Lexical bundles in the academic writing of the Arts and Humanities: from corpus to CALL

1 Introduction

With the development of corpus linguistics, it has become increasingly clear that formulaic language, that is, the use of conventionalised combinations of two or more words, is pervasive in natural language use. Empirical corpus studies have demonstrated that up to 80% of language could be formulaic in nature (see Altenberg 1998; Biber et al. 1999; Erman and Warren 2000; Howarth 1998). The ubiquity of formulaic language, though, does not dovetail with traditional theories of language, in which language is decoded and composed as single words which are stitched together according to the syntactic rules of grammar (e.g., Chomsky 1965). This disjuncture between empirical corpus findings and traditional linguistic theory paved the way for new and radical theories of language. These new theories posit that language is composed of and decoded as multi-word lexical units (rather than as single words), and that grammar is the outcome of this complex lexical structure (rather than vice versa) (see Sinclair 1991; Hoey 2005). These theories reverse the traditional role of lexis and grammar, compelling linguists to elevate the importance of lexis to being at least as important as grammar.

Despite the emergence (and widespread acceptance) of these new theories, language learning was still for a long time construed as the word-by-word generation of sentences with reference to grammatical rules (Wray 2013; Meunier 2012). In fact, it was not all that long ago that practically no published L2 learning materials capitalised on the formulaic nature of language (Wood 2002). This has started to change, though, and it is now well attested that L2 learners need to be exposed to formulaic language in reading and writing for implicit learning to take place, but also that formulaic language needs to be practised extensively to facilitate explicit learning (see Cortes 2006; Jones and Haywood 2004; Ackerley 2017). To that end, a great deal of phraseology research has been dedicated to advancing the teaching-learning of formulaic language in L2 classrooms, with particular attention paid to the formulaic language of academic writing.

In its broadest sense, phraseology research covers a whole range of different word combinations, from collocations to idioms, and plays an important part in language teaching (Autelli 2021). Within this subfield of linguistic research, corpora of academic writing have been used to look at various types of formulaic language, and lists of high frequency phraseological items have been developed for pedagogical purposes. Ackermann and Chen (2013) created the Academic Collocation List containing 2,500 frequent and pedagogically useful two-word combinations that occur in writing across the academy. Simpson-Vlach and Ellis (2010) developed an Academic Formula List, which includes 207 statistically and psychologically significant three- and four-word phrases that are used in speech and writing across a broad range of academic disciplines. Miller (2020) derived a list of 545 frequent idioms from mixed disciplinary corpora of academic speech and writing. Byrd and Coxhead (2010) presented a list of 21 high frequency four-word lexical bundles occurring across academic writing. The latter, lexical bundles, is the focus of the present study.

Lexical bundles are formulaic sequences of three or more words that occur in a given register above a certain frequency threshold (Biber et al. 1999). Lexical bundles are pervasive in academic prose, with three- and four-word bundles, e.g., in order to, one of the, part of the, in the case of, on the other hand, covering around 20% of spoken and written academic discourse (Biber et al. 1999). These high-frequency lexical bundles also serve important discourse functions in academic prose. A number of studies have classified bundles using taxonomies that correspond to the metafunctions of systemic functional linguistics (Biber, Conrad and Cortes 2004; Hyland 2008; Byrd and Coxhead 2010; Durrant 2017). With research consistently confirming that lexical bundles are highly frequent and readily interpretable in functional terms, each study highlights that lexical bundles are a fundamentally important part of the student writers’ communicative repertoire.

It follows, then, that a major strand of lexical bundle studies has been examining them through a disciplinary lens. Corpus research in general tends to show that “specific disciplines […] have different ideas about what is worth communicating and how this should be done” (Hyland 2016: 22), and lexical bundle research is no exception. Studies investigating lexical bundles from a disciplinary perspective have demonstrated that bundle usage is reflective of the distinct social and rhetorical practices of different academic communities, particularly along the hard/ soft division (see Cortes 2004, 2006; Hyland 2008; Durrant 2017). This research highlights the need for a disciplinary approach to the teaching-learning of lexical bundles in academic writing. That being said, comprehensive lists of disciplinary lexical bundles that have been generated for research purposes and published in journal articles (e.g., Durrant 2017) are not widely incorporated into published classroom materials. Many materials dealing with the concerns of academic writing are, after all, designed to be as relevant and saleable to as many students as possible (Hyland 2016), meaning research that distinguishes the discoursal practices of one discipline from others can be an inconvenience to publishers. As a result, research-oriented disciplinary lists of lexical bundles (and other formulaic language, for that matter) are often not transferred to writing instruction. As Conrad (2011: 55) writes, “many teachers […] do not have the time to figure out how to incorporate new information from corpus studies into their teaching”. The aim of this paper, therefore, is not only to develop a disciplinary list of lexical bundles organised by their discourse functions, but also to offer some ready-made computer-based materials to facilitate the teaching-learning of the bundles to student writers in the Arts and Humanities (AH) disciplines.

The AH disciplines have been chosen because they are more discursive than their counterparts in the hard knowledge fields. In the ‘soft’ AH disciplines, knowledge is viewed as mediated and contested, meaning there is certain degree of internal discord (North 2005). By contrast, the ‘hard’ disciplines are characterised by internal unity and a set of shared assumptions (North 2005). These differences are, of course, epistemological, but they have a direct impact on a variety of discourse features. For example, in the soft knowledge fields, writers have to use more hedges because knowledge is contested (Hyland 2005), and more citation to establish context and persuade their readers (Hyland 1999). These differences have practical implications for AH writers throughout their academic careers. The IELTS entry requirements of British universities, for example, are typically higher for students in the AH disciplines than for those in the hard sciences. AH writers are also afforded a maximum of 80,000 words for PhD theses in comparison with 50,000 in the hard sciences. For reasons such as these, the present study focuses specifically on developing a list of lexical bundles and associated CALL resources for students in the AH disciplines.

CALL, or Computer Assisted Language Learning, refers to the application of a computer, or, more broadly, technology, to language teaching-learning. In the present study, the computer-based resources will be developed according to the principles of Data Driven Learning (DDL), a type of CALL. DDL requires students to analyse corpus data and make generalisations (Johns 1991), i.e. it is an inductive approach whereby learners are also language researchers. DDL has been found to improve students’ abilities to draw inferences, create a heightened awareness of language patterns, and facilitate the expansion of vocabulary knowledge (O’Sullivan 2007; Gilquin and Granger 2010). While some advocates of DDL encourage students to look at first-hand corpus data using corpus management software (e.g., Lee and Swales 2006; Charles 2012, 2014), others have stated that the use of professional corpus management software in the classroom can be overwhelming for both teachers and students alike (Warren 2016; Hafner and Candlin 2007). As a result, a number of user-friendly DDL tools have been successfully created and used to aid L2 writing instruction, such as Compleat Lexical Tutor (Horst, Cobb and Nicolae 2005), SKELL (Hirata and Hirata 2018) and Collocaid (Frankenberg-Garcia et al. 2019). Each of these corpus-informed CALL resources provides an excellent example of how a data-driven approach can be taken to writing instruction. However, each of the aforementioned resources comes with a caveat in that they are generic, rather than disciplinary, in nature. In contrast, the current paper will introduce disciplinary computer-based resources for the teaching-learning of lexical bundles specifically in the Arts and Humanities disciplines.

2 Corpus and methods

2.3 Qualitative analysis of the functions of bundles in the AH corpus

The 47 bundles were classified qualitatively according to a functional taxonomy. Because PhD theses are oriented towards the research world (Swales 2004), Hyland’s functional taxonomy that “specifically reflect[s] the concerns of research writing” (Hyland 2008: 13) was chosen as the point of departure for the functional analysis (Figure 1). The Sketch Engine’s random sample function was used to generate samples of 100 concordances for each of the 47 lexical bundles, which were then examined through a functional lens (i.e. a total of 4,700 concordances, approximately one third of the total concordances, were examined). It has been argued elsewhere that random samples of 100 concordances are sufficient for the purposes of categorising lexical items in a taxonomy (Groom 2007: 101, cited in Whiteside 2016). Bundles were categorised according to their primary function in the random samples of 100 (i.e. the function they served in the majority of occurrences).^[2] This process was qualitative, carried out by the researcher alone, and should not be considered absolute. As Durrant (2017: 185) points out in relation to his own categorisation, “this is not intended to be a definitive enumeration […] but rather a means of interpreting the present list”. Two modifications were made to Hyland’s taxonomy (Figure 1) for the purposes of the present study:

1. The names of the three main categories were simplified to (1) presentation of content, (2) organisation of the text, (3) expression of attitudes by the writer. This aims to facilitate use of the list in classroom settings with less-proficient language learners (Byrd and Coxhead 2010: 42).

2. The description sub-category was divided into physical description and abstract description. While some researchers use the sub-category intangible framing attributes to describe bundles indicating an abstract property of something (e.g., Biber et al. 2004; Durrant 2017), it was felt that abstract description was again more teaching-learning-friendly.

Fig. 1:

Hyland’s (2008) functional taxonomy modified

3 Results and discussion

4 Pedagogical implications and applications

With reference to the quantitative and qualitative findings discussed above, this section presents a series of ready-to-use computer-based resources to ensure that this research can be directly applied to the classroom. The first of these resources is the Bundle Finder (O’Flynn 2019) (Figure 2), a powerful but user-friendly and free-to-use tool developed using H5P. This tool can search any text for all 47 lexical bundles and return their discourse functions, enabling both teachers and learners to find examples of the bundles in context. Teachers can use the Bundle Finder to develop classroom activities and materials and set learning goals, while students can use it for independent learning activities. The Bundle Finder could, for example, be used for rhetorical consciousness-raising activities, in which students identify where a writer, be it themselves or a professional, has chosen to use a certain bundle (e.g., a stance feature) and then provide possible explanations for this rhetorical choice (Hyland 2016: 24). In many cases lists of formulaic language items are presented as static and decontextualised lists, but it is hoped that by enabling students to see the bundles functioning in the texts they read and write, they will consider the bundles a valid use of their learning effort. This is important because one of the pedagogical challenges of working with lexical bundles is a lack of face validity for some students (Byrd and Coxhead 2010). It is, however, important that both teachers and learners know how the list of bundles has been developed (Byrd and Coxhead 2010), which is why the Bundle Finder also includes information, in layman’s terms, regarding the corpus from which the list was derived and the criteria by which the bundles were selected for the list.

Fig. 2:

Arts and Humanities Bundle Finder (URL: https://h5p.org/node/438388)

The second set of materials exemplify how the findings of this study could be realised in corpus-based classroom materials. The sample materials presented below (Figure 3) were developed in H5P using the BAWE AH subcorpus and aim to demonstrate how some of the bundles could be used in DDL classroom activities. These sample materials require students to analyse the concordances not only in respect to the bundle itself, but also in respect to the dense academic language to the left and right to draw inferences about function and use in context. These materials, inspired by Johns’ (2000) Kibbitzers, do not require students to analyse a corpus directly but rather to analyse a selection of corpus data presented to them by the teacher. Materials such as these will be well suited to the majority of student writers who do not have prior knowledge of corpus management software, and could be used to introduce students to key DDL concepts, e.g., concordances. The use of corpora and concordances in the classroom by students has been increasingly investigated since it was pioneered by Johns (1991) as Data-Driven Learning (DDL). Charles (2012) found students in the Arts and Humanities disciplines to be particularly receptive to DDL, and several researchers have used corpus-based instruction to improve students’ use of formulaic language in their academic writing (see Thurstun and Candlin 1998; Wu, Witten and Franken 2010; Charles 2012; Ackerley 2017). In fact, Byrd and Coxhead (2010) specifically recommend the use of corpora in the classroom to teach lexical bundles.

Fig. 3:

DDL sample materials (URL: https://h5p.org/node/727097)

The third resource is a set of Quizlet flashcards, which can be used in a variety of games, exercises and tests (Figure 4). Quizlet and other game-based apps have been found to have a range of advantages for vocabulary learning, including a positive effect on acquisition, motivation, and autonomy (see Chien 2015; Anjaniputra and Salsabila 2018; Chen, Liu and Huang 2019; Bueno-Alastuey and Nemeth 2020). The Quizlet sets can be introduced by teachers on short pre-sessional courses and then used independently by students on a long-term basis. This is important because it can be particularly challenging for teachers to help learners improve their use of bundles during short courses of instruction (Byrd and Coxhead 2010). In fact, improving the active use of lexical bundles by learners is likely to be a much longer-term project (Cortes 2006). It is, therefore, important that learners continue their independent study of lexical bundles beyond their short courses of writing instruction and into their full courses of study in AH departments, which is something Quizlet can facilitate.

Fig. 4:

Quizlet sets (URL: https://bit.ly/3NWEwkj)

The final resource for the teaching-learning of the lexical bundles is the list itself. A printer-friendly version of the full list organised by discourse functions is included in Appendix 2. Although this list is static and presents the bundles as decontextualised items, it can be a useful resource if used appropriately. The functional list will be useful as a quick reference guide for teachers to keep a record of the bundles that have been subject to attention in classroom-based teaching-learning activities, or for students to monitor their independent learning activities. Keeping accurate records is important because the bundles will need to be revisited regularly (Byrd and Coxhead 2010). It is also important that the explicit teaching of the bundles on the list is mixed with ample opportunities for implicit learning (Hultsjin 2001), which means “ensur[ing] that students are reading academic prose so that they encounter academic vocabulary” (Byrd and Coxhead 2010: 56). Empirical studies suggest that through implicit learning and explicit instruction, L2 learners can start to develop better strategies for producing lexical bundles more frequently and appropriately in their academic writing (see Cortes 2006; Jones and Haywood 2004; Ackerley 2017).

5 Limitations

It is important to acknowledge some potential limitations of the study. One potential limitation is that the corpus includes only one genre of text produced by university students at one level and one institution. Although it would have been desirable to compile a corpus of texts produced by university students at various levels across institutions, empirical evidence from the BAWE corpus presented in this paper suggests that the list has scope beyond just the genre, level and institution from which it was derived. Another potential limitation is that the qualitative analysis was based on random samples of 100 rather than on every instance of the bundles in the corpus. As such, the functional categorisation, like others, should be treated with some level of caution and seen as a guide to facilitate teaching-learning, rather than as a definitive enumeration.

6 Conclusion

This paper has developed, discussed and presented a functional list of lexical bundles and associated resources for the teaching-learning of lexical bundles in the academic writing of the AH disciplines. At every stage of the development of the list and resources, methodological decisions were guided by pedagogical considerations. The final product is a short and powerful list of lexical bundles. The 47 bundles alone cover up to an impressive 0.9% of academic writing across genres, levels and institutions in the AH disciplines. The ready-made computer-based resources facilitate the immediate incorporation of the list into academic writing programs. It is hoped that by adding further evidence to demonstrate that bundle usage is reflective of the distinct social and rhetorical practices of different academic communities, this paper will spur more researchers and teachers to develop disciplinary lists of lexical bundles (or, for that matter, other types of formulaic language), and to incorporate their findings into their curriculums. Further studies could investigate the effects of incorporating a list of disciplinary lexical bundles into the curriculum on the accurate and appropriate production of lexical bundles in student writing.