What corpora can (and cannot) do to improve graduate thesis writing

Abstract

This paper reports on the outcomes of a 15 hour course on corpus approaches to data-driven learning for the improvement of graduate thesis writing. 50 graduate students were introduced to corpus query software including Sketchengine, SKELL, Antconc and UAMCorpusTool corpus software, and trained in understanding concordance, collocation, word sketch, keywords and webbootcat functions. Students then submitted samples of their thesis writing, receiving highlighted errors of omission, redundancy, collocation, and phrasing as feedback from the teacher, then using the corpus tools to correct these errors in class time. Analyses were performed on the number and type of errors the students were able to correct with/without the corpus software at each stage of training. Students also completed pre- and post-course questionnaires about the effectiveness of using corpora to improve their writing. Results from the writing correction sessions suggest that errors of phrasing or collocation are often successfully resolved by corpus query, but that issues with missing or redundant words in their writing cannot easily be resolved. Moreover, the exact location and scope of the error must be very clearly defined by teachers before students can devise suitable corpus queries. Questionnaire results suggest that students are positive towards using DDL for writing when interfaces are kept simple, corpus query syntax is minimal, and where investment in time is limited, and that students’ ability to create and annotate their own corpora is considered more useful than consulting general or even discipline-specific corpora. However, students still expect the corpus to provide 'the right answer', rather than embracing an inductive approach to learning required for successful DDL. We present recommendations about how to resolve these issues in future DDL course preparation.