Corpus-informed vocabulary instruction in Bangladesh: a practical framework for EFL contexts

2.1 Theoretical foundations in vocabulary learning

Vocabulary knowledge is more complex than simple knowledge of what the words mean. The vocabulary learning process, as described by Nation (2013), is multidimensional and involves form (the way words appear and sound), meaning (what words mean), and use (how words are used). The emphasis should be put on teaching vocabulary to address all these dimensions, not one or two.

The theory of task-induced involvement, proposed by Laufer and Hulstijn (2001), explains why some vocabulary-learning exercises are more effective than others. According to this theory, learning activities that engage learners in evaluating, searching, and actively applying target vocabulary items lead to greater retention than passive exposure or automatic repetition. The implications of this principle for corpus-informed activities are considerable, and these activities engage learners in meaningful work with the target vocabulary.

However, when applying these theoretical frameworks in an EFL context like Bangladesh, one must be aware of local factors. English is taught as a foreign language in Bangladesh, as well as a second language, in an environment referred to as the expanding circle, in which learners are exposed to English to a limited degree outside the classroom. In such situations, students face special difficulties in learning the language because they need to master higher stages of lexical development with minimal real input.

Furthermore, the assessment culture in education in Bangladesh has the potential to cause stress in pursuit of immediate, measurable results, which may be incongruent with the progressive nature of vocabulary development (Rahman 2020). The pressure on both teachers and students is to focus on aspects of English that are readily evaluated, which, in the vast majority of cases, are grammar rules and discrete vocabulary items, rather than patterns of natural language use.

2.2 Corpus linguistics and vocabulary instruction

Corpus linguistics has transformed our understanding of how vocabulary works in real English texts. Researchers have been able to define patterns by analysing millions of words in authentic contexts, patterns that were previously invisible to language teachers and students (Sinclair 1991). These lessons have led to significant advancements in vocabulary education, particularly in mastering collocations, formulaic phrases, and academic vocabulary.

Word pairs that are natural combinations are known as collocations, which are an essential but frequently not well-studied part of vocabulary. Conventional vocabulary instruction generally focuses on individual words, whereas corpus studies reveal that proficient writers combine predictable word sequences to form natural-sounding prose (Nesselhauf 2005). In academic writing, in particular, certain collocations are necessary to achieve the accurate, formal register required for academic discourse (Ackermann and Chen 2013).

Learners processing corpus data directly through data-driven learning (DDL) techniques, which require identifying patterns in language, have demonstrated specific potential in vocabulary acquisition (Boulton and Cobb 2017). These approaches guide learners in developing metalinguistic awareness – understanding how language functions – which, in turn, facilitates long-term vocabulary acquisition. Nevertheless, the majority of DDL studies have been conducted in environments with well-developed technological infrastructure and comprehensive teacher preparation programs.

The difficulty lies in adapting these insights to resource-constrained environments. Conventional corpus-informed teaching typically presupposes access to large-scale corpora, specialized software, teacher training, and flexible curricula, which enable the use of innovative methods (Kennedy and Miceli 2017). These demands pose severe constraints in situations where basic educational materials are scarce.

2.3 L1 transfer in vocabulary learning

The vocabulary difficulties faced by Bangladeshi learners are often influenced by the transfer of Bangla, their first language. L1 transfer occurs when learners transfer patterns from their mother tongue to the target language, and this may lead to mistakes; however, it is also a valuable source of learning (Jarvis and Pavlenko 2008).

Among Bangladeshi learners of English, recurring transfer patterns were observed, particularly in the formation of collocations and the use of prepositions. As a case in point, when learners are asked to write to conduct research, they tend to write to make research, since the word in the Bangla language is a verb that is closer to “make” than “conduct.” These are predictable, rule-governed patterns; hence, they can be taught systematically rather than corrected ad hoc.

However, conventional methods of transfer often have a deficit view of the transfer, and L1 influence is perceived as a nuisance that should be eliminated. Current studies indicate that transfer may become a beneficial phenomenon in language learning when learners become aware of cross-linguistic differences and similarities (Ringbom 2007). This outlook enables a shift in emphasis on the removal of transfer to assist learners in utilising their multilingual resources strategically.

Cross-linguistic studies have shown that patterns of transfer are predictable in systematic ways by examining the structural variations between learners’ first and target languages (Odlin 1989). This predictive ability enables the formulation of proactive teaching strategies that address potential challenges before they become entrenched.

2.4 Assessment and institutional contexts

Contemporary assessment practices should be encouraged to support vocabulary development initiatives in educational settings, as they will help demonstrate institutional support and keep students motivated. However, traditional knowledge-testing models are more likely to emphasise discrete knowledge rather than its application to context, which inherently breaks with the ideal practice of instruction (Read 2000). The accuracy of collocations, the correct application of the formula, and the degree of lexical sophistication are aspects of words-in-context measurement that are prioritised in modern vocabulary testing (Kremmel et al. 2017). Such practices are less inconsistent with instructional practice guided by the corpus since they focus on language patterns in real-life situations rather than knowledge outside of context.

In such a scenario, as in Bangladesh, the challenge is to implement new assessment innovations that focus on vocabulary without requiring extensive changes to existing systems. New teaching must be clearly connected to the organisational goals, and students and educators in the institution will not oppose new approaches, even though they may have pedagogical advantages (Wall and Alderson 1993).

2.5 Research questions

This study investigated three research questions (RQs):

• RQ1: What specific vocabulary patterns characterize Bangladeshi learners’ academic writing compared to target academic discourse?

• RQ2: How can a corpus-informed vocabulary framework, designed from these findings, function effectively within Bangladesh’s educational constraints while producing measurable learning improvements?

• RQ3: What implementation protocols enable teachers to sustain corpus-informed vocabulary instruction independently, and how feasible is the approach for broader adoption across institutions?

3.1 Research design

The research examined in the article used a single-group, mixed-methods feasibility design to assess the feasibility and classroom applicability of corpus-informed vocabulary instruction in Bangladeshi EFL classrooms. The quantitative part considered changes in the written language of the students before and after the intervention period of 36 weeks. In its qualitative part, teacher interviews and student focus group discussions were analysed to provide a picture of the perception of feasibility and pedagogical impact.

The design focused on documentation and implementation assessment, and not statistical generalisation. Lexical gaps and transfer-based errors were first defined in the corpus analysis of the learner’s writing, which informed the intervention’s instructional materials. Follow-up classroom results showed pre- and post-writing outcomes, supplemented by teacher and student comments. The combination of these sources ensured descriptive validity and provided an overview of how corpus-informed academic collocations were targeted in different conditions.

3.2 Participants and setting

The research was conducted in English-language classes at tertiary institutions. The selection of classes was done in consultation with the administrators, based on consent and logistical feasibility. Table 1 presents the total number of enrolled students and those who have completed the pre- and post-writing activities. The inclusion criteria were that the participant had to be enrolled in a mandatory English course at the time of the study; there was no random assignment or control group. The entire teaching was conducted during regular classroom sessions and in accordance with each institution’s existing curriculum.

3.3 Corpus development

In this study, two complementary corpora were created: one a learner corpus indicating student writing, and the other a reference corpus indicating the academic language observed in the local instructional materials. The corpus of learners was estimated to contain 285,000 authentic student writing tokens gathered over 18 months during regular classroom activities. These were academic essays, exam answers, and writing tasks, the typical genres of Bangladeshi tertiary education.

The reference corpus consisted of approximately 420,000 tokens, selected from locally available textbooks, nationwide examination papers, and university-level assignments. This ensured that the corpus was realistic and reflected what the students are exposed to in their coursework. This localised corpus also emphasised contextual relevance, unlike foreign academic corpora, which would not align with local standards.

Preprocessing and analysis of all texts were performed using spaCy 3.7.2 (tokenization, lemmatization, and part-of-speech tagging) and AntConc 3.5.8 (frequency analysis and collocation analysis). A 5 % sample was manually validated to assess tagging accuracy. Several cross-checks were conducted during corpus construction to improve reliability. Although this style focused more on local authenticity, it also meant less access to broader global academic discourse. This weakness is recognised and elaborated in Section 5.4.

3.4 Framework development

The teaching model described in this paper integrated corpus-informed principles with class-based constraints typical in Bangladesh EFL settings. It comprised three components in relation to each other:

1. The Bangladesh Academic Collocation List (BACL). As the lexical basis of the intervention, a localised academic collocation list was gathered. Based on the learner and reference corpus (Section 3.3), statistical filtering was used to identify high-frequency collocations (Mutual Information ≥ 3; minimum frequency = 5). The resulting BACL suggested 800 verb-noun and adjective-noun combinations commonly found in academic discourse, but they were somewhat challenging for local learners.

2. The predictive model of L1 transfer errors. To address the prevalent lexical and collocational errors resulting from L1 influence, a predictive model was developed by comparing the repetitive patterns in the writing of Bangladeshi students with those of native English academic language. The categories of mistakes (e.g., omission of a preposition, miscollocation, overgeneralization) were hand-coded and used to create specific educational resources. This model allowed teachers to predict the types of errors and to draw attention to the instruction.

3. Protocol of Implementation and Teacher Support. An instructional protocol helped teachers follow every instructional stage. It featured weekly lesson models, clear instructional objectives, and peer-reviewing tools for students. Teachers were given brief orientation training on how to use the corpus and on the pedagogical reasoning behind the tasks. The framework focused on manageability-the teachers were invited to match examples and pacing to their own syllabus without compromising the integrity of the core instructional cycle.

The combination of the three elements created a context-dependent system that could be replicated to integrate the teaching of corpus-informed language knowledge with regular curricula. The complete framework materials, including sample collocation lists, weekly guides, and coding templates, are presented in the Appendices.

3.5 Intervention procedures

The vocabulary programme based on the corpus was completed in 36 weeks, with two 45-minute lessons per week (about 54 hrs total). The teaching cycle incorporated corpus analysis, guided discovery, as well as production-based writing tasks. The Bangladesh Academic Collocation List (BACL), created for this study, was used to select 10 academic collocations per week.

3.6 Outcome measures

Three outcome indicators were selected to report changes in students’ written language after 36 weeks of intervention. Operationalization was used to define each measure, enhancing transparency and replicability.

1. Academic vocabulary coverage. The student’s writing was analysed in terms of the percentage of tokens selected from the known academic word lists (e.g., the Academic Word List and the Bangladesh Academic Collocation List). The coverage was measured as a percentage of total words, indicating the extent to which learners incorporated academic vocabulary during a lengthy writing task.

2. Collocation accuracy. Correctness was measured using a predetermined set of verb-noun and adjective-noun collocations that were focused on during teaching. Each case was categorised as correct, wrong, or half correct based on grammatical and semantic correctness. The scores were reported as the number of accurate collocations per 1,000 words.

3. Formulaic sequence use. The number of times multiword phrases and sets of academic words (e.g., as a result of, in terms of, it is essential to note that) appeared was divided by 1,000 words. This was an action that demonstrated productive application of ready-made academic expressions by the learners.

3.7 Data analysis

A combination of thematic analysis and descriptive statistics was used to analyse data from student writing, teacher interviews, and student focus group discussions.

3.8 Ethical considerations

The research followed institutional and international ethics of education research. No formal permission was obtained from an Institutional Review Board (IRB) because the research posed little risk and was conducted in the regular classroom setting. The involvement was voluntary; no experimental interventions or sensitive personal information were gathered.

Participating institutions and course instructors gave their consent before taking part. Students were informed of the study’s purpose, the confidentiality of their responses, and their right to withdraw from the study at any time without penalty. Before data collection, written consent from the teachers and verbal consent from the students were obtained.

Any analysis of all writing samples and interview transcripts was carried out anonymously, and any identifying information was omitted. The research was conducted in accordance with the ethical standards of the British Educational Research Association (BERA) Ethical Guidelines to Educational Research (2018) and conventional procedures of transparency, confidentiality, and voluntary involvement in a classroom-based research.

4.1 Quantitative descriptions of language development

The pre- and post-writing samples were compared descriptively and showed improvement across the three measured indicators. There was a significant change in the percentage of academic-word tokens in written language; more have used target collocations correctly and more often used formulaic academic language. Even though no statistical testing was performed, the descriptive patterns indicate a slow increase in academic vocabulary control and collocational competence over the 36 weeks.

These findings were supported by teachers’ classroom notes and student feedback, which indicated a better understanding of collocational patterns and greater confidence in applying academic vocabulary. Teachers also claimed they used fewer direct translations of Bangla and less repetitive or generic vocabulary.

This descriptive trend was confirmed in a review of sample essays. Indicatively, the common combinations of lexical constructions, e.g., make research or give suggestions, were used in typical sentences before the intervention. Post-intervention samples, in contrast, also included more target-like utterances, such as conducting research and offering tips. Moreover, learners generated a greater number of discipline-appropriate phrases (e.g., “play a vital role”, “conclude”), providing evidence of exposure to and internalisation of more academic collocations.

4.2 Qualitative insights from teacher and student feedback

Thematic analysis of teacher interviews and student focus-group discussions found a few common themes:

1. Relevance of activities of corpus perceived: Both groups ranked the authenticity of examples based on a real academic situation as more significant than the list of examples in a textbook.

2. Greater learner engagement: Students said the discovery-based teaching method made vocabulary learning less robotic and prompted them to pay closer attention to collocations when reading.

3. Practical deliverability: Teachers outlined the materials to be manageable, with an average of 20 mins per week of preparation time, which they stated they could manage.

4. Evidence of writing improvement: Teachers reported fewer inaccurate lexical choices and less unnatural phrasing in the final assignments, consistent with the writing analysis.

Teacher and student statements also supported these quantitative observations. As one teacher remarked, students begin paying attention to collocations even in the reading materials, pointing them out during the discussion. A student also found himself saying, “Now I understand why we say conduct research, not make research, because I noticed that it was repeated in the corpus lines”. These remarks imply increased metalinguistic awareness and enhanced involvement in natural language patterns.

The overall findings suggest that the intervention was pedagogically viable, popular among participants, and capable of generating observable lexical growth within current curricular limitations. A summary of the trends and qualitative findings is shown in Table 4, which indicates the main results of the intervention and their pedagogical explanations.

5.1 Addressing the research questions

The findings directly address the three research questions guiding this study.

Regarding RQ1, the study identified specific vocabulary patterns that characterize the academic writing of Bangladeshi learners in contrast with target academic norms. The pre-intervention analysis revealed several recurring issues: the overuse of high-frequency, general-purpose vocabulary; frequent collocational errors influenced by L1 transfer; and the limited use of formulaic academic expressions. These features collectively highlight the lexical gap between learners’ writing and standard academic English. Such trends are consistent with previous findings in EFL contexts, underscoring the persistent challenge of developing lexical sophistication in education systems that emphasize results over language depth (Nation 2013).

For RQ2, the study outlined practical steps for implementing a corpus-informed vocabulary framework in Bangladesh’s educational setting. The intervention demonstrated that corpus-informed tasks and materials can be feasibly and meaningfully designed and applied even in resource-limited environments. Teachers were trained to interpret collocational data and integrate it into instruction, while learners completed writing tasks derived from locally relevant corpora. Improvements in lexical coverage, collocational accuracy, and phraseological variety indicate that the framework appeared to support the intended learning outcomes – enhancing learners’ ability to produce more academically appropriate writing.

With respect to RQ3, the investigation established implementation protocols that enable teachers to sustain corpus-informed vocabulary instruction independently. The training component played a pivotal role in building educators’ confidence and competence in applying corpus insights in their own classrooms. The model required minimal additional resources and proved sustainable after initial training. Furthermore, the potential for inter-institutional scalability appears promising, though broader trials are necessary to validate its feasibility on a national scale.

5.2 Integration of quantitative and qualitative evidence

Although the findings are primarily descriptive, their validity is reinforced by convergence across multiple data sources, by the overlap between textual data and the study participants’ participation. It was observed that students were often correcting their incorrect collocations, such as “make research”, with standard words like “conduct research”, and this was also reflected in the learner corpus. Students also said they felt equipped to organize their writing better through learning formulaic expressions, such as “on the other hand” and “it should be noted that”. These observations correlate with the diversity of phrases in learners’ texts.

The consistency of the triangulation of evidence based on text analysis, teacher observations, and student reflections enhances the reliability of the findings. Due to the subjectivity inherent in single-method analysis in applied linguistics research, methodological triangulation, as described by Dörnyei (2007), serves as a protective measure against this subjectivity. This work addresses the limitations of failing to present detailed statistical indices by demonstrating convergence across several sources.

5.3 Relevance to existing research and pedagogical contribution

The findings of this feasibility study align with and extend prior research on corpus-informed pedagogy. The Academic Word List, created by Coxhead (2000), demonstrated the utility of corpus-informed lexical resources for academic literacy, and the Academic Vocabulary List by Gardner and Davies (2014) highlighted the need to update word lists to account for disciplinary variation. This work continues the tradition and further develops the Bangladesh Academic Collocation List (BACL), a localised resource that reflects the lexical patterns used in Bangladesh’s academic environment.

This enhancement in collocational accuracy is consistent with Nesselhauf’s (2005) claim that the primary issue for L2 learners with collocations is negative transfer and that collocational patterns cannot be improved without explicit exposure to them. Likewise, Durrant and Schmitt (2010) also emphasize the pedagogical significance of corpus-informed collocation teaching. This research paper contributes to the current body of research by demonstrating that such practices can be successfully translated in a low-resource setting.

The increase in the number of formulaic sequences used confirms the argument by Wray (2002) that formulaic language is the focus of fluency and discourse coherence. Although the concept of phraseology has been employed numerous times in the past, particularly in more resource-intensive or technology-intensive contexts, the current results suggest that even simple corpus-informed exercises can aid learners in acquiring multi-word phrases, which, in turn, facilitate the comprehension of academic texts.

The study demonstrates that corpus-informed pedagogy can be effectively localised without losing theoretical rigour, thereby expanding the global evidence base for context-sensitive CALL practices.

5.4 Limitations and directions for future research

The research has several limitations that cannot be ignored, despite its positive results. On the one hand, the reference corpus was primarily compiled from locally commissioned materials, including textbooks, examination scripts, and student assignments. This was contextually appropriate, but it may have limited access to the broader scope of academic vocabulary found in international corpora. Therefore, students can remain unfamiliar with idiomatic or globally standardised patterns of use. The subsequent study may test the hybrid type of corpora, such as local and international sources, to balance relevance and introduction in the global academic discourse. Second, the researchers did not administer a delayed post-test to evaluate long-term retention. Although immediate gains were realised, it is unclear whether they would persist over time without constant exposure. A longitudinal study with a delayed post-test would also help determine the sustainability of corpus-informed teaching. Third, even though teacher training was practical in the institutions under analysis, the model has not been tested at scale. The implementation may be affected by variability across institutions in resources, teacher preparedness, and curriculum needs. Future research may investigate how flexible the framework is under various institutional conditions, such as rural schools with limited infrastructure. Lastly, the fact that this article qualitatively presents its findings is a drawback, as it lacks descriptive statistical measures. The goal of future work should be to present quantitative estimates of effect sizes and levels of significance to increase the external validity of the findings. Despite these constraints, the study provides a transparent foundation for future controlled trials that could quantify the observed descriptive patterns.

5.5 Pedagogical implications

Despite these limitations, the findings hold meaningful implications for the practice of EFL teaching in low-resource contexts. To begin with, the paper demonstrates that corpus-informed approaches do not require costly technology or specialized software. Properly selected resources, such as the BACL, can help a teacher guide learners to identify real academic patterns using classroom resources. This is particularly important in situations such as Bangladesh, where technological barriers often hinder the adoption of innovation. Second, the findings emphasise the key importance of teacher training. Teachers not only adopted practices with corpus information, but also maintained them without intervention. This implies that teacher education in corpus linguistics should be incorporated into professional development programs, enabling instructors to innovate in their classrooms. Third, the framework also works towards the overall objective of educational equity. The intervention enabled students to access academic language sources that were generally not available in their learning contexts by adapting corpus-informed pedagogy to their local needs. This can help decrease the gap in educational performance and increase learners’ preparedness for higher education and international communication.