Do-it-yourself corpora and discipline-specific writing: A focus on the benefits of corpus building and cleaning (in the age of large language models)

Within the field of data-driven learning (DDL), an increasing number of studies have underscored the benefits of creating small specialized DIY corpora for the teaching/learning of discipline-specific writing. However, in these studies, corpus cleaning is often described as an optional step in the process of corpus creation and is frequently presented as a tedious, unnecessary, and time-consuming task, with the majority of scholars calling for the creation of “quick and dirty” corpora. In this paper, we re-examine corpus creation, namely corpus cleaning and metadata construction, for discipline-specific writing. More specifically, our paper seeks to reframe corpus cleaning and metadata construction as meaningful and purposeful activities that increase learners’ awareness of disciplinary norms and conventions, particularly in a comparative context. We base our analysis on the reflections provided by learners from various disciplines who designed, compiled, cleaned, and analyzed corpora, along with the final papers they drafted for the courses they took in the Department of English at a teaching-focused research university in the Middle East. Corpus cleaning and metadata creation, as we hope to show, not only make visible the invisible writing conventions within disciplines, including the integration of evidence and raw data, but also position language learners as data engineers, promoting their critical awareness of the role and nature of (language) data in the age of Large Language Models (LLMs).