This doctoral thesis presents three studies in collaboration with the open source FLAX project (Flexible Language Acquisition flax.nzdl.org). This research makes an original contribution to the fields of language education and educational technology by mobilising knowledge from computer science, corpus linguistics and open education, and proposes a new paradigm for open data-driven language learning systems design in higher education. Furthermore, the research presented in this thesis uncovers and engages with an infrastructure of open educational practices (OEP) that push at the parameters of policy for the reuse of open access research and pedagogic content in the design, development, distribution, adoption and evaluation of data-driven language learning systems. 

Study 1 employs automated content analysis to mine the concept of open educational systems and practices from qualitative reflections spanning 2012-2019 with stakeholders from an on-going multi-site design-based research study with the FLAX project. Design considerations are presented for remixing domain-specific open access content for academic English language provision across formal and non-formal higher education contexts. Primary stakeholders in this ongoing research collaboration include the following: knowledge organisations – libraries and archives including the British Library and the Oxford Text Archive, universities in collaboration with Massive Open Online Course (MOOC) providers; an interdisciplinary team of researchers; and knowledge users in formal higher education – English for Academic Purposes (EAP) practitioners. Themes arising from the qualitative dataset point to affordances as well as barriers with the adoption of open policies and practices for remixing open access content for data-driven language learning applications in higher education against the backdrop of different business models and cultural practices present within participating knowledge organisations.

Study 2 presents a data-driven experiment in non-formal higher education by triangulating user query system log data with learner participant data from surveys (N=174) on the interface designs and usability of an automated open source digital library scheme, FLAX. Text and data mining approaches (TDM) common to natural language processing (NLP) were applied to pedagogical English language corpora, derived from the content of two MOOCs, (Harvard University with edX, and the University of London with Coursera), and one networked course (Harvard Law School with the Berkman Klein Center for Internet and Society), which were then linked to external open resources (e.g. Wikipedia, the FLAX Learning Collocations system, WordNet), so that learners could employ the information discovery techniques (e.g. searching and browsing) that they have become accustomed to using through search engines (e.g. Google, Bing) for discovering and learning the domain-specific language features of their interests. Findings indicate a positive user experience with interfaces that include advanced affordances for course content browse, search and retrieval that transcend the MOOC platform and Learning Management System (LMS) standard. Further survey questions derived from an open education research bank from the Hewlett Foundation are reused in this study and presented against a larger dataset from the Hewlett Foundation (N=1921) on motivations for the uptake of open educational resources. 

Study 3 presents a data-driven experiment in formal higher education from the legal English field to measure quantitatively the usefulness and effectiveness of employing the open Law Collections in FLAX in the teaching of legal English at the University of Murcia in Spain. Informants were divided into an experimental and a control group and were asked to write an essay on a given set of legal English topics, defined by the subject instructor as part of their final assessment. The experimental group only consulted the FLAX English Common Law MOOC collection as the single source of information to draft their essays, and the control group used any information source available from the Internet to draft their essays. Findings from an analysis of the two learner corpora of essays indicate that members of the experimental group appear to have acquired the specialised terminology of the area better than those in the control group, as attested by the higher term average obtained by the texts in the FLAX-based corpus (56.5) as opposed to the non-FLAX-based text collection, at 13.73 points below.