Most of the work in developing foreign language tutoring systems has focused on the use of Natural Language Processing (NLP) tools to diagnose errors in student input. However, almost all NLP research projects have targeted language produced by native speakers, and the vast majority of approaches to tokenization, POS tagging, parsing, and morphological analysis do not handle well ill-formed input produced by language learners. In this talk I will describe some of the challenges of using NLP technology for analyzing learner language, and present some solutions that I explored during the development of an automatic language tutor for Portuguese (the TAGARELA system).  I will show a small demo of the system, and briefly talk about current research projects related to it.