Creating a learner corpus infrastructure: Experiences from making learner corpora available

Abstract

With language resources being collected in many - also small - projects in learner corpus research with considerate amounts of time and ef- fort spent in this activity, making these types of data available in a FAIR way, with standardized and reasoned methods, would contribute substan- tially to the advancement of the field. Additionally, it would answer current demands in transparency, replicability and reusability. In this article, we dis- cuss some of the challenges when making learner corpora FAIR and report from experiences in fulfilling this aim while creating a learner corpus infra- structure at a research institution hosting five different learner corpora.

DaLAJ - a dataset for linguistic acceptability judgments for Swedish: Format, baseline, sharing

“…Datasets and corpora collected from (second) language learners contain private information represented both on the metadata level and -depending on the topic -in the texts. Presence of personal information makes those datasets non-trivial to share with the public in a FAIR 5 way (Frey et al, 2020;Volodina et al, 2020), to say nothing of a potential to use such data for shared tasks. This is rather unfortunate since collection and preparation of such corpora is an extremely time-consuming and expensive process.…”