JADT 2014: 12es Journées internationales d’Analyse statistique des Données Textuelles Matching the data: developing a multilingual corpus of language development

In recent years, there has been a growing awareness in the field of Second Language Acquisition (SLA) that research into some of the outstanding questions on language development is contingent upon the use of appropriate corpora (Housen & Kuiken, 2009; Larsen-Freeman, 2009; Verspoor et al., 2011; Verspoor et al., 2012; Vyatkina 2012). Previous studies in SLA have too often relied on cross-sectional corpora and are frequently biased towards English (Myles, 2005). Whereas cross-sectional corpora rely on group means to approximate developmental trajectories, longitudinal corpora allow for a more direct examination of linguistic development. However, the challenges posed by the constitution of longitudinal corpora from a practical perspective are numerous: data collection is often limited in terms of time span, group size is difficult to maintain because of participant dropout, and it may be difficult to obtain equivalent data without repetition effects. In multilingual corpora, these challenges are not only repeated for each additional language, but the corpora should also be as equivalent as possible. For corpora intended to reflect linguistic development, this equivalence should ideally be guaranteed for, amongst others, the participant background, the completed task and the proficiency levels or developmental stages. Additionally, research departments have often accumulated a considerable body of data which, though not initially conceived as multilingual corpora, may share enoug