Intelligent CALL Systems for Writing Development: Investigating the Use of Write & Improve for Developing Written Language and Writing Skills

Introduction

Giving feedback to learners to help them to develop both written language (i.e., the language they produce) and writing skills (i.e., the strategies they use to produce writing) is a core facet of language teachers’ professional lives (Hyland & Hyland, 2006; Nassaji, 2020). Technological innovation in supporting this practice is well-established in the literature on computer-assisted language learning (CALL) (cf. Frankenberg-Garcia, 2020), and in the context of writing development, research has found that intelligent CALL systems that can respond instantly to learner language can support the production of better writing output by learners (Tschichold & Schulze, 2016). Such writing technologies that involve learners practicing their writing online and in their own time are particularly pertinent for supporting learner autonomy and self-led online writing development (Ghufron & Nurdianingsih, 2019).

Building on the relevance of such feedback and accuracy development technologies to language production, the field of CALL has seen an increase in the use of data-driven learning technologies that combine language data, language models, and pedagogical theories on feedback and language learning to inform their development. Collocaid, for example, takes a feed forward approach to developing learner knowledge of collocational patterns in academic language by using corpus data to suggest collocations and sentence patterns that learners might not have considered in their writing (Frankenberg-Garcia, 2020). A similar corpus-informed project, designed to give feedback on academic writing, is the BAWE QuickLinks project. This project uses Sketch Engine links to sanitize concordance searches of the British Academic Written English corpus (2008) that direct learners to language models for addressing language errors identified in their written work (Vincent & Nesi, 2018). SKELL (Sketch Engine for Language Learning; Lexical Computing, 2019) is another technology of note that allows for word searches, synonym checks, and collocation analysis. Lexical Computing (2019) reports that SKELL is “a state-of-the-art cloud tool for building, managing and exploring large text collections in dozens of languages. It is used all over the world by many individuals, as well as companies such as Cambridge University Press, Oxford University Press and Macmillan.” Each of these technologies offers different means of engagement and insights to their users. Collocaid feeds forward and avoids corrective feedback, BAWE QuickLinks offers corpus-based feedback with example sentences based on errors, and SKELL is a reference technology used to check how language items are used. This study is based on Write & Improve (2020), which, unlike these other technologies, offers corpus-informed automated and corrective feedback.

Write & Improve uses machine-learning technology and data from the 30-million word error-annotated Cambridge Learner Corpus (Cambridge University Press & Cambridge Assessment, 2020) to identify errors in learner written language. The technology identifies errors for which it is 90% certain and, owing to its design, uses input data from learners on an ongoing basis to further inform its identification of error patterns (Write & Improve, 2020). This technology can determine the level of learners’ language, benchmarked against the Common European Framework of Reference (CEFR), and it delivers summative feedback, indirect and formative feedback, and progression feedback on learners’ writing. The technology seeks to guide learners to notice and address language errors, while facilitating learner autonomy and engagement. Typically, students can respond to writing tasks that reflect Cambridge English language examinations or tasks set by their teacher, who can create virtual classrooms and workbooks for their students. Students work in their own time and receive automated feedback from the technology. A major benefit of this is that students can gain feedback in a non face-threatening (Brown & Levinson, 1987) environment, which can lessen the anxiety they feel during feedback, and in turn, lower the Affective Filter (Krashen, 1985) in order to enhance the learning experience. There is also space for teachers to manually add feedback, and, as a result of this type of feedback mechanism, motivation is also heightened (Golonka et al., 2014).