Crossing the Border Twice: Reimporting Prepositions to Alleviate L1-Specific Transfer Errors

Abstract

We present a data-driven approach which exploits word alignment in a large parallel corpus with the objective of identifying those verb- and adjective-preposition combinations which are difficult for L2 language learners. This allows us, on the one hand, to provide language-specific ranked lists in order to help learners to focus on particularly challenging combinations given their native language (L1). On the other hand, we provide extensive statistics on such combinations with the objective of facilitating automatic error correction for preposition use in learner texts. We evaluate these lists, first manually, and secondly automatically by applying our statistics to an error-correction task.