Combining diagnostic and prescriptive corpus functions for ELT: The HKU-CAES learner corpus

Abstract

The use of corpora as a tool in English language teaching may now be described as a 'marriage', rather than a 'fling', following the parlance of Gabrielatos (2005). In other words, it is now increasingly apparent that if we wish to demonstrate - in quantifiable terms - the impact of SLA and ELT methodology on student performance, it is through corpus analysis that some of the most revealing insights into the efficacy of such methodology may be ascertained (Tsui, 2004; Conrad, 2005; Chau, 2012). Aside from their diagnostic usefulness, corpora are now essential tools driving innovation in ELT (Hyland & Wong, 2013), the development of new course materials (Alexander, 2007; Jones & Durrant, 2010), the sequencing of course curricula (Thorne, Reinhardt & Golumbek, 2008), and for the derivation of course assessment and grading criteria (Barker, 2006, 2010; Hawkins & Buttery, 2009). Increasingly, the use of specialised corpora for 'English for academic purposes' (EAP) have allowed for greater insights into the type and frequency of linguistic features that are representative of particular disciplines (Wright, 2008; Nesi, 2009; Boulton, Carter-Thomas & Rowley-Jolivet, 2012) which can be used to enhance the teaching of these disciplines (Reguzzoni, 2013). Measuring native data alongside the linguistic data of EAP coursebooks can aid the future development of such coursebooks and other materials (Gabrielatos, 2005). For example, data-driven activities using learner, native and cross-corpus data should also be ‘an integral part, rather than an additional option, of the overall language curriculum’ (McEnery & Xiao, 2010:24). In terms of academic writing, the use of professional corpora based on written data published in journals, reports etc. works differently to regular native vs. non-native cross-corpus comparison in that even native (L1) data produced at HKU by non-HK students can be fairly compared against the data of ‘established’ writers, thus making no a priori assumptions of the L2 dataset as deficient to an L1 dataset (Gabrielatos & McEnery, 2005). However, by observing L2 data, we can ‘define areas that need special attention in specific contexts and at different levels of competence, and so devise syllabi and materials’ (Gabrielatos, 2005:6), alongside improved insights into the language learning process through the analysis of learning as a product, evidenced in real, authentic L2 production. This is because, as claimed in McEnery & Xiao (2010), ‘if learner performance data is shaped and constrained by such a mental process, it at least provides indirect, observable, and empirical evidence for the language acquisition process’ (p.18). The proposed paper for IELT-Con 2015 looks at the construction of a new multimillion-word longitudinal corpus of second-language (L2) English entitled ‘The HKU-CAES Learner Corpus’. It will initially collect L2 data sourced from the academic writing of undergraduate EFL students (mostly L1 [native] Cantonese- and Mandarin-speaking students), totalling 6 million words. This data will be collected at three key stages during the semester, making the HKU-CAES corpus a relatively rare case of a longitudinal corpus. The paper asks whether the academic skills taught on current undergraduate English writing programmes over one semester result in any improvement to the linguistic features of students’ L2 English academic writing as evidenced by corpus analysis. In particular, suggestions for the building and annotation of learner corpora that do not involve substantial knowledge of coding (making them suitable for the majority of English language teachers) will be discussed, as well as how learner corpora can inform future developments regarding L2 assessment, curriculum and materials development that can be specifically and quantifiably tailored to ELT students.