Improved Inference of Human Intent by Combining Plan Recognition and Language Feedback