Nina Vyatkina

Data-Driven Learning for Beginners: The Case of German Verb-Preposition Collocations.

Research on data-driven learning (DDL), or teaching and learning languages with the help of electronic corpora, has shown that it is both effective and efficient. Nevertheless, DDL is still far from common pedagogical practice, not least because the empirical research on it is still limited and narrowly focused. This study addresses some gaps in that research by exploring the effectiveness of DDL for teaching low-proficiency learners lexico-grammatical constructions (verb-preposition collocations) in German, a morphologically rich language. The study employed a pretest-posttest design with intact third- and fourth-semester classes for German as a foreign language at a US university. The same collocations were taught to each group during one class period, with one group at each course level taking a paper-based DDL lesson with concordance lines from a native-speaker corpus and the other one taking a traditional rule-based lesson with textbook exercises. These constructions were new to third-semester students, whereas fourth-semester students had been exposed to them in the previous semester. The results show that, whereas the DDL method and the traditional method were both effective and resulted in lexical and grammatical gains, DDL was more effective for teaching new collocations. The study thus argues in favor of using paper-based DDL in the classroom at lower proficiency levels and for languages other than English.

The Kansas Developmental Learner corpus (KANDEL)

This article presents the Kansas Developmental Learner corpus (KANDEL), a corpus of L2 German writing samples produced by several cohorts of North American university students over four semesters of instructed language study. This corpus expands the number of freely and publicly available learner corpora while adding to the depth of these corpora with a unique set of features. It does so by focusing on an L2 other than English, German, targeting beginning to intermediate L2 proficiency levels, and including dense developmental data and annotations for multiple linguistic variables, learner errors, and over twenty learner and task variables. Furthermore, this article reports the procedure and results of an inter-annotator agreement study as well as an in-depth analysis of annotator disagreement. In this way, it contributes to best practices of annotating learner corpora by making the annotation process transparent and demonstrating its reliability.