Exploring big educational learner corpora for SLA research

Abstract

We consider the opportunities presented by big educational learner corpora for Second Language Acquisition (SLA). In particular, we focus on theEF Cambridge Open Language Database(EFCAMDAT), an open access database of student writings submitted toEnglishtown, the online school ofEF Education First. EFCAMDAT stands out for its size (33 million words, 85 thousand learners) and a range of 128 writing tasks covering all CEFR levels with data from learners from varying nationalities. We discuss methodological issues arising from analyzing big data resources generated in educational contexts and argue that Natural Language Processing (NLP) is essential for the automated processing of such datasets. As a study case, we follow the developmental trajectory of relative clauses, a construction that necessitates deeper syntactic analysis. We consider specific issues that can affect the developmental trajectory, including task effects, formulaic language and national language effects.

Modeling Systematicity and Individuality in Nonlinear Second Language Development: The Case of English Grammatical Morphemes

“…The data set itself is also a source of limitations. For example, it is worth looking into the potential effects of other variables, including tasks, teaching materials in Englishtown, and varying progress rates across learners (Alexopoulou et al., ).…”

“…An alternative approach is to use the notion of national language (see Alexopoulou, Geertzen, Korhonen, & Meurers, ).…”

Task Effects on Linguistic Complexity and Accuracy: A Large‐Scale Learner Corpus Analysis Employing Natural Language Processing Techniques

“…However, prior studies have shown that native language taggers and parsers perform fairly well on the learner data in EFCAMDAT (Geertzen et al., ). For example, Alexopoulou, Geertzen, Korhonen, and Meurers () evaluated the accuracy of extraction of relative clauses, reporting an F score of 83.9%, and found that state‐of‐the‐art NLP tools provide reasonable quality.…”

Automatic extraction of subordinate clauses and its application in second language acquisition research

“…Another source of difference might be how different L1 groups use formulaic sequences. In a corpus coming from an EFL teaching environment, it is possible that learners use subordinate clauses lifted from their input or task-prompts (Wray, 2002;Alexopoulou et al, 2015). The within L1 comparison of the development of different types of SCs (Fig.…”

“…These are statistics of the latest version (Version 2).9 The analysis of infinitival clauses requires analysis of unexpressed arguments that is beyond our scope; descriptively many infinitival clauses might not be considered full clauses.10 In this paper we do not consider the case of formulaic uses of subordinate clauses, that is, the possibility that learners use SCs lifted from their task prompts that potentially are unanalyzed and, as such, do not reflect true knowledge of SCs Alexopoulou et al (2015). present a method for automatic identification of formulaic uses.…”