ALBERT, an intelligent problem-solving monitor and coach, has been developed to assist students solving problems in one-dimensional kinematics. Students may type in kinematics problems directly from their textbooks. ALBERT understands the problems, knows how to solve them, and can teach students how to solve them. The program is implemented in the TUTOR language and runs on the Control Data mainframe PLATO system. A natural language interface was designed to understand kinematics problems stated in textbook English. The interface is based on a pattern recognition system which is intended to parallel a cognitive model of language processing. The natural language system has understood over 60 problems taken directly from elementary Physics textbooks. Two problem-solving routines are included in ALBERT. One is goal-directed and solves the problems using the standard kinematic equations. The other uses the definition of acceleration and the relationship between displacement and average velocity to solve the problems. It employs a forward-directed problem-solving strategy. The natural language interface and both the problem-solvers are fast and completely adequate for the task. The tutorial dialogue system uses a modified version of the natural language interface which operates in a two-tier fashion. First an attempt is made to understand the input with the pattern recognition system, and if that fails, a keyword matching system is invoked. The result has been a fairly robust language interface. The tutorial is driven by a tutorial management system (embodying a tutorial model) and a context model. The context model consists of a student model, a tutorial status model and a dynamic dialogue model. ALBERT permits a mixed initiative dialogue in the discussion of a problem. The system has been tested by Physics students in more than 80 problemsolving sessions and the results have been good. The response of the students has been very favourable