This experimental study, and partial replication of Sun and Hu (2020), examined how and to what extent direct (i.e., hands-on corpus use) or indirect (i.e., hands-off corpus use) online data-driven learning (DDL) modules were able to assist Chinese writers from an English as a foreign language (EFL) context in using more appropriate hedging (i.e., in patterns and frequencies closer to native English-speaking writers). Participants' perceptions regarding the challenges and benefits of the online corpus tools, as well as the online instructional modules, were assessed using Likert-scale surveys, open-ended survey responses, and interviews with select participants aided by a stimulated recall procedure. Students participated in four weeks of online DDL instruction, and a pre- and post- instruction writing task. Participants' hedging frequencies and patterns were statistically examined pre- and post- instruction, with log likelihood and mixed-effects modeling. Participants' Likert-scale survey responses were also examined with their open-ended survey responses and interview data. Results showed that, when both groups (direct and indirect DDL groups) were considered together, there was a significant difference across time (pre- and post- instruction) on the frequency of hedge use, indicating that DDL instruction, in general, helped to significantly increase hedge usage frequencies over time. When each group was examined separately, the direct DDL group showed a significant change over time, and the indirect DDL group showed non-significant changes over time. Therefore, while DDL instruction as a whole seemed beneficial, a more detailed analysis revealed that the direct instruction was statistically more effective. Results from the surveys and interviews identified larger patterns, and then more specific sub-themes, that participants found challenging and beneficial regarding the interface, functions, and texts, of the corpora. Challenges included, among others, the lack of easily accessible help information on the interfaces, error messages and account issues, a lack of understanding the role of certain functions, and issues with the compilation of the corpora. Benefits included, among others, the authenticity of the texts, the ability to do comparisons (i.e., across text types and learner English populations), and the ability to see keywords in larger contexts.
Read
