2016, p. 58), with students’ major role in discovering language and the scaffolding from the teacher and peers throughout the process.

Figure 9:

The student-centered data-driven “mini-lecture” instructional design.

Pre-task was the teacher-led training activities. Appropriate training helps learners understand the advantages of corpus so their attitudes can be transformed from using corpus only to complete in-class tasks to using corpus to promote language learning (Chambers, 2019; Lee et al., 2019; Poole, 2022). The in-task activity was based on the four-step strategy proposed by Kennedy and Miceli (2010): (1) formulating the question; (2) devising a search strategy; (3) observing the data and selecting examples; (4) drawing conclusions. Consultation with COCA provided learners with the KWIC display and drew learners’ attention to the target synonyms, thus encouraging learners’ noticing (Hyland & Milton 1997, as cited in Boulton, 2009). Post-task was an after-class quiz and follow-up feedback. It helped verify learners’ query results and construct synonym knowledge.

In this instructional design, some key elements such as task-based learning, teacher guidance, students’ autonomy, sufficient exposure duration, and scaffolding from both teachers and peers, and so on were considered. It creates an immersive learning environment. Therefore, compared to traditional synonym learning, “guided induction could be seen as a promising methodological alternative” (Mizumoto & Chujo, 2016, p. 62).