Detecting Interlingual Errors: The Case of Prepositions

Prepositions pose a particular challenge for many language learners, in part because of their seemingly arbitrary usage with certain verbs that does not necessarily translate directly across languages. As a consequence, many preposition errors in writing can be attributed to a direct transfer from a writer’s native language. While research in scalable tooling for second-language writing assistance has largely focused on automated error detection and correction, relatively little attention has been given to explaining why the errors may have occurred. A system that can distinguish interlingual errors – arising from an over-literal translation from a writer’s native language – from non-interlingual errors could serve as an insightful tool for language learners and teachers alike. In this work, we demonstrate the feasibility of classifying English preposition errors produced by native speakers of Spanish as interlingual. We propose a corpus-based method that exploits translation probabilities to estimate the likelihood that a writer has translated word-for-word from Spanish to English. We then show that this method correlates well with human judgments on the interlingual status of preposition errors and can be a basis for developing a tool for explaining one key source of errors in second-language writing.