This paper describes a system, VP2, that has been implemented to tutor non-native speakers in English. The system applies Artificial Intelligence techniques developed in Natural Language research. In particular, it differs from standard approaches by employing a model of its users to customize instruction based on knowledge of the student's native language. The system focuses on the acquisition of English verb-particle and verb-prepositional phrase constructions. It diagnoses errors that students make due to interference of their native language. VP2 recognizes syntactic variation in English sentences, allowing freer translation. VP2 is a modular system: its model of a user's native language can easily be replaced by a model of another language. Its correction strategy is based upon comparison of the native language model with a model of English. The problems and solutions presented in this paper are related to the more general question of how modeling previous knowledge facilitates instruction in a new skill.